Book of Proceedings 3 RD INTERNATIONAL SYMPOSIUM & 25 TH NATIONAL CONFERENCE ON OPERATIONAL RESEARCH ISBN:

Transcription

1 Hellenic Operational Research Society University of Thessaly 3 RD INTERNATIONAL SYMPOSIUM & 25 TH NATIONAL CONFERENCE ON OPERATIONAL RESEARCH ISBN: Book of Proceedings Volos, June

2 ISBN:

3 Table of Contents Multicriteria decision support for financing Greek agricultural units... 4 Supply Chain Modeling geared to Customer Satisfaction using the simulation s/w package SIMUL8: A linear and a non-linear model Supply chain Linear and Non Linear optimization with regard to customer satisfaction: Solving with GAMS Internet and agro-tourism sector for regional development in Crete: A multicriteria ranking On the way to a reference model for supply chains in the construction industry The problem of robustness in the MUSA method: Theoretical developments and applications.. 39 A combined MCDA approach for facilitating maritime transportation policies evaluation Optimal Strategic Design of Flexible Supply Chain Networks An Integrated Multi-Regional Long-Term Energy Planning Model Incorporating Autonomous Power Systems Comparison of GA-ANN and Traditional Box-Jenkins Methods for Railway Passenger Flow Forecasting Inspection of power grid by periodic vehicle routing formulation Environmental performance evaluation using a fuzzy aggregation-disaggregation approach Rationalizing electricity production investments from renewable energy sources in Greece using a synergy of multicriteria methods Orisma(c): Optimizing long term fleet wide crew assignment Simulation analysis of a pilot handling system for the rail transport of conventional semitrailers Research on internet sufficiency of websites concerning women agricultural co-operatives in Greece: A multicriteria approach Adaptation of ITA for project portfolio selection within a group of decision makers F.W. Lanchester s combat model application in a supply chain in a duopoly An optimization modeling approach for the establishment of a bike-sharing network using Monte Carlo Simulation and stochastic demand: a case-study of the city of Athens Evaluating new service development effectiveness in tourism: An ordinal regression analysis approach New Technologies & Labor Market Regression modeling for spectral data sets: A multi-objective genetic approach Optimal use of non-collaborative servers in two-stage tandem queueing systems

4 GreenYourRoute platform Reducing Waiting Time at Intermediate Nodes for Intercity Bus Transportation Innovation management strategies for organizational performance Country risk evaluation methodology to support bilateral cooperation in the field of electricity generation from renewable sources Sustainable Food Security: A System Dynamics Decision-Making Methodology Towards the implementation of optimal train loading plan in the Athens Thessaloniki freight services An exact method for the inventory routing problem Open Governmental data sources in Europe: A comparative evaluation of semantic and technical characteristics A branch and price solution algorithm for the tail assignment problem A multi-stage column generation solution approach for the bidline aircrew scheduling problem A Calibration Tool for Macroscopic Traffic Flow Models Air Traffic Management: The free flight concept Comparison of pricing mechanisms in markets with non-convexities Development of Optimization Models for Addressing Various Decision and Information Related Issues in Supply Chain Planning Measuring employee satisfaction in a Greek academic environment Identifying factors of bank service quality during economic crisis in Greece Touristic Guide: A prototype software for touristic journey planning FindMyWay: A prototype web-based platform for journey planning in Athens city, Volos city and Crete island

5 Multicriteria decision support for financing Greek agricultural units Athanasios Valiakos University of Piraeus, 80, Karaoli & Dimitriou Street, GR Piraeus, Greece Yannis Siskos University of Piraeus, 80, Karaoli & Dimitriou Street, GR Piraeus, Greece Abstract The European Agricultural Guarantee Fund (EAGF) finances direct payments to farmers in Member States of European Union with specific implementation rules. This is done based on the entitlements, which derived from the total production during the historical reference years. European Union rendered this decoupled financial aid as a Single Direct Payment (SDP) scheme, and the total production in Greece is significantly decreased. In view of Common Agricultural Policy's reform, the evaluation of agricultural units is proposed using robust ordinal regression (ROR) approach. In this paper, a case study of farmers in the industry of the juicing citrus is conducted. A method is proposed as an evaluation tool for financially aid to the farmers, towards the new policy, granting the production based approach more effective and more objectively allocating the direct payments. An additive evaluation model is proposed based on a consistent family of criteria composed be DEA s input and output criteria. The phenomenon of Sofa Farmers could be eliminated, since farmers would be financially aided after been evaluated. The preference information used in UTASTAR method is given in the form of a partial pre-order on a subset of farmers (reference set). In order to obtain robust conclusions, post-optimality analyses are applied by computing complementary robustness measures as well as a goal programming type regression model. Keywords: Multicriteria decision analysis; Common Agricultural Policy; Ordinal regression; Robustness. 4

6 1. Introduction The Common Agricultural Policy (CAP) is the agricultural policy of the European Union (EU), which implements a system of agricultural subsidies and other programs (European Union, 2009). In this paper, an evaluation tool is proposed for rural farms to calculate total production subsidy taking into account other criteria, and not only the total area (European, Commission, 2011). This way the «sofa farmers' phenomenon» (European, Commission, 2010) can be eliminated, in light of the CAP's reform (European, Commission, 2013), since farmers would be financially aided after been evaluated. An evaluation methodology is therefore proposed for estimating the financial aid, as form of direct payment. Furthermore, a multi-criteria additive value model, namely UTAStar, is constructed with a consistent family of criteria composed be DEA s input and output criteria. In addition, a linear regression model is proposed to estimate the direct payment, based on the global utility value. The rest of the paper unfolds as follows. In section 2, UTAStar is reformulated based on DEA s input and output criteria. Post-optimality analyses are applied by computing complementary robustness measures and the estimation of direct payment is calculated as a linear regression based on the global utility values. In section 3, a case study of agricultural units in the industry of the juicing citrus is presented using the proposed methodology. The last section concludes this approach. 2. Methodology UTAStar and Linear Regression The methodology described in this paper is a robust ordinal regression approach (Greco S., Słowiński R., Figueira J., Mousseau V., 2010) and consists of two models to calculate financial aid; the additive value function UTAStar (Siskos Y., Yannacopoulos D., 1985) and the least squared method. The synergy of these models leads to the evaluated direct payment of each agricultural unit. The model of this methodology is the multi-criteria additive value model (Jacquet-Lagrèze E., Siskos J., 1982), which is based on a consistent family of criteria composed by DEA s input and output criteria (Charnes A., Cooper W.W., Rhodes E., 1978). Consider a finite set of agricultural units A = {a 1, a 2,, a z } evaluated by criteria from a consistent family. The criteria are divided as input and output oriented, following (Valiakos A., Siskos Y., 2013) work. Therefore, the family of criteria is the vector G = {g i I, g r O }, i = 1,..., m ; r = 1,, s. We select A R : {a 1, a 2,, a n } A a reference set of n agricultural units. UTAStar is applied on the selected reference set (Jacquet-Lagrèze, E., Siskos, J., 2001). A two-stepped post-optimality analysis is applied. In order to reduce in size the polyhedron, we can maximize the distance δ, of two consecutive agricultural units, by solving the following LP, max D = δ (1 ) under the constraints, All UTAStar constraints (2 ) 5

7 The second step is to check whether the model is significantly robust. The sensitivity analysis is achieved by investigating the extreme maximum and minimum value functions of each criterion. Post-optimality analysis is applied in order to evaluate the robustness of the additive model and to calculate the most representative additive value function. The final solution of the model is the average (barycentral) of the extreme maximum values forming the representative utility value. The solution is extrapolated to the complete set of A, acquiring the global utility values u[g(a i )], i = 1,, z. This utility value is used to assess the amount of subsidy of the agricultural units a regression type method is applied, using the least squares method. Let Y = {y 1,, y z } be the values of traditional direct payments. The estimation of linear regression, using least squared method for (u[g(a 1 )], y 1 ),, (u[g(a z )], y z ) is expressed by the following equation, y i = a + b u[g(a i )], i = 1,, n ( 3 ) where a, b are the estimators of the constants α, β. The estimator y i is the estimation of the direct payment. 3. Case Study - Juicing citrus Although decoupled there could still be a 'connection' of economic aid under the current regulation for the common agricultural policy (CAP) from 2015 to For that reason, farmers in the industry of the juicing citrus, and more specifically oranges as selected in this approach. From 2003 until 2008 the direct payment was based on the total production. During the period the new scheme was implemented for the coupled financial aid of citrus. In Greece a 60% of the single area payment scheme in citrus was connected with the production processing. Pursuit of the European and Greek juicing industry was to maintain the link with the aid of citrus processing. However, European Union rendered this decoupled in the year 2010, and since then the total production is significantly decreased. A case study of 1,789 farmers, who were active in Argolida region, is conducted from the total number of farmers in Greece, which are approximately 8,000 farmers. The case study involves five criteria, three input and two output, which are presented in Table 1, Criteria Name Measurement Min Valu e Max Value Input Criteria g 1 I Labor Cost euro ( ) ,000 g 2 I Production Cost euro ( ) ,000 g 3 I Capital euro ( ) ,000 Output Criteria Description Number of hours of operator, family, and hired farm labor e.g. tilth, carving. Expenses on fertilizers, pesticides, sprayers other chemicals, plant values and others. Agricultural equipment, machinery and buildings. 6

8 g 1 O Trees units (#) 10 3,500 Total number of trees operated. g 2 O Net Production kilos (kg) 1, ,000 Weight of the final product, once processed. Table 1: Evaluation criteria of agricultural units. For the selected case scenario 10 farmers are chosen to form the reference set. Much effort was spent to construct this set in order to be representative of the whole entity of farming categories and to avoid being very complicated for the DM to rank. The constructed reference set, its evaluation by the DM and the confirmation of this evaluation by the UTASTAR method are presented in Table 2, DM's Ranking g 1 I g 2 I g 3 I g 1 Ο g 2 Ο Global Value u(g) 1 2, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , Table 2: Evaluation Criteria of agricultural units. The model, which is finally adequately stable and sufficiently consistent with the preferences of DM, is expressed in Equation 4, u(g) = u 1 (g 1 I ) u 2 (g 2 I ) u 3 (g 3 I ) u 4 (g 1 O ) u 5 (g 2 O ) ( 4 ) Least squared method is applied. The Y i values are the traditional direct payment the agricultural units received with the entitlements. The linear regression line (Eq. 3) and the estimators α, β, for the set of agricultural units extrapolation to the set A, is displayed in Figure 1, 7

9 Figure 1: Estimation - Linear Regression line for Juicing Citrus. In Figure 1, we can observe that most of units are assembled in high utility values. In Figure 2, a comparison is made between the traditional direct payment and the final direct payment from the methodology. For presentation purposes, only the reference set is included. In this case, some of the units are over-financed while others are under-financed. Figure 2: Comparison Traditional/Final Direct Payment of Reference set of Units. It is worth mentioning that from the estimation, using the model of regression UTAStar the total financial aid is 5,684,425.47, while with the EU funded this scheme with the amount of 5,684, Through the evaluation of the agricultural units not only we can achieve rationalizing of the direct payment, but also the overall total direct payment remains the same. 8

10 4. Conclusions In this paper, a framework is proposed to calculate the direct payment, based on evaluation of agricultural units using robust ordinal regression (ROR) approach. The synergy of two methods, the additive evaluation model and the goal programming regression model is proposed to measure the final financial aid. The additive evaluation model is proposed based on a consistent family of criteria composed be DEA s input and output criteria. The phenomenon of Sofa Farmers could be eliminated, since farmers would be financially aided after been evaluated. The robustness of the two phase methodology is controlled by post-optimality analyses by computing complementary measures. The financial aid, as form of direct payment would be provided after evaluation. The decoupled direct payment is still coupled, for evaluation purposes and the EU budget can be controlled. For the direct payment of the scheme "juicing of citrus" the total EU budget remains the same, by financing with the proposed methodology. Acknowledgment This study is funded and supported by the Institute of National Funds of Greece, since the first author is under financial scholarship. Data obtained from N. Samaras, Greek Payment and Control Agency for Community Aid Guidance and Guarantee Fund, December 01, References European, Commission. The CAP towards 2020: Meeting the food, natural resources and territorial challenges of the future, Official Journal of the European Union, 2010, Brussels. European, Commission. Proposal for a regulation of the European Parliament and of the Council: Establishing rules for direct payments to farmers under support schemes within the framework of the common agricultural policy, Official Journal of the European Union, 2011, Brussels. European, Commission. Overview of the CAP reform , Official Journal of the European Union, 2013, Brussels. European Union. Council Regulation (EC) No 1782/ Establishing common rules for direct support schemes under the common agricultural policy and establishing certain support schemes for farmers and amending Regulations, Official Journal of the European Union, 2009, Brussels. Greco S., Słowiński R., Figueira J., and Mousseau V. Robust ordinal regression, Trends in multiple criteria decision analysis, Springer, 2010, pp Jacquet-Lagrèze E., and Siskos J. Assessing a set of additive utility functions for multicriteria decision making. European Journal of Operational Research, Vol. 10, Issue 2, 1982, pp Jacquet-Lagrèze, E., and Siskos, J. Preference disaggregation: 20 years of MCDA experience. European Journal of Operational Research, Vol. 130, Issue 2, 2001, pp Siskos Y., and Yannacopoulos D. UTASTAR: An ordinal regression method for building additive value functions. Investigação Operacional, Vol. 5, Issue 1, 1985, pp

11 Valiakos A., and Siskos Y. From Data Envelopment Analysis to Multi-criteria Decision Support: Application to Agricultural Units Evaluation in Greece. 2nd International Symposium and 24th National Conference on Operational Research, ISBN: , Athens: Hellenic Operational Research Society (HELORS), 2013, pp

12 Supply Chain Modeling geared to Customer Satisfaction using the simulation s/w package SIMUL8: A linear and a non-linear model Papadopoulos T. Chrissoleon* Department of Economic Sciences, Division of Business Economics, Aristotle University of Thessaloniki, University Campus, Thessaloniki, Greece. Gavriel Eleni, Department of Economic Sciences, Division of Business Economics, Aristotle University of Thessaloniki, University Campus, Thessaloniki, Greece. Chrysochoidis-Trantas Panagiotis Department of Economic Sciences, Division of Business Economics, Aristotle University of Thessaloniki, University Campus, Thessaloniki, Greece. Kalotychos Thomas Department of Economic Sciences, Division of Business Economics, Aristotle University of Thessaloniki, University Campus, Thessaloniki, Greece. Bibos Aggelos Department of Economic Sciences, Division of Business Economics, Aristotle University of Thessaloniki, University Campus, Thessaloniki, Greece. address of the corresponding author: hpap@econ.auth.gr Abstract The aim of this paper is the simulation of a general multi-criteria supply chain model. The main objective is to optimize the supply chain geared to customers satisfaction and then the optimization of profit and costs. Specifically, the model is set up for a four echelon supply chain with two objectives which are: to guarantee customers satisfaction and the optimization of the holding cost, the transportation cost and the 3PL cost at central warehouses and therefore the maximization of profit. For the purpose of this research, a linear and a non-linear simulation model were developed. The customer s demand is served immediately from the retailers or in a short time from the warehouses. We present two scenarios in order to achieve this target. In the first one, the 60% of the demand is satisfied immediately from the retailers and the 40% from the warehouses. In the second one, the 30% of the demand is satisfied immediately from the retailers and the other 70% from the warehouses. 11

13 Keywords: Customer satisfaction, simulation of supply chain, simul8. 1. Introduction and literature review Supply chain management and distribution networks have been designed by many researchers during recent years. Most of the papers were typically oriented at cost minimization and profit maximization. The satisfaction of customers demand is the main reason to increase the service level of the supply chain. As a result, lately, many research studies try to optimize the supply chain operations with regard to demand satisfaction. The literature in the area of supply chain simulation is large. Here, due to space limitations, only a few studies are mentioned. The interested reader is addressed to the master s thesis of Gavriel (2014) for more references and a systematic classification of the relevant material. Ganeshan (1999) tried to find near-optimal stocking policies (reorder point, order quantity) on both the retailer and the distribution center level, and lead time conditions, in order to satisfy the demand. Lin et al. (2000) developed an extended-enterprise supply chain analysis tool due to the willingness of IBM to reengineer its global supply chain in order to achieve quick responsiveness to customers with minimal inventory. Huq et al. (2006) used a mathematical and a simulation model to demonstrate that under specific circumstances, an inventory replenishment system with two warehouses and n-retailers provide better customer service without significant changes in the cost than one warehouse. Lim et al. (2006) studied a production-distribution plan taking into account a multi-facility, multiproduct, and multi-period problem in order to determine the optimal production-distribution plan in a network with a bill of material (BOM). The main purpose of this paper is to create a linear and a non-linear simulation model of a general multi-criteria supply chain which serve the demand and satisfy customers. The difference in our study is that we attempt to guarantee that all customers will get the quantity they ordered in the time promised and then we examine the optimization of the financial metrics. To solve the model we use the simulation software simul8, the educational version. The rest of the paper is structured as follows: In section 2, the general structure of the models and their main characteristics are presented, in section 3 the final results of the two scenarios for the linear model are given, and the final results of the two scenarios for the non-linear model are presented in section 4. In section 5, a comparison between a system with one product and a system with two products is given. Finally, the last section concludes the paper and gives a few areas for further research. 2. The Structure of the models and their main characteristics We studied a supply chain consisting of 4 echelons: One supplier, one manufacturer, two warehouses and four retailers. The supplier of raw material is focused only on the execution and the delivery of the orders that receives from the manufacturer and not on the procedure of production of raw materials. We assume that we have only two final products. The first one consists of three items of raw material. The second one needs four items of raw material to be produced. The manufacturer is responsible for the inventory replenishment of the warehouses and of the retailers, through the warehouses. Moreover, he has a production department in order to produce the products that are necessary for his inventory replenishment. Each warehouse serves two retailers. The lead time between echelons is known and constant. We set two goals to achieve: 12

14 the customers satisfaction and the optimization of profit and cost. In this study, we examined the inventory and the replenishment policies (reorder point, order quantity) in each member of the supply chain in order to minimize the holding cost, the transportation cost and the 3PL 1 cost at warehouses and therefore to maximize the profit. Also, in the non-linear model the objective is to calculate the cost of the defective products in the supply chain and how the demand fill rate is affected. We designed our model in the simulation software simul8, which runs for 90 days 2. We suppose that the daily demand for the first product follows the normal distribution and the demand of the second product follows the poisson distribution. The four parameters for the poisson distribution are (the number represents the average): 200, 300, 400, and 150. The four pairs of parameters for the normal distribution are (the first number is the average and the second is the standard deviation): (500,50), (600,100), (800,150), (1000,200). The values of distributions were randomly selected based on historical data. In the real business world the demand is satisfied 100% from the retailers or 100% from the distributors. So, we decided to divide this percentage between the retailers and the distributors in order to see how the supply chain is affected. The two scenarios are the following: Scenario1: 60%-40%- the 60% of the demand is satisfied immediately from the retailers and the 40% from the warehouses in the next period. Scenario 2: 30%-70% - the 30% of the demand is satisfied immediately from the retailers and the rest 70% from the warehouses in the next period. Some businesses, not be able to satisfy immediately customers' demand, sell their products in a discount price. Thus, we assumed that the quantity of products that are delivered from the warehouses directly to customers have 10% discount on the final price. For each scenario we start with a large amount of inventory and big values in the replenishment process (Q, R) for all the echelons in order to be sure that the demand will be satisfied and then we end up to the minimum quantity of inventory and the smaller values for the quantity of order (Q) and the reorder point (R) in the replenishment process for each echelon which is necessary in satisfying the customers demand. In the non-linear model the additional element which describes the non-linearity is that while transporting orders from warehouses directly to customers there are some defective products, which are replenished by retailers in the next period of time. The objective consists of one more value to minimize: the cost of the defective products in the supply chain and then it is examined how the demand fill rate is affected. The percentage of the destroyed items is determined by a uniform distribution with a range of values between (0.1, 0.2) every time that an order is delivered. The mathematical function which is used for the calculation of the cost of the defective products is equal to the percentage of the destroyed products in the power of two multiplied by a penalty cost (constant value). 1 3PL cost is the price that is charged from the distributor (who acts as a 3PL provider) for every inbound and outbound unit of product 2 The simulation model of this paper has the same structure with the GAMS implementation, in the paper Supply chain Linear and Non Linear optimization with regard to customer satisfaction: Solving with GAMS (No.: QP-78-06), 3rd International Symposium and 25th National Conference on Operational Research, Simulation model runs for 90 days such as the GAMS' implementation. Thus, the simulation' results can be compared with the ones of the GAMS implementation. 13

15 R1 R2 R3 R4 W1 W2 F_PRODU F_RAW Book of Proceedings P,W,C,T (Destroyed_Items pwct ) 2 σ pwct p,w,c.t=1 Assumptions: There isn't any constraint about vehicles' capacity. There isn't any constraint about the maximum limit in the inventory that any echelon can store. Lead-Time from the manufacturer to the warehouses is three working days, from the warehouses to the retailers and from the warehouses to the customers is one working day. 3. Linear model Comparison between the final results of scenario1: 60%-40% and scenario 2: 30%-70%. Through the comparison we can see that in the first scenario the holding cost and the 3PL cost at warehouses is smaller than in the second one. Also, the transportation cost is bigger from the warehouses to the retailers but smaller from the warehouses directly to the customers. 150, , , Scenario 1: 60%-40% Scenario 2: 30%-70% Holding Cost 30, , , , , , PL Cost W1 W2 Scenario 1: 60%-40% Scenario 2: 30%-70% Figure1: Comparison of the holding cost two scenarios Figure 2: Comparison of the 3PL cost between the between the two scenarios 14

16 300, , , , , , Transportation Cost Scenario 1: 60%-40% Scenario 2: 30%-70% Figure 3: Comparison of the transportation cost between the two scenarios 4. Non-linear model Comparison between the final results of scenario1: 60%-40% and scenario 2: 30%-70%. Through the comparison of the two different scenarios we can see that the non-linear model reacts like the linear model. The difference is due to the cost of the defective items which is 29% of the total cost of the supply chain in scenario 1 and 59% in scenario 2. This happened due to the shift of the demand from the retailers to the warehouses. The average of customers satisfaction who received a non-complete order is 83% in both scenarios. 3PL Cost Figure 4: Comparison of the holding cost between the two scenarios Figure 5: Comparison of the 3PL cost between the two scenarios 15

17 300, , , , , , Scenario - 1: 60%-40% Scenario 2: 30%-70% Transportation Cost Product P FILL RATE Figure 6: Comparison of the transportation cost between the scenarios Figure 7: Fill Rate of product P1 5. System with one product Vs. system with two products In the comparison between the system with one product and the system with two products, the customers satisfaction is served in both systems. Despite the fact that we had two products in the first system we didn't face any problem during transportation in the system with two products. This is due to the fact that vehicles transfer the executed orders independently from the kind and the quantity of the products. There was only a difference between them in the financial metrics. In a system with N products the operations will become more difficult and a lot of problems may occur in different parts of the system. This is something that we investigate as a continuity of this research. 6. Conclusions and further research In conclusion, we provided a simulation model of a general and easily applicable supply chain in order to ensure customers satisfaction and therefore the optimization of the holding cost, the transportation cost and the 3PL cost at central warehouses and the maximization of profit. Two types of supply chain models were simulated: a linear and a non-linear model. The customer s demand is served immediately from the retailers or in a short time from the warehouses. We presented two scenarios in order to achieve this target. The additional element which describes the non-linearity is that while transporting orders from warehouses directly to customers there are some damages in products, which are replenished by retailers in the next period of time. The objective is to calculate the cost of the defective products in the supply chain and how the demand fill rate is affected. Through the linear and the non-linear model it is verified that choosing the best replenishment process (order quantity, reorder point), the correct level of inventory for each member of the supply chain and the best production rate in the manufacturer the total cost of the supply chain is decreasing despite the fact which scenario will be applied. Through the comparison between the two different scenarios, scenario 1: 60%-40%, retailers kept bigger inventory but the transportation cost from warehouses directly to customers was decreasing. Also, in the non-linear 16

18 model the percentage of destroyed Items is smaller than in the second scenario. A supply chain can use one of the two different scenarios that are mentioned above. This decision depends on the way that the supply chain is willing to satisfy the demand (immediately or after a short time). Also, it must take into consideration other factors as well, such as the nature of the products. In the comparison between the system with one product and the system with two products the customers satisfaction is served in both systems. There was only a difference between them in the financial metrics. This study can be extended by improving the transportation part by adding a specific number of vehicles that are used from each member of the supply chain and with certain capacity. Also, the determination of distances between the different members will result to a better calculation of the transportation cost that may occur. Moreover, including some limitation about the maximum quantity of inventory that each member can keep will lead to better results. Another part that can be improved is the production process by checking if some problems occur during the production, how it can be fixed and what is the effect on customers satisfaction. Finally, adding more than two products the supply chain will be more realistic. This part is currently under investigation. A model with N products is under development. We are in contact with 3 companies in order to cooperate and give us real data to evaluate our generic model through a real case study. Acknowledgement: In this research, Ms. Gavriel Eleni, Mr. Chrysochoidis-Trantas Panagiotis, Mr. Kalotychos Thomas, and Professor Chrissoleon T. Papadopoulos have received a grant from THALES (Project: ASPASIA), a project co-financed by the European Union (European Social Fund ESF) and Greek national funds through the Operational Program "Education and Lifelong Learning" of the National Strategic Reference Framework (NSRF) - Research Funding Program: Thales. Investing in knowledge society through the European Social Fund. References Ganeshan, B. Managing supply chain inventories: A multiple retailer, one warehouse, multiple supplier model. Int. J. Production Economics. Vol. 59, 1999, pp Huq, F., Cutright, K., Jones, V., Hensler, A. D. Simulation study of a two-level warehouse inventory replenishment system. International Journal of Physical Distribution & Logistics Management. Vol. 36, 2006, pp Lim, J. S., Suk J. J., Kim, S. K., Park W. M. A simulation approach for production-distribution planning with consideration given to replenishment policies. Int. J. Adv. Manuf. Technol. Vol. 27, 2006, pp Lin, G., Ettl, M., Buckley, S., Bagchi, S., Yao, D. D., Naccarato, L. B., Allan, R., Kim, K., Koenig, L. Extended-Enterprise Supply-Chain Management at IBM Personal Systems Group and Other Divisions. Interfaces. Vol. 30, 2000, pp

19 Mirzapour Al-e-Hashem, S.M.J., Baboli, A., Sazvar, Z. A stochastic aggregate production planning model in a green supply chain: considering flexible lead-times, nonlinear purchase and shortage cost functions. European Journal of Operational Research. Vol. 134, 2011, pp Seferlis, P., and Giannelos, F. G. A two-layered optimization-based control strategy for multiechelon supply chain networks. Computers and Chemical Engineering. Vol. 28, 2004, pp

20 Supply chain Linear and Non Linear optimization with regard to customer satisfaction: Solving with GAMS Chrissoleon T. Papadopoulos* Department of Economic Sciences, Division of Business Economics, Aristotle University of Thessaloniki, University Campus, Thessaloniki, Greece. Angelos G. Bimpos Department of Economic Sciences, Division of Business Economics, Aristotle University of Thessaloniki, University Campus, Thessaloniki, Greece. Panagiotis Chrysochoidis-Trantas Department of Economic Sciences, Division of Business Economics, Aristotle University of Thessaloniki, University Campus, Thessaloniki, Greece. Eleni Gavriel Department of Economic Sciences, Division of Business Economics, Aristotle University of Thessaloniki, University Campus, Thessaloniki, Greece. Thomas Klotychos Department of Economic Sciences, Division of Business Economics, Aristotle University of Thessaloniki, University Campus, Thessaloniki, Greece. address of the corresponding author: Abstract The purpose of this study is to develop a general multi-criteria supply chain model. The main target is the optimization of a supply chain with regard to customer satisfaction. The optimization of the financial figures such as profit, holding cost, transportation cost etc., are taken into account too. Specifically, a supply chain of three echelons is examined and the model minimizes the holding, the transportation as well as the 3PL costs in every echelon, increasing thus the profits while guaranteeing customer satisfaction. Two mathematical models are developed: A linear and a nonlinear. There are two main alternative scenarios presented as a means of achieving these targets. The first scenario proposes that the retailers satisfy 60% of the daily customer demand while the rest 40% is satisfied by the distributors the next day. The second scenario examines 30% immediate demand coverage from the retailer and 70% demand coverage from the distributor the next day. There is a price discount in the demand that is served from the distributors. In order to solve the developed mathematical models, the GAMS software is used. Key-words: Optimization of Supply Chains, Customer Level-Of-Service and Profit Maximization, Holding-Cost Minimization, GAMS. 19

21 1. Introduction and Literature Review During the past years much research has been devoted to the optimization of supply chains. As soon as the manufacturing breakthroughs came to an end, the majority of the companies tried to maximize their profits by optimizing their transportation processes and by minimizing their storage limits. Nowadays, it is well recognized that the final customer is the most important part in the supply chain. He is the main reason the manufacturing of a product occurs. As a result, lately, many research studies try to optimize the supply chain operations with regard to demand satisfaction. Supply chain optimization has been extensively researched. Due to space limitation, only a few articles will be presented herewith, whereas for a more extensive literature review and a systematic classification of the relevant material, the reader may read the Master s Theses of Kalotychos Thomas (2014) and Chrysochoidis-Trantas Panagiotis (2014). Paschalidis et al. (2004) in their research tried to achieve the minimization of holding cost, under the constraint of the demand coverage. They analyzed the interplay between the Perturbation Analysis and the Large Deviation Analysis in their objective function and its appliance in a supply chain of up to two echelons. Furthermore, Farahani and Elahipanah (2008) believe that the satisfaction of customer s demand will lead to the reduction of all the costs of the supply chain. They created and analyzed a three echelon mathematical model with multiple products, capacity constraints and service times, adopting the JIT distribution model. Behin Elahi et al. (2011) optimized the supply chain, taking into consideration a percentage of defective products, in each echelon of the supply chain. The supply chain consists of four echelons, and each echelon has more than one elements. Their research focused on two main parts. First, they tried to reduce the total cost of the supply chain and then they attempted to reduce the number of defective products. Mirzapour Al-e-Hashem et al. (2013) studied a stochastic production model in a green supply chain, taking into consideration flexible lead times, nonlinear markets and holding cost. The main purpose of this paper is to create an easily applicable, general multi-criteria model which fully serves the demand and satisfies customers. The distinctive element of this study is that all customers get the quantity they order at the promised time. Then the financial values are examined. The aim is to achieve full customer satisfaction by minimizing the holding, the transportation and the 3PL costs in the supply chain as a whole. In order to get the optimal solution GAMS software was used (LINDOGLOBAL solver). The rest of the paper is structured as follows: In Section 2, the two mathematical models are presented whereas in the third Section, a numerical example is given. The fourth Section provides conclusive remarks as well as a set of suggestions for further research. 2. The Mathematical Models A supply chain of 3 echelons was studied: Manufacturer, Distribution Centers and Retailers. More specifically, the supply chain consists of 1 manufacturer, 2 distribution centers and 4 retailers. The Supplier echelon is considered just to define some variables. Each distributor is responsible for serving exclusively two retailers. There are two different products. Three goals were set: Customer satisfaction, profit optimization and cost minimization. The inventory management in order to 20

22 minimize holding, transportation and 3PL costs was studied. Moreover, as a result of reducing the above mentioned costs, the profit is optimized. The objective is the holding, transportation and 3PL costs minimization along the supply chain. The constraints are sub-divided into six categories: The first one includes the constraints I to VII and these are the constraints which guarantee that there will always be enough inventory at any echelon to fulfill the demand. The second category includes constraints from VIII to XI and refers to the maximum limit in the inventory that any echelon can store at any time period. Constraints XII to XV (third category) set the initial inventory at every echelon at the beginning of the examination. The fourth category of constraints (from XVI to XIX) are the equations which calculate at every time period the new available inventory. The fifth category of constraints (from XX to XXV) reveal the inventory replenishment policy (the (Q, R) inventory replenishment policy is used). Finally, the last constraint (XXVI) is the nonnegativity constraint of the variables. Objective 3 Minimization [RM HC at manufacturer + FP HC at manufacturer + TC at manufacturer + FP HC at distributor + TC at distributor + LC at distributor + FP HC at retailer + TC at retailer] 4 And in math terms A,T I,M,T A,S,M,T I,W,T Min h a Q at + h im Q imt + Q asmt TRC sm + h iw QALL iwt a,t=1 i,m,t=1 a,s,m,t=1 i,w,t=1 I,M,W,T I,M,W,R,C,T + (Q imwt + QRET imwt )TRC mw + P imw (Q imwt + QRET imwt ) i,m,w,t=1 i,m,w,r,ct=1 I,R,T I,W,R,T + P iwr Q iwrt + P iwr (1 a)d irct + h ir Q irt + Q iwrt TRC wr i,r,t=1 i,w,r,t=1 I. Q imwt = Must be equal to what distributor wants based on the final customer demand II. QRET imwt = Must be equal to what distributor wants based on the retailers demand III. Q iwrt = Must be equal to what retailer wants based on the final customer demand IV. Q irt ad irct V. Q at Pr i,m Q it I,M,W,T i,m,w,t=1 I,W,R,T VI. Q imt Q imwt 3 + QRET imwt 3 VII. Q iwt i,w,r,t=1 Q iwrt + (1-a) D irct Q imt MaxI im IX. Q at MaxI am VIII. X. Q irt MaxI ir 3 The description of the symbols used are given in the Appendix. 4 RM=Raw Materials, HC= Holding Cost, FP= Finished Product, TC= Transportation Cost, LC= Logistics Cost 21

23 XI. QALL iwt MaxI iw XII. Q irt = IN ir (t=0) XIII. Q iwt = IN iw (t=0) XIV. Q imt = IN im (t=0) XV. Q at = IN am (t=0) XVI. Q irt = Q iwrt 1 + Q irt 1 ad irct XVII. Q at = Q at 1 + Q asmt 1 Q it 1 Pr im XVIII. I,M,W,T Q imt = Q imt 1 + Q it 1 Q imwt 3 + i,m,w,t=1 I,W,R,T i,w,r,t=1 QRET imwt 3 XIX. QALL iwt = Q imwt 3 + Q iwt 1 Q iwrt 1 +(1-a) D irct 1 XX. Q iwrt = n QO irt XXI. R ir Q irt + l=0 QRET imwt l Q iwmt = n QO iwt XXII. 4 3 XXIII. R iw Q iwt + l=0 Q imwt l XXIV. Q asmt = n QO at XXV. R am Q at + Q asmt 1 XXVI. Q imwt, QRET imwt, Q iwt, QALL iwt, Q iwrt, Q imwt, Q imt, Q asmt, Q it, Q at, Q iwrt, Q irt, Q iwrt 0 In addition, an extra cost in the retailers echelon was added, named penalty cost. This cost refers to the loss occurring in the quantity of products that the customers get from the distributors. Then the retailers are obliged to replenish the defective products to the end customers. The nonlinear differentiation-addition is presented below. Also, a new constraint regarding the amount of defective products is presented: I,R,C,T Min + (k irct ) 2 σ i,r,c,t=1 XXVII. k irct = f irt (1 a) Dirct 3. Numerical Example Microsoft Excel (random number generator) was used to input the data regarding customer demand. More specifically, we assumed that the first product follows the Poisson distribution and the second product follows the Normal distribution. The eight pair numbers are presented in the Table below: Product Retailer 1 Retailer 2 Retailer 3 Retailer 4 1 POISSON POISSON POISSON POISSON (200) (300) (400) (150) 2 NORMAL NORMAL NORMAL NORMAL (500,50) (600,100) (800,150) (1000,200) Distributions regarding generated data 22

24 Axis Title Book of Proceedings Two different scenarios were examined. The first one assumes that the retailers are responsible for satisfying 60% of the daily demand and the rest is covered by the distributors the next day. In the second scenario only 30% of the demand is satisfied from the retailers inventory and the rest is covered from the distributors. In the non-linear model, a percentage of products sent from the distributors to the end customers is defective. The amount of the defective products has to be replenished the next day from the retailers. (There is a price discount in the demand that is served from the distributors). Results: It is not fair to qualify one scenario as the best because the two scenarios represent two different supply chains and two different product types (e.g., not all the products can be managed and sold using the 2 nd scenario). In Figure 1, the total cost, the revenues and the profits in the two scenarios are given. Each enterprise would consider its product characteristics, demand pattern and some other clues before choosing between the two scenarios. 25,000,000 Supply Chain 20,000,000 15,000,000 10,000,000 5,000,000 Total cost Revenues Profit Total 0 60% Linear 30% Linear 60% Non Linear 30% Non Linear and Profit Comparison between the two Scenarios Cost, Revenues 4. Conclusion and Further Research In conclusion, we tried to satisfy the customer demand and based on that to achieve all the other targets which can be the maximization of profit, the minimization of holding and transportation cost, etc. We managed to develop a multi-criteria, supply chain model. The supply chain consists of three main echelons before the final customer: The manufacturer, the distributors and the retailers. Each one of the three echelons has three objectives to satisfy: maximization of profit and customer satisfaction as well as minimization of holding, transportation and 3PL costs. The capacity limitation of the warehouses in each echelon was also taken into consideration in the linear model. Through the minimization of the above costs, the profit was maximized while the customer satisfaction was guaranteed. In the linear model of this paper, we tried to research the consequences in each one of the three criteria mentioned earlier, when only a part of the demand (60% or 30%t) is satisfied immediately while the rest of the demand is served afterwards in a predefined time. By comparing the two policies we managed to determine the exact inventory each echelon needs to hold in order to satisfy the customers completely. The nonlinear model of this paper has the same, basic structure as the linear model. Our supply chain consists of three echelons, having the same targets as the linear one. The main difference in this model is that we tried to 23

25 consider the probability that some of the products transported from the distributors to the end customers are defective, in a way that the final customer is not able to use them. A non-linear penalty cost has been added in the objective function of the retailers. Both models were solved using the GAMS (edition ) software package, which is used for solving linear, nonlinear and mixed integer models. Because of the limitations of the GAMS software used (demo version), some simplifications were necessary to be made. This study can be extended by establishing a transportation model, aiming to get the real conditions of the supply chain even closer. In addition, more products could be added, with another supply chain to be combined in the echelon of distributors. A sensitivity analysis about how many unfulfilled orders would not change our result, would be useful. Also it is worthy to extend the models by analyzing the impact of unreliable suppliers to the customer satisfaction. Furthermore, a more complex production model with variable production rates could be examined. The existence of overlapping deliveries from the distributors to the retailers might be a useful addition. Acknowledgement: In this research, Mr. Chrysochoidis-Trantas Panagiotis, Mr. Kalotychos Thomas, Ms. Gavriel Eleni and Professor Chrissoleon T. Papadopoulos have received a grant from THALES (project: ASPASIA), a project co-financed by the European Union (European Social Fund ESF) and Greek national funds through the Operational Program "Education and Lifelong Learning" of the National Strategic Reference Framework (NSRF) - Research Funding Program: Thales is investing in knowledge society through the European Social Fund. Appendix Symbol Description Symbol Description i s m w r Product {1,, I} Supplier {1,, S} Manufacturer {1,, M} Warehouse {1,, W} Retailer {1,, R} ECi,m Cim Qit ha Qat Labor cost per Employee per Working Hour for product i at manufacturer m Machine operation cost per product i at manufacturer m Quantity produced of product i at time t Holding cost for raw materials a Inventory of raw materials a at time t at manufacturer m 24

26 c t a Pimw Qimwt Pa Qasmt WHi,m,t QO irt Ei,m MaxIam INim Customer {1,, C} Time Period {1,, T} Raw Materials {1,, A} Price charged for inbound units of product i at distributor w (Logistics Cost) Quantity of product i transported from manufacturer m to distributor w at time t for customers use Price charged for raw materials from supplier s to manufacturer m Quantity of raw materials a transported from supplier s to manufacturer m at time t Working Hours for product i at manufacturer m at time t Fixed order quantity of product i at time t from retailer r Employee for product i at manufacturer m Maximum inventory capacity of raw materials a at manufacturer m Initial inventory of product i at manufacturer m him Qimt TRCsm QRET imwt Orm MaxTVCmw MaxTVCsm Prim QO iwt MaxIim INiw Bt Holding cost for product i at manufacturer m Inventory of product i at manufacturer m at time t Transportation cost per product per distance value from supplier s to manufacturer m Quantity of product i transported from manufacturer m to distributor w at time t for retailers use Order cost for manufacturer m Maximum transportation vehicle capacity from manufacturer m to distributor w Maximum transportation vehicle capacity from supplier s to manufacturer m Production rate per working hour per employee for product i at manufacturer m Fixed order quantity of product i at time t from distributor w Maximum inventory capacity of product i at manufacturer m Initial inventory of product i at distributor w The minimum value between inventory of product i at distributor w and demand for 25

27 INam Lt Piwr Qiwrt hiw Qiwt TRCmw k irct ORw MaxTVCwr MaxIiw Initial inventory of raw materials a at manufacturer m The minimum value between inventory of product i at manufacturer m and demand for product i from distributor w to manufacturer m Price charged for outbound units of product i at distributor w (Logistics Cost) Quantity of product i transported from distributor w to retailer r at time t Holding cost for product i at distributor w Inventory of product i at distributor w at time t for customer use Transportation cost per product per distance value from manufacturer m to distributor w Quantity of product i transported from retailer r to customer cat time t to replace default products Order cost for distributor w Maximum transportation vehicle capacity from distributor w to retailer r Maximum inventory capacity of product i at distributor w Pirc Dirct hir Qirt TRCwr f irt ORr GCrt MaxIir INir QALL iwt product i from retailer r to distributor w Price charged for product i from retailer r to customer c Demand for product i from customer c at retailer r at time t Holding cost for product i at retailer r Inventory of product i at retailer r at time t Transportation cost per product per distance value from distributor w to retailer r Percentage of defective products i the retailer r has to compensate for at time t Order cost for retailer r Generalized cost at retailer r per time period t Maximum inventory capacity of product i at retailer r Initial inventory of product i at retailer r Total inventory of product i at distributor w at time t 26

28 n σ Binary variable for the decision whether an order occurs or not Constant value used for the determination of penalty cost QO at Fixed order quantity of raw material a at time t from manufacturer Table 1 Notation References Paschalidis I.Ch., Liu Y., Cassandras C. and Panayiotou C. Inventory Control for Supply Chains with Service Level Constraints: A Synergy between Large Deviations and Perturbation Analysis. Annals of Operations Research. Vol. 126, 2004, pp Elahi B., Pakzad-Jafarabadi Y., Etaati L. and Seyed Hosseini S.M. Optimization of Supply Chain Planning with Considering Defective Rates of Products in Each Echelon. Technology and Investment. Vol. 2, 2011, pp Farahani R.Z. and Elahipanah, M. A genetic algorithm to optimize the total cost and service level for just-in-time distribution in a supply chain. International Journal of Production Economics. Vol. 111, 2008, pp Mirzapour A.M.J., Baboli A. and Sazvar, Z. A stochastic aggregate production planning model in a green supply chain: considering flexible lead times, nonlinear purchase and shortage cost functions. European Journal of Operational Research. Vol. 230, 2013, pp

29 Internet and agro-tourism sector for regional development in Crete: A multicriteria ranking Zacharoula Andreopoulou Laboratory of Forest Informatics, School of Forestry and Natural Environment, Aristotle University of Thessaloniki, Box 247, 54124, Greece Christos Lemonakis School of Management & Economics, Technological Educational Institute of Crete, Agios Nikolaos Branch, Greece Christiana Koliouska Laboratory of Forest Informatics, School of Forestry and Natural Environment, Aristotle University of Thessaloniki, Box 247, 54124, Greece Konstantinos Zopounidis Department of Production Engineering and Management, Technical University of Crete, Greece Authors addresses: randreop@for.auth.gr,culemon2004@yahoo.gr,kostas@dpem.tuc.gr, ckolious@for.auth.gr Abstract Nowadays, effective use of Internet provides an opportunity to identify successful practices and policies for innovative business models in order to promote regional development through agrotourism. Agro-tourism sector can exploit natural and rural resources in the context of employment, growth and competitiveness. In Crete, agro-tourism entrepreneurs have developed commercial activities in the Internet where customers and firms are linked up together in the exchange of agrotourism services. This paper aims to assess websites of commercial purpose within agro-tourism sector in Crete and rank them according to multiple criteria using the multicriteria analysis method of PROMETHEE II. Keywords: agrotourism, regional development, internet, website assessment, total ranking, multicriteria analysis, promethee II, business model, e-commerce 1. Introduction Agrotourism constitutes an integrated and sustainable regional development approach. Agrotourism, farm tourism or agricultural tourism is the process of attracting visitors and travellers to agricultural areas (Ezung, 2011), primarily for agricultural purposes. Opportunities for uniqueness and customization are limitless within the Greek context (Zopounidis et al., 2014). Nowadays, the Internet, apart from a channel to collect information, offers enterprises the 28

30 opportunity to market goods and services to more customers than ever before (Griffin, 2000). The enterprises aim at their participation in the internet society since the benefits are high and electronic systems are ready to serve customers all over the world 24 hours per day and 7 days a week (Andreopoulou et al., 2011; Koliouska and Andreopoulou, 2013). Many agro-tourism firms have already developed their commercial websites, because their Internet presence tends to be a fundamental component, in order to sustain a successful enterprise. This paper aims to assess websites of commercial purpose within agro-tourism sector in Crete and rank them according to multiple criteria. Initially, qualitative and quantitative features were identified in the collected websites. Further, the websites were ranked according to these features to be used as criteria using the multicriteria analysis method of PROMETHEE II. Furthermore, the optimum agrotourism firms websites are identified and described to be used as models with the total internet adoption. 2. Materials and methodology The websites that promote agrotourism activities in the region of Crete are retrieved from the Internet through large-scale hypertextual search engines, such as Google, Yahoo, Pathfinder and MSN Search, which bring satisfying results. Various keywords and combinations were used such as agro-tourism sector in Crete, agro-tourism firm in Crete, agro-tourism services in Crete, etc. There were a variety of features introduced in these websites, aiming to promote the agrotourism sector in Crete. The criteria/features were used to describe variables x1, x2,, xn. These criteria are presented in Table 1. Initially, qualitative analysis was implemented in order to examine the type of common criteria/ features, representing internet adoption, found in these commercial websites. Then, a quantitative analysis through a 2-dimentional table was performed in order to examine the presence or absence of these features. The value of 0 and the value 1 were attributed to the variables x1, x2,, xn, for the non-existence and the existence of each criterion respectively. Variable Features Variable Features X1 Information about products, services or activities X8 Online reservation (enabled with online payment) X2 Current prices X9 Online communities (forums, chat rooms, guestbooks) X3 Contact information X10 Additional topics with information on different categories X4 Local information X11 Code access: website areas where access is allowed only for members through codes or passwords X5 Links to other companies X12 Third person advertisement 29

31 X6 Related sources of information X13 Personalization of the page, trace, safety X7 Online reservation (enabled with traditional ways of payment) Table 1. Variables attributed to features, representing internet adoption Then, the total ranking of the websites was studied. The method that was used for the total ranking was the multicriteria analysis named PROMETHEE II. That method applies a linear form of service in the particular case and fits better to the targets of the project even if it is compared to other well-established methods (Andreopoulou et al., 2014). The identified e-marketing services of the websites are used as criteria in order to determine the superiority of one website over another website and the net flow for each website is estimated, which is the final number in order the websites to be ranked from the best to the worst (Zopounidis, 2001). 3. Results and discussion The research on the Internet resulted in the retrieval of 29 websites that support and promote agro-toursim services for regional development in the region of Crete. Based on the application of the PROMETHEE II method, the first and the last 10 cases of the total ranking of the agro-tourism firms websites in Crete are presented in Table 3. In the same Table it is also presented the total net flow that is estimated for each website and it is used for the comparison between the websites in order to obtain the total ranking, as each website with a higher net flow is considered superior in ranking. The findings of the research show that, the values estimated for total net flows φ present a spectrum of values between +0,82 to -0,73 and that indicates a great difference concerning superiority between the first and the last case in the ranking of the agrotourism firm website. The websites with high superiority are the websites which provide a price list (x2) and useful links to other relevant firms (x5). These websites also, allow the users to make on-line reservation (x7) and to pay through the Internet (x8). Some other features that improve the total net flow of a website are the provision of information on different topics (x10), the third person advertisement (x12) and the ability for the users to personalize the website (x13). Furthermore, the eleventh variable (the ability to create a user account) and the third variable (contact information) are not so essential for the website efficiency. The web visitors of these commercial websites are mainly interested in the agro-tourism activities of the firms (x1) and not so much for the local area information (x4). Furthermore, the ninth variable (social media, guestbook) seems not to be critical. The twelfth variable (third person advertisement) accompanied by the thirteenth variable (personalization of the page, trace, safety) refer to more experienced users. 30

32 Total Ranking Website Agro-tourism Firm Name Total Net Flow 1 Vamos Traditional Village 0, Sarris House 0, Xatheri Villas 0, Milia Mountain Retreat 0, Drys Villas 0, Syia Hotel 0, Stratos Villas 0, Melidoni x village 0, Aspalathos Villas 0, Iliopetra Studios 0, Akros Οreon -0, Mohlos Villas -0, Arodamos -0, Kouriton House -0, Agrikies Stratakis Estate -0, Listaros S.A. "ξα σου" -0, Lasinthos -0, Earino -0, Viglatoras -0, Arolithos Traditional Cretan Village -0,7308 Table 2. Ten greater and ten worst in total ranking of Agro-tourism Websites in Crete and Total Net Flows-TNF 4. Conclusions Findings confirm that internet adoption in agro-tourism sector in Crete is still in initial level. The results indicate that the effective internet sites require: price list, useful links to other relevant firms, online reservation and payment system, information on various topics, third person advertisement and personalization of the website. 31

33 The findings are useful in improving internet technologies adoption through the improved design and implementation of an internet presence to accomplish certain characteristics and to generally optimize the internet activities in agrotourism sector in Crete. Given the public awareness for the environment within tourism, eco-agrotourism websites should evolve in further e-service provision (Andreopoulou and Koutroumanidis, 2009). References Andreopoulou, Z., Koliouska, C., Lemonakis, C. and Zopounidis, C. (2014). National Forest Parks development through Internet technologies for economic perspectives. Operational Research, DOI /s (electronic version) Andreopoulou, Z. and Koutroumanidis, Τ. (2009). Assessment of the ICT adoption stage in ecoagrotourim websites in Greece. 6th International Conference of Management of Technological Changes, 3rd-5th September, Alexandroupolis, Greece. Book I. Pp Andreopoulou, Z., Manos, B., Viaggi, D., Polman, N. (2011). Agricultural and Environmental Informatics, Governance, and Management: Emerging Research Applications. IGI Global. USA. Ezung, T. Z. (2011). Rural Tourism in Nagaland, India: Exploring the Potential. International Journal of Rural Management, 7(1-2), Griffin, M. (2000) Emarketing Planning: Accountability and Emetrics. Embelix Software [online] (Accessed 1 July 2013). Koliouska C. and Andreopoulou Z. (2013). Assessment of ICT Adoption Stage for Promoting the Greek National Parks. Procedia Technology, Vol. 8, pp Zopounidis C. (2001). Analysis of financing decisions with multiple criteria. Anikoula Publications, Thessaloniki. Zopounidis, C., Lemonakis, C., Andreopoulou, Z. and Koliouska, C., Agrotourism Industry Development through Internet Technologies: A Multicriteria Approach. 53rd Meeting of the EURO Working Group for Commodities and Financial Modelling (EWGCFM), Chania, Crete, May

34 On the way to a reference model for supply chains in the construction industry Ilias P. Tatsiopoulos National Technical University of Athens, School of Mechanical Engineering, Sector of Industrial Management and Operational Research, Heroon Polytechniou 9, Zografou, Greece (itat@central.ntua.gr) Dimitrios-Robert I. Stamatiou National Technical University of Athens, School of Mechanical Engineering, Sector of Industrial Management and Operational Research, Heroon Polytechniou 9, Zografou, Greece (drstam@mail.ntua.gr) Abstract The construction industry is a project based industry with many particularities that differ in regions, projects and/or cultures. The fact that there are many actors involved at different levels in the construction process, with low quality information exchange caused by restricted use of communication channels, makes it inefficient. In this paper we examine the literature on supply chain reference models for the construction industry. The search demonstrates that there are no universally accepted reference models for the construction supply chain. The lack of findings is probably due to the small amount of research on supply chain management in construction when compared to manufacturing. In order to cover this gap, we propose the adoption of the Supply Chain REMEDY reference model for the project based construction industry. We believe that the efforts construction companies make towards customer delight, one of their basic objectives in every project, could be supported by the existence of a reference model that takes into consideration the particularities of the sector. In the first stage, a generic reference model for the construction industry will be developed, tested and informed. In the second stage, partial models will be developed for groups of construction projects described in the paper. The article presents a brief literature review, followed by the presentation of the current literature on reference models in the literature and the construction industry literature. Conclusions make the final part of the paper. Keywords: Construction, Supply chain management, Reference model, Supply chain remedy, Construction supply chains 33

35 5. 1. Introduction The construction industry is a project based industry. It traditionally involves many actors that work at a local basis and are usually involved in a number of projects simultaneously. The final products have a long life span and, in most cases, the final consumer is unknown during the project execution. Construction markets are closed to the global competition due to government subsidies, national and local regulations and culture (Segerstedt & Olofsson 2010) and competition doesn t work as effectively as it does in other industries. In this paper we propose that a reference model regarding the supply chains of construction projects will help to improve the efficiency of these supply chains. In the following chapters, first, the concept and some problems of construction supply chains are described, second, the description of the reference model is provided and last, conclusions are drawn as to the feasibility of a reference model and the steps to follow Construction supply chain management Supply chain management is a branch that has received plenty of attention for the past decades and will continue to receive attention at the same scale, if not larger, in years to come. Although there is so much research going on, not all industries with supply chains have been studied to the same extent. The construction industry is one of the industries less studied. Eccles (1981) defined construction as the erection, maintenance, and repair of immobile structures, the demolition of existing structures, and land development. Each of these functions usually involves a tuple of actors. These actors do not always have the same amount of information coming their way and, most likely, they do not belong to the same tier of the supply chain. Construction is a sector with many particularities. One of the main problems is that the industry faces high fragmentation with many SMEs, as noted by Briscoe & Dainty (2005). Other problems that impede the adoption of supply chain management in the construction industry include discontinuous demand, uniqueness of each project in technical, financial and sociopolitical terms (Segerstedt & Olofsson 2010), concentration of main contractors exclusively on the clients (Saad et al. 2002) and the fact that coordination is mainly driven through project management techniques and alignment of ICT systems (Briscoe & Dainty 2005). Construction supply chain management offers new approaches to reduce the cost and increase the reliability and speed of construction (O Brien 1999). Persson et al. (2010) defined construction supply chain management as the task of integrating organizational units along a supply chain, including the construction site and subcontractors, and coordinating materials, information and financial flows with the project site plan in order to fulfill the (ultimate) customer demands. Construction supply chains have many particularities, especially when one compares them to manufacturing supply chains. Many times, projects have contracts stating financial clauses in case of delays, material mismatches and declinations from the specified designs that may not exist to such an extent in manufacturing. On the other hand, Cox & Ireland (2002) found that dominant thinking in the construction sector lacks an understanding of contextual factors highly regarded in the manufacturing industry, like Porter s five forces. In the 90s, after the publication of the Latham (1994) and Egan (1998) reports, large construction firms first adopted the partnering concept proposed by Latham and then moved on to formulating 34

36 even tighter relationships with companies that they collaborated with as proposed by Egan. Ever since, there have been more reports (Egan 2002; Wolstenholme 2009; Bourn 2001; Harris 2013) that propagate the need for tighter relationships between construction firms and higher quality information sharing. A reference model for the construction industry supply chains, just as in the manufacturing industry, can provide a basis for improving processes, information sharing and collaboration between constructions firms involved in a project Construction supply chain reference model In order to check the applicability of a reference model to the industry and propose a suitable reference model, we performed a search in three stages (Figure 1). First, we searched the literature on supply chain management for available reference models. Second, we searched the literature on construction supply chain management for previous attempts to adopt or create reference models. Third, we propose what we believe to be a suitable reference model for construction supply chains. Check Supply Chain Management literature for reference models Check Construction Supply Chain Management literature for reference models Proposal of a reference model for construction supply chains Figure 1: Research Process Following the first step of the methodology, a brief literature review was conducted. The results of the review are based on previous works on available reference model presentation (Fettke et al. 2005) and research conducted through the academic databases Scopus and SpringerLink. They include seven reference models which we divide into three groups; process models, IT based models and conceptual academic models. The first group is comprised by the Supply Chain Operations Reference Model SCOR ( and the Global Supply Chain Framework ( the second group is comprised by the SAP R/3 reference model by SAP AG ( and the Collaborative Planning, Forecasting and Replenishment CPFR reference model and the third group is comprised by the Mentzer reference model (Mentzer et al. 2001) and efforts by Verdouw et al. (2011) and Klingebiel (2008). Although none of these models is industry specific, most of them carry a manufacturing industry nature. The literature review on construction supply chains, dictated by the second step of the methodology, yielded a very small number of results, indicating a need for research directed to the matter. On the one hand, there have been some efforts in the literature (Persson et al. 2010; Cheng et al. 2010; Yeo & Ning 2002; London & Kenley 2000) to apply the most popular reference models, SCOR and GSCF, to construction supply chains. On the other hand, the efforts retrieved for industry specific reference models were limited to the Process Protocol (Kagioglou et al. 2000) and to a conceptual model proposed by (Aloini et al. 2012). The results from the first two steps 35

37 show that there is a gap in the literature concerning reference models on construction supply chain management. In order to fill in this gap, at the third step of our methodology, the Supply Chain REMEDY model was selected as a basis. The Supply Chain REMEDY model (Gayialis et al. 2013; Ponis et al. 2014) is being developed under the research project A Holistic Approach for Managing Variability in Contemporary Global Supply Chain Networks. It is a process-based generic supply chain reference model that is being built to be adaptable to different production strategies, from Engineer-to-Order to Make-to-Stock. The model offers multiple views to the user; process, knowledge, risk, decision, algorithmic and IT views. It is comprised of nine main functions that are, respectably, analyzed in over ninety high level processes that focus on demand variability management. The knowledge and risk enhanced views offer a valuable and versatile tool for all industries and key decisions that highly effect the entire supply chain are pinpointed. The other views offer support to IT processes, especially in industries that heavily rely on technology for their demand-supply operations. We propose that the model is adapted to the project-based construction industry in order to cover the gap of construction supply chain reference models. This industry specific approach is going to be developed and updated following the methodology modeling, measuring and improving proposed by Jianyuan & Fan (2006). Firstly, the existing high level processes will be brought closer to the construction industry reality and analyzed to an industry specific level, but still generic enough to be adaptable to any construction project. We recognize that no two construction projects are the same, but there are distinctions to be made and groups to be formed. We divide construction projects in two major categories; public and private projects. Public projects are government owned but, in most cases, subcontracted to private construction firms. They can be divided into infrastructure projects (motorways, gas lines, etc.) and common-wealth projects (parks, hospitals, etc.). Private projects are more heterogeneous but, can be grouped into housing projects (private or apartment buildings) and financially profitable projects (shopping malls, office blocks, factories, etc.). These four groups will respectively provide a basis for partial construction supply chain reference models. Secondly, the model and the partial models described above will be tested on case studies in the literature and, if possible, through real life cases. Its performance will be measured through carefully defined metrics and, thirdly, processes will be aligned accordingly based on the review of the performance measurements Conclusions The research conducted realizes that construction supply chains have many particularities. These particularities have made the adoption of supply chain management principles difficult. We identified a gap in the literature regarding reference models specific to the industry and we sought to contribute. The attempt to create a reference model for this industry s supply chains, that takes into consideration its project based nature, promises to create a starting point and supporting tool for professionals to understand basic and advanced concepts of supply chain management that elude them. The methodology to be followed sees to make the reference model very adaptable to all companies in the sector and present opportunities to create new and manage existing knowledge. 36

38 Acknowledgement The research efforts described in this paper are part of the research project A Holistic Approach for Managing Variability in Contemporary Global Supply Chain Networks in research action: Thales - Support of the interdisciplinary and/or inter-institutional research and innovation, which is implemented under the Operational Programme: Education and Lifelong Learning, NSRF and is co-funded by European Union (European Social Fund) and Greek Government. References Aloini, D., Dulmin, R., Minnino, V. and Ponticelli, S., A conceptual model for construction supply chain management implementation. Proceedings of the 28th Annual ARCOM Conference. Edinburgh, UK, 2012, pp Bourn, J., Modernising construction. London, 2001 Briscoe, G. and Dainty, A., Construction supply chain integration: an elusive goal?. Supply Chain Management: An International Journal, Vol. 10(4), 2005, pp Cheng, J.C.P., Law, K.H., Bjornsson, H., Jones, A. and Sriram, R.D., Modeling and monitoring of construction supply chains. Advanced Engineering Informatics, Vol. 24(4), 2010, pp Cox, A. and Ireland, P., Managing construction supply chains: the common sense approach. Engineering, Construction and Architectural Management, Vol. 9(5/6), 2002, pp Egan, Sir J., Accelerating change, London, Fettke, P., Loos, P. and Zwicker, J., Business process reference models. In C. J. Bussler & A. Haller, eds. Business Process Management Workshops. Springer Berlin Heidelberg, 2005, pp Gayialis, S.P., Ponis, S.T., Tatsiopoulos, I.P., Panayiotou, N.A. and Stamatiou, D.R.I.,, A Knowledge-based Reference Model to Support Demand Management in Contemporary Supply Chains. Proceedings of the 14th European Conference on Knowledge Management. Kaunas, Lithuania, 2013, pp Harris, E.C., Supply Chain Analysis into the Construction Industry: A Report for the Construction Industrial Strategy, London, Jianyuan, Y. and Fan, Q., The System Model for Supply Chain Process Optimization. Proceedings of 2006 International Conference on Management of Logistics and Supply Chain. Sydney, Australia, 2006, pp Kagioglou, M., Cooper, R., Aouad, G. and Sexton, M., Rethinking construction: the Generic Design and Construction Process Protocol. Engineering, Construction and Architectural Management, Vol. 7(2), 2000, pp

39 Klingebiel, K., A BTO Reference Model for High-Level Supply Chain Design. In G. Parry & A. Graves, eds. Build To Order: The Road to the 5-Day Car. Springer London, Limited, 2008, pp London, K. & Kenley, R., Mapping construction supply chains: widening the traditional perspective of the industry. Proceedings 7th Annual European Association of Research in Industrial Economic EARIE conference. Lausanne, Switzerland: European Association of Research in Industrial Economics, Mentzer, J.T., DeWitt, W., Keebler, J.S., Min, S., Nix, N.W., Smith, C.D. and Zacharia, Z.G., Defining supply chain management. Journal of Business Logistics, 22(2), 2001, pp O Brien, W., Construction supply-chain management: a vision for advanced coordination, costing, and control. NSF Berkeley-Stanford Construction Research Workshop, 1999, pp.1 7. Persson, F., Bengtsson, J. and Gustad, Ö., Construction Logistics Improvements Using the SCOR Model Tornet Case. Advances in Production Management Systems, 338, 2010, pp Ponis, S.T., Gayialis, S.P., Tatsiopoulos, I.P., Panayiotou, N.A, Stamatiou, D.R.I. and Ntalla, A.C., Modeling Supply Chain Processes : A Review and Critical Evaluation of Available Reference Models. Proceedings of 2nd International Symposium and 24th National Conference on Operational Research. Athens, Greece, 2014, pp Saad, M., Jones, M. and James, P., A review of the progress towards the adoption of supply chain management (SCM) relationships in construction. European Journal of Purchasing & Supply Management, Vol. 8(3), 2002, pp Segerstedt, A. and Olofsson, T., Supply chains in the construction industry. Supply Chain Management: An International Journal, Vol. 15(5), 2010, pp Verdouw, C.N., Beulens, A.J.M., Trienekens, J.H. and van der Vorst, J.G.a.J., A framework for modelling business processes in demand-driven supply chains. Production Planning & Control, Vol. 22(4), 2011, pp Wolstenholme, A., Never Waste a Good Crisis: A Review of Progress since Rethinking Construction and Thoughts for Our Future, London, Yeo, K. and Ning, J., Integrating supply chain and critical chain concepts in engineer-procureconstruct (EPC) projects. International Journal of Project Management, Vol. 20(4), 2002, pp

40 The problem of robustness in the MUSA method: Theoretical developments and applications Yannis Politis School of Science and Technology, Hellenic Open University, Parodos Aristotelous 18, GR26335 Patra, Greece, Evangelos Grigoroudis School of Production Engineering and Management, Technical University of Crete, University Campus, GR73100 Chania, Greece. Ifigeneia Pologiorgi School of Production Engineering and Management, Technical University of Crete, University Campus, GR73100 Chania, Greece. Abstract The MUSA method is a collective preference disaggregation approach which has been developed in order to measure and analyze customer satisfaction. It follows the main principles of ordinal regression analysis under constraints using linear programming techniques and it is used for the assessment of a set of marginal satisfaction functions in such a way that the global satisfaction criterion becomes as consistent as possible with customer s judgments. Considering that the MUSA method is based on a linear programming modelling, the problem of multiple or near optimal solutions appears in several cases. This has an impact on the stability level of the provided results. The quality of collected data collected and the incapability to interact with customers complicates the task of finding stable solutions. For this reason different ways to overcome this problem may include asking customers to give additional information (e.g. information about the importance of the criteria along with the usual satisfaction questions) or introducing additional constraints in the basic LP of the method which will reduce the polyhedron of feasible solutions. This study presents the implementation of an extension of the MUSA method in a real case study concerning the evaluation of customer satisfaction from Greek mobile service providers. More specifically, additional constraints regarding special properties of the assessed model variables and additional customer preferences about the importance of the criteria have been incorporated in the LP of the original MUSA method and have been modelled as a Multiobjective Linear Programming (MOLP) problem. The main aim of the study is to show how the introduction of these additional constraints and information can improve the stability level of the estimated results. Different stability and fitting measures have been used in order to analyze and compare the provided results. The application showed that the introduction of the additional constraints and information in the original MUSA method has improved the method s robustness, still without affecting the conclusions drawn by the implementation of the basic MUSA model, enhancing thus the proposed conclusions and improvement actions. Keywords: MUSA Method, Robustness Analysis, Mobile Services, Satisfaction analysis. 39

41 1. Introduction The MUSA (MUlticriteria Satisfaction Analysis) method is a preference disaggregation approach following the main principles of ordinal regression analysis. It measures and analyzes customer satisfaction assuming that customer s global satisfaction is based on a set of criteria representing service characteristic dimensions. The data collected to measure customer satisfaction can have such characteristics that make it impossible to find a feasible solution consistent with customers preferences. Incapability to interact with customers further complicates the task of finding stable solutions. For this reason actions are required in order to improve the stability of the results. Such actions may include asking customers to give additional information (e.g. information about the importance of the criteria along with the usual satisfaction questions) or introducing additional constraints in the basic LP of the method which will reduce the polyhedron of feasible solutions. Based on an extension of the MUSA method with the introduction of additional constraints, a real-world application in Greek mobile service providers is presented. 2. The MUSA Method 2.1 Mathematical Development According to the MUSA method, customer s global satisfaction is based on a set of criteria representing service characteristic dimensions. The main object of the MUSA method is the aggregation of individual judgments into a collective value function. The method is an ordinal * regression-based approach used for the assessment of global and partial satisfaction functions Y and X respectively, given customers judgments Y and * i Grigoroudis and Siskos, 2010): n n * * i i i i1 i1 Y b X with b 1 X (Grigoroudis and Siskos, 2002; (1) where b i is the weight of the i-th criterion and the value functions i * Y and * X i are normalized in the interval [0, 100]. Introducing a double-error variable, the ordinal regression equation becomes as follows: n * * i i i1 Y b X (2) * * where Y is the estimation of overall value function Y and and are the overestimation and underestimation error, respectively. The following transformations which represent the * * successive steps of the value functions Y and X i can be introduced in the model: * m1 * m zm y y for m= 1, 2,..., 1 * k1 * k wik bi xi bi x i for k = 1, 2,..., i 1 and i = 1, 2,...,n The final LP of the basic MUSA model has the following form: (3) 40

42 M [max] F j j j1 subject to n tji 1 tj 1 wik zm j j 0 j 1,2,, M i1 k 1 m1 (4) 1 zm 100 m1 n i 1 wik 100 i1 k1 zm, wik, j, j 0 i, j, k, m where t j and t ji are the judgments of the j-th customer globally and partially for each criterion i 1,2,, n and M is the number of customers. 2.2 Stability Analysis The stability analysis is considered as a post-optimality analysis problem, taking into account that the MUSA method is based on a LP modeling. During the post-optimality analysis stage, n LPs (equal to the number of criteria) are formulated and solved. Each LP maximizes the weight of a criterion and has the following form: i 1 max F wik for i 1, 2,, n k 1 subject to (5) * F F all the constraints of LP (4) * where F is the optimal value of the objective function of LP (4) and is a small percentage of * F. The average of the optimal solutions given by the n LPs (5) may be considered as the final solution of the problem. In case of instability, a large variation of the provided solutions appears and the final average solution is less representative. 2.3 Basic Results The method estimates a set of useful indices for benchmarking purposes such as the average global and partial satisfaction indices S and S i, which can be assessed according to the following equations: α 1 m * m S p y 100 m1 αi 1 k * k Si pi xi for i 1,2,, n 100 k 1 where m p and k p i are the frequencies of customers belonging to the m y and (6) k x i satisfaction levels respectively. Similarly, the average global and partial demanding indices, D and D i, 41

43 represent the average deviation of the estimated value curves from a normal (linear) function and reveal the demanding level of customers. They are normalised in the interval [ 1, 1] and, are assessed as follows: α1 100m 1 * m y m1 α 1 D for α 2 1 m m1 α 1 (7) α1 100k 1 * k x i k 1 αi 1 Di for α 2 and 1, 2,, i 1 i i n k k 1 αi Fitting and Robustness Indicators The fitting level of the MUSA method refers to the assessment of a preference collective value system (value functions, weights, etc.) for the set of customers with the minimum possible errors. Three fitting indices are proposed which are normalized in the interval [0, 1]. The Average Fitting Index ( AFI 1) depends on the optimum error level and the number of customers: * F AFI 1 1 (8) 100 M An alternative fitting indicator is based on the percentage of customers with zero error variables, i.e., the percentage of customers for whom the estimated preference value systems fits perfectly with their expressed satisfaction judgments. This average fitting index AFI 2 is assessed as follows: M 0 AFI (9) 2 M where M 0 is the number of customers for whom 0. This is a rather strict indicator considering that it examines only the existence of non-zero errors, without taking into account the values of these error variables. A third fitting indicator AFI 3 examines separately every level of overall satisfaction and calculates the maximum possible error value for each one of these levels: AFI 3 * F 1 M p max y,100 y m1 m * m * m The Average Stability Index ( ASI ) is an indicator of the MUSA s method stability level. ASI is nothing else than the mean value of the normalized standard deviation of the estimated weights during the post-optimality stage and is calculated as follows: i1 2 n 2 n j j nbi bi n 1 j1 j1 ASI 1 (11) n 100 n 1 (10) 42

44 j where b i is the estimated weight of the i-th criterion in the j-th post-optimality analysis LP. ASI is normalized in the interval [0, 1]. 3. Modeling Additional Information and Properties 3.1 Preferences on Criteria Importance Introducing additional constraints in the MUSA method can limit the space of feasible solutions and therefore increase the stability of the method. These constraints may concern preferences on criteria importance or desired properties of the provided results. Particularly, a customer satisfaction survey may include, besides the usual performance questions, preferences about the importance of the criteria. Using such questions, customers are asked either to judge the importance of a satisfaction criterion using a predefined ordinal scale, or rank the set of satisfaction criteria according to their importance. The evaluation of preference importance classes C is similar to the estimation of thresholds T i. An ordinal regression approach may also be used in order to develop the weights estimation model. The WORT (Weights evaluation using Ordinal Regression Techniques) model is presented in LP (12) in which the goal is to minimize the sum of errors under a set of constraints according to the importance class that each customer j considers that a criterion i belongs (Grigoroudis and Spiridaki, 2003): [min] F2 Sij Sij j i subject to ai 1 w 100 ˆ it T1 Sij 0, bij C 1 t1 ai 1 wit 100 Tl 1 Sij 0 t1, bˆ, 2,..., 1 ai 1 ij Cl l q wit 100 Tl Sij 0 t1 ai 1 w 100 ˆ it Tq 1 Sij 0, bij C q t1 n a i 1 wik 100 i1 k1 Tq 1 Tq2Tq 1 T 1 T 2 wik, Sij, Sij 0, i, j, k i i, j (12) 43

45 where b ˆij is the preference of customer j about the importance of criterion i, is a positive number used to avoid cases where bij Ti i, and is a small number introduced to increase the discrimination of the importance classes. 3.2 Desired Properties of the Results A linkage between global and partial average satisfaction indices may be assumed, since these indices are considered as the main performance indicators of the business organization. In particular, the global average satisfaction index S is assessed as a weighted sum of the partial satisfaction indices S i : n n i m * m k * k i i i i i i1 m1 i1 k 1 or S b S p y b p x m1 n i k 1 m k p z p w t i it m2 t1 i1 k 2 t1 (13) Similarly, a weighted sum formula may be assumed for the average demanding indices: n D b D or i1 i i 1 m1 i1 i1 k1 100( m 1) ( 1) z ( k 1) w ( 1) w (14) ( 1) ( 1) t n it i it m1 t 1 k 1 t 1 t 1 i1 i i Equations (13) and (14) may be introduced as additional constraints in the LP (4). However, these additional constraints should be used carefully, since they do not guarantee a feasible solution of the LP, especially in case of inconsistencies between global and partial satisfaction judgments. For this reason, the aforementioned equations may be written using a goal programming formulation and used alternatively as post-optimality criteria. A double error variable es, es and ed, ed can be introduced in each one of the equations for this reason. 3.3 Extension of the MUSA Method Using both customers performance and importance judgments and introducing, at the same time, additional constraints about the average satisfaction and demanding indices, an extension of the MUSA method may be modeled as a Multiobjective Linear Programming (MOLP) problem. M minf j j j1 n M min Sij Sij i1 j1 (15) min es es ed ed subject to all the constraints of LPs (4) and (12) constraints (13) and (14) The above problem may be solved using any MOLP technique (e.g., compromise programming, global criterion approach). Here, an alternative heuristic method, consisting of four steps (lexicographic approach), is presented: Step 1. Min F subject to all constraints of the examined problem. 44

46 Step 2 (and 3). Min (or ) subject to all constraints of the examined problem and * F F 1. Final step (stability analysis). Max b i subject to all constraints of the examined problem and * F F 1, * 2, * 3, where * F, *, and * are the optimal error values for the basic MUSA model, the WORT model, and the desired properties of the produced results, and 1 2, 3 are small percentages of * F, *, and *, respectively. 4. Real-World Application 4.1 Satisfaction Criteria and Survey Conduct The above procedure has been implemented for the analysis of service quality in the Greek mobile service industry. A sample of 80 questionnaires from the three mobile service providers of Greece (Cosmote 57.5%, Vodafone 22.5%, Wind 20%) has been used. The collected data included mostly customers judgments about the performance of the companies globally and partially (i.e., foe each criterion), as well as a ranking of these satisfaction criteria from the most to the least important one. The selected satisfaction criteria included the following: offers, provided services, provided devices, network, webpage, charges, branch network, and company image. 4.2 Results and Comparison As it can been observed in Table 2, AFI 1 and AFI 3 are particularly high for the basic as well as for the extension of the MUSA method. These indices are slightly worse for the extension of the method which is quite expected as according to the heuristic method applied in the MOLP problem, there is a small decrease of the optimal solution of the basic problem in order to achieve a better consistency regarding the other two objective functions. AFI 2 is particularly low in both cases but as already mentioned this is a rather strict index. Regarding ASI, there is an significant increase of 9.90% with the introduction of additional information and constraints in the basic MUSA model. AFI AFI AFI ASI Original MUSA method 98.08% 17.50% 93.39% 79.11% Proposed extension 94.58% ( 0.53%) 7.50% ( 57.14%) 91.92% ( 2.25%) 86.94% (+9.90%) Table 2. Fitting and stability indices The action diagram combines the relative importance and the performance of the criteria as they have been estimated according to the MUSA and the extension of the MUSA methods. 45

47 High Book of Proceedings According to this diagram the company s profile is regarded as their competitive advantage considering that it has both the highest importance and satisfaction level. The devices provided by the companies could potentially regarded as a competitive advantage while the rest of the criteria have a relatively low satisfaction and importance level but could potentially be regarded as first priority improvements if there is an increase of their importance. Generally, the conclusions provided by both the basic MUSA and the extension of the MUSA method are similar proving that the introduction of the additional constraints has improved the stability of the provided results without however diversifying the conclusions and the strategic actions that the companies should follow. Company's profile Low PERFORMANCE Provided devices Branch network Provided services Network Offres Webpage Charges MUSA Extension Low IMPORTANCE Figure 2. Action diagram High 5. Concluding Remarks The MUSA method is a rather flexible approach and thus several extensions may be developed taking into account additional information or data. The implementation of the proposed extension of the MUSA method in a real case study revealed that the stability of the provided results has been improved. Furthermore, the introduction of additional constraints in the original MUSA method has not diversified the strategic actions that the companies should follow, leading to more stable conclusions. Finally, developing additional measures of robustness may facilitate the investigation of various extensions of the MUSA model, while future research could include the study of the impact of the model parameters or of different extensions of the MUSA method through an extended simulation. 46

48 Acknowledgement This research has been co financed by the European Union (European Social Fund ESF) and Greek national funds through the Operational Program "Education and Lifelong Learning" of the National Strategic Reference Framework (NSRF) Research Funding Program: THALES. Investing in knowledge society through the European Social Fund. References Grigoroudis, E. and O. Spiridaki. Derived vs. stated importance in customer satisfaction surveys. Operational Research: An International Journal. Vol. 3 No. 3, 2003, pp Grigoroudis, E. and Y. Siskos. Preference disaggregation for measuring and analysing customer satisfaction: The MUSA method. European Journal of Operational Research, Vol. 143 No. 1, 2002, pp Grigoroudis, E. and Y. Siskos (2010). Customer Satisfaction Evaluation: Methods for Measuring and Implementing Service Quality. Springer, New York, Siskos, J. Analyses de régression et programmation linéaire. Révue de Statistique Appliquée, Vol. 23 No. 2, 1985, pp

49 A combined MCDA approach for facilitating maritime transportation policies evaluation Eliza GAGATSI Department of Civil Engineering, Aristotle University of Thessaloniki, University Campus Thessaloniki Greece, George GIANNOPOULOS Hellenic Institute of Transport- Centre for Research and Technology Hellas, 6th km Charilaou Thermi Rd, Thessaloniki, Greece Christos PYRGIDIS Department of Civil Engineering, Aristotle University of Thessaloniki, University Campus Thessaloniki Greece Georgia AIFANDOPOULOU Hellenic Institute of Transport- Centre for Research and Technology Hellas, 6th km Charilaou Thermi Rd, Thessaloniki, Greece Abstract This paper presents a methodology developed at the frame of an on-going PhD research, for supporting decision making in maritime transportation, based on a combination of two different MCDA methods in a multi-actors environment. In the first part, it elaborates on the development of the methodology with emphasis on the evaluation process and the operational synergy of the two multi-criteria evaluation techniques exploited (PROMETHEE and AHP) and the key benefits of this mixed approach on improving both methods applicability and limiting their deficiencies. A stakeholders mechanism is exploited for the evaluation of the compared alternatives (policies) to facilitate the consensus building among stakeholders with conflicting objectives through the provision of a transparent policy selection process and to ensure the selection of realistic policy measures evaluated under criteria that correspond to the actual needs of the relevant stakeholders. In the second part the methodology is applied in a real case, namely to the evaluation of 4 existing policy proposals aiming to support the Greek coastal sector. The key advantages of the combined methodology along with the preliminary results and main messages from the real case application are discussed in the concluding part of the paper. Keywords: MCDA, Maritime Transport Policy, PROMETHEE, AHP, multi-actors evaluation 48

50 1. Introduction Policy making is a challenging and highly complex process. Decision making, as a process linking policy formulation to the actual policy implementation, is characterized by a high level of complexity and arduousness. Reaching the appropriate decision involves the optimization of a multitude of parameters and a complex interplay of information, interests and opinions of a variety of affected social groups (Hey et al, 1997). To support decision making, a variety of decision supporting methods have been developed over the years suitable for various evaluation environments. Examples of some well know methods are those of Cost-Benefit Analysis (CBA), Cost-effectiveness analysis (CEA), Economic Effects Analysis (EAA), balance sheets, Multicriteria analysis (MCA) etc. In the particular sector of maritime transportation- that is examined at the frame of this paper-, policy is made and implemented under conditions of multiple objectives (deriving from the variety of involved stakeholders) and constraints (Frankel, 1992) in an environment characterized by strong complexity in the relations between jurisdictions, administrators, politicians and the industry (Roe, 2009). Being a strong analytic tool that supports decision making in an under uncertainty environment and at can facilitate the building of consensus among all involved actors in a well explicit way, MCDA became over the years a popular evaluation method with numerous recorded applications in complex problems. The lack of one single & central goal, common to all integrated policies such as those in maritime transport, combined with a great number of impacts that cannot always be monetised, regards MCDA methods more suitable for supporting policy evaluation than other widely used financial-economic evaluation methods (eg CEA,CBA) that fail to capture the holistic view of a problem the necessary incorporation both tangible & intangible (or fuzzier ) aspects. The most popular among the various techniques to conduct a MCDA, applied in the field of transport are multi-attribute theory variants (AHP, MAUT, MAVT, SMART, SMARTER), outranking methods (PROMETHEE, ELECTRE) & regime analysis. The selection of the appropriate evaluation methodology is crucial and needs to be carefully examined in relation to the problem particularities, needs and constraints (eg in terms of time, available data etc) ; using different methods can sometimes even lead to divergent results especially when, as Finco and Nijkamp highlights (Finco&Nijkamp,1997), a complete ranking of the under examination alternatives is needed. Recently, the combination of various MCDA methods starts gaining ground facilitated by the advancing technologies that ease their use. Applying a multi-method approach can facilitate policy making by reviewing preferences and judgments derived from more than one MCA method (Mysiak, 2006). Τhis paper proposes an operational synergy of two MCDA methods to facilitate policy making in the area of maritime transportation. The proposed combination of PROMETHEE & AHP seeks to improve both methods applicability, decreasing their deficiencies, while the application in a multi-actors environment facilitates the building of consensus among actors through the provision 49

51 of a transparent participatory policy selection mechanism. A real case application of the methodology is discussed in the second part of the paper along with some preliminary results and key messages on the methodology applicability. 2. The methodological Framework The evaluation mechanism that is presented in the following is based on a multi-method approach applied in a multi-actors environment and is structured around 3 building blocks, namely, the stakeholders analysis, the combination of two MCDA methodologies (PROMETHEE & AHP) and the exploitation of two independent mechanisms in the evaluation process, namely the experts and the stakeholders group. The methodological approach is depicted in the next figure while its main building blocks are described in the following paragraphs: Figure 1: Methodological Framework The Stakeholders Analysis (SA): The term stakeholders refers to people or groups who have an interest, financial or otherwise, in the consequences of any decision taken (Macharis, 2011) or any policy selected in the specific case. In the proposed evaluation methodology, the SA is used as an aid to properly identify the range of stakeholders which needs to be consulted and views should be taken into account in the evaluation process (Macharis et al, 2012). More specifically it facilitates: the evaluation matrix structuring, through the preliminary identification of the evaluation criteria (representing the stakeholders priorities) and of the evaluation alternatives which refer to key policy proposals deriving from the various stakeholders the definition of the stakeholders group which comprises the evaluating body as described below Combination of 2 MCDA methods: the methodology proposes a combination of 2 MCDA methods namely the AHP and the PROMETHEE methods, an operational synergy that aims at reducing both methods weaknesses and enhancing their strengths. 50

52 AHP supports the design of the decision making hierarchy and the definition of the criteria weights, eliminating the main disadvantage of PROMETHEE, the lack of a structured way of defining the criteria weights. The evaluation methodology was initially structured under the AHP logic in a multi-actors environment. The ease of applicability and the structure of AHP, which follows the intuitive way in which managers solve problems (Ishizaka A& Labib A(2009) were among the reasons for examining AHP applicability in the particular research. However the assumption of the structural elements independence that is prerequisite in AHP, turned to be a critical limitation not easy to be secured in a complex environment such as this of maritime transportation. The Analytic Network Process, proposed by Saaty (Saaty, 2008) for overcoming AHP limitations by handling interdependence among elements through its supermatrix was examined as an alternative but also considered not suitable for the examined case. Its application led to a pretty complex network of elements with increased data requirements (resulting from the large number of pairwise comparisons required to capture the interdependencies and the among the network elements relations) difficult to be collected in the particular case which involves many actors coming from different environments that are usually not familiarized such methods. PROMETHEE on the other hand was considered more advantageous compared to both methods since it helped overcoming the interdependencies requirement (AHP) facilitating at the same time the various stakeholders in the alternatives evaluation through a more simplified and easy to use evaluation matrix (compared to the numerous pairwise comparisons required to solve the ANP network). Both methods support the group level decision making process. AHP exploits the geographical mean of the individual pairwise comparisons (Zahir 1999) while in the case of PROMETHEE, that is used in this case, the final evaluation results from the calculation of the weighted sum of the individual net flows (Figure 2). Tools and supporting mechanisms: the multi-actors methodology relies on the exploitation of 2 independent mechanisms in the evaluation process. In particular, an experts group is consulted for the selection of the evaluation criteria (through a DELPHI process) and a stakeholders group, defined through the SA supports the evaluation by providing the final ratings of the alternatives vs the evaluation criteria. Both groups are also providing weights to the criteria (by an AHP application), supporting the examination (by means of the PROMETHEE method) of different scenarios leading to useful conclusions and results. 3. Application of the proposed methodology The combined MCDA methodology is applied for the evaluation of 4 policies aiming to support the Hellenic coastal transportation sector. The evaluation case is described below focusing on the problem structuring, the data collection and the data analysis procedures. 3.1 Structuring of the evaluation problem The first step towards the evaluation problem structuring was the stakeholders analysis. Following the SA, 6 evaluation criteria were identified, based on the analysis of the key stakeholders group priorities as identified following a bibliographical review. The proposed evaluation criteria were 51

53 examined by an 8-member experts group (coming from the academic and research community) following an application of a- two round Delphi method. This process led to the development of a 4 4 evaluation matrix. Following the PROMETHEE requirements, for each criterion a preference function (PF) has to be defined. The PF is used to translate the difference between the evaluations obtained by two examined alternatives into a preference degree that ranges from zero to one. According to Vincke and Brans (1985) there are 6 basic types of PF: (1) usual criterion, (2) U-shape criterion, (3) V- shape criterion, (4) level criterion, (5) V-shape with indifference criterion and (6) Gaussian criterion. In the examined case all 4 criteria are qualitative; also, in the evaluation matrix, a small number levels on the criteria scale (5-point scale) is used for all 4 criteria. Based on the above properties of the evaluation criteria, the Usual (type I)PF was selected (Deshmukh S.C. (2013)). Alternatives/criteria C1: Transpor Cost Preference function/ Min/max Unit/measurement type Weights S1(Scenario 1= equal weights senari Weights S2(Scenario 2- Stakeholders grou Weights S3(Scenario 3- Experts grou A1: Re-design of the national ferry network under the logic of a network composed by several Hub ports&many peripheral ports around them (Hub & Spoke) A2: Reduction of VAT and other non-remunerative taxes in passenger and vehicle fares A3: Application of less strict coastal fleet manning regulations through measures A4: Application of Road Equivalent Tariff- RET methodology on the un-profitable lines network. Table 2: Evaluation matrix Usual min C2: Trip cost(cost to the user) Usual min C3: Subsidy Cost Usual min 1-5 Likert scale w1=w2=w3=w4 w1(s1),w2(s1),w3(s1),w4(s1) w1(s2),w2(s2),w3(s2),w4(s2) C4: Level of Service Usual max Score 11 Score 21 Score 31 Score 41 Score 12 Score 22 Score 32 Score 42 Score 13 Score 23 Score 33 Score 43 Score 14 Score 24 Score 34 Score Data collection The data collection was based on a survey to the stakeholders group comprising of 4 categories namely: shipping lines, ports, users representatives and labor representatives. The survey sample is presented in the next table. Stakeholder category Sample No Shipping lines 4 Associations,27 individual companies 31 Ports 1 Association, 22 ports/port authorities 23 Users representatives 5 National/regional Associations, 18 local representatives 23 Labor representatives 1 National Association 1 TOTAL 78 Table 3: Survey sample synthesis (stakeholders group) A dedicated questionnaire was developed based on the evaluation matrix presented in the previous session. Further to the alternatives evaluation towards the 4 criteria, the stakeholders were asked to allocate weights to the evaluation criteria through a set of 6 pairwise comparisons (AHP). Further to the pair comparison, a direct (proportional) allocation of weights was requested as an 52

54 internal inconsistency control mechanism. The personal (structured) interview method was used for the data collection along with the on-line survey option. 3.2 Data analysis For the data analysis, the Visual PROMETHEE software was exploited under a GDSS function. The process for data entry of the group decision problem is depicted in the next figure. For each stakeholder, an individual PROMETHEE based evaluation takes place based on his/hers evaluation table and relevant weights allocated to each evaluation criterion. For each action, the positive, negative and net flows are calculated. The positive and negative flows represent the intensity with which one action is preferred ore overcome by the others respectively, while the net flows (φi(ai)) represents the balance between the positive and negative flows. The net flows of each evaluation are used to formulate the evaluation matrix of each Stakeholder group (ie SG1: Shipping Lines, SG2: ports etc). The net flows of each group are again calculated and entered into the final evaluation matrix of the group decision problem (figure 2). The Decision maker can provide weights on each group. The examined case was performed for an equal based weights scenario. The above process was repeated for three scenarios; in the first scenario all criteria are given the same weights while in the second and third scenario, the evaluation is based on the weights allocated to the criteria by the experts group and the stakeholders group as described above Stakeholders Group I Stakeholders Group II Stakeholders Group n Evaluati on matrix of Stakehol der 1 Evaluati on matrix of Stakehol der 2 Evaluatio n matrix of Stakehol der n Evaluati on matrix of Stakehol der 1 Evaluati on matrix of Stakehol der 2 Evaluatio n matrix of Stakehol der n Evaluati on matrix of Stakehol der 1 Evaluati on matrix of Stakehol der 2 Evaluatio n matrix of Stakehol der n φ11 φ12 φ13 φ14 φ21 φ22 φ23 φ24 φν1 φν2 φν3 φν4 φ11 φ12 φ13 φ14 φ21 φ22 φ23 φ24 φν1 φν2 φν3 φν4. φ11 φ12 φ13 φ14 φ21 φ22 φ23 φ24 φν1 φν2 φν3 φν4 SGI evaluat ion matrix φ11 φ12 φ13 φ14 φ21 φ..1 φ22 φ..2 φ23 φ..3 φ24 φ..4 φν1 φν2 φν3 φν4 SGII evaluat ion matrix φ11 φ12 φ13 φ14 φ21 φ..1 φ22 φ..2 φ23 φ..3 φ24 φ..4 φν1 φν2 φν3 φν4 SGn evaluat ion matrix φ11 φ12 φ13 φ14 φ21 φ..1 φ22 φ..2 φ23 φ..3 φ24 φ..4 φν1 φν2 φν3 φν4 φ(g1)1 φ(g1)2 φ(g1)3 φ(g1)4 φ(g2)1 φ(g2)2 φ(g2)3 φ(g2)4 φ(gn)1 φ(gn)2 φ(gn)3 φ(gn)4 Evaluation matrix- Stakeholders Group φ(g1)1 φ(g2)1 φ(g..)1 φ(gν)1 φ(g1)2 φ(g2)2 φ(g..)2 φ(gν)2 φ(g1)3 φ(g2)3 φ(g..)3 φ(gν)3 φ(g1)4 φ(g2)4 φ(g..)4 φ(gν)4 Weights wο1,wο2.wον (of the various groups) Figure 2: Application of the PROMETHEE GDSS 4. Conclusions The policy making process is a highly complex process, requiring not only in-depth knowledge of the sector, but also the employment of a methodology that can facilitate the identification of alternative policy measures and the selection of the most appropriate ones, under conditions of multiple objectives, through a consensus building mechanism. The present paper presents such a 53

55 methodology based on a combination of two multicriteria methods namely AHP and PROMETHEE under a multi-actors evaluation environment. The methodology is designed to support policy making in maritime transportation, a particular complex and important for the economy and employment sector that is characterised by strong power of the involved stakeholders that directly or indirectly can affect policy making and implementation. The methodology proposed is built upon three main elements, the Stakeholders Analysis, the operational synergy of two MCDA method and the exploitation of 2 independent mechanisms in the evaluation process. The combination of the above elements provides key advantages in the policy evaluation process: stakeholders & independent experts' involvement throughout the process, ensures the identification of policy measures that are realistic and the employment of assessment criteria that correspond to their actual needs. a major part of the vagueness usually characterizing policy formulation is removed as the method provides a structured, step-wise approach for identifying and selecting policy measures based on transparent and easy to use (ie PROMETHEE) methods it facilitates consensus building among stakeholders with conflicting objectives, as it provides a transparent process for commonly reaching conclusions on the policy measures to be employed the stakeholders participation improves the ownership of results leading to the stakeholders engagement, necessary component for a successful policy implementation The application of the methodology to the evaluation of 4 policies aiming to support the viability of the Greek coastal transportation that is currently on-going, confirms the applicability of the proposed mechanism in the selected case highlighting however some difficulties in the application of AHP in the stakeholders environment. ACKNOWLEDGEMENT This research has been co-financed by the European Union (European Social Fund-NSF) & Greek national funds through the Operational Program "Education and Lifelong Learning" of the National Strategic Reference Framework(NSRF)-Research Funding Program: Heracleitus II. Investing in knowledge society through the European Social Fund. References Bakker P, Koopmans, C. &Nijkamp, P. (2009). Appraisal of integrated transport policies. VU University, Serie Research Memoranda (No 0052). Deshmukh S.C. (2013) Preference Ranking Organization Method Of Enrichment Evaluation (Promethee), International Journal of Engineering Science Invention ISSN (Online): , Volume 2 Issue 11, November2013,PP ,S.C.Deshmukh Finco, A. & Nijkamp, P. (1997). Sustainable land use: methodology and application. Research Memorandum

56 Frankel G E (1992). Hierarcical Logic In Shipping and Decision Making. Maritime Policy Management, Vol 3, Hey C, Nijkamp P, Rienstra S& Rothenberger D, (1997),Assessing Scenarios on European Transport Policies by Means of Multicriteria Analysis, ECON Papers, III, Tinbergen Institute Discussion Papers Ishizaka A& Labib A(2009) Analytic Hierarchy Process and Expert Choice: Benefits and limitations, OR Insight (2009) 22, doi: /ori Macharis C& Nijkamp P (2011). Possible bias in multi-actor multi-criteria transportation evaluation: Issues and solutions. Research Memorandum Macharis C, Turcksin L (2012). Multi actor multi criteria analysis (MAMCA) as a tool to support sustainable decisions:state of use. Decision Support Systems(54), pp Mysiak J. (2006). Consistency of the results of different MCA methods: a critical review. Environment and Planning C: Government and Policy 2006, Volume 24, pages Roe, M. (2009). Maritime governance and policy-making failure in the European Union. Int. J. of Shipping and Transport Logistics, 2009 Vol.1, No.1, pp.1 19 Saaty,T.(2008). Decision making with the analytic hierarchy process. Int. J. of Services Sciences, 1, pp

57 Optimal Strategic Design of Flexible Supply Chain Networks Magdalini A. Kalaitzidou Aristotle University of Thessaloniki, University Campus Thessaloniki, Thessaloniki, Greece, Pantelis Longinidis University of Western Macedonia, Karamanli & Lygeris Street, 50100, Kozani, Greece. Panagiotis Tsiakis Wipro Consulting Services, 3 Sheldon Square, London W2 6PS, United Kingdom. Michael C. Georgiadis Aristotle University of Thessaloniki, University Campus Thessaloniki, Thessaloniki, Greece. Abstract This paper presents a mathematical programming model for the optimal design of Generalized Supply Chain Networks (GSCNs) that incorporates strategic flexibility in network s configuration. The model is formulated as a deterministic Mixed-Integer Linear Programming (MILP) problem and solved to global optimality using standard branch-and-bound techniques. Optimality is assessed in terms of SCN s overall cost while its applicability, benefits, and robustness are illustrated by using a real case study. Keywords: Supply chain network design, Generalized nodes, MILP, Deterministic. 1. Introduction In recent years, the problem of designing the SCN has gained much interest from business as its contribution to sustainable competitive advantages is universally acknowledged. Facility location is the core decision within strategic design of SCNs. According to Drezner and Hamacher (2004) facility location problems involve a set of spatially distributed customers whose location is known and a set of facilities to satisfy their demands, whose locations are to be determined. Melo et al. (2009) conduct a remarkable review on facility location models and demonstrate how their characteristics affect strategic SCN management. Likewise, Melo et al. (2006) revealed how the structure of the network is strongly affected by external supply of materials, inventory opportunities, storage limitations, relocation, expansion or reduction of capacities. Thanh et al. (2008) presented a dynamic model in the design of production-distribution system by making strategic decisions: supplier s selection; opening-closing facilities etc. over a planning period. In a very recently work, Cardoso et al. (2013) proposed a model for the design and planning of SCNs with reverse flows and demand uncertainty. 56

58 The vast majority of the relevant works on the research stream of SCN design assumes a structure of the network with distinct and consecutive echelons, consisting of nodes with predetermined function, where product flow moves from an echelon s nodes to subsequent echelon s nodes. The aim of this paper is to introduce a flexible composition to network s structure, as the function of the proposed generalized nodes are optimally defined rather than selected from a set of potential alternatives. Moreover, intra-layer material flow connection is permitted among these generalized nodes. To the best of our knowledge, this is the first work that provides such a flexibility option to SCN. 2. Mathematical formulation 2.1 Problem description This work addresses the design of a multi-product, multi-echelon SCN. The model proposes an innovative configuration to network s structure by entering a level consisted of generalized production/warehousing nodes (P/W) whose function is not a priori assumed, as in mainstream fixed echelon SCNs. These nodes can receive material from any potential supplier or any other P/W node and deliver material to any customer zone or any other P/W node, as shown in Figure 1. a Suppliers (fixed location) Generalized P/W nodes (location & function to be selected)?? Customer zones (fixed location) b Suppliers (fixed location) Plants (only location to be selected & no vertical flows allowed) Warehouses (only location to be selected & no vertical flows allowed) Customer zones (fixed location)?? Figure 3 The proposed GSCN structure (a) against the typical fixed echelon SCN structure (b). We denote the set of all nodes in the network as n N. This includes not only the generalized nodes n N P/W but also suppliers nodes n N S and customer zones nodes n N C. Overall we have N=S P/W C. The objective is to minimize the overall capital and operational cost and determine the optimal structure of the network. The model defines: (i) suppliers; (ii) generalized node s location and role; (iii) material flow among SCN s levels; and (iv) functional elements (capacity, material flow, purchases etc.). 2.2 Mathematical model A deterministic MILP model is formulated where each product can be produced at several generalized P/W nodes in different location with known and time-invariant product demand (Kalaitzidou et al., 2014). All transportation flows determined are considered to be time-averaged 57

59 quantities whereas customer zones are single sourced. The objective is to minimize the overall capital and operational cost of the network and is as follows: min n P/W Q inn n C P/W {C P n Y P P n + e E en γ en + C W n Y W n + γ W WH n W n + i C in ( n S P/W Q in n + ξ kn ) + δ P e en k Kn λ ek + i( n S C in n + C inn T n C C inn Q inn )} + n S (C S n Y S n + i C S in S in ) T Q in n T n P/W Q inn + Capital cost is consists of infrastructure cost whereas handling, production, transportation and purchasing contribute to operational cost. Infrastructure cost is related to the establishment of a warehouse or a production plant at a particular node n P/W. If a production capability is established at a node n P/W then its infrastructure cost has a stable element (C P n Y P n ) and a variable element ( e E en γ P en ). The former is the product of the annualized fixed cost required to establish a production capability (C P n ) with the binary variable that expresses the establishment of this capability (Y P n ). The latter element is the sum of the products of the continuous variable expressing the total rate of availability of manufacturing resource e (E en ). Similarly, if a warehousing capability is established at node n P/W then its infrastructure cost has a stable element (C W n Y W n ), the product of the annualized fixed cost required to establish a warehousing capability (C W n ) with the binary variable that express the establishment of this capability (Y W n ), and a variable element (γ W n W n ), the product of a coefficient expressing the unit cost associated with the warehousing capacity (γ W n ) with the continuous variable expressing the warehousing capacity (W n ). Regarding operational cost, handling cost is expressed as a linear function of the total throughput at node n P/W. By multiplying the total throughputs with the unit handling cost for material i (C WH in ) and summarising the resulting products we gain the handling cost. Production cost is related to the utilization of various resources e at node n P/W and is determined as the sum of the products of the unit cost of consumption of resource e at node n P/W (δ P en ) with the total utilization of each resource e ( k Kn λ ek ξ kn ). Utilization is the product of the amount of manufacturing resource e required to perform unit amount of task k (λ ek ) and the continuous variable expressing the rate of operation of task k at node n P/W (ξ kn ). Transportation cost is decomposed into three terms each of which sums the products of unit transportation cost of material i from a node n to another node T T n' (C in n ), and vise versa (C inn ), and the corresponding continuous variables expressing the rate of flow of material i that arrives at a node n from another node n' (Q in n) and vise versa (Q inn ). The first term expresess transportation cost of material i transferred from node n' S to node n P/W T ( n S C in n Q in n), the second term expresses the transportation cost of material i transferred from T node n P/W to other node n' P/W ( n P/W C inn Q inn ), and the third term expresses the T transportation cost of material i transferred from node n P/W to node n' C ( n C C inn Q inn ). Finally, purchasing cost has a stable element (C S n Y S n ), the product of the annualized fixed cost of establishing a relationship with node n S (C S n ), and the binary variable that expresses the selection of node n S as a material provider in the network (Y S n ), and a variable element (C S in S in ), the product of the unit purchase price of material i from node n S (C S in ) and the continuous variable expressing the purchased amounts of material i from the selected node n S (S in ). By summarising this variable element for all materials we gain the purchasing cost from each node n S and then 58

60 by adding the resulting variable element with the stable element and summarizing for all suppliers we reach purchasing cost. The MILP optimization model has six sets of constraints that formulate the structure of the network, the flow of materials within the network, the core operations in the network (purchasing, production, and warehousing), and customer satisfaction. Constraints (1) and (2) demonstrate the conditions for the establishment of a node n P/W. In specific, constraint (1) states that if a production capability is established at a node n P/W (Y n P = 1) then the corresponding node n P/W should be established as the binary variable that expresses its establishment is forced to take the value of one (Y n = 1). In the same fashion, constraint (2) states the conditions for the establishment of a warehousing capability at node n P/W (Y n W = 1). Y n Y n P, n P/W (1) Y n Y n W, n P/W (2) If a node n P/W is established (Y n = 1) it should receive material from at least one other node n' S P/W and should provide material to at least one other node n' P/W C. As shown in constraint (3), if a node n P/W is established (Y n = 1) the binary variable that expresses the establishment of a material transportation link (X n n) is forced to take the value one for at least one pair of n' S P/W with n P/W and provided that n n'. In the same manner, constraint (4) shows that if a node n P/W is established (Y n = 1) the binary variable that expresses the establishment of a material transportation link (X nn ) is forced to take the value one for at least one pair of n P/W with n' P/W C and provided that n n'. Y n X n n, n P/W (3) n S P/W\{n} Y n X nn, n P/W (4) n C P/W\{n} A connection between a node n' S and a node n P/W can exist only if both the supplier is contracted and the generalized node is established. Constraint (5) forces the binary variable expressing the contracting of node n' S (Y S n ) to be unity when the material transportation link, between a node n' S and a node n P/W, is established (X n n = 1). On the other hand, constraint (6) forces the binary variable expressing the establishment of node n P/W to be unity when the material transportation link, between a node n' S and a node n P/W, is established (X n n = 1). X n n Y S n, n S, n P/W, n n (5) X n n Y n, n S, n P/W, n n (6) Similarly, a connection between two nodes n P/W and n' P/W can exist only if both nodes are established. Constraints (7) and (8) stress this condition while constraint (9) requires the establishment of node n P/W if it is going to transfer material to node n' C. X nn Y n, n P/W, n P/W, n n (7) X nn Y n, n P/W, n P/W, n n (8) X nn Y n, n P/W, n C, n n (9) 59

61 As the model does not allow reverse flows, intra-layer flows between suppliers and customer zones, and direct flows from suppliers to customer zones appropriate fixing to zero takes place for the binary variables (X nn ) expressing the establishment of the above prohibited transportation links. Moreover, the flow of materials (Q in n) lies between upper and lower bounds provided that the corresponding transportation connection has been established. In nodes n P/W where production capability is established the overall balance for the production of material i is the inflow ( n S P/W\{n} Q in n) minus the outflow ( n C P/W\{n} Q inn ) of material i plus the rate of production of material i at that node, as shown in constraint (10). Q in n + v ik ξ kn = Q inn, i, n P/W (10) n S P/W\{n} k Kn n C P/W\{n} The term (v ik ) expresses the amount of material i produced by unit amount of task k and multiplied with the continuous variable expressing the rate of operation of task k at node n P/W (ξ kn ) we have the rate of production. The total utilization of each resource e ( k Kn λ ek ξ kn ) is limited to the total rate of availability of resource e at node n P/W (E en ) as shown in constraint (11). λ ek ξ kn E en, e, n P/W (11) k Kn Upper and lower bound are imposed for both the total rate of availability of resource e at node n P/W (E en ) and purchased amounts of material i from the selected node n S (S in ). Additionally, appropriate constraints force the model to transfer all purchased material to generalized nodes and also to satisfy all demand. Finally, warehousing capacity (W n ) lies between higher and lower limits, provided that warehousing capability is established while it is approached as linear function of handled material flow as shown in constraint (12) with (a in in )/(a out in ) expressing the relationship between capacity of warehouse at node n P/W, to material i handled that enters/leaves the node. W n Q in n i,n S P/W in + Q inn a in i,n C P/W a out in, n P/W, n n (12) 3. Case study The applicability of the GSCN design and operation model is illustrated by using a real case study in the European area (Tsiakis et al., 2001). This study is being held for the interests of a European company. This network is comprised by total thirty-eight nodes, whose locations are sited among the European area. More specific there are, five potential suppliers, fifteen potential production plants/warehouses and eighteen customer zones. The number of materials/products provided by the suppliers is fourteen. 4. Results The proposed model (GSCN) is compared with a counterpart model with fixed-echelons (FSCN), both of which were implemented in GAMS software, using CPLEX 12 solver. Identical data were used for both models and production process is approached in the same way. Figure 2, present the optimal network for GSCN and FSCN, respectively. The former establishes 3 P/W nodes with both capabilities in location countries: ES, IT, BE and 1 P/W node with only 60

62 warehousing capability (CH), all of which are provisioned from 2 suppliers (BG, RO), while the latter establishes 3 plants (PT, BE, CH) and 5 warehouses (IT, DK, PT, BE, CH) all of which are provisioned from the same suppliers. The GSCN model shows its superiority firstly from the objective function, and secondly from the network s flexibility. Both models employ the same objective function that was counted 742,016 and 1,051,214 relative money units (rmu) for GSCN and FSCN, respectively. This cost gap, is due to the fact that FSCN model is forced by the a priori structure to build more facilities (sum of plants & warehouses) in order to satisfy customer demand. Furthermore, FSCN s model ends up in a structure where the more sizeable material flow connections are among the plants and warehouses that are located and build at the same country-area. This fact shows the necessity of a generalized node with both warehousing/manufacturing capabilities. Additionally, a sensitivity analysis was performed and the outcome revealed that the GSCN model reacts fairly enough in demand changes and is insensitive to all other parameters. a BG RO Supplier Manufacturing cap. Warehousing cap. Customer zone Flow from supplier to P/W Flow from P/W to P/W Flow from P/W to customer zone Figure 2 Optimal GSCN configuration (a) against the optimal FSCN configuration (b). 5. Results ES ES 898 CH 190 This paper introduces a mathematical model that provides flexibility options on designing and operating SCNs. The model is capable of deciding the appropriate suppliers and material flow connections including intra-layer flows but mainly the location and role/capability of the generalized nodes. It is concluded that, giving the network the option to have nodes that act with both manufacturing and warehousing capability (or choose among them) and simultaneously to avoid having separated manufacturing and warehousing layers, minimizes the overall cost, but mostly benefits in material handling cost. 6. Acknowledgement TR BE BE 2133 IT IT PT ES BE PL 3807 CH AT GR UK DE NO IT FR b FI SE BG IE NL DK RO This research has been co-financed by the European Union (European Social Fund ESF) and Greek national funds through the Operational Program "Education and Lifelong Learning" of the National Strategic Reference Framework (NSRF) - Research Funding Program: Thales. Investing in knowledge society through the European Social Fund PT 616 Supplier Plant Warehouse Customer zone BE CH Forward flow between echelons PT IT BE CH GR BE 6074 NO DK 4230 CH DK TR FI SE FR AT UK ES PT DE IE IT NL PL 61

63 References Cardoso S.R., Barbosa-Póvoa A.P.F.D., and Relvas S. Design and planning of supply chains with integration of reverse logistics activities under demand uncertainty. European Journal of Operational Research, Vol. 226, 2013, pp Drezner Z., Hamacher H.W. Facility location: Applications and Theory, Springer, New York Kalaitzidou, M. A., Longinidis, P., Tsiakis, P., and Georgiadis, M. C. Optimal Design of Multiechelon Supply Chain Networks with Generalized Production and Warehousing Nodes. Industrial & Engineering Chemistry Research, Vol. 53, 2014, pp Melo M.T., Nickel S., and Saldanha-Da-Gama F. Dynamic multi-commodity capacitated facility location: A mathematical modeling framework for strategic supply chain planning. Computers and Operations Research, Vol. 33, 2006, pp Melo M.T., Nickel S., and Saldanha-Da-Gama F. Facility location and supply chain management A review. European Journal of Operational Research, Vol. 196, 2009, pp Thanh P.N., Bostel N., and Péton O. A dynamic model for facility location in the design of complex supply chains. International Journal of Production Economics, Vol. 113, 2008, pp Tsiakis, P., Shah, N., and Pantelides C. C. Design of multi-echelon supply chain networks under demand uncertainty. Industrial & Engineering Chemistry Research, Vol. 40, 2001, pp

64 An Integrated Multi-Regional Long-Term Energy Planning Model Incorporating Autonomous Power Systems Nikolaos E. Koltsaklis Department of Chemical Engineering, Aristotle University of Thessaloniki, University Campus, Thessaloniki, Greece Pei Liu Department of Thermal Engineering, Tsinghua University, State Key Laboratory of Power Systems, Beijing, China. Michael C. Georgiadis Department of Chemical Engineering, Aristotle University of Thessaloniki, University Campus, Thessaloniki, Greece. Abstract This paper addresses the long-term generation expansion planning (GEP) problem of a large-scale, central power system incorporating the possible interconnection with various autonomous power systems. A multi-regional, multi-period, Mixed Integer Linear Programming (MILP) model was developed to determine the optimal power capacity additions per time interval and region and the power generation mix per technology and time period. The model is tested on the Greek power system taking also into consideration the scheduled interconnection of the mainland power system with those of some autonomous islands (Cyclades and Crete), and aims at providing full insight into the composition of the long-term energy roadmap at a national level. Keywords: MILP, GEP, Autonomous power systems interconnection, Renewable energy sources, Power sector. Introduction For decades, strategic long-term planning of a power system has been focused on guaranteeing security and quality of supply, reduction of dependence from imported fuels and power grid s stability and reliability. Nowadays, deregulation of electricity markets along with the introduction of environmental issues have made this task more complicated, requiring a great deal of parameters to be taken into consideration. Long-term GEP determines the optimal type of energy technologies to be installed, the optimal electricity production mix per time period, as well as the location and time construction of the newly built units. There has been extensive research regarding the GEP problem of the Greek power system. Agoris et al. (2004) presented an analysis in order for the Greek power system to achieve the Kyoto targets using the R-MARKAL and WASP IV models. Dagoumas et al. (2007) examined the evolution of 63

65 the Greek power sector up to 2020 by developing several scenarios concerning the implementation of a post-kyoto target. Rampidis et al. (2010) simulated the Greek power sector using the module BALANCE of the Energy and Power Evaluation Program (ENPEP) in order to investigate the feasibility of a variety of investment plants announced by power companies. Roinioti et al. (2012) presents an analysis of the Greek energy system for the period between 2009 and 2030 using the Long range Energy Alternatives Planning (LEAP) modeling tool. Our work constitutes an extension of our previous approach (Koltsaklis et al., 2014) on the grounds that: (i) incorporates possible interconnection with autonomous power systems, (ii) gives the option of electricity (imports and exports) and CO2 emissions (purchases and sales) trade to the model, (iii) provides a more detailed representation of units operation (by dividing total capacity of each unit into blocks with specific technical characteristics), and (iv) includes an annual budget constraint for possible limitations on the total amount of money to be allocated. 1. Problem statement and mathematical formulation The problem under consideration is formally defined in terms of the following items: A given planning horizon is divided into a set of uniform time periods t T. Electricity demand can be described by means of a load duration curve. In our work, eight load blocks b B are taken into account, each of which with specific load duration dur b. The overall power system is divided into a number of sectors s S, each of which is characterized by a specific electricity demand Dem s,b,t, existing initial capacity P max m,t, as well as interconnection parameters (e.g., injection losses INJ s,b,t and Load Loss factor LLF s,b,t ). There is also the option in the model to interchange (imports and exports) electricity with other neighboring grids/countries, offering more flexibility to the operational planning of the network. A set of power generation technologies m M is available to be installed in each sector s S based on different technical (e.g., availability Avf m,t ), economic (investment cost, InvCost m,t, fixed operational and maintenance cost, FOM m,t, variable operational and maintenance cost, VOM m,t, fuel cost, FC m,t, CO2 emission cost, CO2Cost t ), and environmental (CO2 emission rate, CO2EF m,bl ) criteria. These technologies include both existing capacity, m M ex, new candidate power plants, m M new, and new units with firm commissioning plans, m M fx. Technical efficiency, ef m,bl, and CO2 emission rate CO2EF m,bl, are divided into a number of blocks bl BL to represent more realistically the operation of power generating units. Grid s stability is also taken into account by incorporating minimum peak reserve requirements, Rsmg t, and maximum penetration rates of renewable energy technologies (RES), MaxRen t. Furthermore, pumped storage is incorporated in the model to facilitate load balancing. The energy policy tools to be utilized in order to promote the use of RES include a CO2 emission cap CO2Cap t, a mandatory production of a share of the total electricity generation form RES, Trg t, as well as CO2 emission pricing, CO2price t. Finally, CO2 emissions trade is taken into consideration providing the opportunity to the administrator to determine the optimal strategy concerning the total environmental performance of the system. 64

66 The objective function to be optimized concerns the minimization of the total power system expansion cost so as to satisfy the electricity demand of each sector s S in each time period t T. The objective function is given by (1), and the demand balance is expressed by Equation (2). A detailed presentation of the initially developed mathematical model can be found in Koltsaklis et al. (2014). Investment cost Min InvCost m,t + FOMCost m,t + VOMCost m,t m t CO2 trade cost m Fixed O&M cost + (CO2Exp t CO2Rev t ) + FC m,t t t m Fuel cost t Variable O&M cost + (ImpCost nc,t ExpCost nc,t ) PumCost pum,t nc t Electricity trade cost pum t m t Pumping cost (1) existing power units generation pinj m,b,t m (M EXIS M s ) net electricity flow rates new candidate power units generation + pinj m,b,t fixed units generation + pinj m,b,t m (M NEW M s ) m (M FIXED M s ) net electricity imports efl s,s,b,t efl s,s,b,t + ( iminj nc,b,t exwdr nc,b,t ) = s s s s pumping load pum (M PUM M S ) pwdr pum,b,t nc (M NC M S ) electricity demand + Dem s,b,t nc (M NC M S ) s, b, t (2) 2. Case study The applicability of our model has been tested on a case study of the Greek power system. The country has been divided into two sectors, i.e., North and South sector, based on their geographical locations. The model takes also into consideration the interconnection of Cyclades with mainland in 2016, as well as Crete s interconnection with the central power system in When it comes to electricity requirements projection, Greece s severe economic crisis has been taken into account, since electricity demand starts from 54 TWh in 2014, rises with a moderate rate to almost 59 TWh in 2020, and reaches 68.2 TWh in

67 MW Book of Proceedings 3. Results and discussion 3.1 Power capacity additions Time period (year) Lignite Natural Gas Oil Hydro Coal Photovoltaic Wind Diagram 3: Power capacity mix per technology during each time period The Reference scenario examines the case where the majority of old lignite power generation units are to be decommissioned earlier than their expected lifetimes. The studied period is between 2014 and Between 2014 and 2028, an amount of 3,817 MW (3,306 MW by 2021) is to be withdrawn from the Greek power system due to the fact that these units are characterized by very low electrical efficiency and significant carbon footprint. The results of the model indicate the construction of four new lignite plants (two units in 2021, one unit in 2022, and one in 2028) with a total installed capacity of 2,400 MW along with a lignite power station having a firm commissioning plan (615 MW in 2019). This investment plan and strategy can be explained by the fact that lignite constitutes a domestic, low cost fuel and the candidate new lignite power generating units have better technical performance and lower carbon emission factor when compared to the existing conventional lignite units. In total, lignite power capacity is reduced by 715 MW between 2014 and 2030, as depicted in Diagram 1. With the exception of a new, large natural gas unit with a firm commissioning plan (811 MW in 2015), no new natural gas plants are to be constructed since there was an overinvestment of gas units during the previous years when there was no sign of the country s deep economic recession. The model does not determine the construction of coal power plants on the grounds that coal price in not as competitive as domestic lignite s one. Hydro units report a small increase in their capacities since they start from 3,175 MW in 2014 and reach almost 3.5 GW in Oil capacity is reduced by 431 MW due to the gradual decommissioning of some old heavy fuel oil units. Since there is a national target stating that the share of RES in electricity generation must be at least 66

68 TWh Book of Proceedings equal to 40% of the total from 2020 onwards, the results indicate significant capacity additions in wind turbines and photovoltaic units. Wind turbines start from around 1.7 GW in 2014, rise to 5.3 GW in 2020 and reach 6.2 GW by % of that capacity in 2030 is to be installed in both Crete and Cyclades. They comprise the largest power generation technology, in terms of installed capacity, since Following a similar pattern, photovoltaic units report an increase of 2 GW in their capacities between 2014 and The largest share of the newly installed renewable technologies is observed in Cyclades and Crete, being interconnected to the mainland by 2016 and 2020 correspondingly. 3.2 Power generation Time period (year) Lignite Natural Gas Oil Hydro Wind Photovoltaic Imports Exports Pumping Electricity Demand Diagram 2: Power generation mix per technology during each time period As can be observed in Diagram 2, lignite units maintain a constant rate in their electricity production, since it starts from around 23 TWh in 2014 and results in 25.6 TWh in Natural gas units report an almost constant production of around 14 TWh between 2014 and 2018, while their electricity generation is reduced during the next five years due to the large penetration of RES units. Finally, they increasingly contribute to the rising electricity demand satisfaction during the last years of the studied period since almost 12.5 TWh are generated by these units. Hydro units play a balancing role in the electricity demand satisfaction with an average of almost 5 TWh during the whole period. Oil power plants are characterized by a decreasing contribution to the demand balance due to the fact that a significant amount of RES units are to be installed in the interconnected islands (Cyclades and Crete) taking the place of the old, carbon intensive diesel and heavy fuel oil plants which continue to operate up to levels that are necessary for the stability of the local power grid. 67

69 TWh Book of Proceedings The wide variation in the electricity production mix comes from renewable energy technologies, and especially from wind turbines. Their generation begins from 3.6 TWh in 2014, rises to almost 12 TWh in 2020 and reaches 14.5 TWh in Similarly, photovoltaic units almost double their production from 3.8 TWh in 2014 to 7.2 TWh in Almost 32% of that generation is produced in Crete and Cyclades. Net electricity imports account for a significant share of the total electricity demand satisfaction starting from 2.6 TWh in 2014 and approaching almost 5 TWh in Note that the northern interconnections of the country (Albania, FYROM, Bulgaria and Turkey) are mainly importing, while the southern interconnection, i.e., Italy, has an exporting profile. 3.3 Electricity flow rates Time period (year) FROM NORTH TO SOUTH FROM CYCLADES TO SOUTH FROM CRETE TO SOUTH Diagram 3: Electricity flows among domestic zones during each time period The combined effect of the binding target regarding mandatory electricity production from RES and the fact that currently autonomous islands, which are to be interconnected to the mainland, are characterized by higher availability in terms of wind potential and solar irradiation, converts them into net electricity exporters to the mainland after their interconnection. A mitigation of electricity flows, or even reverse flow during the last two years, is also observed from the North to the South system. Diagram 3 illustrates this trend depicting the electricity flows among domestic zones during each time period. 4. Conclusions This work presents an integrated, multi-regional, long-term generation expansion planning model incorporating possible interconnection with autonomous power systems. Our approach enables decision makers to develop and design alternative pathways for a power system by providing detailed calculations on the specific characteristics of each sector of the studied power grid. The 68

70 results highlight that advanced lignite units will continue to play a strategic role in the Greek power mix accompanied by natural gas units and rapidly penetrated RES units. Islands interconnection leads to the mitigation of environmental impact due to the decommissioning of old oil units as well as to significant cost savings by offering more flexibility to the total power system. Current research is focused on developing a demand response mechanism along with a Monte Carlo analysis to investigate the influence of several crucial and uncertain parameters into the long-term development of a power system. Acknowledgement Financial support from the European Commission s FP7 EFENIS project (Contract No: ENER/FP7/296003) Efficient Energy Integrated Solutions for Manufacturing Industries and Marie Curie Energy Systems Engineering (ESE) project is gratefully acknowledged References Agoris D., Tigas K., Giannakidis G., Siakkis F., Vassos S., Vassilakos N., Kilias V., and Damassiotis M. An analysis of the Greek energy system in view of the Kyoto commitments. Energy Policy. Vol. 32, 2004, pp Dagoumas A.S., Kalaitzakis E., Papagiannis G.K., and Dokopoulos P.S. A post-kyoto analysis of the Greek electric Sector. Energy Policy. Vol. 35, 2007, pp Koltsaklis N.E., Dagoumas A.S., Kopanos G.M., Pistikopoulos E.N., and Georgiadis M.C. A spatial multi-period long-term energy planning model: A case study of the Greek power system. Applied Energy. Vol. 115, 2014, pp Rampidis I.M., Giannakopoulos D., and Bergeles G.C. Insight into the Greek electric sector and energy planning with mature technologies and fuel diversification. Energy Policy. Vol. 38, 2010, pp Roinioti A., Koroneos C., and Wangensteen I. Modeling the Greek energy system: Scenarios of clean energy use and their implications. Energy Policy. Vol. 50, 2012, pp

71 Comparison of GA-ANN and Traditional Box-Jenkins Methods for Railway Passenger Flow Forecasting Nataša Glišović Department for Mathematical Sciences, State University of Novi Pazar Vuka Karadzica bb, Novi Pazar, Serbia, Miloš Milenković Division for Management in Railway, Rolling stock and Traction, The Faculty of Traffic and Transport Engineering, University of Belgrade Belgrade, Serbia Nebojša Bojović Division for Management in Railway, Rolling stock and Traction, The Faculty of Traffic and Transport Engineering, University of Belgrade Belgrade, Serbia Rešad Nuhodžić Railway Infrastructure of Montenegro Trg golootockih zrtava 13, Podgorica, Montenegro Abstract The exact prediction of the traffic conditions has become more and more significant due to the vital role in the basic functions of the management of the traffic and railway processes of decision making. This study presents an integrated genetic algorithm (GA) and artificial neural network (ANN) for Railway Passenger Flow Forecasting using stochastic procedures. This paper aims at showing the comparison of the hybrid model of the genetic-neural networks with traditional Box- Jenkins model. The technique of the genetic algorithm is used for determining the design of the neural networks. We compared the performances of proposed methods for multi step ahead prediction of passenger flows on Serbian railways. The performance of this approach is explored and results are presented. Compared performances shown that the proposed hybrid model gives better results than the traditional Box-Jenkins model. Keywords: SARIMA, Genetic algorithm, Artificial neural network, Railway passenger flow forecasting. 70

72 1. Introduction In this paper we studied the railway passenger demand in Serbia. We used the data representing monthly passenger flows on all lines of Serbian railway network provided by the Statistical Office of the Republic of Serbia. At the time of our analysis a time series of monthly passenger flows from January, 2006 to March 2014 was available. We proposed comparison of two methods, traditional Box Jenkins method and Genetic algorithm-artificial neural networks for forecasting the total number of passengers on Serbian railways using stochastic procedure. We compared the performances of proposed methods for multi step ahead prediction of passenger flows on Serbian railways. In the traditional parametric techniques, historical average (Smith and Demetsky, 1997), smoothing techniques (Williamset al., 1998), and autoregressive integrated moving average (ARIMA) (Hansen et al., 1999; Lee and Fambro, 1999) have been applied to forecast transportation demand. Particularly, ARIMA has become one of the common parametric forecasting approaches since the 1970s. The ARIMA model is a linear combination of time-lagged variables and error terms. The ARIMA model has been widely applied in forecasting short-term traffic data such as traffic flow, travel time, speed, and occupancy (e.g., Ahmed and Cook, 1979; Hamed et al., 1995; Lee and Fambro, 1999). However, the applications of ARIMA or seasonal ARIMA models are limited because they assume linear relationships among time- lagged variables so that they may not capture the structure of non-linear relationships (Zhang et al., 1998). Artificial neural network works in the same way as a human brain does, human brain consist of number of neurons connected with each other, in the same way ANN consists of artificial neurons, called nodes in network, connected with each other. The idea of Artificial Neural Network was presented in late 1943 by Walter Pitts and Warren S.McCulloch as a data processing unit for classification or prediction problems (F. Rosenblatt, 1958). The back-propagation learning algorithm (BPLA) were proposed by Rumelhart et al Many studies have indicated that genetic algorithms (GA) can be successfully applied to identity global optimizations of multidimensional functions (Chung and Alonso, 2004; Chu et al. 2008). GAs are widely applied in the optimization of the parameters spaces of neural networks. The aim of the research was to show success of the hybrid model (genetic artificial neural networks (GANN)) for Railway Passenger Flow Forecasting. This study presents an integrated genetic algorithm (GA) and artificial neural network (ANN) for Railway Passenger Flow Forecasting using stochastic procedures. The technique of the genetic algorithm is used for determining the design of the neural networks. This paper aims at showing the comparison of the hybrid model of the genetic-neural networks with traditional Box-Jenkins model. The paper is divided into several sections. In Section 2. will present the mathematical concept hybrid models. The application of the results of the research will be presented in Ssection 3. wwhile in Ssection 4. to the conclusions and opportunities are given for further research are given. 71

73 2. Methodology 2.1 SARIMA model Seasonal autoregressive integrated moving average (SARIMA) processes have been introduced in the literature to model time series with trends, seasonal pattern and short time correlations. The generalized form of SARIMA ( p, d, q) ( P, D, Q) s model for a series Y t can be written as (Box et al. 2008; Cryer and Chan 2008): ( B) ( B s )(1 B) d (1 B s ) D Y ( B) ( B s ) (1) p P t q Q t where s is the length of the periodicity (seasonality) and t is a white noise sequence. ( ) 1 (2) 2 p p B 1B 2B pb ( Bs) 1 B B B (3) P s 2s Ps 1 2 P are the non-seasonal and seasonal autoregressive (AR) polynomial term of order p and P, respectively ( ) 1 (4) 2 q q B 1B 2B qb ( B ) 1 B B B (5) Q s s 2s Qs 1 2 Q are the non-seasonal and seasonal moving average part (MA) of order q and Q, respectively. (1 B) d is the non-seasonal differencing operator of order d used to eliminate polynomial trends s D and the seasonal differencing operator (1 B ) of order D used to eliminate seasonal patterns. B d is the backshift operator, whose effect on a time series Y t can be summarized as B Y. 2.2 Neural Networks The main characteristics of the neural networks is their ability to learn when we have a complex nonlinear relationship between input and output. We use sequential procedures for training and adapting them to the data (Anil Jain et al. 2000). Back propagation is currently the most widely used techniques for training neural networks (Randall Sexton and Robert Dorsey, 2000). BP (back propagation) neural network relies on a gradient algorithm to obtain the weights of the model and uses back propagation algorithm to minimize the objective function. BPNN (back propagation neural networks) typically consists of three layers: an input layer, a hidden layer and an output layer. The most basic treatment processes called artificial neuron in the BN network and simulated on the basis of biological neurons. The summation function of a neuron is done as follows (Cheng and Liu, 2014): bj T rjiai j (6) i t Yt d 72

74 y j f wjixi j (7) i zl g vlj y j l (8) j E 1 t z 2 2 l l (9) l Where bj is the activation level of neuron j. T is the transfer function, r ji is weight value, j is bias. So the output of hidden layer and output layer described by the following equation (7) and (8). The error of output neuron is given by Equation (9). Where x i and z l are the input and output signals. y j is the output of the hidden layer. wji is the weight between input neuron j to hidden neuron i. vlj is the weight between hidden neuron l to output neuron j. j and l are the biases for the hidden layer and output layer. f and g are transfer functions for hidden and output layers. t l is the expected output. E is the error between the expected output and calculated output. In back propagation algorithms resilient BP was used (for details see (Riedmiller and Braun 1993)). 2.3 Genetic algorithm GA are robust and adaptive methods that can be used for solving combinatorial optimization problems. The basic structure is a population of individuals, each individual represents a possible solution in the search space for a given problem (the space of all solutions). In doing so, each grants fitness function which assesses the quality of the given individual, represented as a single solution in the search space. GA must provide a way to continuously from generation to generation, improves the absolute adjustment of each individual in the population, and the average adjustment of the entire population. This is achieved by successive application of genetic operators of selection, crossover and mutation, to give all a better solution given the specific problem (Hansen et al. 1999). 2.4 Hybrid model Applying this idea of genetic algorithms in neural networks, integration is used so that on the basis of input data, select a population that represents the number of neurons in the middle. In this way determine the architecture of neural networks where the work is carried out on the basis of prediction and prediction error MAE select the best architecture that represents the final output. In this research, the stopping criterion is of 200 the population (see pseudo code). 73

75 ****************************************************************************** /* pseudo code for the algorithm to predict hybrid GA-ANN */ Input_Data(); Initialization_of_the_Population(); for(i=0; i< Npop; i++) pi= a_values_function(); Fitness_Function(); Selection (); Intersection (); Mutation (); MAEmin = ANNresilientBackPropagation(); while(!finishing_criteria _GA() ) { for(i=0; i< Npop; i++) pi= a_values_function (); Fitness_Function(); Selection (); Intersection (); Mutation (); MSE=ANNresilientBackPropagation(); If(MAE < MAEmin) { MAEmin=MAE; CurrentArchitecture =ANNresilientBackPropagation(); } } Print_the_Output_Data (the current architecture); ****************************************************************************** 74

76 In the GA, reproduction is implemented by a selection operator. Selection is the population improvement or survival of the fittest" operator. It duplicates structures with higher fitness values and deletes structures with lower fitness values. The crossover, when combined with selection, yields good components of good structures combining to yield even better structures. Crossover forms n/2 pairs of parents if the number of population is n. Each pair produces two offspring structures to the mutation stage. The offspring is the outcome of cutting and splicing the parent structures at various randomly selected crossover points. The approaches for selecting crossover points are one-point crossover, two-point crossover, and uniform crossover. Mutation creates new structures that are similar to current ones. With a small, pre-specified probability, mutation randomly alters each component of each structure. The reason for using mutation is to prevent missing some significant information during reproduction and crossover. This procedure would avoid the local minimum. In this research, where GA neural network model is implemented in the programme language C#, the population represents the input time series, which is the output of SARIMA model obtained as the output of SPSS. Using GA, the best generation is received which represents the structure of neural network for which MSE (mean absolute error) is the smallest. 3. Results We used the data representing monthly passenger flows on all lines of Serbian railway network provided by the Statistical Office of the Republic of Serbia. The learning period of the GA-neural network and traditional Box and Jenkins was from January, 2006 to July, The prediction was performed the following months (from July 2013 to March 2014.) (Figure 1 and Figure 2) Figure 1. Shown SARIMA prediction passenger flows from July 2013 to March Red line presented observed values while blue line presented forecast. Figure 2. Shown GA-ANN prediction passenger flows from July 2013 to March Green line presented prediction values used Hybrid model and blue line presented observed values. 75

77 Comparison GA neural network is made with the traditional Box and Jenkins method and the prediction success is shown through the prediction errors (see the table 1). Table 4 Shows the performance of the prediction through all errors. In this case you can see the advantage of the hybrid model predictions compared to the traditional, as errors of prediction significantly less. Models Errors MAE RMSE MAPE SARIMA Hybrid model Conclusion The study shows that the performance of railway passenger flow forecasting can be significantly enhanced by using proposed hybrid method. These results show that the proposed model can be useful tool for railway passenger flow, better then SARIMA prediction. By comparing the prediction values with the real value of the railway passenger flows and by calculating MAE, RMSE and MAPE the satisfactory prediction results have been achieved. These results show that the proposed hybrid model can be useful tool for the railway passenger flows prediction then traditional Box-Jenkins. 76

78 Acknowledgement The work presented here was supported by the Serbian Ministry of Education and Science (project No. III44006 and No ). References Ahmed, M.S., Cook, A.R., Analysis of freeway traffic time-series data by using Box-Jenkins techniques. Transportation Research Board Record 722, 1 9. Anil K Jain, Robert PW. Duin, Jiangchang Mao. Statistical pattern recognition: A review. IEEE transaction on pattern analysis and machine intelligence. 2000; 22(1): B. Chu, D. Kim, D. Hong, J. Park, J. T. Chung, J.-H. Chung and T. H. Kim, GA-based fuzzy controller design for tunnel ventilation systems, Automation in Construction, vol.17, no.2, pp , Box G.E., Jenkins G.M. and Reinsel G.C., Time Series Analysis, Forecasting and Control, New Jersey: John Wiley and Sons. Cryer J.D. and Chan K.S., Time Series Analysis: with Application in R, New York: Springer. F. Rosenblatt, The perception: A probabilistic model for information storage and organization in the brain, Psychological review, 65(6):386, H. S. Chung and J. J. Alonso, Multi objective optimization using approximation model-based genetic algorithms, The 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, vol.1, pp , Hamed, M.M., Al-Masaeid, H.R., Bani Said, Z.M., Short-term prediction of traffic volume in urban arterials. Journal of Transportation Engineering 121 (3), Hansen, J.V., McDoald, J.B., Nelson, R.D., Time series prediction with genetic-algorithms designed neural networks: an empirical comparison with modern statistical models. Journal of Computational Intelligence 15 (3), Lee, S., Fambro, D.B., Application of subset autoregressive integrated moving average model for short-term freeway traffic volume forecasting. Transportation Research Board 1678, Li Cheng, Jin Liu, An Optimized Neural Network Classifier for Automatic Modulator Recognition, TELKOMNIKA Indonesian Journal of Electrical Engineering, Vol. 12, No. 2, February 2014, pp

79 Riedmiller, M. and Braun, H. (1993) A Direct Adaptive Method for Faster Back propagation Learning: The RPROP Algorithm. In: Ruspini, H., (Ed.) Proc. Of the ICNN 93, San Francisco, pp Randall S Sexton, Robert E Dorsey. Reliable classification using neural networks: a genetic algorithm and back propagation comparison. Decision Support Systems. 2000; 30(1): Rumelhart, D. E., Hinton, G. E., and Williams, R. J Learning internal representations by error propagation. In Parallel Distributed Processing, D. E. Rumelhart, J. L. McClelland, and the PDP Research Group, eds., Vols. I and II, Bradford Books and MIT Press, Cambridge, MA. Smith, B.L., Demetsky, M.J., Traffic flow forecasting: comparison of modeling approaches. Journal of Transportation Engineering 123 (4), Williams, B.M., Durvasula, P.K., Brown, D.E., Urban freeway traffic flow prediction: application of seasonal autoregressive integrated moving average and exponential smoothing models. Transportation Research Record 1644,

80 Inspection of power grid by periodic vehicle routing formulation Dr. Vasilis Spathis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Ioannis Forlidas Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Erotokritos Skordilis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Dr. Georgios Saharidis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Abstract The inspection of unattended power substations by trained personnel has been surprisingly important regardless of the enhanced remote supervision today s technology offers. In this paper, we present a mixed integer linear programming formulation that examines the optimized scheduling of these inspections with respect to the particularities and the nature of the inspection, taking also into consideration the multiple origins of the inspections and the scattered facilities to be inspected. Keywords: periodic vehicle routing problem, multi-depot, multi vehicle, network clustering 1. Introduction The substations that need to be inspected are scattered throughout the country and are essentially demoted voltage substations. These facilities include equipment that costs several million Euros and are essential for the stability of the providing power and the quality of service of the power industry. Specifically throughout mainland Greece are 291 substations connected by 11,300 km of transmission lines. The particularities of the technical inspection of the equipment and moreover the importance of the facilities imply the regularly physical inspection by technical personnel. These inspection have the advantage of the natural presence of highly trained personnel that will not only observe, but also hear or sense a possible malfunction hours or days before it become 79

81 crucial for the system. Their report will introduce an early maintenance, or even replacement of suspicious equipment reducing the probability of a major discontinuance of the power supply. Every route of inspections will include the departure station, the substation that need to be visited, the time of inspection at each substation and the total available time for inspection. Our goal is to reduce the total travel time for the inspection of the substations. The solution of the problem presented in this paper is based on a more general class of problems, those of vehicle routing (VRP) and more specifically of the multi depot VRP. 2. Real case study The mathematical formulation of this problem is based on the formulation of vehicle routing optimization problems with multiple depots and time windows presented by Dondo et al. [1]. The formulation was altered to satisfy the requirements of our problem and then applied on real data. The substations of Central Greece were fed to the algorithm and routes that can cover the inspection to that substation were calculated. Total number of substation in Central Greece is 26 of whom 24 need to be inspected from inspectors departing from 4 depots, with no more than 5 routes from each depot and total time of each route of inspection not exceeding 400 minutes. The formulation was realized in C++ code and solved using CPLEX libraries. Due to the large number of the substations it was not possible to have complete results for the whole problem. Therefore it was necessary to group the substations (clustering) around a given starting point (depot) and solve different sub problems (VRP). The grouping took part considering geographical criteria. PICTURE 1: Electrical substations of central Greece 80

82 3. Results DIAGRAM 2: Re-clustering with respect to partial optimized total solution The chart above shows the total time required for completion of the inspection of five different cases. In the first column (un-optimized) is presented the domestic solution which has not been optimized, already used by the maintenance company. In the second column (fixed clusters) it is shown the time-optimized solution for the existing clustering, which the company is using. The last three columns present the results for three different scenarios of clustering carried out. It should be noted that under each column the CPU time and the total progress of the solution is displayed. Examining the above figure it is clearly shown that the time for the inspection without optimization is about 3000 minutes, and the optimized routing decreases the total time to 1900 minutes. Similarly we observe that for scenarios A, B and C we have no important improvement compared to the fixed clustering solution. Lastly we tested the complete problem and after 48 hours of execution and without leaving the 100% GAP, the best solution provided had an improvement of 60 minutes (1 hour) compared to the clustering scenarios tested. 81

83 4. Conclusions Further Investigation It is clear that the optimized solution provided by the model discussed in this paper can provide savings up to 40% in hourly wages and significant savings in driven km. Furthermore, the creation of organized routes allows smooth implementation of the inspections, solves several personnel management issues, while making the audit process more efficient without ignoring the factors and characteristics of the inspection. One of the real problems that we could improve is the time resolution required by the computer. This could be accomplished by the following ways: Constrains for near-by substations Constrains for single route covering more than one substation Constrains for incompatible substation References R. Dondo, J. Cerda, A cluster-based optimization approach for the multi-depot heterogeneous fleet vehicle routing problem with time windows, European Journal of Operational Research 176 pp , J-F. Cordeau, M. Gendreau, G. Laporte, A Tabu Search Heuristic for Periodic and Multi-Depot Vehicle Routing Problems, European Journal of Operational Research 119 pp , S. Nanda Kumar, R. Panneerselvam, A Survey on the Vehicle Routing Problem and Its Variants, European Journal of Operational Research 74 pp.66-74, Vehicle Routing Problem from Wikipedia, the free Encyclopedia (last access 22/6/2014). 82

84 Environmental performance evaluation using a fuzzy aggregationdisaggregation approach Zouboulia Sbokou School of Production Engineering and Management, Technical University of Crete, University Campus, GR73100 Chania, Greece sbokou.lina@gmail.com Evangelos Grigoroudis School of Production Engineering and Management, Technical University of Crete, University Campus, GR73100 Chania, Greece. Michael Neophytou Mills of Crete, 40 Eth. Venizelou, Souda, GR73200 Chania, Greece. Abstract An important tool for the evaluation and the documentation of a successful environmental management system is the Environmental Performance Evaluation (EPE). The EPE is defined as a continuous internal process and a management tool that uses indicators in order to evaluate the environmental management system of a business organization and to compare past and present environmental performance. International standards ISO describe the categories of performance indicators; however they do not determine a specific framework for the development and measurement of these indicators. The main aim of this study is to present an EPE methodology based on a fuzzy multicriteria analysis approach. In particular, the Fuzzy UTASTAR method is applied in order to evaluate the environmental performance of a mill industry. It is an extension of the well-known UTASTAR method capable to handle both ordinary (crisp) and fuzzy evaluation data. To evaluate the environmental performance of the industry, the production processes are analyzed and the environmental indicators related to the environmental impact of the industry are defined. The environmental indicators are related to the products of the industry, the material consumption, the consumption of natural resources and waste management. The five groups of the indicators are: air emissions, solid waste, natural resources and energy, environmental education and third parts, recycling and improvement measures. The main steps of the presented approach include the following: Criteria assessment (definition of the final set of indicators and their measurement units), Definition of fuzzy sets (fuzzy values that reflect the low, medium and high performance of each indicator), Development and ranking of alternative scenarios for each group of indicators, Application of the Fuzzy UTASTAR model (estimation of fuzzy utility functions, overall utilities for scenarios). The actual value of each indicator is used in the estimated utility function of the corresponding indicator and the result is normalized in order to calculate the environmental performance of the industry. The environmental performance of the industry can be measured per dimension (group of indicators) or per total. Keywords: Environmental Performance Evaluation, Fuzzy UTASTAR, Environmental Management Systems, Aggregation-Disaggregation Approach. 83

85 1. Introduction According to the international standard ISO (ISO, 1999) environmental performance evaluation (EPE) is a process to facilitate management decisions regarding an organization s environmental performance by selecting indicators, collecting and analyzing data, assessing information against environmental performance criteria, reporting and communicating, and periodically reviewing and improving this process. Environmental performance indicators (EPIs) measure the current or past environmental performance of an organization and compare it to the targets set by the organization s management (Jasch, 2008). International standard ISO describes the categories of performance indicators; however it does not determine a specific framework for the development and measurement of these indicators. Multi-Criteria Analysis (MCA) methodologies may be considered as tools that can handle a set of indicators for the environmental evaluation. MCA applications are applied mainly to sustainability appraisals; that is, they include indicators from economic, social and environmental categories, as well as technical criteria (Herva and Roca, 2013). The main aim of this study is to present an EPE methodology based on a fuzzy MCA approach. Although fuzzy methods have been applied in the examined problem, the presented study is the first attempt in the context of preference disaggregation. Fuzzy UTASTAR is a method for inferring fuzzy utility functions from a partial preorder of options evaluated on multiple criteria. It is an extension of the well-known UTASTAR method (Siskos and Yannacopoulos, 1985) capable to handle both ordinary (crisp) and fuzzy evaluation data (Patiniotakis et al., 2011). Fuzzy UTASTAR builds fuzzy additive value functions taking as input a partial preorder on a subset of the options, called reference set, along with their associated scores on the criteria. The resulting fuzzy utility functions can subsequently be used to estimate the (fuzzy) utility of each option, thus allowing their ranking, prioritization, selection or classification. The ranking of the options in partial preorder is as compatible as possible to the original one (Patiniotakis et al., 2011). 2. Environmental Performance Evaluation in a Mill Industry 2.1 Environmental Indicators In order to evaluate the environmental performance of the industry, the production processes are analyzed and the environmental indicators related to the environmental impact of the industry are defined considering the input-output analysis (Figure 4), the environmental aspects and the environmental policy of the industry. The environmental indicators are related to the products of the industry, the consumption of materials, the consumption of natural resources, and the generated wastes. The five groups of indicators are: air emissions, solid waste, natural resources and energy, environmental education and third parts, recycling and improvement measures. 2.2 Criteria Assessment It is necessary to reduce the large set of indicators in order to be applied and controlled by the industry. For this reason, a technique based on the assessment of environmental impacts of the selected indicators is applied (Karavias, 2008). The first three categories of indicators related to 84

86 environmental pollutants (air emissions, solid waste, natural resources and energy) are evaluated on five criteria: 1) the severity of the impact, 2) the likelihood to happen, 3) the frequency of occurrence, 4) whether the indicator is controllable, and 5) if there is legislation demanding to measure or not the indicator. The other two categories of indicators (environmental education and third parts, and recycling and improvement measures) are evaluated on three criteria: 1) reportability, i.e., how and the frequency of exposure of the various documents and processes, 2) the interest from various stakeholders and third parts, and 3) if the indicator is in the organization s objective. A five-level scale is used for the criteria of severity, likelihood, frequency, controllability, reportability, and interest from stakeholders (with 1 the lowest and 5 the highest value). A two-level scale is used for the other two criteria of legislation and organization s objective (with 1 for positive and 0 for negative response). The decision-maker (DM) scored the indicators and then the impact of each indicator is calculated, as shown in Table 5 and Table 6. The impact of each indicator is calculated as follows: 3 1 x 1 t 1 1 x4 1 t 1 Impact x5 for the three first categories of indicators and x6 1 1 x7 1 1 Impact x8 for the two last categories of indicators, where x, i 1,2,,8 are the criteria of severity, likelihood, frequency, controllability, i legislation, reportability, interest from stakeholders, and organization s objective, respectively. Inputs Outputs Raw materials Fuel Energy Water Packaging materials Flour Animal feed Energy consumption Air emissions Solid waste Wastewater Noise Heat release Figure 4. Input-Output analysis of the mill industry The indicators with the highest impact of each category are chosen for the final set of indicators (bold and underline values in Table 5 and Table 6). There is no single threshold for the impact so each category will have at least two indicators. The final set of indicators consists of 17 indicators. 85

87 Indicator Criteria Impact Severity Likelihood Frequency Controllability Legislation Amount of CO Amount of NO x Amount of SO x Amount of SS Amount of CO Amount of PM Table 5. Impact assessment of air emissions Indicator Criteria Impact Severity Likelihood Frequency Quantity of auxiliary materials from recycled materials Annual quantities of recycled products Number of products with environmental friendly instructions Percentage of environmental goals achieved Number of vehicles with eco - friendly technology Number of planned audits and inspections completed Number of findings in inspections per year Number of emergency exercises that have taken place Response time for the corrective actions Costs due to penalties and fines from infringements Cost of environmental improvement actions/the total budget Table 6 Impact assessment of recycling and improvement actions 2.3 Assessment of Fuzzy Sets and Development of Scenarios In cooperation with the management of the mill industry, fuzzy values that reflect the low, medium, and high performance of each indicator are defined. Next, different scenarios with fuzzy values for each indicator are developed for each category of the selected indicators. An example for the case of air emissions is shown in Table 7. In order to apply the Fuzzy UTASTAR method, the DM is asked to rank these scenarios. 86

88 2.4 Application of Fuzzy UTASTAR Based on the aforementioned scenarios, the Fuzzy UTASTAR method is applied in order to estimate the fuzzy value of each scenario and the utility function of each indicator. The fuzzy value of each scenario is a triangular fuzzy number ( Ux, U y, U z), which is converted to a crisp number according to following formula: R( a) ( a1 2 a2 a3) 4, where R() is a ranking function and ( a1, a2, a 3) is a triangular fuzzy number. The calculation of R() is necessary in order to obtain comparable values for all the examined scenarios. Table 8 shows the utilities of the alternative scenarios, using the previous formula. The estimated value functions for the criteria in the category of air emissions are shown in Figure 5. All these value functions refer to decreasing criteria, since they are related to undesirable outputs of the production process. As it can be observed, the most important indicator for the environmental performance of the mill in this particular category is the amount of SOx. No. of scenario Amount of CO2 (tn CO 2 / month) Amount of NOx (kg NO x / month) Amount of SOx (kg SO x / month) Ranking by the DM 1 (30, 40, 46) (75, 80, 85) (140, 150, 160) 1 2 (47, 50, 55) (55, 60, 63) (162, 175, 182) 2 3 (31, 35, 37) (64, 70, 74) (185, 190, 200) 3 4 (38, 40, 46) (55, 60, 63) (185, 190, 200) 4 Table 7 Alternative scenarios for the category of air emissions No. of scenario U x U y U z R Table 8 Estimated overall utilities for the scenarios of air emissions 87

89 Figure 5 Value functions for the criteria of air emissions 2.5 Overall environmental performance The actual value of each indicator, as measured in the mill industry, is used in the estimated utility function of the corresponding indicator and the result is normalized in order to calculate the environmental performance of the industry. The environmental performance of the industry can be measured per dimension (group of indicators) or overall. For example, Table 9 presents the calculations in the case of air emissions. As shown, based on the estimated utility functions ( X corresponds to the values of the criteria scale, while Yi refers to the corresponding utility), the utility of the current value of the industry is calculated through linear interpolation, since Y are piecewise linear functions. The overall performance score is the sum of these utilities, and can be used to normalize the estimated value of each indicator. Following the previous procedure, the environmental performance of the industry may be estimated. Figure 6 shows the EPE results in the category of air emissions, as well as the overall performance of the industry. As it can be observed, the results in the dimension of air emissions are relatively high, while the category of resources and energy appear to have the lowest performance. i Indicator Value Utility Current value X 1 X 2 X 3 Y 1 Y 2 Y 3 Linear interpolation Normalized value Amount of CO Amount of NO x Amount of SO x Overall value Table 9 Calculation of environmental performance evaluation 88

90 Figure 6 Environmental performance of the mill industry (air emissions and overall) 3. Conclusions The applied method is an easy to handle and flexible tool for evaluating the environmental performance of a business organization. This method can be applied in any type of organization, as each organization can choose the suitable combination of evaluation criteria that are appropriate for its own operations and activities. Also, the results are able to determine the strong and the weak points, as well as potential improvement action regarding the environmental management system. Another important advantage is that the proposed approach may take into account the DM s preferences (environmental strategy of the organization) and it can help the DM to handle the uncertainty of the data. Finally, the method is able to give a clear picture of the rate of the environmental objectives and targets achieved from the total environmental objectives and targets set by the senior management. References Herva, M. and E. Roca. Review of combined approaches and multi-criteria analysis for corporate environmental evaluation. Journal of Cleaner Production, Vol. 39, 2013, pp ISO. Environmental management-environmental performance evaluation-guidelines: ISO International Standard Organization, Geneva, Jasch, C. Environmental performance evaluation and indicators. Journal of Cleaner Production, Vol. 8 No. 1, 2008, pp Karavias, P. Development of a methodology for the evaluation of environments performance. MSc Thesis, School of Production Engineering and Management, Technical University of Crete, Chania, 2008 (in Greek). 89

91 Patiniotakis, I., D. Apostolou D., and G. Mentzas. Fuzzy UTASTAR: A method for discovering utility functions from fuzzy data. Expert Systems with Applications, Vol. 38 No. 12, 2011, pp Siskos Y. and D. Yannacopoulos, D. UTASTAR: An ordinal regression method for building additive value functions. Investigação Operacional, Vol. 5 No. 1, 1985,

92 Rationalizing electricity production investments from renewable energy sources in Greece using a synergy of multicriteria methods Eleftherios Siskos National Technical University of Athens, 9, Iroon Polytechniou Str., , Athens, Greece Dimitrios Peronikolis National Technical University of Athens, 9, Iroon Polytechniou Str., , Athens, Greece John Psarras National Technical University of Athens, 9, Iroon Polytechniou Str., , Athens, Greece Abstract The necessity for disengagement from conventional energy sources, along with the increasingly strict measures of the European Union (EU) towards this direction, lead to the promotion of the renewable energy sources (RES). In particular, Greece, although having high potential in electricity production from RES, mainly hydroelectric, wind and solar, is still behind in comparison to other EU countries in the area of RES adoption. Nevertheless, the incentives provided by Europe and the Greek government, during the last few years, for investing on electricity production from RES, are multiple and significant. The aim of this paper is to evaluate and rank medium-scale investments on electricity production from RES, of approximately 10MW, in the broader area of Greece. Specifically, a multicriteria evaluation system is elaborated, based on four points of view: (i) technical (ii) economic (iii) social, (iv) environmental, and (v) political. The investments assessed are categorized with respect to the type of RES invested upon (i.e. solar, biomass, geothermal, etc.) and the area of implementation (mainland, islands or offshore). The selection is supported by the ELECTRE IS method, which takes into account the presence of pseudo-criteria through the proposed evaluation system. The criteria weights are elicited with the aid of a revised Simos methodology. The overall objective of this research work is to support energy policy decision making in Greece and trigger sustainable development. Keywords: Multicriteria decision support, Simos method, Renewable energy sources, Energy policy measures, Electricity production investments. 91

93 1. Introduction Scope of research The necessity for disengagement from conventional energy sources, along with the increasingly strict energy policy measures applied by the European Union (EU) in this direction, lead to the promotion of renewable energy sources (RES). In particular, Greece despite having enormous natural wealth, and all forms of RES, is still behind in matters of electricity production, compared with other EU countries in this area. As said, geologically and by its climate, Greece can support almost all types of electricity production from RES, increasingly attracting investors. There are, though, different investment opportunities, which are differentiated by their type, scale, location and other characteristics. The challenge therefore, is to investigate what energy power investment by RES is the most appropriate at the moment, based on the national needs and objectives. Until now, this selection problem has been approached by various surveys and studies, assessing the dynamics of each RES per category and quality, which deviates from reality. However, few studies have been engaged to the evaluation of RES investments with the use of Multicriteria Decision Analysis (MCDA). This paper addresses the problem of the selection of the optimal investment for electricity production from RES for the case of a public investor, such as the Greek government. Initially, an analysis of the possible investment scenarios in electricity production from RES in Greece is examined and elaborated. Nine alternatives are chosen and evaluated on five points of view. These points of view are (i) technical (ii) economic (iii) social, (iv) environmental, and (v) political each of whom consists of multiple dimensions. In the following phase the MCDA method, Electre IS, is applied. The parameters of the problem, used by the method (weights, thresholds e.t.c.) are determined by the Investor/Decision maker (DM). The method results in an outranking graph of the alternatives, the results of which are examined prior to selecting the most preferred candidate, based on the investor s preferences. 2. Literature Review A number of studies have been published in the field of evaluation of different energy planning solutions. These works, which approximate the current study, can be categorized as follows: Quantitative or qualitative comparison of all type of alternative energy sources Qualitative assessment of RES investments Quantitative evaluation of RES investments Evaluation of different technologies of a particular RES investment Multicriteria evaluation of energy planning investments 92

94 Table 1 presents a general sample of the literature review on evaluations of energy planning alternatives, with a small description and the method that has been applied. Table 1: Indicative applications Scientific study Description Method used Kaya and Kahraman, 2010 Polatidis et al., 2006 Tsoutsos et al., 2008 Karakosta et al., 2012 Georgopoulou et al., 1996 Kahraman et al., 2008 Naim et al., 2001 Burton and Hubacek, 2007 Energy planning evaluation using MCDA (all energy sources) Analysis and comparison of multicriteria methods over their suitability on energy planning from RES. Evaluation of the energy planning of Crete using MCDA Comparison of RES technologies and investments to nuclear ones Evaluation of RES technologies for the island of Crete using MCDA Multicriteria evaluation of all energy sources Multicriteria evaluation of alternative power plant technologies Multicriteria evaluation of small scale power plants TOPSIS Outranking, utility based models, goal programming Promethee Ι & ΙΙ Qualitative evaluation Εlectre ΙΙΙ AHP Additive value model using variable weights Macbeth & Costbenefits Cavallaro, 2005 Energy planning evaluation Promethee Ι & ΙΙ For an elaborative literature review, on sustainable energy decision making using multicriteria decision making methods, the reader is prompted to read the paper of Wang et al. (2009). 3. Description of the problem The purpose of this study is to select the preferentially optimal among the electricity production investments (scale of 10 MW) from RES in Greece. Firstly, the alternative options of investments that can be supported by Greece s geological, climatic and technical conditions and limitations are 93

95 distinguished. Based on the previews studies the following alternatives are finally selected (see Table 2). To simplify the procedure that follows, each alternative is represented by a letter. Table 2: Alternatives Photovoltaic station (Interconnected) Photovoltaic station (Not interconnected) Wind plant (Interconnected) Wind plant (Not interconnected) Small hydroelectric plant Solar thermal plant (Interconnected) Solar thermal plant (Not interconnected) Geothermal power plant Biomass power plant a b c d e f g h i These alternatives are evaluated based on a consistent family of criteria that is built according to the classical modeling methodology (Roy, 1985). In fact, a set of dimensions is grouped in five points of view which in turn are all sub-aggregated to lead to a set of nine evaluation criteria (see Table 3) (Løken, 2005). The values in the parentheses show the type of each criterion and their worst and best levels. The criteria fabrication process was implemented after careful and thorough planning in order to cover all different aspects that may affect the disposition of the investor towards his final decision. Table 3: The evaluation system of RES investments Points of View Dimensions Criteria Social Economic Technical Jobs creation Social acceptance Additional social benefits Investment cost Operational and maintenance cost Electricity selling price % Subsidy Efficiency Compatibility g1: Social criterion (quality scale, 1-5) g2: Cost criterion (M / M,W 2-18,1) g3: Revenue criterion ( /ΜWh, ) g4: Effective operation (quality scale, 1-5) 94

96 Reliability Environmental Political Know how Operational safety Preparation time Life Cycle Effect on soil Effect on water Noise pollution Landscape degradation Required area Penetration margin Stability & policy bureaucracy g5: Expertise (quality scale, 1-5) g6: Project cycle (years, 18-48) g7: Environmental criterion (quality scale,1-5) g8: Required area (w/m 2, 4-60) g9: Political support (quality scale, 1-5) 4. Methodological Frame The MCDA method that was selected for the purposes of this decision problem is the ELECTRE IS, due to the fact that it that can also manage pseudo-criteria. For a pair of actions a,b the primacy relationship is defined (see Figueira et al., 2005): asb (a, b) The concordance check is described by two sets (Roy and Bouyssou, 1993): J s = { j J / gj (a) + qj[gj(a)] gj(b)} And J Q = { j J / gj(a) + qj[gj(a)] < gj(b) gj(a) + pj[gj(a)]} It is positive when C (a,b) = wj jεjs + φjwj jεjω s 95

97 where φj= gj(α) + pj[gj(α)] gj(b) pj[gj(α)] qj[gj(a)] w: criterion importance weight g: value of each alternative of a criterion p: preference threshold q : indifference threshold The discordance check is positive when gj(b) gj(a) vj[gj(a)] ηjqj[gj(b)] V: veto threshold Where nj= 1 C(a,b) wj 1 s wj For the elicitation of the criteria weights, the Simos method is used. It is a simple but effective method that facilitates the Decision Maker express easily his/her preferences over the importance of criteria, with the use of a deck of cards, (see Simos 1990a, for more details on the method). 5. Implementation Before the implementation of the algorithm the parameters of the model need to be determined. First of all, the weights, which are calculated based on the Simos method (Simos, 1990a and Simos, 1990b), the results of which are presented in Table 4. Figure 1 below illustrates the deck of criteria cards, as arranged by the investor in an ascending order of importance. It should be noted here, that the after the arrangement of the cards, the weights were not calculated using the mathematical procedures of Simos, due to a number of robustness issues they bear (see Figueira and Roy, 2002, and Siskos and Tsotsolas, 2014). Instead, the system of inequalities was solved maximizing the most importance criterion, namely the revenues. 96

98 Figure 7: Simos results Secondly, the veto thresholds, indifference and preference levels are estimated after dialogue with the DM. All these values are presented along with the criteria values in Table 4. Table 4: Criteria values and thresholds Alternative g1 g2 g3 g4 g5 g6 g7 g8 g9 a 4 2, b 4 2, c 2 5, d 3 6, e 3 7, f 3 4, g 3 4, h 3 15, i 3 18, Weights 0,072 0,172 0,224 0,112 0,132 0,132 0,072 0,032 0,052 Indifference threshold - 1, Preference threshold Veto threshold {2,5} {1,5}

99 After the execution of the two steps of the algorithm, namely the concordance and discordance checks, we obtain the following matrices, (see Tables 5 and 6). Table 5: Concordance check (s =0,7) a b c d e f g h i a x 0,82 0,72 0,70 0,55 0,89 0,75 0,50 0,53 b 0,70 x 0,73 0,69 0,42 0,90 0,87 0,61 0,53 c 0,43 0,43 x 0,82 0,62 0,43 0,43 0,48 0,29 d 0,41 0,43 0,74 x 0,65 0,54 0,43 0,48 0,29 e 0,45 0,58 0,74 0,88 x 0,69 0,69 0,67 0,48 f 0,47 0,48 0,74 0,60 0,37 x 0,82 0,48 0,38 g 0,46 0,48 0,74 0,74 0,48 0,97 x 0,48 0,38 h 0,50 0,50 0,74 0,74 0,68 0,72 0,65 x 0,41 i 0,47 0,47 0,71 0,71 0,52 0,69 0,62 0,76 x Table 6: Discordance check Pairs of alternatives g2 g3 g4 g5 a,b a,c a,d a,f a,g b,a b,c b,f b,g c,d

100 d,c e,c e,d f,c f,g g,c g,d g,f h,c h,d h,f i,c i,d i,h All the pair values in the concordance check, surpassing s =0,7 take part in the discordance check. In particular four pairs (red colored) exhibit positive check. The rest of the outranks between the alternatives are presented at the outranking graph below (see Figure 2). The Figure also showcases the core of Electre IS (Π) including the best alternatives, which are not outranked by any other. After dialogue with the investor and a final analytical review of the specific values of the four best alternatives over the evaluation criteria, the alternative b is rejected. Finally, the decision maker, after having been informed about all the information for a,e and I, selected the photovoltaic (interconnected) as the best alternative. The procedure followed is presented in Figure 3. 99

101 Figure 8: Outranking graph Figure 9: Decision support procedure and final selection 6. Conclusions The study presented in this short paper, achieves the modeling and solution of the power generation from RES investment evaluation problem, using a synergy of MCDA methods. It also achieves an effective implementation of the Simos method to infer indirectly the criteria weights. The 100

102 evaluation system (points of view, dimensions and criteria) includes all the parameters that affect an investment on electricity production. The optimal alternative, as resulted using the preferences of a fictitious decision maker/public investor, is the interconnected Photovoltaic power plant. The use of an MCDA method, such as the ELECTRE IS, that is able to handle pseudo-criteria, gives rational results, taking into account the indifference in minor differentiations of the values of the alternatives on certain quantitative criteria. The final results emerged after the appropriate implementation of the synergy of multicriteria methods, namely the ELECTRE IS and the Simos. They were then set under careful and meaningful dialogue with the investor, who expressed her/his preferences regarding the importance of the criteria and the final selection of the one among the four final optimal alternatives. As a result, the solution surfaced as an algorithmic-procedural solution combined with the decision maker s judgment. Regarding some potential future perspectives of this study, it could be implemented for the case of another country or area of interest, where it would be interesting to test how different data and parameters diversify the final results. Also a different modeling of the problem or differentiation of the objective (i.e. selection of the optimal investment portfolio) can be applied or even an extension of the alternatives by adding incorporation of RES power plants, energy saving actions, cogeneration of heat and electricity and nuclear power. Finally, the problem could also account the preferences of multiple DMs, with different viewpoints and magnitudes in their opinions. In that case, it is critical to integrate adequately these preferences, using elaborative mathematical models, and result to a compromise solution that is to be accepted by all. References Burton, J. Hubacek, K. Is small beautiful? A multicriteria assessment of small-scale energy technology applications in local governments, Energy policy, 35, 2007, pp Cavallaro, F. An Integrated Multi-Criteria System to Assess Sustainable Energy Options: An Application of the Promethee Method, (2005). Retrieved online: Figueira, J., S. Greco, and M. Ehrgott (Eds.). State-of-Art of Multiple Criteria Decision Analysis, Dortrecht: Kluwer Academic Publishers, Figueira, J. and Roy, R. Determining the weights of criteria in the ELECTRE type methods with a revised Simos procedure, European Journal of Operational Research, 139, 2002, pp

103 Georgopoulou, E., Lalas, D., Papagiannakis, L. A multicriteria decision aid approach for energy planning problems: The case of renewable energy option, European journal of Operational Research, 103, 1996, pp Kahraman, C., Kaya, I., Cebi S. A comparative analysis for multiattribute selection among renewable energy alternatives using fuzzy axiomatic design and fuzzy analytic hierarchy process, Energy, 34, 2008, pp Karakosta, C., Pappas, C., Marinakis, V., Psarras, J. Renewable energy and nuclear power towards sustainable development: Characteristics and prospects, Renewable and Sustainable Energy Reviews, 22, 2012, pp Kaya, T., Kahraman, C. Multicriteria decision making in energy planning using a modified fuzzy Topsis methodology, Expert systems with applications, 38, 2010, pp Løken, E. Use of multicriteria decision analysis methods for energy planning problems, Renewable and sustainable energy reviews, 11, 2005, pp Naim, H., Afgan, M., Carvalho, G. Multi-criteria assessment of new and renewable energy power plants, Energy, 27, 2001, pp Polatidis, Η., Haralambopoulos, A., Munda, G., Vreeker, R. Selecting an appropriate multicriteria decision analysis technique for renewable energy planning, Energy sources, Part B: Economics, Planning, and Policy, 1, 2006, pp Roy, B. Bouyssou, D. Aide multicritère à la décision: Méthodes et cas, Paris: Economica, Simos, J. L évaluation environnementale: Un processus cognitif négocié. Thèse de doctorat, DGF EPFL, 1990a, Lausanne. Simos, J. Evaluer l impact sur l environnement: Une approche originale par l analyse multicritère et la négociation, 1990b, Presses Polytechniques et Universitaires Romandes, Lausanne. Siskos, E., Tsotsolas, N., Christodoulakis, N. Elicitation of criteria importance weights through Simos method: A robustness concern, Under Revision. Tsoutsos, T., Drandaki, M. Frantzeskaki, N., Iosifidis, E., Kiosses I. Sustainable energy planning by using multi-criteria analysis application in the island of Crete, Energy policy, 37, 2009, pp Wang, J-J., Jing, Y-Y., Zhang, C-F., Zhao J-H. Review on multi-criteria decision analysis aid in sustainable energy decision-making, Renewable and sustainable energy reviews, 13, 2009, pp

104 Orisma(c): Optimizing long term fleet wide crew assignment Takis Varelas Danaos Research Centre Akti kondili 14, Piraeus 18545, Greece Sofia Archontaki Danaos Research Centre Akti kondili 14, Piraeus 18545, Greece Myrto Livadioti Danaos Shipping Co Ltd Akti kondili 14, Piraeus 18545, Greece Abstract We present the unique long term maritime crew planning and assignment optimization that Danaos Corporation envisaged, Danaos Management has implemented and Danaos Shipping deployed as an enrichment of its ORISMA 5 (Operation Research In Ship Management) toolkit. The major novelty in this system is the extension of the two coordinates, the number of vessels from one to whole fleet and the time horizon from couple of weeks to several months. Another initiative is the addition of a third dimension the teamwork index of the vessels management team. As assignment optimization problem the definition of an objective assignment function that should be optimized is required. We analyzed the problem and found out the formulas and the variables that are needed for the calculation of coefficients in the identified individual objectives that are combined in a weighted multi-objective assignment penalty function. The extension of coordinates and the team working dimension increases the problem complexity and is hard to achieve optimal solution with conventional heuristics. So we combine operation research genetic and muliti-index axial integer models, efficient assignment algorithms and, new developed ones into one model adjusted to specific problem requirements. System also supports strategic decisions regarding the depth determination of the availability officers pool, the entries attributes such as rank, performance, availability and nationality avoiding in one hand unfeasible solutions and keeping on the other the safety pool level as less as possible. Furthermore an alert mechanism generates the appropriate triggers for actions whenever safety levels are reached and suggests mitigation plan. 5 ORISMA is awarded with the highest distinction in Operation research from INFORMS, the Institute of Operations Research and Management Science, the Franz Edelman finalist award for

105 From design point of view the most important novelty of the ORISMA approach is the usage of ship officers quadruplet as the monitoring entity instead of the individual ship officer. Keywords Assignment problems, Integer programming, quadratic assignment, multi-criteria analysis 1. Problem description Compared with other human resources management systems maritime crewing is highly complex because of particularities of marine profession. Seafarers usually are not work permanently and are employed on period basis. A typical employment period is on about 4 months. Crew synthesis may be multi-national and multicultural. Company compiles crew scheduling taking into account the maximization of the crew retention factor. Furthermore a crew appraisal scheme is utilized to evaluate the seafarers performance as well as the vessel-seafarer suitability. The main goal is to find the proper person for the proper vessel in right time with the highest confidence, keeping the seafarers availability pool as small as possible as well as maximizing the seafarers satisfaction and reducing the non-productive cost of idle time between the ESO (estimated sign off) and AF (available from) dates. In our days the most known conventional approach is the monitoring of the debarkation list and the assignment from the stand-by (availability list) of the most suitable seafarer per vessel. Additionally crew operator, based on his expertise, may take another decision even shifting in or out the debarkation date of the onboard officers, even shifting in or out the seaman availability date. He also takes into account the co-existence coefficient of the eventual synthesis that may raise communication and/or performance problems. Despite the fact that is approach is quite adequate for one vessel in a small time horizon that we need is to view the whole picture to support the holistic better solution. In this direction we develop, implement and deploy the ORISMA(c) solution with impressive results specifically when the fleet population is larger than 20 vessels. Indicatively the rejoin cost may be reduced on about 2M $ /year for a fleet with 50 vessels. Companies utilize their own appraisal schemas to evaluate the performance of the officers. Sophisticated schemas incorporate several assessors and assessment criteria. Evaluation grade may be expressed qualitatively or quantatively. The evaluation quantitative grade for each officer may be formulated as weighted multi-criteria function or by simple assignment. 104

106 vessel officer oficer Anyway an officer has a performance grade based on his professionalism, experience and other attributes. Additionally each officer has a different suitability factor for different vessels. Finally each officer has different coexistence potential with other seafarers. Hereafter is substantial to formulate the assignment index taking the aforementioned factors the individual officer performance, the officer-vessel suitability and the officer-officer coexistence level (the teamwork index). The ship-officer suitability depends on nationality, experience in the same, sister or similar vessel type or engine type in case of engineers. Significant suitability factor is the relation of EOS (estimated sign on-off) of the officer, who is going to be replaced and the AF the availability date for embarkation. There are several objectives in crew scheduling. One of them is to keep the retention factor (RF) in the highest level. According to TMSA (tanker management self-assessment) a more than 80% crew retention is required to achieve the highest assessment stage (Crew retention KPI key performance indicator=4). Another important objective is the definition of officers pool per rank size. The size, or otherwise the depth, of each pool should be as small as possible but adequate enough to assure the embarkation requirements fulfillment. The smaller feasible size assures the higher crew confidence level for next in proper time employment. Of course important object is the crew assignment cost in monetary terms that may be formulated according to crew management policy. All the above objectives (optimization of retention, satisfaction, cost and quality) are combined in one goal programming function adjusted to each company that should be optimized. 2. Conclusions assignment index The solution has been tested successfully. System is utilized from several companies and the feedback evaluation is impressive in terms of tangible as well intangible benefits. The system power originates from the efficiency of the incorporated algorithms which provide the assignment map within seconds. So operators are able to run the program with different scenarios using a user friendly interface. In these scenarios may select criteria such as vessels, vessel nationalities, fleets, ranks and any combination of them. They may also alter system proposals. This interactivity provides the appropriate feedback for system calibration. System is also customizable. In the setup the availability window may be changed and, penalty tables may be altered. Finally an intelligent module may provide not only the optimum solution but proposed solutions in descending order based on specific criteria. A build-in multi criteria analysis model would be useful in the definition of weights of the used individual attributes of the objective function coefficients. 105

107 3. Objective, Indexes and formulas 3.1 Teamwork index (The highest index the lowest co-existence) Four senior officers (n=4) are employed at any time in each vessel. The seniors quadruplet q for a given period consists of the Captain (m), the Chief Engineer (c), the chief officer (o) and the second engineer(s). Each of them has his individual g grade as an integer value (1 to 4). There are three correlation matrices mc; mo. and cs were the value of each matrix cell is the penalty of coexistence of the two senior grades as coordinates. We assume the penalty coexistence index as the sum of the three correlation cells: (1) For each q{m,c,o,s) { pq= mc(g(m),g(c))+mo (g(m),g(o))+cs(g(c),g(e)) } The quadruplet is changed whenever a senior demarcates and is replaced. Assuming that employment period is fixed (f) then we will have at most n+1 different q s. Their corresponding time intervals are as follows: (2) ed0 = current date: edn+1=ed1+f : For i=1 to n+1 { di=edi-edi-1 } The team-working index from initial to final stage when all the quadruplet members will have replaced should take into account the duration of each q and is calculated as (3) P= i=1σ n+1 (Pi* di)/ (ed1+f - ed0) 3.2 Availability index Whenever we have an open position for embarkation among two candidates with the same specification we reasonably select the seafarer who is first in the waiting list. So we define an availability bonus and consequently an associated penalty of short waiting time. If a seafarer is waiting more than one month it is urgent to find an employment for him. On the other hand in case we are unable to find an available officer for an open decision operator may contact a seafarer who is in vacation period to assess the possibility for embarkation. So we need to incorporate in the availability penalty index this alternative with some penalty of course. 4. The ORISMA(c) optimized mode The developed model improves dramatically the elapsed time and at the same time assures the optimum solution. The main initiatives are as follows: Step-1: Generate variables For each v vessel in a fleet there are feasible i alternative vessel officers quadruplets qi with assignment penalty piv. Feasible quadruplets are generated from the entries of the four (one for each rank) availability pools that could be assigned to each vessel. With this filtering q s number is much less of the generated n 4 quadruplets from the combinations of all elements of the mentioned normalized pools (n=max of pools depth). Consequently the number of variables x is iv and denote the of the corresponding v-q assignment truth (x=1) or not (x=0). 106

108 Step-2: Formalize objective v q objective min p ij x ij i=1 j=1 Step-3: Formalize assignment constraints one for each officer The sum of variables, where an officer or is involved, should be 0 or 1 and not only 1 or in other words a candidate officer may be assigned once or may not be assigned. It is in practice explainable because usually we have more candidates that open positions. So we replace the right hand side of all assignment constraints for all officers from =1 to <=1. o r ε q ij i = 1: v, j = 1: q(i) x ij { 0,1} & constraint r = constraint r + x ij constraint r 1 r Step-4: Formalize assignment constraints one for each vessel It is obligatory to assign on quadruplet to each vessel. But in practice whenever availability pools have been kept small sometimes we may not be able to find quadruplets to all vessels without conflicts and the mentioned model will only notify infeasibility without explanations. This infeasibility is too hard to be analyzed manually. To overcome this weak point we introduce in the objective function a dummy xio =x[v] where v is the vessel number with a high value h coefficient and we insert this variable in the corresponding vessel assignment constraints. 4.1 Final model q (i) : number of feasible q s for vessel i, h: high value penalty pij xio : assignment penalty of quadruplet j to vessel I : dummy variable for vessel i, v : number of vessels, or : officer r v i=1 q j=1 v i=1 objective min p ij x ij + h q(i) V j=1 x ij + I=1 x i0 = 1 i = 1: v x i0 Q={qij o r ε q ij i = 1: v, j = 1: q(i) x ij { 0,1} & constraint r = constraint r + x ij constraint r 1 r 107

109 References Balas, E. and Saltzman, M.J. (1991), An algorithm for the three index assignment problem. Operations research 39: Hahu, P, Grant, T., and Hall, N. (1998), Solution of the quadratic assignment problem using the Hungarian method. European Journal of Operational Research, 108: Haln, P.and Grant, T. (1998) Lower bounds for the quadratic assignment problem based upon a dual formulation. Operations Research 46: Karp,R.M (1980) An algorithm to solve the m x n assignment problem in expected time o(mn logn). Networks, 10: Varelas O, Archontaki S (2011) Intelligence in crew option systems. 3 rd International. Symposium on. Ship operations, Management and Economics. (Society of Naval Architects and Marine Engineers (SNAME), Jersey city, NJ pp j Varelas P et al. (2013): Optimizing ship routing to maximize fleet revenue at Danaos Interfaces 43(1), pp

110 Simulation analysis of a pilot handling system for the rail transport of conventional semitrailers A. Ballis Department of Transportation Planning and Engineering National Technical University of Athens 5, Iroon Polytechniou, GR-15773, Athens, Greece Tel: Fax: abal@central.ntua.gr Abstract The transport of semi-trailers by rail is of foremost importance due to the fact that it allows for the smooth shifting from road haulage to the intermodal/combined transport operations. ISU is a pilot system allowing for the loading/unloading of semi-trailers on trains by picking them from their wheels (therefore by applying the same forces as when they are sitting-on during road transport). Within the European project CREAM effort was given to the analysis and improvement of the operations and the configuration of the ISU system aiming to develop the next generation of this equipment. Field observations and a simulation - based analysis were performed by the NTUA research team. The scope of the current presentation is to outline the methodological framework used for the above simulation analysis as well as to present the early results of the research work. Keywords SIMULATION, HANDLING SYSTEM, RAIL TERMINAL, SEMI-TRAILERS. 1. INTRODUCTION Semi-trailers are essential components of the road traffic and thus also of the pre- and post-haulage legs of the intermodal/combined transport chains. The use of semi-trailers as loading units in the railway transport has a long record, either as cranable semi-trailers (that are loaded on special wagons by cranes or reach stackers) or as conventional (non-cranable) semi-trailers that are loaded on trains using various horizontal handling techniques (rolling motorways, the Modalohr system or the ISU system). The scope of the current presentation is to outline the methodological framework used for the simulation analysis of the ISU system and to present the early results of the research work. Section 2 presents the above conventional and innovative systems revealing the need for research in this sector. In Section 3 the methodological approach (field observations, model structure, scenarios investigated) is presented together with the early results of the analysis. The last Section hosts the conclusions and the future steps of the research. 2. THE NEED FOR INNOVATION IN THE RAIL TRANSPORT OF SEMITRAILER A semi-trailer is a goods road vehicle with no front axles designed in such a way that part of the vehicle and a substantial part of its loaded weight rests on the road tractor (i). The use of semi- 109

111 trailer as a loading unit in the European railway transport has a long record, yet while in road transport the number of semi-trailers has been constantly growing, in rail transport the share of semi-trailers is declining (ii,iii). The rail transport of semi-trailers is of primary importance due to the fact that allows for the smooth shifting from road haulage to the intermodal/combined transport operations, given that certain organizational problems are solved (iv). To this aim, legislative measures can be used (and are used) to reduce the transport of semitrailers by road (e.g. lorry tolling systems according to heavy goods vehicle emissions, strict monitoring of maximum driver working hours (v,vi), etc. In parallel, research is carried out in order to improve technical and organisational aspects that improve the efficiency of the rail transport of semi-trailers. One critical technicality concerns the methods used to load/unload semi-trailers on railway wagons. The most known methods are: o Cranable semi-trailers. The term cranable semi-trailer (or grapple arm semi-trailer) stands for a special semi-trailer type having adequate strength to be engaged in a lifting operation by a crane having a grapple-arm spreader. Cranable semi-trailers are in accordance with the normal roadgoing specifications and legislative dimensions (length: 13.6m, height: 4m, width: m) but in addition, are being reinforced to withstand the stresses of being lifted in a laden state from road into rail and vice versa. For this purpose they have strengthen chassis, modified lift suspension, hinged underrun protection device and are fitted with lifting pockets in the underside to host the crane s grapple arms. The disadvantages of the cranable semitrailers in comparison with the conventional ones are the higher cost and (mostly) the higher tare weight that reduces the maximum permitable cargo weight by about 500 kg. o Rolling Motorways. A Rolling Motorway or Rollende Landstrasse or Ro-La is a system where complete road vehicles are driven onto special rail wagons with small wheels that are forming a long platform where road vehicles are rolling on and off (vii). Rolling motorways are used in the European transalpine services across Switzerland and Austria which restricted the number of heavy vehicles transiting the country each year. The disadvantage of Rolling Motorways is that (a) the lorry must also travel with the semitrailer and (b) that the small wheels are wearing much more than the wheels of the ordinary wagons. o The Modalohr system. This innovative system can carry trucks or semi-trailers by use of specifically designed low-floor articulated railway wagon. The upper deck of each wagon can rotate so that semi-trailers can roll on by use of road tractor (viii). The Modalohr system was used in the France-Italy transalpine corridor and in the Luxembourg to Perpignan (French) route. The disadvantage of this system is that it requires special expensive wagons and dedicated terminals (with ramps allowing the roll on and off of semi-trailers to wagon decks) in both ends of the journey. o The ISU System. This innovative system was designed to allow the transport of existing conventional semi-trailers (non-cranable) by picking them from their wheels. A prototype has been developed and demonstrated in Wien Nordwest terminal as well as in Wells terminal in Austria. The pilot run of the system has been performed on the Austria Turkey route as a part of activities performed within the European CREAM project (xi). The detailed description of ISU system is included in the following Section. Other innovative ideas (mostly on conceptual stage or as drawings) exist, yet the thorough presentation of all concepts and ideas published about semi-trailer handling equipment, is outside the scope of the current presentation. 110

112 Figure 10: Conventional tractor-semitrailer combination (upper part) and components of ISU system 3. TRAIN SERVICE PROCESSES AND SIMULATION OF ISU SYSTEM ISU is the abbreviation of the German Innovativer Sattelanhaenger Umschlag that means innovative semi-trailer transhipment. ISU system was invented by two German rail experts (ix) that reached an agreement with Rail Cargo Austria for the pilot use of the system. The system was presented in the framework of the BRAVO project (x) and further was investigated and improved within CREAM project (xi). To allow for the transport of conventional semi-trailers ISU is picking them from their wheels, e.g. by applying the same forces as when they are sitting-on during road transport. The system (see Figure 1) requires a specific (but existing) wagon type, a number of wheel-packer elements, a transverse beam and an adapter (an auxiliary frame that can be mounted to any spreader, hoisting ropes are connected to a traverse beam and wheel-packer elements). The wheel-packer elements are located in a special ramp or in the ground (ISU-ramps); the semi-trailer is moved into the wheel-packer elements and is lowered to the jockey wheels. After the ropes have been mounted by the groundsman the vertical transhipment is performed by the crane into the pocket wagon, where the ropes are detached and loading is completed (x). The wheel-packer elements remain in the wagon to facilitate the unloading procedure at the destination terminal. The handling system of ISU deviates a lot from the common conventional handling operations (due to the necessity for wheel-packer elements and the transverse beam that must be moved between wagons and the ISU ramp). The service cycle of ISU is largely dependent on the initial conditions (wheel grippers in ramp slots or in the wagon, reach stacker near the ramp or near the railway line) as well as by the sequence of handling requests (loading or unloading operations). Field observations and preliminary analysis revealed that any complex train service operations can be analysed in combinations of 4 elementary processes. Figure 2 presents graphically the initial conditions and operations of ISU systems handling a loaded truck after having served another loaded truck. Different operations exist in other cases (loaded truck followed by empty truck, empty truck followed by empty truck, empty truck followed by loaded truck). In order to investigate the parameters affecting the performance of ISU system and to propose improvement for ISU equipment, a simulation-based analysis has been performed by a research team of the National Technical University of Athens (NTUA) participating in the CREAM project consortium. Field observations and evaluation of the analysis was facilitated and supported by a research team of Rail Cargo Austria that also participated in the consortium. 111

113 The simulation model was developed in ARENA software with certain complex modules in Simon language. The model has advanced animation capabilities: (a) tractors and trailers are presented using computer graphics (see Figure 2, middle part) while handling operations are presented using videos. Each time a handling operation is performed the programme retrieves and displays the relevant video recording of the pilot ISU operation in Wells. Video playing time is synchronised with the simulation clock so that tractor and trailer movements are synchronised with video pictures. The simulation analysis revealed that the performance of the system is largely dependent on tractor arrival sequence therefore, proper queuing discipline rules have been introduced to improve system s performance. Furthermore, the following technical modifications have been simulated for various tractor arrival patterns. o Special gripping apparatus to reduce handling time and labour fatigue o Stand alone spreader with supporting legs. This way Reach Stacker can decuple from the spreader (that remains on the ramp) to perform other tasks using a second ISU spreader. This option affects positively the service cycle time. o Anti-sway system o New ramp design allowing for fast repositioning in various terminal locations The embedded Table in Figure 3 outlines the results of the analysis. The Moderate revision option has a fair cost and good performance yet the additional weight of the extra apparatus (legs, anti-sway) increases the weight of the ISU spreader and therefore reduces the maximum weight that can be handled by the Reach Stacker. To overcome this disadvantage, a new lighter frame design is required to compensate for the additional weight. The Major revision option includes all proposed modifications at the expense of higher cost. Initial Conditions for the current operation ISU ramp Empty slots Vehicle Tractor with Semitrailer Reach Stacker Near railway line Previous operation that sets the initial conditions (semitrailer to be unloaded) Current operation (semitrailer to be unloaded) Final Condition for the current operation Wheel grippers and lifting beam in place Departs Near railway line 1. Empty Travel 112

114 2. Reach Stacker picks up wheel grippers and lifting beam 3. Reach Stacker detaches wheel grippers and lifting beam in the associated ramp slots 8. Handling (Reach Stacker lifts semi-trailer into the pocket wagon) 4. Tractor with semi-trailer rides the ramp 5. The driver disconnects the semi-trailer from the tractor 6. Handling (Reach Stacker picks up semi-trailer and moves Figure 2: Initial conditions and operations of ISU systems handling a loaded truck after having served another loaded truck. Different operations exist in other cases (loaded truck followed by empty truck, empty truck followed by empty truck, empty truck followed by loaded truck). 113

115 Minor Revision Moderate Revision Major Revision Spreader with supporting legs Anti sway Gripping apparatus New frame design Mobile Ramp Service cycle -1% -30% -35% Labour fatigue -25% -50% -50% Figure 3: Proposed Loading capabilities Reduced Cost Fair high modifications, simulation analysis and early results 4. CONCLUSIONS Semi-trailers are important elements of the intermodal transport system as they allow for the smooth shifting from road haulage to the intermodal transport chain. The rail transport of semitrailers have been implemented in Europe through various methods, yet more research is required as the existing handling techniques suffer from certain drawbacks. ISU system is a new proposal that is currently in pilot stage. Field observations and simulation-supported analysis performed 114

116 within the framework of a European research project concluded to a number of modifications that will improve significantly the performance of the system. The research teams of NTUA and Rail Cargo Austria have agreed to continue the research towards the development and testing of a prototype derived from the current analysis. ACKNOWLEDGEMENT The current research was performed as a part of the European research project CREAM ( ) which was financed by the 6th EC Framework Programme. References 1 United Nations, Glossary for transport Statistics, 3 rd edition S A I L project. Semitrailers in Advanced Intermodal Logistics, Analysis of the State of the Art Deliverable 1, November 2000, 1 UIRR. Statistics 2007, UIRR International Union of combined Road-Rail transport companies, 1 National Technical University of Athens, UIRR and CEMAT. Investigation of Greek Transport demand for the Combined Transport corridor Greece-Italy-Germany, Pilot Project Financed by European Commission DG VII, RoadTransport.com EU Drivers' Hours explained, 06 May 2008, 1 EC Regulation 561/2006 on the Harmonisation of Certain Social Legislation Relating to Road Transport and Amending Council Regulations (EEC). No 3821/85 and (EC) No 2135/98 and Repealing Council Regulation (EEC) No 3820/85, Lowe D. Intermodal Freight Transport, Elsevier, Lohr Groupe, Modalohr Presentation. Available at: 1 Official site of ISU system at: 1 BRAVO project. Information available at: 1 CREAM project Customer-driven Rail-freight services on a European mega-corridor based on Advanced business and operating Models. Final Report available at : 115

117 Research on internet sufficiency of websites concerning women agricultural co-operatives in Greece: A multicriteria approach Athanasios Batzios Lab. of Agricultural Informatics, School of Agriculture, Faculty of Agriculture, Forestry and Natural Environment, Aristotle University of Thessaloniki, Greece, [thanos.batzios@gmail.com] Thomas Bournaris Lab. of Agricultural Informatics, School of Agriculture, Faculty of Agriculture, Forestry and Natural Environment, Aristotle University of Thessaloniki, Greece, [tbournar@agro.auth.gr] Zacharoula Andreopoulou Lab. of Forest Informatics, School of Forestry and Natural Environment, Faculty of Agriculture, Forestry and Natural Environment, Aristotle University of Thessaloniki, Greece, [randreop@for.auth.gr] Christos Batzios Laboratory of Animal Production Economics, School of Veterinary Medicine, Faculty of Health Sciences, Aristotle University of Thessaloniki, Greece, [batzios@vet.auth.gr] Basil Manos Lab. of Agricultural Informatics, School of Agriculture, Faculty of Agriculture, Forestry and Natural Environment, Aristotle University of Thessaloniki, Greece, [manosb@agro.auth.gr] Abstract This paper deals with the research of the internet sufficiency of websites concerning women agricultural co-operatives, through empirical research for the assessment of criteria/characteristics of relative websites. Towards this direction, the basic criteria/characteristics of a website were identified and then, an empirical research was performed in a sample of 30 websites of women cooperatives. These websites evaluation was accomplished through specific navigation characteristics, design, interactivity, accessibility, e-services and usefulness of the information provided, that reflect the internet sufficiency of these websites. The data that derived from the empirical research were used to evaluate the fulfillment rate of the various criteria/characteristics for the internet sufficiency of the websites of women co-operatives. Multicriteria analysis was further performed aiming to hierarchically classify and rank the websites towards the total net flow of internet sufficiency. The results of the research out sketch the profile of internet presence and promotion of women agricultural co-operatives. The conclusions of the paper can comprise useful consulting tools and contribute in a more rational organization of websites for the promotion of women co-operatives and generally in the effective development of women entrepreneurship in the agricultural production sector. Key words: women agricultural co-operatives, websites, website evaluation criteria, multicriteria analysis 116

118 1. Introduction Women agricultural cooperatives constitute an agricultural business sector that functions under specific rules and legislation due to its collaborative character, that differentiates that sector from other business sectors (Gidarakou et.al., 2000). However, women co-operatives often lack in effective management and they face problems on product packaging, quality control, brand name, etc. affecting the marketing of the products. The problem of products disposal seem also to be important for most of women co-operatives due to their distance from urban centers and their markets. Within the current economic recession and the high unemployment rates in Greek economy, women co-operatives can play an economic and social role as a factor for women employment, local economy and agro-tourism. Lately, rural agro-business tend to adopt new flexible ways for their promotion, while their corporate website is the most contemporary means for their internet promotion. In order to successfully accomplish a contemporary and competitive internet presence for the women agricultural cooperative business, effective web design and structure is a prerequisite within the context of powerful cooperative promotion. Today, women s co-operatives presence in the internet is not satisfactory, as their majority does not even have a website or they are hosted in other local websites while there is recent research findings and proposed models for website evaluation (Andreopoulou et.al. 2014, Chatzinikolaou et.al, 2013; Tsekouropoulos et.al, 2012; Kargioti et.al, 2006; Patsioura et al., 2004). The necessity for analytical research on the characteristics of internet sufficiency of women cooperatives websites in combination to their special role in the socio-economic status of our country was the trigger for that research. The aim of the paper is the research on the internet sufficiency of websites of women agricultural co-operatives, through empirical analysis and evaluation of the websites criteria/characteristics. The research focus in the following targets: Identification of the basic criteria/characteristics for the evaluation of women agricultural co-operatives, Evaluation of the accomplishment rate for the various criteria /characteristics for the internet sufficiency of websites and Multicriteria analysis and hierarchical classification and ranking of these websites towards their sufficiency. 2. Research Methodology Based on bibliography, some basic identified criteria/characteristics were chosen and grouped in six basic multi-criteria categories (Table 1). The research on the internet of women co-operatives has been made using specific relevant key-words and their combinations through «Goοgle» search engine. According this, the proper population of active women co-operatives with (proper) website is Ν=39 co-operatives, in a total of Ν =159 co-operatives active today in Greece. Through random sampling, 30 websites were collected to form the research sample using random numbers. In the form-questionnaire for the research evaluation were also included general info questions and 30 questions about the basic criteria/characteristics of internet sufficiency structured in 6 categories in YES/ NO type. The total accomplishment rate for the various criteria/characteristics is evaluated through the evaluation form. Further, multicriteria analysis PROMETHEE II was performed to generate a total ranking of the websites in the sample. 117

119 3. Results Table 1: Criteria for the evaluation of women agricultural co-operatives websites Bellow are the results of Statistical Data analysis in Criteria/ Characteristics of: Navigation: Average accomplishment Rate: 41,34 ±28,85% Web-design: Average Accomplishment Rate: 57,34 ±30,40% Interactivity: Average Accomplishment Rate :38,02±31,77% Accessibility: Average Accomplishment Rate 52,00±45,49% E- Services: Average Accomplishment Rate : 8,02±5,07% Information Usefulness: Average Accomplishment Rate: 31,34±25,09% The total ranking of the websites using multicriteria analysis (PROMETHEE II) was based on the score of total net flow of their internet sufficiency (Doumpos & Zopounidis, 2004; Roy, 1991). PROMETHEE II, applies a linear function having as criteria the websites characteristics that reflect navigation, web design, interactivity, accessibility, e-services and information usefulness (M. De Marsico and S. Levialdi, 2004). The results from multicriteria analysis using PROMETHEE II are presented in Table 2. AA Website Rank Total Net Flow: Φ (k i) = Φ + (k i) - Φ - (k i) 1 Women Cooperative Arnissas Voras P. Pella 3,

120 2 Women Agro-touristic Cooperative Mesotopos Lesvos 2, Women Agricultural Cooperative Agras Lesvos 1, Women Agricultural Cooperative Leonidion 0, Women Cooperative Kokkinogia Dramas 0, Women Agro-touristic Cooperative Zagoras P. Magnesia 0, Women Cooperative Gerakari P. Larissa 0, Women Agricultural Cooperative Glossa Skopelos 0, Women Agro-touristic Cooperative Portaria, Pelion 0, Women Cooperative Kato Asiton Traditional Asitians dishes P. Heraklion 0, Women Cooperative Rachon Ikaria 0, Women Agro-touristic Cooperative Pteleus Ftelia - P. Magnesia 0, Women Cooperative Ioannina 0, Women Agricultural Cooperative Aigionion Armonia P. Pieria 0, Women Agro-touristic Cooperative Parakila, Lesvos -0, Women Cooperative Erganos P. Heraklion -0, Women Agricultural Cooperative Agios Antonios P. Thessaloniki -0, Women Agro-touristic Cooperative Gardikion Anemona -0, Women Agro-touristic Cooperative Sesklon Ftasma -0, Women Cooperative Archanon Geuseis P. Heraklion -0, Women Cooperative Kalon Agron Dramas Kaloagritissa -0, Women Agricultural Cooperative Argolida Kianon Erga P. Argolida -0, Women Agro-touristic Cooperative Trigonon P. Evros -0, Women Agricultural Cooperative Velvendus P. Kozani -1, Women Agricultural Cooperative Zakron Mehlion P. Lasithi -1, Women Cooperative Poroion P. Serres -1, Women Cooperative Melissanthi P. Heraklion -1, Women Agricultural Cooperative Archontissa tou Aigaiou P. Hydrousa, Andros -1, Women Agricultural Cooperative Anatolikon, Valmada - P. Thessaloniki -1, Women Agricultural Cooperative Gimnotopos P. Preveza -1, Table 2: Multicriteria analysis PROMETHEE II for the ranking of the websites in the sample, based on the total net flow of their internet sufficiency. 119

121 4. Conclusions The websites lag significantly as to their characteristics of Navigation and they present a high accomplishment rate only for criterion fast website loading (86,5%), while the average accomplishment rate in this category is estimated to 41,34±28,85%. Satisfactory Accomplishment is found for the criteria of the Design, only according to neutral website background (96,7%) and the color combination while all the other criteria are accomplished in low rate. A lag is observed in Interactivity characteristics except of the criterion Telephone and/or (86,7%), with another accomplishment rate estimated to 38,02±31,77%. Accessibility criteria achieve a low accomplishment rate (52,00±45,49%), e.g.: 70% of the Websites do not support other languages and connection through social media while they do not have mobile view ή mobile application. However, all the Websites support the criteria various navigation platforms and full access no log-in required. According to E-services criteria the results are totally disappointing (8,02±5,07%). Finally according to Information Usefulness criteria a satisfactory rate is accomplished for the criteria co-operative history info (73,3%). In terms of total net flow of internet sufficiency of the web sites, with a low positive total net flow are found 14 women co-operatives. Optimum Website is of women co-operative in Arnissa, named Voras in prefecture of Pella (total net flow= 3,0489), and then following 2 Websites in Lesvos. The rest 16 Websites of the sample present a negative sign in relation to the total net flow of their internet sufficiency. Findings of the research show that the internet presence of women co-operatives is nor sufficient neither functional, with the majority being characterized of poor content and design. The majority in the sample have only created a basic website with certain characteristics. e-shop with shopping cart, audiovisual content- music/sound/video/slideshow, Social media connection, multiple language support, online payment security, mobile view/application e.g. Android are some of the basic characteristics found in a contemporary women cooperative website, though they were found in limited websites. Only a few cases of websites follow the website s structure and design rules causing a positive interest to internet users/potential clients. Based on this, we have the belief that conclusions can become useful consulting tools for a functional structure of similar websites, in the context of effective promotion of women entrepreneurship. Results of research, lead us to make the following suggestions: Improvement of the existing websites, functional enrichment with attractive content according to the co-operative character. Professional design of the website. Independent website promotion in the internet. Links in relative portals such as Ministry of Agricultural Development, Chambers, Prefecture, etc. Consulting through government stakeholders and motivation to co-operatives without website to proceed in uploading their functional website. Initiatives for informing the co-operatives on the benefits of their internet promotion and probably organization of short term seminars on web design, internet marketing and entrepreneurship. 120

122 References Andreopoulou, Z., Koliouska, C., Lemonakis, C., & Zopounidis, C. National Forest Parks development through Internet technologies for economic perspectives. Operational Research, 1-27 Chatzinikolaou, P., Bournaris, T., & Manos, B. (2013). Multicriteria analysis for grouping and ranking European Union rural areas based on social sustainability indicators. International Journal of Sustainable Development, 16(3), Doumpos M., and Zopounidis K. (2004). Multicriteria Analysis of Decisions. Methodological approaches and applications, ISBN: New Technologies Publications, 2004 Gidarakou I., Xenou A. and Theofilidou K. (2000). Farm Women's New Vocational Activities: Prospects and Problems of Women's Cooperatives and Small On-Farm Businesses in Greece. Journal of Rural Cooperation, 28(1): Kargioti E., Vlachopoulou M. and Manthou V. (2006). Evaluating customers online satisfaction: The case of an agricultural website. Proceedings of the International Conference on Information Systems in Sustainable Agriculture, Agroenvironment and Food Technology (HAICTA2006). ISBN , Ed. Nikolas Dalezios, University of Thessaly Publications, 2006 M. De Marsico and S. Levialdi (2004). Evaluating Web Sites: exploiting user s expectations, Int. J. Human-Computer Studies, 60 (2004), Patsioura, F., Vlachopoulou, M., & Manthou, V. (2004). Evaluation of an agricultural web site. Proceedings of International Conference on Information Systems & Innovative Technologies in Agriculture, Food and Environment, (pp ), Thessaloniki, March Roy B. (1991). The outranking approach and the foundations of ELECTRE methods. Theory and Decision, Vol. 31, pp Tsekouropoulos, G., Andreopoulou, Z., Seretakis, A., Koutroumanidis, T., & Manos, B. (2012). Optimising e marketing criteria for customer communication in food and drink sector in Greece. International Journal of Business Information Systems, 9(1),

123 Adaptation of ITA for project portfolio selection within a group of decision Olena Pechak *, George Mavrotas, Danae Diakoulaki, makers Laboratory of Industrial and Energy Economics, School of Chemical Engineering, National Technical University of Athens, Iroon Polytechniou, 9, Athens John Psarras Energy Policy Unit, School of Electrical and Computer Engineering, National Technical University of Athens, Iroon Polytechniou, 9, Athens Abstract Project portfolio selection is a problem of selecting a subset of projects from a wider set, optimizing one or more criteria and satisfying specific constraints. Unlike in financial problems, these projects are integer variables which are not divisible. Multiple Criteria Decision Analysis and mathematical programming are most common tools to model such problems. When selection process takes place within a group, preferences of multiple decision makers (DMs) are not unique and negotiations should take place to find a needed balance between different points of view. In present work we use a version of the Iterative Trichotomic Approach (ITA) adjusted to group decision making with the focus on convergence. Group-ITA provides a possibility to draw conclusions about the consensus over each individual project as well as on the final portfolio. The basic idea is a classification of projects into three sets: the green projects (selected by all decision makers in the consensus portfolio), the red projects (rejected by all decision makers from the consensus portfolio) and the grey projects which are selected by some (but not all) decision makers. Then a mathematical model is developed, where preferences of decision makers are incorporated and a process of step-by-step convergence of these preferences takes place. As the iterative process moves from one round to the next one, green and red sets are enriched while the grey set shrinks. The emptiness of the grey set means end of calculations. Final outcome is the consensus portfolio of projects, as well as the degree of consensus on each project and the consensus index for the whole portfolio according to the convergence path. The Consensus Index expresses the easiness to arrive at a final conclusion within a group. The more green projects we have from early rounds the greater is the degree of concordance among DMs. On the contrary, if the majority of green projects is identified on last rounds, it means the need to elaborate in the convergence process in order to agree at selected projects. In other words, the consensus is hardly attained. Besides the Consensus Index, we can extract the degree of consensus for each project according to the round that it enters or exits the final portfolio. The method is illustrated with an example based on real data for renewable energy projects. Keywords: Project Portfolio Selection, Multiple Criteria, Integer Programming, Group Decision Making, Consensus *corresponding author, opechak@hotmail.com 122

124 1. Introduction In current research, project portfolio selection is defined as problem of selecting one or a subset from a wider set of proposals, in other words a subset of projects is considered as a portfolio of projects. In past, the usual approach was to rank projects using one or more criteria and choose the top ranked ones that cumulatively satisfy a budget limitation. However, the existence of various limitations to be satisfied by the final selection destroys the independence of projects, which is one of main assumptions in Multiple Criteria Decision Analysis (MCDA) ranking (see e.g. Belton and Stewart (2002)). Specifically, the top ranked projects may only by chance satisfy imposed constraints. On the other hand, Integer Programming (IP) is an appropriate tool for such combinatorial problems where 0-1 (binary) variables express incorporation (Xi=1) or exclusion (Xi=0) of i th project in/from the portfolio. Involvement of several decision makers leads to even more complicated selection process. In an attempt to address the selection problem within a group, we modify and test the Iterative Trichotomic Approach (ITA) under the title Group-ITA. The ITA method has been originally developed for project portfolio selection in order to deal with uncertainty either in problem s parameters (Mavrotas and Pechak, 2013a) or in preference parameters (Mavrotas and Pechak, 2013b). 2. The ITA Method The core idea of ITA method is a separation of initial set into three parts (trichotomy) according to the projects membership in the final portfolio, given the current level of information. The selection is performed iteratively in computation rounds (R), until the process converges to a final set. A predetermined number of these rounds may be initially set and every round feeds its subsequent until the final selection is built. According to ITA the initial set of candidate projects is divided into three subsets (classes): the green projects that are present in final portfolio under all circumstances, the red projects that are absent from final portfolio under all circumstances, and the grey projects that are present in some final portfolios. From round to round the grey set is reduced as a result of convergence or uncertainty reduction. Such process flow helps the Decision Maker (DM) to identify and focus on projects that are really at stake. The sure projects (green and red sets) are determined and the attention can be shifted towards ambiguous projects (the grey set). Similar approach of splitting initial set of projects on 3 subsets has been proposed by Liesiö et al. (2008), however, the membership principles were different. Our method provides quantitative and qualitative information that cannot be acquired using e.g. just expected values of distributions. In the latter case, the DM is provided only with final optimal portfolio where the information about certainty degree of selected projects is missing. For the group project portfolio selection we adopt the combination of Multiple Criteria Decision Analysis and Integer programming (MCDA IP). At the beginning we use MCDA to assign scores to projects and then feed these evaluations in objective function as coefficients in the IP model that incorporates constraints of the project selection problem. In presence of multiple DMs we assume that each DM expresses his preferences by assigning his own weights 123

125 to the criteria of project evaluation. Therefore, each DM calculates his own multi-criteria scores for projects. In general this means that he usually has an objective function that differs from the others. As a consequence, obtained optimal portfolios are usually different. In the so called Group-ITA the membership of each project in the green, red or grey sets is determined according to the concordance of decision makers. Namely, the green set includes projects that are present in the final portfolio according to all DMs, the red set those ones that are absent from the final portfolio according to all participants, and the grey set includes projects that are present in the final portfolios according to some group members. Assume that there are N projects, P DMs and K criteria of evaluation. Therefore the weight of importance that p-th DM assigns to k-th criterion is wpk with p=1..p and k=1..k. For each group member we calculate multi-criteria scores mspi for every project i=1..n. The objective function of the IP problem for the p-th DM is then: max N i1 ms pi X i where Xi is the binary variable that indicates if the i-th project is selected (Xi=1) or rejected (Xi=0). Solving the formulated P integer programming problems we obtain at most P different optimal portfolios (some of them may be identical). Subsequently, the members of the green, red and grey sets are identified. Items of green set are those projects present in all P optimal portfolios (green projects). Accordingly, within red set are projects that are absent from all P optimal portfolios (red projects) and the grey projects (i.e. members of grey set) are those that are present in some of P optimal portfolios. 3. Case study Group-ITA is applied in a group decision making problem dealing with energy projects. There are 133 energy projects from three RES technologies (wind - W, small hydro - SH, photovoltaic - PV) distributed across 13 regions of Greece. The 5 criteria for evaluation are Regional development, Employment, Economic Performance (expressed with IRR), CO2 emission reduction and Land use. All but the 5 th objective are to be maximized. The data for this problem are available in Makrivelios (2011). There are also specific policy constraints for the project portfolio selection problem that must be respected, namely: Available budget is 150 M (the total cost of the 133 projects is 659 M ), Cost of projects in Central Greece should be less than 30% of the total cost, Cost of projects in Peloponnese should be less than 15% of the total cost, Cost of projects in East & West Macedonia, Northern & Southern Aegean, Epirus should be greater than 10% of the total cost, Number of projects by technology should be between 20% and 60% of selected projects, Total capacity of selected projects should be greater than 300 MW. 124

126 % of green projects in the final portfolio Book of Proceedings Cost (<=150 M ) W (20%- 60%) SH (20%- 60%) PV (20%- 60%) MW STE PEL Other projects (>=300) (<=30%)(<=15%)(>=10%) round % 3.2% 22.5% 15.1% 35.6% 49.3% round % 9.2% 21.1% 16.2% 35.1% 48.6% round % 8.2% 20.8% 16.9% 33.8% 49.4% round % 7.9% 20.0% 17.9% 33.3% 48.7% round % 10.3% 25.6% 20.5% 32.5% 47.0% Table 1. Characteristics of consensus portfolio (i.e. green projects only). The results and particular characteristics of the portfolio created by green projects in each round (consensus portfolio) are shown in Table 1. By studying it decision makers may decide to select a consensus portfolio prematurely, i.e. before arriving to Round 10, if they accept the respective constraints violations, which are denoted with bold fonts. Further, we develop a measure of consensus for the final portfolio according to the degree of concordance among group members, which actually expresses the easiness of convergence to final consensus. In order to calculate a consensus index (CI) we draw the so called consensus chart where the percentages of green projects that have already been found in r th round are plotted as a function of respective decision round. The resulting curve is called consensus curve. In Figure 1 observe that until round 3 there are no new projects added in the green set. This may happen especially when the maximum number of rounds (R) is relatively high. In addition, the DM is aware of projects prioritization given that he knows in which round a project enters the green set. 100% α 10 98% 96% 94% α 6 α 7 α 8 α 9 92% 90% α α 5 4 α 0 α 1 α 2 α 3 88% 86% 84% 82% 80% Rounds a a a a a a CI a a CI a R R1 R (... ) / R1 0 R [ r ] / 2 r1 2 a 1 CI a R R1 0 [ r ] / 2 r1 2 R Fig. 1. Consensus chart for the application. CI takes values between 0 and 1 and is calculated using the trapezoid rule for piecewise linear functions according to the following equations where ar is the percentage of green projects that have already been found in r th round: CI [ ]/10 91%

127 4. Conclusions Project portfolio selection is a complex problem that usually involves numerous criteria and constraints (budget, policy, allocation etc) that should be satisfied. Moreover, multiple decision makers from different positions, with different backgrounds and usually with conflicting views participate in the selection process. In present paper we propose a systematic procedure towards convergence of different points of view. For this method we assume that the preferences of group members are expressed with a set of weights used in the MCDA for overall project evaluation. The cornerstone of the iterative process is the systematic way for weights convergence that guarantees the convergence of the whole process to a final portfolio. The outcome of Group-ITA is not merely the final portfolio, but also the course towards it that may provide fruitful information about the project selection problem and may be used to reconsider some of the initial assumptions. After all calculation we not only converge to the final portfolio, but also measure the degree of consensus for each project that is selected or rejected. Moreover, it provides a measure of consensus for the final portfolio as a whole. References Belton, V., Stewart, T.J. Multiple Criteria Decision Analysis: An Integrated Approach. Kluwer Academic Publishers, Boston Liesiö J., Mild P., Salo A. Robust portfolio modeling with incomplete cost information and project interdependencies. European Journal Operations Research, Vol. 190, 2008, pp Makrivelios E. Multi-criteria evaluation of projects based on renewable energy sources in Greece. Master thesis in Management of Industrial Systems, joint program of National technical University of Athens and University of Piraeus, Greece, Mavrotas G., Pechak O. (a) The trichotomic approach for dealing with uncertainty in project portfolio selection: Combining MCDA, mathematical programming and Monte Carlo simulation. International Journal of Multicriteria Decision Making, Vol. 3 (1), 2013, pp Mavrotas G., Pechak O. (b) Combining Mathematical Programming and Monte Carlo simulation to deal with uncertainty in energy project portfolio selection. In: F. Cavallaro (Ed), Assessment and Simulation Tools for Sustainable Energy Systems, Springer-Verlag, London 2013, pp

128 F.W. Lanchester s combat model application in a supply chain in a duopoly Miltiadis Chalikias Applied Economic Statistics and Operations Research Laboratory, Department of Business Administration, School of Business and Economics, Technological Education Institute of Piraeus, 250 Thivon & P. Ralli, 12244, Egaleo, Greece, mchalikias@hotmail.com Michalis Skordoulis Management Information Systems and New Technologies Laboratory, Department of Business Administration, School of Business and Economics, Technological Education Institute of Piraeus, 250 Thivon & P. Ralli, 12244, Egaleo, Greece mskordoulis@gmail.com Abstract The purpose of this study is to investigate the possibility of applying some of the most widely known mathematical theories of war in firms. In this research, Frederick William Lanchester s combat models were examined that seemed to be particularly useful in the U.S. Army at the Pacific campaign against the Japanese fleet during World War II. These mathematical models were based on differential equations and its main purpose was to predict the outcome of battles. Keywords: Frederick William Lanchester, mathematical theories of war, differential equations, supply chain in duopoly. 1. Introduction Frederick William Lanchester was born in 1868 in London and studied engineering (Ricardo, 1948). In 1916, he invented the operations strategy for the Royal Air Force of England, formulating model based on differential equations, the two models estimate the forces that are required for winning in a military battle (Bracken, 1995). In the first Lanchester combat model, it is considered that two forces with the same martial ability, R(t) and G(t), initiate a military conflict between them at the time (t). R(t) neutralizes g number of soldiers, and G(t) neutralizes r number of soldiers respectively (MacKay,2006). The numbers g and r, are called efficiency ratios of forces R and G respectively (Daras, 2001). We have an initial situation where applies the following system: dg gg dt dr (1) rr dt Apart from the first case, there exists the more complex mathematical case, Lanchester s second model, where two forces participate at the military conflict, one of which has greater military capacity than the other, creating the so-called asymmetric warfare (Lanchester, 1956). 127

129 As far as business administration is concerned, Lanchester s combat models have been applied in several cases. A new approach to Lanchester s combat models, was applied for the first time in Japan by Taoka and Yano in marketing strategy (Oudrhiri, 2005). 2. Construction of the mathematical model We consider two competing firms A and B which coexist in a common market and sell the product P. This market is characterized as an oligopoly, as we deem that there aren't other similar firms, while the input of such firms is very difficult. Because in this oligopoly we consider the existence of only two firms, we are eventually also led to the special form of duopoly just as in the case of the implementation of Richardson s arms race model (Chalikias & Skordoulis, 2014). It is, at first phase, assumed that the technology that is used by the two firms is the same. The firms know each other's moves. Let x(t) to be the number of available product units for sale of firm A and y (t), the number of available product units for sale of firm B at time t. During the operation of firms A and B that are competitive with each other, the rate of change of the quantities x(t) and y(t) equals the rate of growth of refueling at the distribution points, minus the rate of their reduction. The rate at which the available for-sale product units are increasing and decreasing is denoted by f(t) for A and g(t) for B respectively. The rate of available for sale product units for firm A equals ay(t) and for firm B with bx(t), where a and b are appropriate positive constants. As in the case of the two warring conventional forces that is analyzed by Daras (2001), the mathematical model that is based on Lanchester's combat models and describes the above situation is the following: dx ay f(t) dt dy (2) bx g(t) dt Any solution of the system of differential equations (2) for x 0 0 and y 0 0 will be given by the formulas: x x x x a e e e e x(t) y 0 ( abt) 2 ab(t s) f(s)ds b 2 t 0 (3) x x x x b e e e e y(t) x 0 ( abt) 2 ab(t s) g(s)ds a 2 t 0 (4) The mathematical model may be applied on several examples of supply chains in duopoly. It is proposed that this model is applied on the market of cola type drinks in Greece. The competition between Coca-Cola and Pepsi has motivated the interest of many researchers who have studied and implemented models like this. Chintagunta & Vilcassim (1992), used Lanchester s combat models in order to examine the effects of advertising expenditure on consumer demand for Coca- Cola and Pepsi in the level of competition between these two firms. A similar model using statistical data on advertising expenditure of Coca-Cola and Pepsi in order to analyze advertising strategies that are used by them was also applied by Erickson (1992). In the same context, Wang & Wu (2001) resulted in the fact that consumers respond to the advertisements of Coca-Cola and Pepsi in the same way. Finally, using Lanchester s combat models, Fruchter & Calish (1997) 128

130 described the competition between Coca-Cola and Pepsi and solved the problem of determining the optimum advertising strategy for maximum profits. In their entirety all the researches have not taken into account the possible influence of other firms that sell cola type products. As far as the previous literature analysis is concerned, it would be very interesting that the constructed model will be applied in the duopoly of Coca-Cola and Pepsi in the Greek market. 3. Conclusion From the above research, it was concluded that the mathematical model that was constructed based on Lanchester s combat models may be applied in the case of duopoly that was examined. In a next step, it is proposed the construction of similar models that take into account several factors such as the price and quality of products that may affect consumer preferences and ultimately the firms production. References Bracken, J. (1995). Lanchester models of the Ardennes campaign. Naval Research Logistics. 42(4): Chalikias, M. & Skordoulis, M. (2014). Implementation of Richardson s arms race model in advertising expenditure of two competitive firms. Applied Mathematical Sciences. 8(81): Chintagunta, P. & Vilcassim, N. (1992). An empirical investigation of advertising strategies in a dynamic duopoly. Management Science. 38(9): Erickson, G. (1992). Empirical analysis of closed-loop duopoly advertising strategies. Management Science. 38(12): Fehlmann, T. (2008). New Lanchester theory for requirements prioritization. In: Proceedings of the Second International Workshop on Software Product Management. Barcelona, September Bacelona: I.E.E.E, pp Fruchter, G. & Kalish, S. (1997). Closed-loop advertising strategies in a duopoly. Management Science. 43(1): Lanchester, F.W. (1956). Mathematics in warfare. The World of Mathematics. 4: McKay, N. (2006). Lanchester combat models. Math Today. 42: Oudrhiri, R. (2005). Six Sigma and DFSS for IT and Software Engineering. The Quarterly Journal of the TickIT Software Quality Certification Scheme. 4: 7-9. Ricardo, H. (1948). Frederick William Lanchester. Obituary Notices of Fellows of the Royal Society. 5(16):

131 Wang, Q. & Wu, Z. (2001). A duopolistic model of dynamic competitive advertising. European Journal of Operational Research. 128(1):

132 An optimization modeling approach for the establishment of a bike-sharing network using Monte Carlo Simulation and stochastic demand: a case-study of the city of Athens Zygouri E. Department of Mechanical Engineering, School of Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece, el.zigouri.haf@gmail.com Fragkogios A. Department of Mechanical Engineering, School of Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece, fragkogiosantonios@gmail.com Saharidis G.K.D. Department of Mechanical Engineering, School of Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece and Kathikas Institute of Research & Technology, Paphos, Cyprus saharidis@gmail.com Mavrotas G. Department of Process Analysis and Plant Design, School of Chemical Engineering, National Technical University of Athens, Zographos Campus, Athens, Greece, mavrotas@chemeng.ntua.gr Abstract This study introduces a novel mathematical formulation that addresses the strategic design of a bicycle sharing network with stochastic estimated demand. The developed pure integer linear program takes into consideration the available budget of a city for such a network and optimizes the location of bike stations, the number of their parking slots and the the bicycle fleet needed in order to meet as much demand as possible and to offer the best services to the users. The methodology used for the simulation of demand is the Monte Carlo simulation, which is combined with the Iterative Trichotomic Approach (ITA). The proposed method is implemented on the very center of the city of Athens, Greece. Keywords: Bike, Sharing System, Monte Carlo, Integer 131

133 1. Introduction Bike-sharing networks have received increasing attention during the last decades and especially in the 21st century as a no-emission option in order to improve the first/last mile connection to other modes of transportation, thus facilitating the mobility in a densely populated city. The bike-sharing network consists of docking stations, bicycles and information technology (IT) interfaces that have been recently introduced to improve the quality offered to the users. This expanding trend of bike-sharing networks necessitates their better planning and design in order that they are successful. The goal of this paper is to propose a novel mathematical formulation to design such networks incorporating stochastic estimated demand, the fixed costs of infrastructure, the proximity and density of stations, as well as their size. Given a set of candidate locations of stations and with a predefined available construction budget the model decides the number and the location of the stations, how large they will be and how many bikes should they have in order to meet the assumed demand. 2. Model Formulation Given a set of candidate locations of bike stations and the value of demand for bikes at these locations it is necessary to know where to place the bike stations and how many parking slots and bikes should each one have. The available budget of a city for the construction of the whole bikesharing system is predefined and so are the costs of a single bike, a single parking slot and a single station. So it is a matter of optimization for the model to decide how many stations, bikes and parking slots it will include in its solution. The walking time between the locations is another parameter of the problem used to ensure the proximity of the constructed stations as far as this is possible. As regards demand in each location, it is split into Demand for Pick-Ups and Demand for Drop- Offs. The first one depicts how many users would like to take a bike from a station and the second one shows how many riders would like to leave a bike at a station. In Figure 1 the thorough consideration of the problem is explained. N locations i are predefined together with their Demand for Pick-Ups and Demand for Drop-Offs. The walking time between these N locations is, also, known. It is a matter of optimization how many bike stations will be established and where, so that every location has a nearby station. The locations k, where stations are established, is a subset of the locations i. Figure 10 Network structure of bike-sharing system. 132

134 If budget is not enough to construct stations at all N locations, at some locations i there will be no station. These locations should have a nearby station k no more than a specific walking time away and only a percentage of their demand is considered to be passed to this station k. The rest of their demand is not served supposing that this part of citizens will not take a bike due to the distance of the station k from their location i. In this way, it is assumed that location i is served by station k. On the one hand, this transfer of the demand is inevitable as the restricted budget does not allow stations to be built at all locations. On the other hand, it is not desirable because it means that the users of the network will have to walk from location i, where they would rather a station to be present, to the established station k and vice versa. This would result into poor service quality offered to the users of the bike-sharing network, as some potential customers will not eventually use the network. This consideration is accomplished through the following objective function. The objective function of the model is a minimization of three terms. The first term expresses the amount of demand that is transferred from a location i to its allocated station k, which are a specific walking time away from one another. Thus, the model will propose a dense distribution of stations, establishing no station at locations with low demand ensuring that they are as close to a station as possible. This term is multiplied by the penalty unit cost to differ its importance from the other two terms. The second and the third term of the objective function are introduced in order to minimize the unmet demand. There is a difference between the parameters that express the Demand for Pick-Ups from location i during time interval t and the Demand for Drop-Offs at location i during time interval t and the variables that express the number of bicycles that are available at station k at the beginning of time interval t and the number of bicycles that could leave station k during time interval t. The former express the users who would like to pick up and drop off a bike from and to a candidate station location respectively. However, the station k may not have the required bikes or free parking slots to meet these two types of demand respectively. So the number of bikes that eventually leave or arrive at a station k is expressed by the two mentioned variables. Both the parameters and the variables refer to each time interval t. These two terms are multiplied by the same penalty unit cost meaning that no different importance is given to either of them. In the mathematical model there is a constraint, which warrants that the total available budget is not exceeded. Other constraints ensure that the bicycle parking slots at each constructed station are between the permissible minimum and maximum value and that each station cannot have more bikes than the number of its parking slots. Some other constraints guarantee that a location i cannot be served by location k, if a station is not built in location k, also, that if a station is constructed at location k this location will be served by its own station. Furthermore, that each location i may be served by exactly one bike station k and that a constructed station k can serve only locations which are located within a maximum walking time from it. Finally, among others there are some constraints, which guarantee that the bicycles that can leave the station can be no more than the available ones and the bikes that can come to a station can be no more than the free parking slots. 3. Monte Sarlo Simulation and Iterative Trichotomic Approach In order to find the optimal solution the Iterative Trichotomic Approach is applied, which consists of decision rounds. Each of these rounds includes a series of 1000 Monte Carlo simulations - IP 133

135 optimizations, from which 1000 optimal solutions arise. In each of the simulations values for every demand of every candidate location are randomly generated within the probability distributions that are estimated to describe them. In each decision round the candidate locations are categorised into green, red and grey subsets, depending on how many times they are included in the optimal solution for the establishment of a bicycle sharing station. Specifically, if a candidate location is selected as a station in more than 990 iterations of a decision round, it is allocated to the green subset. Respectively, if a candidate location is selected as a station in less than 10 iterations of a decision round, it is allocated to the red subset. If a candidate location is selected as a station in more than 10, but less than 990, iterations of a decision round, it is allocated to the grey subset. At the end of each decision round, the green and red locations are fixed with constraints in order to be selected and not selected respectively for the establishment of bike stations and the variance of the demand of the grey locations is reduced, assuming that their demand becomes less stochastic (more information is gained for these stations). The Iterative Trichotomic Approach is completed when all candidate locations are categorised into the green and red subsets and no grey locations exist, as depicted at Figure 2. Figure 2 The Iterative Trichotomic Approach Of particular importance is the method used to estimate the stochastic demand characteristics of each candidate location. Originally, the hourly usage of an already established Bike Sharing Network, Velib in Paris, is used as a basis. The candidate locations are categorized into clusters, whose demand is multiplied by standard rates adjusted on the center of Athens. However, each location is evaluated for 11 parameters that affect the expected demand for shared bike rental. These 11 parameters are Population Density, Universities (facilities, student dormitories, etc.), Jobs Density, Retail Jobs Density, Tourist Attractions, Parks and Recreation areas, Metro, train stations, Buses, trolleys stations, Bicycle lanes, Bicycle friendly roads and Topography. Thus, the upper and lower limits of the probability distributions of every demand of each location are obtained. These distributions are used as the basis of the stochastic Monte Carlo simulations. 134

136 4. Athens Case-Study 4.1 Data Settings The authors chose 50 candidate locations in the 1 st Municipality District of Athens, where bikesharing stations could be constructed. These candidate locations were categorized into four clusters ( Housing, Employment, Subway and Spare Time ) depending on their location. The stochastic demand is assumed to follow the uniform distribution. The upper and lower limits of the probability distributions of every demand of each candidate location are obtained through the evaluation of the location on the 11 preciously mentioned parameters. The walking time between these locations was calculated using Google Earth. As regards the costs of the network, two already implemented networks were taken into account, the first one in Greece (Karditsa) and the second one in Cyprus (Nicosia). Examining the budget and the dimensions of each city and its network the following data were assumed for the case of Athens. The cost of establishing a station is 12,000. The cost of each slot in a station is 900. The cost of a bike is 500 and the total available budget is 1,000,000. Furthermore, it is assumed that a location with no station cannot stand off a location with a station within more than 7 minutes of walking time. The minimum and the maximum parking slots that a station can have are as many as in the Velib network (between 8 and 70 per station). 4.2 Results Decision rounds of the Iterative Trichotomic Approach were executed, in each of which 1000 Monte Carlo simulations were made. In each decision round the candidate locations are categorised into in the green, red and grey subsets, as mentioned above. The convergence is obtained after four decision rounds. The above decision making process determined the locations where bike stations will be established. However, when designing a Bike Sharing Network, it is necessary to determine the number of parking slots of each station and the number of bikes in the whole network. The methodology followed for this calculation each station is as follows: Initially, having added constraints for establishing stations on the green and not the red locations, another round of 1000 Monte Carlo iterations was performed, where bikes demand was stochastic for all sites (uniform distribution with the largest range of values as in the 1st decision round). The 1000 iterations with stochastic demand led to a distribution of parking slots for each station. Afterwards, the less probable values of slots at each station were excluded and the reduced range that remained are included in the mathematical model as constraints. Finally, the mathematical model is solved for the last time and the demand of every location equals the centre value of its uniform distribution. Thus, one optimal solution is gained with the parking slots and bikes of every station. Figure 3 depicts the proposed established bike stations. The shape of each dot corresponds to the station s cluster, while its size represents the number of parking slots each station should have. The total number of docking stations is 38 and the number of parking slots is 470 making a mean value of 470/38=12.4 slots per station. Looking at the parking slots of each station, one can notice 135

137 that the larger stations belong to the cluster Subway, which is typical of the increased demand in the metro stations. The total number of bikes in the network is 242. Of particular importance is the knowledge of the decision round, in which each location is included in the green or red subset. The sooner a location enters the green or red subset, i.e. the sooner her condition ("go" or "no go") is concluded the larger the degree of certainty of that decision. This information is very useful for decision-makers given the uncertainty of demand of each location. Housing Employment Subway Spare Time Figure 3: The established stations of the solution of the case of Athens categorized in clusters and with their size 5. Conclusions Solving the proposed mathematical model many times with different values of stochastic demand each time, can lead to the optimal design of a Bike Sharing Network so as to meet as much demand as possible during its usage afterwards. The knowledge gained from the already implemented networks can and should be used for the design of future ones. This model reclaims the usage data from the Velib network of Paris to predict demand in Athens and designs a suitable bike-sharing network to meet that demand. However, some parameters could be altered to notice how the solution changes. Such parameters could be the available budget or even the demand profiles to approximate the seasonal differences (winter-summer) or the week differences (weekdays-weekend). The larger application of the model is, finally, another work to be done concerning, for example, the whole Municipality of Athens. 136

138 References DeMaio, P., 2009, Bike-sharing: History, Impacts, Models of Provision, and Future, Journal of Public Transportation, Vol. 12, No 4, pp Mavrotas, G. and Pechak, O., 2013, The trichotomic approach for dealing with uncertainty in project portfolio selection: combining MCDA, mathematical programming and Monte Carlo simulation, Int. J. Multicriteria Decision Making, Vol. 3, No. 1, pp Lin, J-R., T-H. Yang., 2011, Strategic design of public bicycle sharing systems with service level constraints, Transportation Research Part E, Vol. 47, pp Martinez, M. L., L. Caetano, T. Eiro, F. Cruz., 2012, An optimization algorithm to establish the location of stations of a mixed fleet biking system: an application to the city of Lisbon, Procedia- Social and Behavioral Sciences, Vol. 54, pp Krykewycz G. R., Puchalsky C. M., Rocks J., Bonnette B., & Jaskiewicz F., 2010, Defining a Primary Market and Estimating Demand for Major Bicycle-Sharing Program in Philadelphia, Pennsylvania, Transportation Research Record, Lathia N., S. Ahmed, L. Capra., 2011, Measuring the impact of opening the London shared bicycle scheme to casual users, Transportation Research Part C, Vol. 22, pp Etienne C., L. Oukhellou., 2012, Model-based count series clustering for Bike-sharing system usage mining, a case study with the Velib system of Paris, Transportation Research-Part C Emerging Technologies, Vol. 22, pp

139 Evaluating new service development effectiveness in tourism: An ordinal regression analysis approach Fotis Kitsios School of Information Sciences, Department of Applied Informatics, University of Macedonia, P.O. Box 1591, GR54006 Thesaloniki, Greece, Evangelos Grigoroudis School of Production Engineering and Management, Technical University of Crete, University Campus, GR73100 Chania, Greece. Abstract Innovation development is an important factor for the viability and profitability of service businesses operating in modern markets. The importance of the service sector in developing economies and the specific characteristics of services compared to tangible products require further investigation in the New Service Development (NSD) process and effectiveness assessment when innovations applied. The innovation development process will be significantly improved and the contribution of innovations in company s goals will be substantial. The purpose of this study is to evaluate the effectiveness of the NSD process in the tourism economy and in particular the Greek hotel sector. For this purpose, factors influencing the process of developing new services in the hospitality sector were explored and correlated with the financial results of the hotel enterprises through an ordinal regression analysis model. The model adopts a mathematical programming approach in order to estimate the efficiency of this process. In the presented study the Greek tourism industry and its importance to the national economy is discussed. The study explores in detail the factors influencing the NSD process. The questionnaire developed for the purpose of the survey included a large number of variables related to all the stages of the NSD process (from idea generation till the service launch). All variables are measured in a 5point Likert type scale and data was collected using in depth structured and questionnaire-based interviews with 77 hotel managers for 147 new services in a representative sample of 99 hotels in Greece. Several financial ratios covering different aspects of the business (e.g., profitability, liquidity, activity) are used in order to evaluate the NSD process for three years after the services innovation had been launched. The main results of the ordinal regression model include the estimated contribution of each factor to the financial performance of the hotels studied. Keywords: New Service Development, Efficiency Evaluation, Ordinal Regression Analysis, Business Performance, Service Innovation Strategies 138

140 1. Introduction The rapid changes in today's business environment such as globalization and competition have a direct impact on the conditions in which businesses operate. The evolution of management science, the development of technology, and the fact that customers over time becomes increasingly demanding, creating the need for companies to constantly seek new ways to differentiate what they offer to the market. They aimed to gain a competitive advantage in order to get profitable and sustainable. The ability to create innovative products and services is the key to sustainability and growth. The rapid development of technology, the management systems that ensure quality of products and services, the pretentiousness of consumers, the competition from non-traditional sources, such as the Internet, and the emerging hybrid industries, increase the need for business innovation. In recent years there have been many discussions and research to develop new products and services and to evaluate their effectiveness. In the field of products, new product development is accepted as requirement for business development and prosperity, and the effectiveness is measured and reflected in the company's turnover. However, new service development (NSD) is not so widely studied, given the particular characteristics of the service industry (e.g., differences between products and services, deficient knowledge available in NSD). The purpose of this paper is to evaluate the effectiveness of the NSD process in the Greek hospitality industry. For this purpose an exploration of the factors influencing the process of developing new services in the hospitality sector is conducted and an ordinal regression analysis linking these factors with the financial results of hotels is applied. 2. Literature Review The development of new services and the orientation towards innovation are components of success for modern enterprises. In this context, Dolfsma (2004) emphasized that service firms are more profitable when they are innovative. In addition, a service business innovator may also have high performance in non-financial assets, such as reputation, trust building, and good relations establishment with existing and new customers (Avlonitis et al., 2001). A strong correlation between innovation strategy and financial outcomes has been found by Zahra and Covin (1994), who suggested that companies should avoid investing in innovations that do not fit with the strategic goals of the business. Moreover, they found that the relation between financial performance and different types of innovation may vary. The relation between innovation and business performance may be studied through several variables, like patents acquired by enterprises, innovations in processes, and particularly investment in R&D departments (Nås and Leppãlahti, 1997). All previous research efforts emphasize that innovation is always associated with the business profile, the philosophy adopted, as well as the sector and the size of the organization. The evaluation of business performance can be based in different datasets. For example, some variables that measure the profitability of a business-oriented innovation are: efficiency, growth, profit, liquidity, success / failure, and market share. 139

141 Generally, all studies report a significant correlation between innovation and business efficiency. However, the adoption of innovations may result to business competitiveness, only if the company can defend itself in the market against competitors. In addition, innovations may lead to better flexibility, which is an extremely important advantage in market conditions with strong competition. Numerous studies show that innovative firms may have improved financial (assets, sales, exports, etc.) or business performance results (growth, number of employees, etc.) (see for example Thwaites and Wynarczyk, 1996; North and Smallbone, 2000). However, all these researches focus on the relationship between innovation and business performance, without measuring the efficiency of adopted innovations, especially in the service sector. Furthermore, most studies are cross-sectoral, and thus presented findings cannot be easily extrapolated to the service sector. This justifies the necessity of conducting more sector-oriented studies in order to have comparable and representative results (Kitsios, 2005). 3. Methodology a. Data and Variables As far as the Greek hotel sector is concerned, a first and thorough approach was conducted in 2005, aiming to record and comprehend the decision-making process followed by hotel managers (Kitsios, 2005). Collecting determinant factors of success in NSD defined in Greek and international literature, this research formed a 126 factor questionnaire that was applied in interviews with 99 Greek hotels of a wide geographical range. The study used a Likert-type data collection process and applied several statistical analysis methods. The initial large set of factors reduced in 24 determinant new factors, 6 statistically significant. These factors have been included in a predictive model which may be used as a guide by the hotel managers. Based on the aforementioned framework, this study analyzes the efficiency of NSD in the hotel performance. For this reason, two sets of variables are used: Drivers: These variables are based on the NSD process and can be considered as the causes of financial results. As shown in Table 11, the 24 variables used in the study are categorized into 6 main groups: 1) Enterprise s behavior for the service innovation, (2) Idea generation sources for the provided service, (3) Actions for developing the provided service, (4) Organizational structure impact, (5) Enterprise s resources allocation impact, and (6) Market impact (see details in Kitsios, 2005; Kitsios et al., 2009) Outcomes: These variables are based on the financial balance sheets of the examined hotels. A total of 8 financial ratios are used in this study, covering profitability, turnover, efficiency, as well as solvency ratios (Table 12). The final dataset of the presented study consists of 77 hotels and a total of 153 new service projects, both successes and failures. Data were collected by direct in depth interviews with the hotel managers. 140

142 b. Ordinal Regression Analysis The applied model is an ordinal regression analysis approach, assuming a set of explanatory and a set of response variables. In particular, given a set of result variables Y and a set of driver variables X i, the main principle of this approach is that the weighted average of the values of Y j can be written as a weighted average of the values of X i, according to the formula: j Dimension Variables Dimension Variables Strategic Focus 1. The strategic objectives had been clearly identified Integrated Market Launch 15. An integrated promotion plan had been implemented (e.g. brochures, advertisements, direct sales, exhibitions, conferences and seminars for clients) 2. Expression of objectives as contribution to the income of the company 3. The areas of strategic focus were clearly identified After Launch Review and Assessment 16. The measurements and forecasts had been successful for the performance of the new service 17. The advertising, promotion and communication efforts were targeted to the right customer segment 4. The strategic action plans were clearly identified Market Potentiality 18. Previous knowledge of the potential market size Idea Generation 5. There was a mechanism and a systematic effort to capture and collect new ideas for development Market Synergy 19. An analysis of how the product meets customers needs was conducted as opposed to competing products Preliminary Market Assessment 6. Preliminary market assessment had been undertaken prior to any major investment 20. The service was aligned with the overall image of the hotel 7. Enough time and money were spent on preliminary market assessment 21. The potential needs of customers were appreciated in the commercialization stage of the new service 8. A clear and focused definition of the target market was given during the preliminary market assessment 22. The customers purchase decision process and behavior was clearly perceived by the hotel Operational Analysis 9. A realistic business analysis had been carried out 23. There were strong support for the new product after its launch 10. A comprehensive analysis of the competition had been carried out 24. Potential customers had showed a great need for this class of product 11. Forecasts of expenses and sales had been conducted 12. Discount cash flow analysis 141

143 13. Breakeven and return on investment analysis 14. Informal analysis (guesses and estimates) Table 11 NSD variables (drivers of efficiency) Dimension Profitability Variable 1. Gross profit margin (Gross profit/sales) 2. Net profit margin (Net income/sales) Turnover 3. Asset turnover (Sales/Total Assets) 4. Equity turnover (Sales/Equity) 5. Liability turnover (Sales/Liabilities) Efficiency 6. ROE- Return On Equity (Net income/equity) 7. ROA - Return On Assets (Net income/total Assets) Solvency 8. Solvency ratio (Equity/Total Liabilities) Table 12 Financial variables (efficiency outcomes) where m n wj bi 1 j1 i1 m n * * wjyj bi X i j1 i1 * Y j and * X i are the value functions of Y j and X i, respectively, b i and coefficients, n and m are the number of factors. It should be noted that linear, monotone value functions, normalized in the interval [0, 1], thus: where *1 * j y j 0, y j 1 for j 1,2,..., m *1 * i xi 0, xi 1 for i 1,2,..., n *k y j and of functions *k x j are the values of the * Y j and k y j and * X i, respectively. The scales k x j level, j and j and * Y j and (16) w j are the weight * X i are piecewise (17) i are the number of scale levels i are defined by the analyst and, when necessary, linear interpolation is used in order to calculate in-between values. According to the previous assumptions, and using a goal programming approach, the ordinal regression analysis equation may take the following form (Wagner, 1959; Siskos, 1985): m n * * j j i i j1 i1 w Y b X (18) 142

144 where and are the overestimation and the underestimation error, respectively. The parameters of the model may be estimated using linear programming techniques, where the objective is to minimize the sum of errors, subject to the following constraints: Equation (18) for each case (in our application for each new service project) Monotonicity constraints of * Y j and Normalization constraints of b i and * X i. w j. Non-negative constraints for all the variables of the model. This model is similar to canonical correlation analysis, under the philosophy of ordinal regression. A detailed presentation of ordinal regression analysis principles, including a discussion about model stability may be found in Grigoroudis and Siskos (2002, 2010). c. Results The most important results of the ordinal regression model refer to the estimation of importance ( * b i and w j ) and performance (average of Y j and X ) of the examined variables. These results are normalized in [0, 1] and thus it easy to identify the strong and weak points of the innovation drivers and results. It should be noted that the ordinal regression model has been applied in three different periods (in 2004, the year that innovation has been developed, and in 2005 and 2006 after one or two years), in order to examine potential hysteresis in the relation between the variables. Table 13 presents the weights and the average performance of the main NSD dimensions. As it can observed, operational analysis is the most important dimension in all years, it appears to have one of the lowest average performance indices. In general, the importance, as well as the performance, of these NSD factors appear unvaried the examined period. Similarly, the weights and the average performance indices of the financial variables are shown in Table 14. Regardless of the influence of the general economic environment, it seems that the most important impacts refer to the improved performance of profitability and turnover. Analytical results of the ordinal regression model may be found in Charalambous (2009). * i Dimension Weights (%) Performance (%) Strategic Focus Idea Generation Preliminary Market Assessment Operational Analysis Integrated Market Launch After Launch Review and Assessment Market Potentiality

145 Market Synergy Table 13 Weights and average performance for the drivers Dimension Weights (%) Performance (%) Gross profit margin Net profit margin Asset turnover Equity turnover Liability turnover ROE ROA Solvency ratio Table 14 Weights and average performance for the outcomes 4. Conclusions This study examined the linkage between drivers and outcomes of NSD applying an ordinal regression model. The results reveal specific factors that seem to play an important role in innovation efficiency performance. For example, operational analysis appear as the most important NSD dimension in all of the examined years. This category refers to one of the initial stages of NSD and can be characterized as a test of project feasibility and profitability. Market synergy is also an important dimension. It is not a particular stage of the NSD process, but represents the control of the market, targeting mainly to the consumer. Market synergy concerns the harmonization of new services with the market and the customer desires and needs. Another important dimension refers to strategic focus, which is the first stage of the NSD process, where the innovation strategy is formulated in agreement with business strategy and objectives. Its importance is justified by the relevant literature, since a company that identifies appropriate areas of interest, can set long term goals in the market. Regarding the financial variables, the most important ratios refers to equity turnover and solvency ratio, which appear very important in all of the examined years. Other important variables, although their weights may vary during the examined period, include: gross profit margin, asset turnover, liability turnover, and ROE. These findings reveal the importance of financial liquidity and managerial efficiency for the hotel industry (i.e., the ability of a firm to use available resources in order to achieve specific sale goals). The aforementioned variables can determine how quickly and effectively assets are converted to cash. In general, the findings show the emphasis that should be given on the one hand to the customer needs, and on the other to the effective management of a NSD project. 144

146 References Avlonitis, G., P. Papastahopoulou, and S. Gounaris. An empirically-based typology of product innovativeness for new financial services: success and failure scenarios. Journal of Product Innovation Management, Vol. 18 No. 5, 2001, pp Charalambous, M. Evaluation of innovation efficiency in tourism businesses. Diploma Thesis, Technical University of Crete, Chania, 2009 (in Greek). Dolfsma, W. The process of new service development: Issues of formalization and appropriability. International Journal of Innovation Management, Vol. 8 No. 3, 2004, pp Grigoroudis, E. and Y. Siskos. Preference disaggregation for measuring and analysing customer satisfaction: The MUSA method. European Journal of Operational Research, Vol. 143 No. 1, 2002, pp Grigoroudis, E. and Y. Siskos. Customer satisfaction evaluation: Methods for measuring and implementing service quality, Springer, New York, Kitsios, F. Innovation management in new service development, PhD Thesis, Technical University of Crete, Chania, 2005 (in Greek). Kitsios, F., M. Doumpos, E. Grigoroudis, and C. Zopounidis. Evaluation of new service development strategies using multicriteria analysis: Predicting the success of innovative hospitality services. Operational Research: An International Journal, Vol. 9 No. 1, 2009, pp Nås, S.O. and A. Leppãlahti. Innovation, firm profitability, and growth, STEP Report 1/97, The STEP Group, Oslo, North, D. and D. Smallbone D. The innovativeness and growth of rural SMEs during the 1990s. Regional Studies, Vol. 34 No. 2, 2000, pp Siskos, J. Analyses de régression et programmation linéaire. Révue de Statistique Appliquée, Vol. 23 No. 2, 1985, pp Thwaites, A. and P. Wynarczyk. The economic performance of innovative small firms in the South East region and elsewhere in the UK. Regional Studies, Vol. 30 No. 2, 1996, pp Wagner, H.M. Linear programming techniques for regression analysis. Journal of the American Statistical Association, Vol. 54, 1959, pp Zahra, S.A. and J.G. Covin. The financial implications of fit between competitive strategy and innovation types and sources. The Journal of High Technology Management Research, Vol. 5 No. 2, 1994, pp

147 New Technologies & Labor Market Nikolaidis K. of author: Abstract Today the new working relationships have major changes. The new forms such as telecommuting, workers by lease, part-time employment, the fourth shift give another dimension to the workplace. Telecommuting is any form of work that includes electronic data processing and the use of media for multiple/cross communication so that the employee can produce the work he was asked in an area outside the space where the business is located. There are alternative names for telecommuting in the relevant bibliography such as teleworking at home or distance working. Keywords TELECOMMUNICATION, TELEWORKING, Labour Market 1. INTRODUCTION What led to the development of telecommuting? This question can be answered by the development of IT and telecommunications. The new forms of communication open new possibilities in computing via high-speed transmission of the data with the VDSL at the speed of 50 mbps. At that speed we can have perfect image in HD and high quality sound. So the potential weakness in the communication of the past have been overcame. The globalization of the economy is another fact which has led to the development of telecommuting. In our day and time, the economy and consequently the firms operate globally, with the result that workers face a flexibility issue. Both the business forms of work have changed. The new trends that have appeared nowadays in the field of teleworking regarding the new forms of telework are as follows: The international literature identifies the following types of telework : Home Based Teleworking : Teleworking is made home-based (exclusively or on a regular basis). An area of the house converted into office with the proper equipment (computer, telephone, modem, fax and stationery). Satellite Centers : These centers are used by the employees of the same organization and are located in remote areas near the homes of the workers. 146

148 Telework Centers : They are well organized spaces with access to telecommunication and electronic equipment, in the form of offices used by the employees of different companies, or employees of the same company-who belong to different fields of work-or even by self-employed ones with some basic lease. Televillages : It is the modern form of telecottage. Entire villages are equipped with the appropriate technological apparatus, whose houses are wired to be able to communicate with each other and with other villages. Teamwork from distance: Some examples are as telemedicine, tele-education, e-commerce and research from a distance. 2. SECTION Ι The employer- employee working relationship may be based on a contract of dependent employment where the former has at his disposal the entire labor power of the worker, who is considered employed. In any other type of contract the worker is self-employed. Based on the above we have 3 types of employment relations. Full-time employment is perfomed at home and concems one employer by this term we mean that the work is done entirely at home and is not related to working hours. In part-time employment, work is carried out partly at home and the rest at the employer s premises. The self-employed type is more flexible as the employee works at home for more than one employers. In this case the worker has his own working model and determines his employment relationship. Another division regarding the forms of teleworking is related to the use of IT, and whether it is necessary for the worker or to be online offline. In the case online work the worker is online with the company and there is not enough freedom in the time and pace of their work, which means that he should abide by company s actual working hours. In offline work the employee has greater freedom and flexibility in the management of his work since he can link to the company s network only when necessary. So he can manage his time the way he thinks is the best. Whether the contact is online or offline is also a very important factor in the process of telecommuting. Both styles have positive and negative effects on this process. In online communication, the worker directly depends on the understanding of the presentation and discussions as they are conducted and by whether he takes good notes or has a good memory. 147

149 During the same period, the contributions of the project leader or keynote speaker as well as those of the participants are almost spontaneous. On the other hand, in offline communication, the distance worker has more time to think about his contribution and less pressure to respond immediately. Which form of communication is most appropriate depends on what kind of activities it will support. For example, offline communication is better suited for file transfer, information retrieval, etc., while realtime communication is very useful for the communication and discussion of specific implementation issues or employment problems. Thus, depending on the case, both forms of communication may be used by an entity that implements telecommuting procedures. Many different disciplines and fields are now ripe for the implementation of flexible working arrangements. The general factors that can serve as criteria for diversification and broad axes of direction to introduce teleworking schemes in business operations are generally working without personal contact, task management through profit results and tasks related to the management and electronic processing of data. These factors generally cover effectively the organizational and physical side of the work, but work as a social institution has a social dimension too. So when we analyze key factors for conducting business through flexible working arrangements we should also include on the analysis level those factors which are directly related to social interaction (such as sales, insurance). In general, the global literature indicates that the characteristics that make a job suitable for integration into telecommuting shapes are: 1. The ability to be handled without constant personal contact and interaction with other people. 2. The ability to organize the necessary social contacts on a periodic basis. The work that currently require daily meetings can be reoganised aiming at the integration of partial work at home into the working pattern. 3. The ability to be manageable through profit results or by agreeing to meet specific objectives in a given time. Experience has shown that teleworkers need short-term goals if they are to work effectively. 4., A possible access from distance (either electronically or by telephone conference) via remote device / PC or with a permanent connection to a specific database in cases where access to information is needed on a daily basis. 5. The possibility of a task, where a job is directly dependent in cases where a job on the time of delivery of the products or to be delivered by electronic means or "hand in hand " or via courier. On the basis of the above factors, the key sectors that have already adopted some forms of flexible working procedures in Greece and are expected to continue such practices are summarized as: Distance education, Telemedicine 148

150 Marketing & advertising. At this point, it must be stressed that teleworking should not be considered the ideal working method for all occupations. Occupations having as a prerequisite personal contact or manual labor cannot implement telework. As far as enterprises are concerned the main advantage of telecommuting is the increase in productivity. This increase is mainly due to lower vacation periods during labor and greater concentration of workers, increased motivation and job satisfaction, greater commitment to work in the absence of lost time and hassle when traveling. Arguably, the shift to teleworking and systematic use creates and projects the image of an innovative and advanced company which streamlines the organization of work, benefits from the information society and adjusts to modern developments. The greatest and most immediate benefit is the gain from the reduction of operating costs. We save in salaries and travel costs, we limit the need for premises and therefore the fewer the buildings the less the maintenance costs. With the implementation of telework we greatly contribute to the reorganization of enterprises by increasing the workers productivity and by improving the management of their tasks. All these have as a result the increased competitiveness of the company. With this new form of work, the company has more flexibility in the rational management of staff. The term "office" as a fixed spatial point ceases to exist. The company is no longer defined by the offices occupied, but as a network of relationships, which are connected through telecommunication networks. In this way, the opportunity for access to the labor market is given even to geographically remote areas. The desire for greater self-determination and control of time that the employee has leads to the adoption of flexible forms of work. The possibility offered by telecommuting to workers not to make unnecessary movements,or need to communicate with their colleagues in the narrow sense of an office and a specific timetable makes telework very attractive to a large number of workers Sub Section Ι At this point we can see what the resulting problems resulting of telework are We have problems with the educational system, the level of penetration and the use of technologies. There are also many questions about the social context of work which are not sufficiently circumscribed. Another important problem is that companies have not integrated their information infrastructure due to the large initial financial cost needed for the initial installation. 149

151 The employer cannot control and supervise the employees due to the fact that they are not constantly at the workplace. At this point a significant question arises. Whether companies can continue to require commitment from employees while they themselves do not commit; I believe the answer to this question is not easy especially nowadays when in Europe there is massive unemployment, plaguing mostly young people aged up to 35 years. Let s not forget that teleworking and the new forms of work under discussion refer mainly to younger age groups of workers. This research we are conducting ends up with some suggestions that, I think, will improve the employment frame of telework and the new forms of work related to telecommunications and computing. These proposals do not necessarily cost money but they are important in order to improve and institutionalize necessary and sufficient conditions for the proper implementation of telework. So, I will indicatively mention the implementation of teleworking in the public administration. As well as the education of young people on flexible types of work even at school, for example in the subject of professional orientation. Of course, there are some suggestions related to businesses. The field of teleworking is new and thus there is considerable margin for optimization as well as proposals that will contribute to the better implementation of new technologies. For instance, the application of pilot programs, the study of all the matters related to human resources and new technologies. Another point that we will focus our proposals in the field of telework on has to do with the employee. It is equally important to make suggestions and take steps in industry associations aiming at the collective representation of workers. Teleworkers are entitled to claim the same rights as employees who are on the premises Sub Section ΙΙ With basis to the investigations Eirobserver, Social Partners sign teleworking accord, Ecat-IST Programme Key Action II, SBIS General Population Surveys, 2002 EMERGENCE , which took place in 2002, we will present some tables and statistics presenting the situation in E.E. from 2002 until today. First we should note that the Netherlands and the Nordic countries are those that the precede in Europe. The Great Britain is above the average, followed France, Italy and Spain, while Germany is very close to average. COUΝΤRY Home Based Home Based Total Teleworking (full time) Teleworking (additional work) teleworking AUSTRIA BELGIUM DENMARK FINLAND of 150

152 FRANCE GERMANY GREECE IRELAND ITALY LUXEMBOURG HOLLAND PORTUGAL SPAIN SWEDEN GREAT BRITAIN AVERAGE E.E From the above table we see that Greece is close to the average of E.U. The Nordic countries have greater penetration to telework in relation to the countries of the South. Then the thesis will try to study the behavior of employees in relation to the evolution of telecommunications and the development of the internet from 1 Mbps up to 100Mbps. We will study how the labor cost is affected by the speed of the internet and the possible scenarios that arise from this study. 3. Conclusions In this paper we propose a methodology that may be useful at improving the current framework as an additional tool in the sector of telecommunication. At this point we can see what the resulting problems of telework are. We have problems with the educational system, the level of penetration and the use of technologies. There are also many questions about the social context of work which are not sufficiently circumscribed. Another important problem is that companies have not integrated their information infrastructure due to the large initial financial cost needed for the initial installation. The employer cannot control and supervise the employees due to the fact that they are not constantly at the workplace. This thesis will contribute to scientific research approaching the following questions. Is there potential improving for the way to use the telecommunications? How can we make the substitution of labor by telecommunications? Is there clear evidence that this resource will be used by this research will reduce the overall cost of labor. 151

153 References (1) Ali, M.S. (2002). Information resource centre : mainstream for the flow of information for lifelong learning. Paper presented at the XV annual conference of the Asian Association of Open Universities (AAOU), 21G23 February 2002, New Delhi, India. (2) Anastasiades P.A. (2003). Distance learning in elementary schools in Cyprus: the evaluation methodology and results. Computers & Education, 40(1), pp. 17G40. (3) Bates, A.W. (1993). Theory and practice in the use of technology in distance education, in Keegan, D. (Ed.), Theoretical Principles of Distance Education, London: Routledge, pp.213g233. (4) Garrison, D. R. (2000). Theoretical challenges for distance education in the 21st Century: A shift from structural to transactional issues. International Review of Research in Open and Distance Learning 1(1) (pp. 7G13), (5) Dabholkar, P.A. (1994), "Technology-based service delivery: a classification scheme for developing marketing strategies", in Swartz, T.A., Bowen, D.E., Brown, S.W.(Eds),Advances in Services Marketing and Management, JAI Press, Greenwich, CT, Vol. Vol. 3 pp

154 Regression modeling for spectral data sets: A multi-objective genetic approach Loukas Dimos Agricultural University of Athens, Department of Food Science & Human Nutrition, Laboratory of Microbiology and Biotechnology of Foods, Iera Odos 75, Athens 11855, d.loukas@aua.gr Ropodi Athina Agricultural University of Athens, Department of Food Science & Human Nutrition, Laboratory of Microbiology and Biotechnology of Foods, Iera Odos 75, Athens Nychas George-John Agricultural University of Athens, Department of Food Science & Human Nutrition, Laboratory of Microbiology and Biotechnology of Foods, Iera Odos 75, Athens Abstract In prediction problems, the finding of the best regression model in terms of quality-of-fit and parsimony is of great importance. Thus, the selection of a subset of the most informative and uncorrelated variables (Variable Selection problem-vs) is critical for the model's performance. This work focuses on the multi-objective modeling of the VS problem especially for spectroscopic applications. The proposed methodology considers the VS for the Partial Least Squares Regression (PLS-R) modeling as a two objective task, minimizing the number of selected variables as well as the Mean Square Error MSE) of prediction. The model selection is a two step procedure: (i) the NSGA-II genetic algorithm is applied in order to generate the frontier of Pareto-optimal solutions with respect to the multi-objective formulation of the problem, and (ii) a decision making process, based on information metrics, enables the selection of the final regression model. The aforementioned method was applied on spectral data related to the microbiological quality of minced meat samples, which were acquired by means of a spectroscopic instrument (Fourier Transform InfraRed - FTIR). Keywords: multi-objective optimization, genetic algorithm, variable selection, spectral data, meat quality 153

155 1. INTRODUCTION VS is one of the most critical steps in statistical modeling. Suppose Y a dependent variable and X={X1,..,Xp} a set of potential predictors, are vectors of n observations. The problem of variable selection, or subset selection, arises when one wants to model a relationship between Y and a subset of X, but there is uncertainty about which subset to use (George, 2000). This becomes harder in cases of high complexity or irrelevant information(noise). In regression modeling, VS can viewed as a special case of the model selection problem, where each model under consideration corresponds to a distinct subset of X of the most 'significant' variables. The 'best' model corresponds to the most parsimonious model, regarding the biasvariance trade-off (Hastie et al., 2009). More specifically, the low complexity models result in high bias in regressors and an underfit model that fails to identify all important variables, while the high complexity models result in high variance in regressors and an overfit model that cannot be generalized beyond the observed sample data. Typically we need to choose the 'best' balance between complexity(to be minimized) and quality-of-fit (to be maximized). However, it is impossible in practice to optimize simultaneously both objectives and we usually select a good subset of variables instead of the optimal one. VS is mainly applied in areas for which datasets with hundreds or thousands variables are available. In biological sciences, high dimensional multivariate data sets are frequently generated by means of analytical instruments, or 'sensors', that are based either on vibrational spectroscopy or surface chemistry. Among the most popular spectroscopic techniques is FTIR, which involves the rapid acquisition of absorbance values within a specific range of 3700 wavenumbers (400 cm cm -1 ) of the IR spectrum. The information contained in the spectral data can be used for the prediction of the molecular fingerprint of the sample (Nychas et al., 1998). Supervised linear modeling techniques such as PLS-R (Marten & Naes, 1989) are ideal for the analysis of spectroscopic data. PLS-R is based on latent variables and produces reliable full-spectrum models that are almost insensitive to noise (Leardi, 2003). Nevertheless, VS can be highly beneficial for the regression modeling when the number of variables to select is large compared to the number of samples, which is the typical biological case. The simplest method for model selection would be to examine all possible combinations of variables by means of an exhaustive search. However, for large number p of initial variables the VS problem is known to be NP-hard with time complexity O(2 p ) (Amaldi & Kann, 1998). A large number of heuristics methods have been proposed for providing solutions in a reasonable amount of time, most of the them can be roughly categorized as following: (i) Information criteria methods (AIC; Akaike, 1974), (BIC; Shwarz, 1978), (ii) Stepwise selection methods (forward selection, backward elimination and stepwise regression), (iii) Evolutionary Algorithms (Leardi, 2003; Jarvis & Goodacre, 2005), (iv) Statistical approaches (cross-validation; Shao,1993), (bootstrap; Efron,1986), (PLS- based methods; Nørgaard et al., 2000) (v) Regularization techniques (Lasso; Tibshirani, 1996), (vi) Stochastic approaches (Bayesian Model Averaging; Clyde et al., 2011), (Competitive Adaptive Reweighted Sampling - CARS; Li et al., 2009), and (vii) Decision Trees ( CART; Breiman et al., 1984, Random Forests; Breiman, 2001). For more details, please refer to the Guyon & Elisseeff (2003), Mehmod et al. (2012) and Fan & Lv (2010). Although widely used, most of those classical approaches treat VS as a single-objective problem (i.e. minimizing MSE) with a complexity penalty scheme to represent the trade-off between empirical risk and model complexity. Due to difficulty of the penalty choice, the sensitivity of the single solution might be quite critical especially for the case of high-dimensional data. For this reason, a two-stage multi-objective approach is proposed for the VS by this study. 154

156 In the first stage, an optimization procedure based on the application of the Multi-Objective Genetic Algorithm (MOGA) NSGA-II (Nondominated Sorting Genetic algorithm) (Deb, 2001) is introduced in order to provide the Decision Maker (DM) with the Pareto front of the optimal solutions in terms of quality-of-fit and complexity. Subsequently, in the decision-making stage, the AIC criterion is used for choosing the appropriate model among a reduced set of good models. In the next three paragraphs, we will firstly introduce the multi-objective formulation of the model selection problem. Secondly, the proposed methodology for PLS-R modeling will be presented and an application for the FTIR data of a microbiological experiment will be finally given. 2. MODEL SELECTION AS A MULTI-OBJECTIVE PROBLEM Regression modeling is a special case of supervised learning. Let X a fixed input space, Y the output space and S={Xi,Yi} n training samples drown independently and identically distributed from an unknown distribution D(X,Y). The problem of model selection is to find the appropriate model f : X Y with minimal error on the training set with respect to D. More formally, the model is assumed to belong to a predefined hypothesis-space H, which for linear regression is the following: H x k xk / J 1,..., p, xk X k (1) kj H forms a nested structure H1... Ht... H where Ht represents the subset of models with t many variables. By interpreting the model selection problem as finding the 'best' trade-off between complexity and quality-of-fit, we can formulate the following bi-objective problem where the objectives are jointly minimized (Sinha et al., 2013): Definition 1. Let : H, 1, ) a bi-objective function where: ( 2 i. the first objective : H min d : f 1 represents the complexity of the model in terms of the number of variables; and n : H min Y i f Xi represents the quality-of-fit n i1 of the model in terms of its generalization error. ii. the second objective The optimization problem is given by: H d min ( f ) ( ( f ), 2 ( f H s.t. f C, C H 1 f )), (2) Both objectives of (2) conflict to each other in a sense that the improvement of the one leads to deterioration of the other. The proposed multi-objective framework provides us with the set of the best trade-off solutions, called Pareto optimal solutions. The Pareto optimality concept is formally defined as follows (Deb, 2001): 155

157 Definition 2. A model f (1) is said to dominate the other model f (2), denoted f (1) f (2), if (1) (2) (1) (2) f f f f, j 1,2. and j j j j Definition 3. A feasible solution f * C of problem (2) is called a Pareto optimal solution,if g C that g f. The set of all the Pareto optimal solutions is called the Pareto set: PS= f C / g C : g f. The image of the PS in the objective space is called the Pareto Front: PF= f / f PS. 3. THE PROPOSED MULTIOBJECTIVE METHODOLOGY FOR PLS-R MODELING Due to its latent origins, PLS-R requires specific treatment. For this reason, the proposed methodology will be analyzed in two parts. In the first part, the optimization of the &2 problem will be introduced by means of the application of an hybrid version of the NSGA-II. In the second part, the overall procedure for the PLS-R modeling will be given by an analytical scheme. 3.1 The hybrid NSGA-II for model selection In order to avoid overfitting, an out-of sample estimator of the MSE, the MSE of cross-validation (MSE-CV) has been chosen as the second objective. Cross-validation is a method for model selection based on the predictive ability of the models. In its simplest form, the leave-one-out cross-validation (LOOCV; Stone, 1974), it considers an initial set of n data points and splits it in two parts. The first part contains n-1 points for fitting a model (model construction), whereas the last one point is used in the second part for the validation of the model. LOOCV selects the model with the best average predictive ability calculated, based on all the n divisions of the initial data set (Shao, 1993). Although computational expensive, the hybridization of the NSGA-II with the LOOCV optimization technique ensures that only the best models will be selected in every step of the procedure and also that the final solutions will be as accurate as possible in terms of the generalization error. The steps of the proposed MOGA for model selection are the following (Deb, 2001; Sinha et al. 2013): 1. Initialization: A binary vector of the size of the number of the p variables is initially introduced. It is denoted as a chromosome. If a particular variable is present, the bit value is 1; otherwise it is zero. The random combination of the zero-one genes formulates an individual. For the model selection problem, each one individual corresponds to a PLS-R model for a specific subset of the p variables. A parent population Pt, of size p, is initialized by picking the regression variables with uniform probability. 2. Crossover: A single point crossover of the binary strings of two randomly chosen members of the Pt is used for the creation of two offsprings. The procedure is repeated with different parents until a Ot population of p offspring members is produced. 3. Mutation: A binary mutation on each offspring is performed by flipping the bits with probability pm=1/p. 4. Evaluation & Non-Domination Sorting: The combination of the Pt and Ot populations results in an intermediate population Rt of size 2p. Each member of the Rt is evaluated with respect to the two objective functions. Those fitness results are used for the ranking of the individuals into different non-dominated fronts. Subsequently, a new parent population 156

158 Pt+1 is created by choosing firstly individuals of the best ranked fronts followed by the next-best and so on, until p individuals are obtained 5. Stopping criteria: Steps 2 to 4 are repeated until a number of 500 generations is reached. 3.2 The overall methodology for PLS-R modeling According to the principles of supervised learning, the initial data set is divided into training and test set. For the training set, the maximum number of PLS components is firstly extracted by an application of the PLS-R for the set of the initial variables. Subsequently, by a sequential application of the MOGA approach of &3.1 and the AIC criterion a single solution is resulted, which is validated by the test set. The whole procedure is graphically described by the next figure (Figure 1). Train Set Maximum number of PLS components Pareto front of optimal solutions Post-Optimality Analysis Final Solution Test Set 4. APPLICATION Figure 15: The overall multi-objective methodology for PLS-R modeling The proposed methodology was applied on experimental data concerning the microbiological quality of minced meat samples. For this experiment, minced beef meat of normal ph was obtained from a central retail of Athens. It was divided in portions of 70-80g, placed onto styrofoam trays and stored aerobically and under Modified Atmosphere Packaging (MAP) at 4 and 10 o C. Microbiological analysis (total viable counts - TVC) was performed every 12 and 24 hours for samples stored at 10 and 4 o C, respectively. In parallel, Fourier Transform Infrared Spectroscopy spectra were collected from which the fingerprint area ( cm -1 ) of the FTIR spectrum was selected. In order to deal with the multicollinearity of the data, and also to increase the performance of the algorithm, we performed sub-sampling (of window 5) and reduced the number of variables (wavenumbers) to 207. Subsequently, a Savitzky-Golay smoothing was introduced for noise reduction (Brown & Wentzell, 1999) and autoscaling was selected in order to enhance the variation of the most uninformative variables (Leardi, 2003). Autoscaled FTIR data were used for the prediction of microbial counts, regardless of storage and packaging conditions. For an initial number of 12 PLS components, the application of the hybrid NSGA-II of the &3.1., provided the DM with a Pareto set of 20 non-nested PLS-R models (Figure 2a). The resulted variations in complexity (from 3 to 27 variables) and in MSE (from to ) of the training data, enhance the DM to understand more thoroughly the requested trade-off that separates the alternative models. At this point, DM should decide either to keep the best model in MSE (Model of 27 variables: training (0.0327), test (0.0404)) or further examine the solutions of the optimal frontier. In second case, the application of the corrected AIC (AICc) for small samples (n<40): RSS 2kk 1 AIC c n log( ) 2k (3) n n k 1 157

159 ( RSS MSE n and k=number of the model variables) selects the model of 12 variables (Model 12) as the optimum one in terms of the trade-off between complexity and the quality-of-fit. The application of the PLS-R for the provided 12 regressors results to MSEtest= (Figure 2b). Figure 2: (a) Pareto plot; and (b) Plot of observed versus predicted log values of TVC (Model 12) 5. CONCLUSIONS This work proposed a methodology for PLS-R model selection for spectral data set. A hybrid version of the multi-objective genetic algorithm NSGA-II was introduced for the identification of the Pareto front optimal models. In case of the single solution, a second stage decision making process which is based on the AIC criterion was suggested. The computational results showed, verified that the proposed approach constitute a good alternative for the model selection problem. However, a further validation with many more data sets is required. ACKNOWLEDGEMENTS This work has been supported by the project Intelligent multi-sensor system for meat analysis - imeatsense_550 co-financed by the European Union (European Social Fund ESF) and Greek national funds through the Operational Program "Education and Lifelong Learning" of the National Strategic Reference Framework (NSRF) - Research Funding Program: ARISTEIA-I. REFERENCES Akaike, H. "A new look at statistical model identification". IEEE Transactions on Automatic Control, Vol. 19, 1974, pp Amaldi, E., and Kann, V. "On the approximation of minimizing non zero variables or unsatisfied relations in linear systems". Theoretical Computer Science, Vol. 209, 1998, pp Breiman, L., Friedman, J. H., Olshen, R. A., and Stone, C. J. Classification and Regression Trees. Wadsworth & Brooks,

160 Breiman, L. Random Forests, Berkeley, CA, USA: Statistics Department, University California, 2001 of Brown, C.D., and Wentzell, P.D. "Hazards of digital smoothing as preprocessing tool in multivariate calibration". J. of Chemometrics, Vol. 13,1999, pp Clyde, M., Ghosh, J., and Littman, M. "Bayesian Adaptive Sampling for Variable Selection and Model Averaging". Journal of Computational and Graphical Statistics, vol. 20, no. 1, 2011, pp Deb, K. Multi-Objective Optimization Using Evolutionary Algorithms, John Wiley & Sons Ltd., 2001 Efron, B. "How biased is the apparent error rate of a prediction rule?". J. of American Statistical Association, Vol. 81, 1986, pp Fan, J., and Lv, J. "A selective overview of variable selection in high dimensional feature space". Statistica Sinica, Vol. 20, 2010, pp George, E.I. "The Variable Selection Problem". Journal of American Statistical Association, Vol.95, No. 452, 2000, pp Guyon, I., and Elisseeff, A. "An Introduction to variable and feature selection". Journal of Machine Learning Research, Vol. 3, 2003, pp Hastie, T., Tibshirani R., and Friedman J. The Elements of Statistical Learning (2nd edition). Springer, New York, USA, 2009 Jarvis, R.M., and Goodacre, R. "Genetic algorithm optimization for pre processing and variable selection of spectroscopic data". Bioinformatics, Vol. 21, No 7., 2005, pp Leardi, R. "Genetic algorithm-pls as a tool for wavelength selection in spectral data sets", in Nature-inspired Methods in Chemometrics and Artificial Neural Networks (Leardi, R., ed.). Elsevier, Amsterdam, 2003 Li, H.-D., Y.-Z. Liang, Q.-S. Xu, & D.-S. Cao. "Key wavelengths screening using competitive adaptive reweighted sampling method for multivariate calibration". Anal. Chim. Acta 648, 2009, pp Marten, H., and Naes, T. Multivariate Calibration, Wiley, Chichester, 1989 Mehmod, T., Liland, K. H., Snipen, L., and Sabo, S. "A review of variable selection methods in Partial Least Squares Regression", Chemometrics & Int. Lab. Systems, Vol. 118, 2012, pp Nørgaard L., Saudland A.,. Wagner J, Nielsen J.P., Munck L. and Engelsen S.B. "Interval Partial Least Squares Regression (ipls): A Comparative Chemometric Study with an Example from Near-Infrared Spectroscopy", Applied Spectroscopy, Vol. 54, 2000, pp

161 Nychas, G. J. E., Drosinos, E. H., and Board, R. G. "Chemical changes in stored meat", in The microbiology of meat and poultry (Board, R.G., and Davies A. R., eds), pp London: Blackie Academic and Professional, 1998 Schwarz, G. E. "Estimating the dimension of a model". Annals of Statistics, Vol. 6, Νο. 2, 1978, pp Shao, J. "Linear Model Selection by Cross-Validation". J. of the American Statistical Association, Vol.88, No.422, 1993, pp Sinha, A., Malo, P., and Kuosmanen, T. "A Multi-objective Exploratory Procedure for Regression Model Selection". Journal of Computational and Graphical Statistics, (In Press), Stone, M. "Cross-validatory choice and assessment of statistical predictions". J. B Met, Vol. 36, No. 2, 1974, pp Roy. Stat. Soc. Tibshirani, R. "Regression shrinkage and selection via the lasso". J. Roy. Statist. Soc, Ser. B58, 1996,pp

162 Optimal use of non-collaborative servers in two-stage tandem queueing Dimitrios Pandelis systems Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece, Ioannis Papachristos Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Abstract We consider two-stage tandem queueing systems with a dedicated server in each queue and a flexible server that can attend both queues. We assume exponential interarrival and service times, and linear holding costs for jobs present in the system. We study the optimal dynamic assignment of servers to jobs assuming a non-collaborative work discipline with idling and preemptions allowed. We formulate the problem as a Markov decision process and derive structural properties of the optimal policy. For larger holding costs in the upstream station we show that i) non-idling policies are optimal, and ii) if there is no dedicated server in the first station, the optimal allocation strategy for the flexible server has a threshold-type structure. We also provide numerical results that reveal that under the non-collaborative assumption the optimal policy may have counterintuitive properties, which is not the case when a collaborative service discipline is assumed. Keywords: Tandem queues, Flexible servers, Markov decision processes. 1. Introduction We study the optimal dynamic server assignment in a two-station tandem queueing system with one dedicated server for each station and one flexible server that can work in both stations. In particular, we seek server allocation strategies that minimize linear holding costs for systems with Poisson arrivals and exponential service times. The problem we consider is motivated by the increasing use of flexible resources, such as cross-trained workers and reconfigurable machines, in order to cope with varying demand and changing operating conditions. We refer interested readers to Hopp and Van Oyen (2004) and Andradottir et al. (2013) for extensive literature surveys on workforce flexibility. Because of the complexity of the mathematical models involved, research on the optimal use of flexible servers with holding costs has focused on two-stage systems. Farrar (1993), Wu et al. (2006), and Pandelis (2007) considered different versions of a system without arrivals (clearing system) with dedicated servers in each station and one flexible server that is either constrained to work in the upstream station or can work in both stations. They showed that the optimal policy is 161

163 characterized by a switching curve; the flexible server is idled or assigned to the downstream station if the number of jobs there exceeds a certain threshold. Wu et al. (2008) and Pandelis (2008a) showed the optimality of such switching-curve policies for systems with arrivals and no dedicated server in the upstream station. Finally, Pandelis (2008b) studied a model with server operating costs in addition to holding costs and identified conditions under which the switchingcurve structure of the optimal policy is preserved. With the exception of Pandelis (2007) (constrained version), a common assumption in all the papers referred to in the previous paragraph was that different servers could collaborate to work on the same job, in which case the total service rate was equal to the sum of the individual servers rates. Moreover, a non-idling discipline for at least the dedicated servers was assumed in all of these papers. Both of these assumptions were relaxed by Pandelis (2014) in the case of clearing systems, where the optimality of threshold-type policies was established for certain special cases. In this paper we also relax the aforementioned two assumptions and show the optimality of nonidling policies for larger holding costs in the upstream station. Moreover, we show that when there is no dedicated server in station 1 and the dedicated server in station 2 is faster than the flexible server, the optimal policy is determined by a switching curve. In addition, we provide a condition under which priority is always given to station 1. The paper is organized as follows. In section 2 we formulate the problem as a discrete-time Markov decision process. The structure of the optimal policies is derived in section 3. Finally, we discuss our results in Section Problem formulation We consider a two-stage tandem queueing system where jobs arrive according to a Poisson process with rate. After their service is completed in the upstream station (station 1), jobs move to the downstream station (station 2) where they receive additional service and then leave the system. Each job in station i, i 1,2, incurs linear holding costs at rate h i. There are dedicated servers, one for each station, that are trained to work only in their corresponding station, and one flexible server that can work in both stations. We assume that this server can transfer from station to station, for instantaneously without any cost. We assume exponential processing times with rates 1 2, for jobs served by the flexible server in station 1,2, jobs served by the dedicated server and 1 2 respectively. At most one server can be assigned to each job, that is, server collaboration is not allowed, but two servers can work simultaneously on different jobs in the same station. Our objective is to find a server allocation strategy that minimizes the total expected discounted holding cost over a finite time horizon. Allowing preemptions at times of arrivals and service completions, we formulate the problem as a Markov decision process with state space {( x1, x2) : x1, x2 0}, where x 1, x 2 are the number of jobs in station 1 and station 2 respectively, including those in service. Instead of the continuous time problem we consider the equivalent discrete time problem obtained by uniformization, where without loss of generality we assume V x x the minimum n -step expected cost starting from state 1 2. We denote by (, ) n 1 2 ( x, x ), discounted by a factor, and by 162

164 A( x1, x 2) the set of feasible service rates at state x1 x 2 following optimality equations. where and (, ) Vn ( x1, x2) h1 x1 h2 x2 Vn 1( x1 1, x2 ) min Wn; 1, ( x 2 1, x2) ( 1, 2 ) A( x1, x2 ), 1 2. Then, the value function satisfies the Wn ;, ( x1, x2) 1V n1( x1 1, x2 1) 2Vn 1( x1, x2 1) (1 1 2) Vn 1( x1, x2), V0 ( x1, x2) The optimal policy We start this section with a broad characterization of the optimal policy that reduces the search for the optimal allocation in a subset of the feasible service rates. First, we provide a monotonicity property of the value function, which can be proved by a straightforward induction on n. Lemma 1. V ( n x, ) 1 x 2 is non-decreasing in its arguments. Because there is nothing to gain by keeping jobs in the downstream station, it is reasonable to allocate as much service rate as possible to that station to push jobs out of the system. This is established in the following proposition. Proposition 1. For given 1, Wn ; 1, ( x 2 1, x2) is minimized by maximizing 2. Proof. For 2 2 we have Wn ;, ( x1, x2 ) Wn ;, ( x1, x2) ( 2 2) Vn 1( x1, x2) Vn 1( x1, x2 1), which is positive by Lemma 1 and the proposition is proved. A consequence of Proposition 1 is that the optimal policy does not idle the dedicated server in station 2 when there are at least two jobs there. Next, regarding the optimal allocation in the upstream station, we define function f ( n x, ) 1 x 2 as follows: fn( x1, x2) Vn ( x1, x2) Vn ( x1 1, x2 1), x11, x2 0. Note that for some initial rate allocation 1, 2 the incentive to allocate additional rate in station 1 is given by W ( x, x ) W ( x, x ) f ( x, x ). n;, 1 2 n;, 1 2 n

165 f (, ) n x x determines whether we should add service rate to station 1 or not. Therefore, the sign of 1 2 This is formalized in the following proposition. Proposition 2. For given 2, Wn ; 1, ( x 2 1, x2) is minimized by maximizing 1 if fn 1 x1 x2 and by minimizing 1 if fn 1( x1, x2) 0. Proof. For 1 1 we have Wn ;, ( x1, x2 ) Wn ;, ( x1, x2) ( 1 1) fn 1( x1, x2), (, ) 0 which proves the proposition. f (, ) n x x, the optimal policy should either According to Proposition 2, depending on the sign of 1 2 allocate as much service rate as possible to the upstream station or not serve that station at all. In particular, as far as the dedicated server in station 1 is concerned, the optimal policy should not idle him when 1 2 f ( n x, ) 0 1 x2. f (, ) 0 n x x and there are at least two jobs upstream, and idle him when When it is not cheaper to have jobs in station 1 compared to station 2, it makes sense not to idle resources to keep jobs upstream. This is a consequence of the following lemma. Lemma 2. Let h1 h2. Then, 1 2 f (, ) 0 n x x for all x11, x2 0. Proof. The proof is by induction on n. Let, be the optimal allocations in state ( x1, x 2). Then, * * 1 2 f ( n x, 1 x ) 2 h1 h2 fn 1( x1 1, x2) W * * ( x1, x2) W * ( x ; 1 1, x2 1) n 1, 2 n;0, 2 * h1 h2 fn 1( x1 1, x2) 2 fn 1( x1, x2 1) by the induction hypothesis. * * (1 1 2) fn 1( x1, x2) 0, Based on the properties obtained so far, the optimal server allocations for h1 h2 are either explicitly determined (in certain cases with one job in one or both stations) or are restricted to two choices (see Table 1 below). On the other hand, when 1 2, the possible optimal allocations are those of Table 1 with the addition of allocation (0,max{ 2, 2}) if x2 1 and (0, 2 2) if x2 1. h h

166 (1,1) ( 1, 2) ( 1, 2) ( 1, 2) ( 1, 2) (1, 2) x ( 1, 2 2) (, ) or 1 2 or 1 2 (, ) (, ) ( 1,1) x (, ) (, ) or ( 1, 2) (, ) (, ) or 1 2 (, ) ( x1, x 2) ( 1 1, 2) or (, ) Table 1 Optimal allocations for 1 2 In the rest of the section we further characterize the structure of the optimal policy when h1 h2, there is no dedicated server upstream, and the dedicated server in station 2 is faster than the flexible server. Specifically, we show that the optimal policy is determined by a switching curve and identify a condition under which it always assigns the flexible server to station 1. This is stated in the following theorem. h Theorem 1. Let 1 2 1) For each 1 1 h, 1 0, and 2 2. Then, h h x there exists a threshold value tx ( 1) 2 such that the optimal policy assigns the t( x ) flexible server to station 2 if 2 1 2) If x, and to station 1 otherwise. ( h h ) h, it is optimal to assign the flexible server to station 1. Proof. For x1 1, x2 0 we define function f ( x, x ) V ( x,( x 1) ) V ( x, x ). d ( n x, ) 1 x2 1 n n 1 2 n 1 2 For x2 1 this function gives the incentive to assign the flexible server to station 1 instead of station 2 and can be computed by the following recursive expressions: d ( n x,0) 1 ( ) 1 h1 h2 dn1( x1 1,0) 2dn 1( x1,0) 2 1 fn 1( x1 1,1) 1 ( x1 1) 1 2 Vn 1( x1,0) Vn 1( x1 1,0) for x1 0, d ( n x,1) 1 ( ) 1 h1 h2 2h2 dn1( x1 1,1) 2dn1( x1,0) 1d n1( x1 1,2) 1 ( x1 1) 2dn 1( x1,1) for x1 0, d (1, n x ) 2 ( ) 1 h1 h2 2h2 dn1(2, x2) 2dn1(1, x2 1) 2dn 1(1, x2 1) ( x2 2) 1d n 1(1, x2) 2dn 1(1, x2) 1 for x2 1, 165

167 d ( n x, 1 x ) 2 ( ) 1 h1 h2 2h2 dn1 x1 x2 2dn1 x1 x2 1d n1 x1 x2 ( 1, ) (, 1) ( 1, 1) 2dn 1( x1, x2 1) ( x2 2) 1d n 1( x1, x2) 2dn 1( x1, x2) 1 for x1 1, x2 1. d ( x, x ) d ( x, x 1), Applying induction on n we can use the above equations to show that i) n 1 2 n 1 2 which suffices to prove the first part of the theorem, and ii) 1 2, which proves the second part. d (, ) 0 n x x when 1( h1 h2 ) 2h2 4. Discussion We considered optimal server allocations in two-stage queueing systems with dedicated servers at each stage and one flexible server. In contrast to similar models studied in the past we assumed a non-collaborative work discipline and allowed server idling. We derived properties of server allocation strategies that minimize the expected discounted holding cost over a finite time horizon. It can be shown by standard dynamic programming arguments that these properties are also valid for the infinite horizon discounted cost criterion and, assuming stable systems, for the average cost per unit time criterion as well. The results we managed to obtain were for jobs that accrue at least as much holding costs when present at the first station compared to the second station, in which case we proved that non-idling policies are optimal. For this case, when there is no dedicated server upstream and the flexible server is not faster than the dedicated server in the downstream station, we showed the optimality of a policy that assigns the flexible sever to the downstream station when the number of jobs there exceeds some threshold value. This value becomes infinite, that is, the optimal policy always gives priority to the upstream station, when the holding cost saved from a service completion in station 1 is greater than or equal to the cost saved from a service completion in station 2. When server collaboration is allowed, it has been shown that when there is no dedicated server downstream and the holding cost saved from a service completion in station 1 is less than the cost saved from a service completion in station 2, priority should be given to station 2. Although this is a reasonable property of the optimal policy, it does not necessarily hold when collaboration is not allowed. For example, consider a system with arrival rate equal to 0.07, service rate for the dedicated and flexible server in station 1 equal to 0.1 and 0.05, respectively, service rate for the flexible server in station 2 equal to 0.85, and holding cost rate for station 1 and station 2 equal to 17 and 1, respectively. The minimum average cost policy obtained by the value iteration algorithm assigns the flexible server to station 1 when there are two jobs in each station. 166

168 References Andradottir S., Ayhan H., and Down D.G. Design principles for flexible systems. Production and Operations Management, Vol. 22, 2013, pp Farrar T.M. Optimal use of an extra server in a two station tandem queueing network. IEEE Transactions on Automatic Control, Vol. 38, 1993, pp Hopp W.J. and Van Oyen M.P. Agile workforce evaluation: a framework for cross-training and coordination. IIE Transactions, Vol. 36, 2004, pp Pandelis D.G. Optimal use of excess capacity in two interconnected queues. Mathematical Methods of Operations Research, Vol. 65, 2007, pp Pandelis D.G. Optimal stochastic scheduling of two interconnected queues with varying service rates. Operations Research Letters, Vol. 36, 2008, pp Pandelis D.G. Optimal control of flexible servers in two tandem queues with operating costs. Probability in the Engineering and Informational Sciences, Vol. 22, 2008, pp Pandelis D.G. Optimal control of noncollaborative servers in two-stage tandem queueing systems. Naval Research Logistics, Vol. 61, 2014, pp Wu C.-H., Down D.G., and Lewis M.E. Heuristics for allocation of reconfigurable resources in a serial line with reliability considerations. IIE Transactions, Vol. 40, 2008, pp Wu C.-H., Lewis M.E., and Veatch M. Dynamic allocation of reconfigurable resources in a two-stage tandem queueing system with reliability considerations. IEEE Transactions on Automatic Control, Vol. 51, 2006, pp

169 GreenYourRoute platform Georgios K.D. Saharidis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece, Abstract The objective of the proposed research is to develop a Decision Support System (DSS) for a web based platform which will help individuals and companies move commodities in the most environmental friendly way, minimizing environmental externalities (e.g. CO2 emissions) and transportation costs. The developed platform which is the final outcome of an FP7 European research project and the national Operational Program "Education and Lifelong Learning", referred to as GreenRoute and EnvRouting projects, uses existing information systems (e.g. geographical, weather, real time traffic information systems) and emission calculation models as a basis to apply three main scientific outcomes. Keywords: Green, vehicle routing, web platform 1. Introduction EU-15 greenhouse gas emissions (GHG emissions) decreased in all sectors between 1990 and 2006 except in the transport sector, where an increase of 26% was documented (EC, 2008). The carbon footprint of transport has been constantly increasing inside and beyond the EU, currently reaching nearly a quarter of the overall greenhouse gases. Governments, organisations and companies want to better monitor and optimise the environmental impact of all the logistics operations and movements across the supply chain and the transportation corridors of the EU, reversing the current trends without jeopardizing international trade and movement. The objective of the proposed research, henceforth referred to as GreenYourRoute, is to develop a Decision Support System (DSS) for a web based platform which will help individuals and companies move commodities in the most environmental friendly way, minimizing environmental externalities (e.g. CO2 emissions) and transportation costs. The developed platform which is the final outcome of GreenRoute and EnvRouting projects will use existing information systems (e.g. geographical, weather, real time traffic information systems) and emission calculation models (ECMs) as a basis to apply their main three scientific outcomes, detailed in the following section. 2. Scientific outcomes The first scientific outcome of the GreenRoute project is the development of a function that assigns a score to each arc of EU transportation network referred to as the arc environmental externalities score (EESarc). EESarc approximates the potential environmental externalities if this arc is used. EESarc is estimated undependably of the vehicle but depended on the environment (i.e. transportation network) where the vehicle moves. The second scientific outcome is the development of a revised version of existing emission calculation models where the EESarc 168

170 function is incorporated and used as a correction coefficient. The revised emission calculation models predict the emissions at an arc under different conditions (e.g. traffic conditions). The third scientific outcome is the development of a novel approach for solving the general Vehicle Routing Problem (VRP) whose objective is to find the most environmental friendly route. The new solution approach is the first based on decomposition techniques. Two main types of decomposition techniques are studied: a) structural decomposition and b) mathematical decomposition. These three scientific outcomes play the role of a black box which is used by the platform. The GreenYourRoute platform is published to the web and is available with no cost to the user seeking to find optimal environmental friendly routes for its vehicle fleet. Modeling approaches to predict the environmental externalities (e.g. estimate CO2 emissions) coming from a move of a vehicle, are based on the estimation of fuel consumption by the vehicle. The fuel consumption is based on several factors with the most popular being distance, time, type of vehicle (e.g. body style, model year, type of engine etc.), weight load and mode of operation (e.g. different engine management concepts, gear-shift philosophies etc.). Certainly, the vehicle, which is the source of emissions, is a very important component for the evaluation and estimation of environmental externalities coming from freight transportation. However, there is also another important component which affects the amount of emissions produced by freight transportation that has been taken partially (e.g. the case of traffic by using average speed instead of real time traffic information) or not at all under consideration (e.g. the case of wind condition are not taken under consideration for the estimation of environmental externalities). This component is the characteristics of the transportation network. The transportation network is the recipient of the emissions but at the same time its specific characteristics are one of the sources of emissions (the other is the vehicle). The transportation network is composed by nodes which correspond to intersections and arcs which connect two nodes. Its arcs are characterised by many factors (from now on referred to as Transportation Network (TN) factors) which definitely influence the fuel consumption and the emission production if this arc is used. The first TN factor taken under consideration in the framework of GreenYourRoute is the traffic conditions which are taken by real time traffic information systems instead of mean speed. It is noted that the use of mean speed distributions in emission modeling does not explicitly take into account the effect of different driving dynamics at a particular mean speed (e.g. constant speed versus high levels of speed fluctuation) on vehicular emissions. This affects the accuracy of emission predictions especially for urban areas where the traffic jam is high. The second TN factor is the infrastructure profile. Even in the case of large-scale considerations, it cannot be assumed that for example - extra emissions when travelling uphill are balanced by a corresponding reduction in emissions when travelling downhill. Finally, the third TN factor, taken under consideration, is the weather conditions. It is certain, that when, for example, a vehicle travels against the wind the environmental externalities are higher than when there is no wind or the wind follows the traveling direction of the vehicle. Summarizing, TN factors taken under consideration in the framework of GreenYourRoute are (without be limited to): the traffic conditions, the 169

171 infrastructure profile and the weather conditions. By traffic conditions we mean the real time traffic conditions, by infrastructure profile we define the road gradient (uphill downhill), the type of road (road width, number of traffic lanes) and the traffic lights system and by weather conditions we consider the wind, the temperature and the humidity conditions of an arc. A screenshot of the user interface of GreenYourRoute platform is presented in the following figure. Figure 1: GreenYourRoute platform - User interface 3. Conclusion In this research work three prototype platforms for three regions of Greece are presented. The next step of this work is to collect transport information for additional regions of Greece and introduced them in the database developed in the frame of FindMyWay project. The final outcome of this research will be the development of a journey planner for the entire Greece connecting all cities and villages having a population greater or equal to Acknowledgements The author gratefully acknowledges financial support from the European Commission under the grant FP7-PEOPLE-2011-CIG, GreenRoute, and the Action «Supporting Postdoctoral Researchers» of the Operational Program "Education and Lifelong Learning" (Action s Beneficiary: General Secretariat for Research and Technology, Greece), and is co-financed by the European Social Fund (ESF) and the Greek State. 170

172 Reducing Waiting Time at Intermediate Nodes for Intercity Bus Transportation Dimitropoulos Charalampos Department of Mechanical Engineering, School of Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece, Skordilis Erotokritos Department of Mechanical Engineering, School of Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece, Saharidis George K.D. Department of Mechanical Engineering, School of Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece and Kathikas Institute of Research & Technology, Paphos, Cyprus Abstract Scheduling of transit networks is one of the most addressed problems in the mathematical optimization science, due to the increase of public transportation in the last decade. Researchers have introduced various formulations to address the problem of timetabling, using different objectives like bus synchronization and passenger demand. In this paper, we present a mixedinteger linear programming formulation with the objective of minimizing passenger waiting times at transitional transfer nodes, taking also into consideration high passenger demand that occurs at certain times. Keywords: minimizing waiting time, public transportation, mixed integer linear programming, transitional, nodes. 1. Introduction The foundation of the Greek public transport bus service consists of 61 regional bus cooperatives of individual owners of 4,175 vehicles with a public coach license. These cooperatives are called KTEL, an abbreviation of Kentrikon Tameion Eppagelmation Leoforion (roughly translated into Central Union of Bus Operators). The KTEL companies provide for about 80% of all passenger transportation in Greece. Interregional transport is provided by most of the KTEL companies. According to the Study of passenger transport by coach, undertaken in January 2009 by the European Commission [2007], 180 million passenger journeys were conducted by KTEL only in 2002, whereas 5,710 thousand passenger kilometers were covered by KTEL buses in Despite the existence of routes and roads, there are still several Greek major and smaller cities not directly connected by KTEL buses, mostly because of the arrangements made between the companies about the transportation of passengers from one city to another. As a result, passengers have to transfer from one bus to another, in order to reach their final destination, leading to substantial waiting times at the transfer/transfer nodes. 171

173 The case mentioned above, along with the increasing use of public bus transportation, leads to the necessity of a coordinating synchronization mechanism, the purpose of which is to minimize the waiting time for the passengers at the transfer nodes. Increased number of passengers using KTEL services leads to a large number of passengers waiting at these nodes, so a big issue for KTEL companies is how to mitigate this waiting time. Unfortunately, at the moment, there is no synchronization mechanism, although it would be for the benefit of both the passengers and the KTEL cooperatives. This paper presents a novel approach for timetabling, in which the center of the formulation is the minimization of waiting times at transfer nodes of bus networks. In contrast to previous studies, the bus routes are predetermined and the available number of buses remains unmodified, which is the most applicable and realistic approach for KTEL companies. 2. Model formulation One basic assumption to our modeling approach is that travelling time is assumed to be constant and equal to the travel time of traversing a link at the posted speed limit. One could argue that various factors (that are volatile like traffic congestions and weather conditions) could affect trip travel times. These factors were not considered as all buses use national roads where these factors are not as influential to travel time. The same principles apply to a number of attributes that describe the bus network in order to make it more flexible and easier to modify. These are the starting and ending time of the bus routes, the time a bus has to wait at a transfer-transfer node and the time interval between successive routes. Furthermore, the number of routes between nodes is also considered as fixed, as the purpose of the formulation is to not to minimize the existing number of bus routes, but rather to schedule the buses on these routes. Additionally, it is assumed that transfer/transfer nodes have sufficient capacity to accommodate the demand for bus parking. Thus, no additional constraint regarding limited number of bus terminals needs to be considered. It is also assumed that buses travelling have to return to their originating node (similar to the traveling salesman problem). However, in cases of bi-directional routes, this assumption is no longer necessary. Additional data will be used to describe some extensions to the primary formulation, such as the passenger demand pattern. The formulation described in this paper includes both continuous and binary variables. The objective function is the minimization of the total waiting time for the first and second available buses departing from all transfer nodes of the network. Furthermore, during the day, fluctuation of passenger demand is observed with distinct peak periods. These periods are referred to as highpriority periods and penalty factors are used in the objective function to account and assign higher priority to routes with this demand. Additionally, time restrictions for departure times for a subset of buses are considered. This is necessary in order to define the buses departure time window. There are also certain cases where this time window needs to be defined for specific bus routes to accommodate the schedule of the bus drivers. Figure 1 illustrates the concept of the formulation. It presents a layout of how the waiting time at the transfer nodes is calculated based on the arrival and departure times of all routes concerning these nodes. 172

174 Fig. 1 Illustration of the formulation. For the mathematical formulation a number of constraints were developed. The first constraint defines the combination of routes to and from an intermediate node as active or inactive, based on whether the bus from the intermediate node has already left or not. In conjunction to that constraint, the second constraint calculates the waiting time between the arrivals on each intermediate node and the first two available departing routes from that node towards the destination node. Furthermore, a constraint which assures that a bus departing from any node will eventually return to this node was developed, since our assumption suggests that each bus will return to their respective starting nodes. There is also a continuity constraint which assures that every itinerary must begin after the previous one regarding the same destination. Finally, a constraint which defines the time window in which any bus must depart was also implemented. Regarding the extended objective function that takes into consideration high passenger demand on certain time periods during a day, two additional constraints were utilized. The first one calculates the difference between the departing time of a route and the time the high passenger demand occurs. The second constraint defines the number of extra buses that need to be used in order to accommodate this high passenger demand. 3. Numerical examples This section describes a real-life example of a bus network in Greece. The example is based on the bus network of the island of Crete. We introduced the data needed for the formulation based on information obtained from Crete s KTEL. Note that time values introduced both as parameters and estimated as decision variables in the models, are expressed in minutes and based on a 24-hour conversion (e.g. 10:00 am is translated into the 600 th minute). The intercity bus network of Crete consists of 122 nodes, representing cities and villages accessible through the network. Out of these, 13 are regarded as transfer nodes. Among them are the four main cities at the island (Chania, Rethymnon, Heraklion, Ag. Nikolaos). The routes connecting the nodes of the network vary in number, from two to twenty five, depending on the population density at each node. The formulation was applied to the full extent of this intercity bus network. 173

175 Numerical Results & Discussion In this section, results of numerical examples are presented. These include a comparison between the total waiting time from the existing bus schedule and the total waiting time produced by the proposed model. Problem sizes and different values of the objective functions for all different aforementioned model variations are also presented. Aside from the values of the objective functions, the total waiting time for each extension is presented as the sum of the waiting times from all nodes to all nodes and for the total number of routes. The average waiting time is calculated by dividing the total waiting time that is obtained as the sum of the minimized waiting times yielded from the objective function, with the total number of active connections between routes to and from the transfer node. The resulting values of decision variables, as well as the objective function, are all measured in minutes, accordingly with all the parameters describing time in the network. All tables presented below include certain cases regarding the initial objective function and the extended one, all applied on the bus network of Crete. Case 1 represents the actual waiting times based on the existing bus schedule and act as a benchmark for the effectiveness of the formulation. Furthermore, case 2 corresponds to the values of the waiting times yielded by the formulation without taking into consideration an increased passenger demand. Finally, case 3 accommodates the results of utilizing the extension for increased passenger demand at all routes that depart from all transfer nodes. We must again mention that cases 1, 2 and 3 correspond to the minimization of the sum of waiting times only for the first and second available departing buses from any transfer node. Table 1 includes the number of constraints, variables and CPU solution times for all the cases that will be presented next. Table 2 presents the initial values of the total and average waiting time and the number of active routes based on the existing timetables, as well as the values extracted from the formulation. Furthermore, a percentage of improvement derived from the simulation compared to the existing timetables is also included. Table 1 Problem s Size Decision Variables Constraints CPU Solution Time (seconds) Gap (%) Case Case Case Table 2 Numerical Results Total Waiting (%)* Total Number of (%)* Average Waiting (%)* Time Active Connections ( Y i,k,j ) Time Case Case Case *Percentage of improvement over original bus timetables. 174

176 In our attempt to approach the existing bus network more accurately, the time window during which the routes occur was divided into five time zones, each one representing: early morning, late morning, noon, afternoon and night respectively. Each one of them can be represented by a specific time of the day around which all existing departing routes are focused. It should be noted that a typical timetable covers the time window from 5:30 am to 11:00 pm, or 330 to 1380 (in minutes). The first time zone is between the 330 th and the 540 th minute of the day, the second is between the 540 th and the 750 th minute of the day, the third is between the 750 th and the 960 th minute of the day, the forth is between the 960 th and 1170 th minute of the day and the last one is between the 1170 th and the 1380 th minute of the day. For these time zones, table 3 and figure 2 present the number of departing routes that occur during these zones, both for the formulation and the existing bus schedule. For case 3, we introduced five high passenger demand periods, each one corresponding to one of representing the five time zones, and we assumed it occurs in the middle of each zone (i.e. for time zone between 330 th and 540 th minute, the high demand period occurs at 435 th minute). The penalty factor values were deduced from the number of departing routes as distributed for time zones 1-5 for the existing bus schedule. This will lead to a timetable closer to original one, although the total waiting time will be increased as a result. Table 3 Distribution of departing routes Time zone 1 Time zone 2 Time zone 3 Time zone 4 Time zone 5 Case Case Case Timezone 1 Case 1 Case 2 Case 3 Timezone 2 Timezone 3 Timezone Timezone 4 5 Fig. 2 Graph representing cases 1,2,3 It can be deduced from cases 2 and 3, that there is a significant decrease of the waiting times at all transfer nodes of the bus network, compared to the existing bus schedule (case 1). Specifically, comparing cases 1 and 2, there is a 75.9% decrease in the average waiting time, despite the fact that the active number of connections (Y m,n i,k,j ) is significantly smaller. Case 3 produced a time table that minimizes the total waiting time, without deviating too much from the existing time table, since it takes into consideration only the first and second available departing routes from every intermediate node, leading to more active connections for the passengers to use. 175

177 Although the total waiting time is increased compared to case 2, it remains smaller compared to case 1 by 26.2%. In both cases of objective functions, it is prudent that case 3 yielded increased waiting times compared to case 2, but greatly decreased compared to the existing bus schedule (cases 1). Meanwhile, the increased number of active connections in case 3 make the results more suitable for realistic examples, where the passengers need to have more available options concerning their departing from a transfer node. 4. Conclusions This paper has addressed the problem of minimum waiting times when creating bus timetables. A mathematical formulation was proposed to develop a more desirable and passenger-friendly transit system. This was achieved by minimizing the waiting time spent by passengers at transfer nodes on the bus network. This approach was tested by utilizing data obtained from the bus network of the Greek island of Crete. By comparing existing timetables to the ones extracted from the proposed model, a definitive improvement was observed. This improvement remained despite our effort to remain closer to the existing timetables, through the use of the various extensions that were introduced. References 1. Steer Davies Gleave. Study of passenger transport by coach. Publication TREN/E1/ European Commission, Ceder, A, Golany B, Tal O. (2001) Creating Bus Timetables with Maximal Synchronization. Transportation Research Part A. 35: Eranki, A. A model to create bus timetables to attain maximum synchronization considering waiting times at transfer stops. Master s thesis. Department of Industrial and Management Systems Engineering, University of South Florida, Ibarra-Rojas, O. J., and Y. A. Rios-Solis. Synchronization of bus timetabling. Transportation Research Part B, Vol. 46, 2012, pp Hall R., Dessouky M. and Lu Q. Optimal Holding Times at Transfer Stations. Computer and Industrial Engineering. Vol. 40, 2001, pp Bussieck M. R., Winter T., and Zimmermann U. T. Discrete Optimization in public rail transport. Mathematical Programming. Vol. 79, Issue 1-3, 1997, pp Goverde R. M. P. Synchronization Control of Scheduled Train Services to Minimize Passenger Waiting Time. Transport, Infrastructure and Logistics, Proceedings 4 th TRAIL Congress, Chen D. and Wu K. Research on Optimization Model and Algorithm of Initial Schedule of Intercity Passenger Trains based on Fuzzy Sets. Journal of Software, Vol. 7, No. 1, 2012, pp Reinhardt L. B., Clausen T., and Pisinger D. Synchronized dial-a-ride transportation of disabled passengers at airports. European Journal of Operational Research. Vol. 225, 2013, pp Wong R. C. W., Yuen T. W. Y., Fung K. W. and Leung J. M. Y. Optimizing Timetable Synchronization for Rail Mass Transit, Transportation Science, Vol. 42, No. 1, 2008, pp

178 Innovation management strategies for organizational performance Dimitrios Mitroulis School of Informatics, Department of Applied Informatics, University of Macedonia P.O. Box 1591, GR54006 Thessaloniki, Greece Fotios Kitsios School of Informatics, Department of Applied Informatics, University of Macedonia P.O. Box 1591, GR54006 Thessaloniki, Greece Abstract Innovation has always been a critical factor for every kind of entrepreneurial achievement and performance. However, most of the organizations, which were supposed to innovate, have focused on better short term efficiency, ignoring the chance of getting competitive advantage over their competitors. The organization s relationship with either customers or competitors could improve its knowledge over the market conditions and gain market-oriented information. The successful management of the market-oriented inflow and organizational innovation leads to the improvement of the organizational performance. The purpose of this paper is to collocate a main framework which a business or industry could use in order to identify whether its organizational innovation could be the joint between market-orientation and organizational performance. The whole research focalizes on the questions that could unveil the organizational performance, by evaluating its innovation capabilities. Other studies have shown the importance of innovation in today s organizations, giving emphasis on market-orientation, better efficiency, innovativeness and organizational performance. Therefore, in order to evaluate and examine the theoretical assumptions, a questionnaire, addressed to Greek SMEs, is cited. It is used to examine and evaluate capabilities, operations and competitive advantages which could lead SMEs to organizational performance. All variables are measured in a 5point Likert type scale. The results of this study are examined with multicriteria methods. Keywords: Market-orientation, Innovation, Performance, Dynamic Hybrid Strategy 177

179 1. Introduction Globalization, economic crisis, technological changes, lack of opportunities threaten business sustainability. In fact, they try to follow the unceasing ongoing changes so that they could gain stability. Customers seek the best, possible, combination of quality and price for products and services that are offered. Competition has reached the highest limits and the organizations are now focused to find a solution to this questionings. How are we supposed to gain competitive advantage? How are we going to seize our opportunities? How could we differentiate from the others? Is innovation the key to success? How could we have a better performance? In order to gain a sustainable competitive advantage, firms are supposed to follow a differentiation strategy. Additionally, market orientation is the factor that connects the organization with the market preferences (Medina & Rufın, 2008). However, this is not the only factor that affects the organizations. Many researchers agree that innovation is the mediator for better organizational performance. They, moreover, realize that innovation is the link between market orientation and organizational performance (Jin K. Han et al., 1998). How is it possible to identify and understand this kind of organizational assets? Motivated by the previously written aspects, this study tries to identify the main perspectives of reclaiming advantage. Strategy drivers, market orientation, innovation and organizational performance are the four factors that are investigated and analyzed. The rest of the paper presents the conceptual framework (Section 2), methodology, data and findings (Section 3) and finally, conclusion (Section 4). 2. Conceptual framework 2.1. Market Orientation: Jin K. Han et al., (1998) identify market orientation as a part of the organizational culture. The three basic forces: i) customers behavior and the ability to foresee their needs, ii) competitors behavior and the ability to compete them in capabilities and technology, and iii) the interfunctional coordination. Market-driven business is always ready to anticipate the continuously changing needs of its customers and respond to them through innovation (Salavou et al, 2004). A market oriented culture could improve the organizational intelligence and agility by creating an interactive relationship ( Fátima Evaneide Barbosa de Almeida et al, 2013). Customer satisfaction should be considered as a part of the market orientation concept (Avlonitis & Gounaris, 1999) Dynamic Hybrid Strategy: Strategy is the main framework of the organization. Having in mind all the changes in their environment, firms set hybrid strategies which give emphasis on differentiation, flexibility and adaptability (Pertusa-Ortega, 2008). Combining the strategy with market orientation, we could be led to a dynamic or behavioral hybrid strategy (Avlonitis & Gounaris, 1999). Hybrid competitive strategies are related to better organizational performance (Pertusa-Ortega, 2008). Therefore, a market oriented culture along side with dynamic hybrid strategy offer a stable base for higher rates of performance (Avlonitis & Gounaris, 1999; Fátima Evaneide Barbosa de Almeida et al., 2013). Relating the above with innovation strategy, Dynamic Hybrid Strategy appears. In fact, it includes components of innovation strategy, competitive strategy and differentiation strategy. Hybrid strategies might be more sensitive on customer needs. 178

180 The difficulty of gaining and using business intelligence is the main problem in this relationship. The pursuit of hybrid competitive strategies may help obtaining several sources of advantage and, therefore, make it possible to put over higher performance levels (Pertusa-Ortega, 2008). H1: There is a positive relationship between Market Orientation and Dynamic Hybrid Strategy. H2: There is a positive relationship between Dynamic Hybrid Strategy and Business Performance Innovation Management: Organizational innovation is influenced by the market orientation. Jimenez et al, (2008) mention that innovation is the organization s means of response to the market. Additionally, innovation has to be developed and executed as an internal part of the business strategy (Gerald & Emamisaleh, 2014). In addition, they support the importance of innovation to the organizational performance, indicating that it should be followed by the appropriate action plan. Creating innovation capabilities and giving emphasis to every innovation type (organizational, administrative, process, production, technical and marketing innovation) is going to make a successful introduction of competitive advantage. Competition is the force of innovation (Salavou et al., 2004). Innovation management is the means of creating innovation capabilities. (Salavou et al, 2004; Han et al, 1998; Avlonitis & Gounaris, 1997; Jimenez et al, 2008). It reveals the firm s ability to obtain advantage of the appropriate innovation management inside the business terms. H3: There is a positive relationship between Innovation Management and Dynamic Hybrid Strategy Organizational Performance: Ferraresi et al., (2012) support that organizational performance could be counted with both financial and non-financial criteria. These are market share, sales of new products, the rates of return on investment and evaluation of internal factors such as operational improvements and reducing the time of response against the changes imposed by the market. This is more or less the main idea of organizational performance that dominates the literature. Additionally, in this study the relationship between innovation performance and organizational performance is tested, so that it would be clear whether it is worth to invest on an innovation which is successful for the firm and its effect on performance. H4: There is a positive relationship between Innovation Performance and Business performance. 179

181 Figure 1: Conceptual model 3. Methodology, Data and Findings This survey has started in Greece, in order to test the hypotheses of this study. Greek SME s have been asked to participate in the survey. It was designed to cover a wide range of industries. However, it is limited to the companies that create or offer IT services. This is an ongoing research which until now does not have enough evidence to prove its hypotheses and the wanted results. A questionnaire has been developed following the 5-point Likerty-type scale. The variables and the measurement items were adapted and combined from existing scales in the literature. The questionnaire has been examined by a group of experts. It is addressed to general managers or marketing directors. Additionally, further information was collected via personal interviews with the general manager or the marketing director of each company, using a structured questionnaire. All questionnaires were ed to the companies. The structural equations model (SEM) is going to be used in order to test the validity and the relationship among the variables. Until now, 20 questionnaires have been given and 10 of them have been obtained, yielding a response rate of 50 percent. 4. Conclusion The harmonious existence of innovation strategy within the general strategy of the organization could provide a stable base for more successful business strategy, operations and higher levels of performance (Wong, 2013). In addition, the influx from market orientation could yield a positive impact for gaining a sustainable competitive advantage and offering added value to the business. Considering innovation as a link between customer satisfaction and organizational performance, it is easily understood that strategically, organizations could be able to gain intelligence and weaponize its strategy, creating dynamic hybrid strategy, adjusted to its needs, resources and capabilities. Performance is a fundamental factor for a successful organization. In other words, every innovative organization acquires competitive advantage against its competitors. If an organization succeeds in innovation management, producing innovation capabilities and competitive advantage, it would be able to achieve the superior organizational performance. 180

182 References Alex A. Ferraresi, Carlos O. Quandt, Silvio A. dos Santos and Jose R. Frega. (2012). "Knowledge management and strategic orientation: leveraging innovativeness and performance". Journal of Knowledge Management. 16 (5), p Affendy Abu Hassim, Asmat-Nizam, Abdul-Talib, Abdul Rahim Abu Bakar. (2011). "The Effects of Entrepreneurial Orientation on Firm Organisational Innovation and Market Orientation Towards Firm Business Performance". International Conference on Sociality and Economics Development. 10 (2011), pp Cayetano Medina and Ramo n Rufı n. (2009). "The mediating effect of innovation in the relationship between retailers strategic orientations and performance". International Journal of Retail & Distribution Management. 37 (7), p Daniel Jimenez-Jimenez, Raquel Sanz Valle, Miguel Hernandez-Espallardo. (2008). "Fostering innovation, the role of market orientation and organizational learning". European Journal of Innovation Management. 11 (3), pp Eva M. Pertusa-Ortega. (2008). "Hybrid Competitive Strategies, Organizational Structure and Firm Performance". George J. Avlonitis and Spiros P. Gounaris. (1999). "Marketing orientation and its determinants: an empirical analysis". European Joumal of Marketing. 33 (11/12), p Gloria L. Ge and Daniel Z. Ding. (2005). "Market Orientation, Competitive Strategy and Firm Performance: An Empirical Study of Chinese Firms.Journal of Global Marketing".18 (3/4), p Gurhan Gunday, Gunduz Ulusoy, Kemal Kilic, Lutfihak Alpkan. (2011). "Effects of innovation types on firm performance". International Journal of Production Economics. 133 (2), p H. Salavou, G. Baltas and S. Lioukas. (2004). "Organisational innovation in SMEs: The importance of strategic orientation and competitive structure". European Journal of Marketing.38 (9/10). p Jin K. Han, Namwoon Kin and Rajendra K. Srivastva (1996). "Market orientation and organizational performance; Is innovation the missing link". Lokman Mia and Lanita Winata. (2014). "Manufacturing strategy and organisational performance The role of competition and MAS information". Journal of Accounting & Organizational Change. 10 (1), p Masood Ul Hassan and Sadia Shaukat. (2013). "Effects of Innovation Types on Firm Performance: an Empirical Study on Pakistan s Manufacturing Sector". Pakistan Journal of Commerce and Social Sciences. 7 (2), p Parvaneh Gelard and Korosh Emamisaleh. (2014). "The Evolution of Innovation Types Towards Production Performance". International Business Management. 8 (4), p Stanley Kam Sing Wong. (2013). "The role of Management involvement in innovation". Management Decision. 51 (4), pp

183 Fátima Evaneide Barbosa de Almeida, João Veríssimo Lisboa, Mário Gomes Augusto, Paulo César de Sousa Batista. (2013). "Organizational Capabilities, Strategic Orientation, Strategy Formulation Quality, Strategy Implementation and Organizational Performance in Brazilian Textile Industries". 182

184 Country risk evaluation methodology to support bilateral cooperation in the field of electricity generation from renewable sources Aikaterini Papapostolou National Technical University of Athens 9, Iroon Polytechniou str., 15780, Zografou, Athens, Greece, Charikleia Karakosta National Technical University of Athens 9, Iroon Polytechniou str., 15780, Zografou, Athens, Greece Vangelis Marinakis National Technical University of Athens 9, Iroon Polytechniou str., 15780, Zografou, Athens, Greece John Psarras National Technical University of Athens 9, Iroon Polytechniou str., 15780, Zografou, Athens, Greece Abstract Renewable energy sources (RES) cooperation within the European Union, as well as with EU neighboring countries is high on Europe s political agenda. According to the EU Directive 2009/28/EC, one or more Member States could cooperate with one or more developing countries in joint projects, regarding the generation of electricity from renewable sources. This paper outlines a multicriteria methodology to evaluate country opportunities and risks for the successful implementation of the cooperation mechanisms with third countries. The proposed evaluation criteria are built on three points of view: (1) investment framework/country risk profile, (2) social, and (3) energy security. The overall evaluation of countries is obtained through a multicriteria additive value model, which is assessed using an ordinal regression approach. Five countries of North Africa are evaluated and ranked considering the latest criteria data. Keywords: Country risk; Renewable energy; multiple criteria; robust ordinal regression. 183

185 1. Introduction Renewable Energy Sources (RES) cooperation within the European Union (EU), as well as with EU neighboring countries is high on Europe s political agenda. EU Renewable Energy Directive (DIRECTIVE 2009/28/EC) allows Member States to achieve their 2020 Renewable Energy (RE) targets by implementing joint projects in neighboring countries and thus to import electricity from RE sources from them in order to meet domestic demand in Member States. In literature, several studies exist that try to identify the perception of risk in RES investments for the North Africa countries. Hawila et al. (2014) apply a consistent methodology (Weighted Sum Method (WSM)) across all the North African countries to assess the present state of infrastructure, institutions and human capital factors to adopt and deploy RE technologies. Komendantova et al. (2012; 2011) address the perception of risks in RES projects, which is considered as regulatory, political, and force majeure. Corruption and inefficient and unpredictable bureaucracies are matters of great concern for the project developers in North Africa. In the case of Morocco and Egypt, political instability and regulatory barriers are issues that discourage the development of clean development mechanism (CDM) projects, however well-organized mechanisms have already been established to support RES development in both countries so far (Karakosta et al., 2013; Karakosta & Psarras, 2013). An important number of Multiple Criteria Decision Making (MCDM) methods have been used so far in order to evaluate the feasibility of RES projects, such as the Multi-Attribute Utility, the ELECTRE and the PROMETHEE methods, the Analytical Hierarchy Process and the TOPSIS (Cavallaro & Ciraolo, 2005; Doukas et al., 2009; 2010; Haurant et al., 2011; Pohekar & Ramachandran, 2004; Rosso et al., 2011). To the best of our knowledge, there are only very few studies using UTASTAR method for RE and energy sector problems. The aim of this study is to evaluate country risk in order to support bilateral cooperation in the field of electricity generation from RES. To this end, a multicriteria decision support methodology has been developed taking into account three evaluation points of view, the investment framework/country risk profile, the social, and the energy security point of view. An additive value model has been constructed and the UTASTAR disaggregation method has been implemented to infer the criteria weights. The obtained ranking of alternatives has been subjected to robustness analysis and finally the proposed methodology has been applied to North Africa countries. Apart from this introductory section, the paper is organized as follows: the second section elaborates the multicriteria evaluation system obtained as regards the evaluation points of view and the criteria proposed. The third section introduces the methodological framework and in section four the implementation of the proposed methodology is presented. Finally, the fifth section summarizes the main points of the study and presents the future perspectives. 184

186 2. Designing a multicriteria evaluation system In order to assess the opportunities and barriers for RE development in third countries, the current situation in the analyzed countries has to be assessed as it is likely to condition the future RES deployment path. In this framework, a multicriteria evaluation model is proposed. It comprises nine evaluation criteria based on three evaluation points of view to support decisions for the implementation of RES investments (Figure 1): (i) investment framework/country risk profile, (ii) social, and (iii) energy security. Figure 16: Evaluation points of view and criteria This framework will determine which countries are more suitable or likely to attract a larger share of foreign investment in RE. The definition of criteria is presented below while the scales used to evaluate countries are given in Table 1. OECD country risk rating (g1): This criterion is an OECD (Organization for Economic Cooperation and Development) country risk classification index which classifies countries into one of eight categories (0-7) through the aggregation of two dimensions: A quantitative assessment of country credit risk produced by Country Risk Assessment Model (CRAM) and a qualitative and subjective assessment of the CRAM results made by country risk experts from OECD members, integrating political risk and/or other risk factors. Ease of doing business rank (g2): The ease of doing business rank has been developed by the World Bank and ranks economies from 1 to 189. A high ranking on this index indicates that the regulatory environment is more conducive to the operation of a local firm. Global competitiveness index (g3): This index, developed by the World Economic Forum, captures the competitiveness conditions of 148 economies, integrating environmental and social sustainability issues. Global competitiveness index includes a weighted average of many different components, each measuring a different aspect of competitiveness. 185

187 Social Hotspot Database index for the energy sector (g4): This index assesses the existing social risk (social hotspots) as regards a variety of sectors in terms of: community infrastructure, labor rights and decent work, health and safety, human rights and governance. In this paper the sector that has been examined is the electricity sector. Energy Development Index (g5): The Energy Development Index (EDI) is a multidimensional index built by the International Energy Agency (IEA) that includes the principal indices related to energy development (access to electricity, access to clean energy cooking facilities, access to energy for public services, access to energy for productive use). Rate of electricity transmission and distribution losses (g6): This indicator captures the quality of electrical infrastructures and networks and includes losses in transmission between sources of supply and points of distribution and in the distribution to consumers. The indicator is developed by the World Energy Council. Energy demand growth (g7): This indicator captures the stress of a system (understood as its capacity to meet the energy demand). The higher the annual energy demand growth, the greater the risk that the existing supply and new energy infrastructure investments will not be able to provide the required energy supply. Age of technology fleet (g8): This is a qualitative indicator from 1 to 5 intending to capture the need to replace the existing infrastructure (as the older the fleet is, the higher the technical failure risk). Share of fossil fuels in electricity production (g9): This indicator intends to capture the dependency of the country on fossil fuels (oil, coal, natural gas) in the electricity production. Criterion Worst Level Best Level Source g OECD, 2014 g World Bank, 2013 g World Economic Forum, 2013 g Social Hotspot Database, 2014 g International Energy Agency, 2011 g 6 20% 0% World Energy Council, 2011 g 7 30% 0% BETTER project report, 2013 g BETTER project report, 2014 g 9 100% 50% Trading Economics, 2011 Table 17 Criteria evaluation scales 186

188 3. Methodological Framework The aim of the proposed methodology is to assess an overall evaluation model that aggregates the nine evaluation criteria. The multicriteria preference model is assumed to be an additive value function of the following form: u n g p u g i1 i i i (1) subject to the following normalization constraints: u u * 1 0 i g i i g i i 1,2,..., n (2) n i1 p i 1 (3) pi 0 i 1,2,..., n (4) where g = (g1, g2, gn) is the performance vector of a country on the n criteria, gi* and gi* are the least and most preferable levels of the criterion gi, respectively, ui, i = 1,2,,n, are non-decreasing real valued functions, named marginal value functions, which are normalized between 0 and 1, and pi is the weight of ui. Thus, for a given country a, g(a) and u[g(a)] represent the multicriteria vector of performances and the global value of country a respectively. Due to objective difficulties to convince decision makers (DM) in externalizing trade-offs between heterogeneous criteria, it is usually preferred by the analysts to infer the additive value functions from global preference structures, by applying disaggregation or ordinal regression methods (Greco et al., 2008; 2010; Jacquet-Lagrèze & Siskos, 1982; 2001). In this study the method used is the UTASTAR method (Siskos & Yannacopoulos, 1985). 4. Implementation of the model In this section the implementation of the UTASTAR method is presented as well as the application of the model obtained to North Africa countries. 4.1 Implementation of the UTASTAR method and robustness analysis According to UTASTAR ordinal regression method, a ranking of reference real or fictitious countries is required to infer an additive model that is compatible with the ranking. A set of 17 virtual countries that differ on two or three criteria is then constructed in order to be easier for the decision maker (DM) to compare them. The DM makes pairwise comparisons and inserts 187

189 progressively each alternative country into a global preference ranking. The constructed reference set, and the DM s given ranking are presented in Table 2. Reference country g 1 g 2 g 3 g 4 g 5 g 6 g 7 g 8 g 9 DM s Ranking Global Values A B C D E F G H I J K L M N O P Q Table 2 Multicriteria evaluation of the 17 fictitious countries The application process of UTASTAR method is terminated when a full compatibility between the additive value model and the DM s ranking is achieved. Because of the infinity of such compatible value functions the most representative value function is obtained (barycenter value function, see Hurson & Siskos, 2014). The last column in Table 2 shows the global value of each fictitious country that is compatible with the DM s ranking. The obtained barycenter value function is defined by the following equation (5) and the average marginal value functions of Figure 2 (in yellow): u(g) = 0.231u1(g1) u2(g2) u3(g3) u5(g5) u6(g6) u7(g7) u8(g8) u9(g9) (5) 188

190 In order to test the existence of multiple or near compatible additive value models, robustness analysis is conducted (see Hurson and Siskos, 2014). Figure 2 and Figure 3 present graphically the minimum, maximum and most representative marginal value function and weights of each criterion respectively, revealing the variation of these functions. The chart shows considerable robustness with the exception of the first and third criterion. 1 0,9 0,8 0,7 0,6 0,5 0,4 0,3 0,2 0, ,9 g 1 0,8 0,7 max 0,6 0,5 min 0,4 average 0,3 0,2 0, ,9 g 2 0,8 0,7 max 0,6 0,5 min 0,4 average 0,3 0,2 0, g 3 max min average 1 0,9 0,8 0,7 0,6 0,5 0,4 0,3 0,2 0, ,9 g 5 0,8 0,7 max 0,6 min 0,5 average 0,4 0,3 0,2 0, ,9 g 6 0,8 0,7 max 0,6 min 0,5 average 0,4 0,3 0,2 0, g 7 max min average 1 0,9 0,8 0,7 0,6 0,5 0,4 0,3 0,2 0, ,9 g 8 0,8 g 9 0,7 max 0,6 0,5 max min 0,4 min average 0,3 average 0,2 0, Figure 2 The assessed marginal value functions of the criteria 0,3 0,25 p i 0,2 0,15 0,1 min avg max 0,05 Figure 3 The assessed weights of the criteria In order to examine further the stability of the proposed model the ASI robustness index (Grigoroudis and Siskos, 2010) is computed. The ASI is defined as the mean normalized standard deviation of the estimated weights of the criteria and is calculated in the case of UTASTAR as: m 2 m 2 n 1 m p j ij p 1 j1 ij ASI 1 (6) n i1 m n 0 m 1 g 1 g 2 g 3 g 4 g 5 g 6 g 7 g 8 g 9 189

191 where m is the number of weighting instances of the system, n the number of the criteria and pij the weight of the i-th criterion for the j-th instance. The ASI is equal to 0.998, a value indicating that the nine criteria s weights are adequately stable. 4.2 Application to North Africa countries Taking into account the assessed most representative value model five North Africa countries are evaluated. The score of the countries on the nine criteria, the global value and the final ranking obtained are presented in Table 3. Country g 1 g 2 g 3 g 4 g 5 g 6 g 7 g 8 g 9 Global Value Ranking Morocco Tunisia Egypt Algeria Libya Conclusions Table 3 Evaluation of five North Africa countries This paper evaluates the opportunities and risks for the implementation of cooperation mechanisms in the field of RES electricity production based on a multicriteria evaluation model. The model obtained is fully compatible with the DM s ranking on a set of fictitious countries and enables each individual to assess the current situation of a country as regards the investment, social and energy security profile of it. After the implementation of the model to North Africa countries, the obtained ranking shows that Morocco is considered to have the most suitable conditions among the five North Africa countries for successful implementation of RES projects. Considering the above, a decision support system would be a useful tool for the implementation of the proposed methodology supporting potential users/ DMs in evaluating/ benchmarking energy country risk. Further research could also address the integration of new multicriteria robustness analysis techniques to take into account the ranking variation due to lack of sufficient preference information. Furthermore, it would be valuable the development of a formal protocol and adequate techniques for the construction of reference set of countries, in order to infer the weights of the criteria. Finally, the study would be enhanced by applying the obtained model to other regions worldwide. Acknowledgement The current paper was primarily based on the research conducted within the framework of the project BETTER - Bringing Europe and Third countries closer together through renewable 190

192 Energies (project number: IEE/11/845/SI ), supported by the Intelligent Energy Europe program. References BETTER, D2.5: Indicators and Methodologies to Assess Key Issues for the Implementation of the Cooperation Mechanisms. Deliverable of the BETTER project (IEE/11/845/SI ). BETTER, D3.2.1: Demand Development Scenarios for North Africa. Deliverable of the BETTER project (IEE/11/845/SI ). BETTER, D3.1: Inventory of RES-E in North Africa. Part A: Power Plant and Grid Inventory in Morocco, Algeria, Tunisia, Libya and Egypt. Deliverable of the BETTER project (IEE/11/845/SI ). Cavallaro F., Ciraolo L., A multicriteria approach to evaluate wind energy plants on an Italian island, Energy Policy, Volume 33, Issue 2, Pages Coplin W., O'Leary M., Introduction to political risk analysis, Policy Studies Associates, New York, USA. Doukas H., Karakosta C., Psarras J., RES technology transfer within the new climate regime: a helicopter view under the CDM, Renewable and Sustainable Energy Reviews, 13 (5), pp Doukas H., Karakosta C., Psarras J., A Linguistic TOPSIS Model to Evaluate the Sustainability of Renewable Energy Options. Energy Efficiency, Environmental Performance and Sustainability The International Journal of Global Energy Issues (IJGEI), 32(1/2): Doukas H., Karakosta C., Psarras J., Computing with Words to Assess the Sustainability of Renewable Energy Options. Expert Systems With Applications (ESWA) 37(7): EC - European Commission, Directive 2009/28/EC of the European Parliament and of the Council of 23 April 2009 on the promotion of the use of energy from renewable sources and amending and subsequently repealing Directives 2001/77/EC and 2003/30/EC. Official Journal of the European Union Belgium. Greco S., Mousseau V., Słowiński R., Ordinal regression revisited: Multiple criteria ranking using a set of additive value functions, European Journal of Operational Research, 191 (2), pp Greco S., Słowiński R., Figueira J., Mousseau V., Robust ordinal regression, in: M. Ehrgott, S. Greco, and J. Figueira (eds.), Trends in multiple criteria decision analysis, Springer, Berlin. Grigoroudis E., Siskos Y., Customer satisfaction evaluation, Springer, New York, USA. 191

193 Haurant P., Oberti P., Muselli M., Multicriteria selection aiding related to photovoltaic plants on farming fields on Corsica island: A real case study using the ELECTRE outranking framework, Energy Policy, Volume 39, Issue 2, Pages Hawila D., Mondal M.A.H., Kennedy S., Mezher T., Renewable energy readiness assessment for North African countries, Renewable and Sustainable Energy Reviews, Volume 33, Pages Hurson Ch., Siskos Y., A synergy of multicriteria techniques to assess additive value models, European Journal of Operational Research, Volume 238, Issue 2, Pages Jacquet-Lagrèze E. and Siskos J., Assessing a set of additive utility functions for multicriteria decision making, European Journal of Operational Research, 10 (2): Jacquet-Lagrèze E. and Siskos J., Preference disaggregation: 20 years of MCDA experience, European Journal of Operational Research, 130: Karakosta C., Marinakis V., Psarras J., RES Cooperation Opportunities between EU and MENA Countries: The Case of Morocco. Energy Strategy Reviews, 2(1): Karakosta C., Psarras J., Understanding CDM Potential in the Mediterranean Basin: A Country Assessment of Egypt and Morocco. Energy Policy, 60: Komendantova N., Patt A., Barras L., Battaglini A., Perception of risks in renewable energy projects: the case of concentrated solar power in North Africa, Energy Policy, 40, pp Komendantova N., Patt A., Williges K., Solar power investment in North Africa: Reducing perceived risks Review Article, Renewable and Sustainable Energy Reviews, Vol. 15, Issue 9, Pages Pohekar S.D., Ramachandran M., Application of multi-criteria decision making to sustainable energy planning A review, Renewable and Sustainable Energy Reviews, 8 (4), pp Rosso M., Bottero M., Pomarico S., La Ferlita S., Comino E., Integrating multicriteria evaluation and stakeholders analysis for assessing hydropower projects, Energy Policy, Volume 67, pages Siskos J. and Yannakopoulos D., UTASTAR, and ordinal regression method for building additive value functions, Investigacao Operacional, Vol. 5, No. 1, pp

194 Sustainable Food Security: A System Dynamics Decision-Making Methodology Keramydas Ch. Department of Mechanical Engineering, Aristotle University of Thessaloniki, University Campus P.O. Box 461, Thessaloniki, Greece, chkeramy@auth.gr Tsolakis N. Department of Mechanical Engineering, Aristotle University of Thessaloniki, University Campus P.O. Box 461, Thessaloniki, Greece. Vlachos D. Department of Mechanical Engineering, Aristotle University of Thessaloniki, University Campus P.O. Box 461, Thessaloniki, Greece. Iakovou E. Department of Mechanical Engineering, Aristotle University of Thessaloniki, University Campus P.O. Box 461, Thessaloniki, Greece. Abstract In this research we first discuss the role of Small Farms (SFs) in enhancing food security focusing in developed countries. We then present a more generic framework for Small Farms policymaking, which embraces sustainability and food security aspects. A System Dynamics methodology is employed that captures the effect of regulatory interventions on the diffusion of SFs products by consumers. Keywords: Small Farms, Food Security, Sustainability, System Dynamics, Agrifood Supply Chains 1. Introduction A major concern of modern world relates to the sustainability of food systems (UN, 2013). This concern stems from the projections that indicate a global population growth to 9.1 billion in 2050 with a corresponding increase in food demand by 70% (FAO, 2009). Nevertheless, the challenge that the agricultural sector faces is not the capability to increase food production capacity by 70% within the forthcoming 40 years, but to make 70% more food available for households (FAO, 2009) and to meet the diverse eating and dietary habits in the developed world (Lin et al., 2014). Despite the fact that agrifood production and distribution systems have been extensively investigated thus far, major problems still exist. For example, estimations suggest that nearly 870 million people around the world suffer from undernourishment or chronic hunger (FAO, 2013), about 2 billion people suffer from micronutrient insufficiency or hidden hunger (FAO, 2012) and 193

195 approximately 500 million people suffer from obesity (WHO, 2011). Hence, such records have positioned food security to the top of the public agenda not only in developing countries, but also in developed countries. In addition, the ignorance of specific food related incidents may be detrimental for the industrialized nations. Indicatively, incidents like the: (i) sudden and sharp increase in world food prices in due to climate change, (ii) oil shortages, (iii) increased use of biofuels, (iv) trade embargos, and (iv) rapidly growing food demand in China and India, have challenged the ability of the developed world to maintain adequate food supply (Hubbard and Hubbard, 2013). To that end, Small Farms are suggested to be the key to resolve the developed world s food security and sustainability challenges (Eurovia, 2013). However, an integrated framework that could assist the assessment of the SFs impact upon the triple helix of sustainability (i.e. economic, environmental and social) in the developed world does not yet exist. Such a quantitative model would be of great interest to policy-makers and enterprises in order to support related decisions and/or regulations towards SFs development. In this work, the diffusion of agricultural commodities produced in SFs is particularly addressed. The objective of this study is two-fold: (i) to provide a policy-making support tool at the strategic level, and (ii) to identify policies that could support the development of SFs for ensuring food security in a sustainability context. The rest of the paper is organized as follows. First, in Section 2 generic characteristics that highlight the significance of SFs towards sustainable food security is provided. Following, in Section 3 a System Dynamics (SD) modeling framework is developed for managing the adoption of SF products by consumers, while incorporating an extension of the Bass diffusion model. The application of the proposed framework is further illustrated on the real-world case study of Greece, and interesting policy-making interventions are analyzed. Finally, in Section 4 conclusions and suggested areas for future research are discussed. 2. Sustainable food security and small farms: theoretical background There is a myriad of reports highlighting the challenges of sustainable food security in the developing countries. Nevertheless, this issue has been ineffectively tackled for the case of the developed countries. Economic Challenges: In the develop world, the continuously rising food prices along with the low economic recovery-rates from the global financial crisis highlight potential food insecurity. Specifically, in Europe citizens spend one fifth of their income on food supplies, thus further deepening social inequality in the region (FAO, 2011). Moreover, Europe is not food sufficient considering that it suffers from 70% protein deficiency. This means that the European rural development policy-makers need to promote the domestic production of protein crops at the expense of other arable crops (Noleppa and Cartsburg, 2013). Moreover, in Europe a 2% decrease in total agricultural output has been observed during the last decade (ETH, 2014), while in the United States a 20% decline in farm holdings is reported (USDA, 2005). Environmental Challenges: Loss of biodiversity, combined with water scarcity from overuse, soil erosion and depletion as well as climate change may reduce agricultural yield by at least 5-25% by 2050 (Dimitri, Effland and Conklin, 2009). Additionally, every year 1.3 billion tons of food are wasted globally and in the industrialized world over 40% of this wastage occurs at the retail and consumer levels (NRDC, 2012). 194

196 Social Challenges: In the developed nations live about 15.7 million people who suffer from chronic hunger and undernourishment (FAO, 2013). The most worrying fact is that the rate of chronic hunger and undernourishment in developed countries has risen since 1990 by 15.7% (FAO, 2013). Furthermore, obesity almost doubled between 1980 and 2008 globally and in Europe nearly 50% of both men and women are overweight (WHO, 2008). Also, statistics highlight that in Europe 1 in 3 of 11-year-old children are overweight (WHO, 2008). Another concerning fact is the ageing population in Europe (Eurobarometer, 2012), that increases the appetite for a diet rich in carbohydrates and animal protein (ETH, 2014). Taking into account the aforesaid challenges, a radical transformation is required to promote sustainable agricultural intensification (UNEP/IFAD, 2013). To that end, the need to enhance the role of smallholder farming towards food production and natural resource stewardship is critical. This is further highlighted by the contemporary funding schemes of the European Union (EU) to support initiatives for the development of short food supply chains and local food systems in the Community (EAFRD, 2013). In addition, the EU recognizes the significance of SFs in food and nutrition security and has already approved funds to support small food businesses (European Commission, 2013). Europe hosts around 14 million farms with the SFs to account for 2.5% of the total used agricultural area (EU, 2013). Nevertheless, SFs in Europe disappear; in Europe around 3 million farms (20% of the total number of farms) have disappeared during the last eight years, mainly SFs (Eurovia, 2013). However, SFs could stimulate local business and job creation (Diao et al., 2007). Furthermore, SFs have been found to reduce poverty gap more intensively than other sectors (Christiaensen, Demery and Kuhl, 2011). Additionally, SFs are reported to promote welfare through effective nutrition intake (Faber and Wenhold, 2007). Smallholdings are also documented to be more resource-efficient (Altieri and Koohafkan, 2008), and more productive per hectare than large-scale plantations (Borras, Kay and Akram-Lodhi, 2007). 3. System dynamics framework In this section, a novel modeling approach for managing the diffusion of commodities produced in SFs is developed, merging: (i) the theory of new product diffusion adapted from the field of marketing, and (ii) the theory of SD, which has a proven track record for tackling strategic decision-making problems. The main goals of the model are to (a) study the adoption of SFs products by consumers, (b) predict the SFs evolution during a given time horizon, and (c) evaluate the impact of alternative policy interventions on SFs. The model is based on an extension of the Bass Diffusion Equation (Bass, Trichy, and Dipak, 1994): dn dt = {[p+ q N(t)] [m-n(t)]} x(t) m where, N(t) is the cumulative number of SFs products consumers (adopters) in a specific geographical region at time t, dn/dt is the rate of change for the SFs products consumers (adopters) at time t, m stands for the sum of potential consumers (fraction of regional or national 195

197 population), p is the coefficient of innovation, q is the coefficient of imitation, and x(t) is an general intervention function that describes the current effect of the time-dependent external decision variables on the probability of SFs product consumption at time t. Figure 1 presents a simplified conceptual model of the system under study. The input parameters include the innovation and imitation factors, p and q, that represent the corresponding trends for the first time consumers (innovators) of SFs products, and the word-of-mouth effect, i.e. consumers that purchase the product after being influenced by previous adopters (imitators), respectively. Moreover, the utilized agricultural area (UAA) that refers to the land used for farming in a given region is used as a measure of the aggregate large and small farms size, while the standard gross margin (SGM), that is the difference between the value of the agricultural output (crops or livestock) and the cost of inputs required to produce that output, is used as a measure of the economic size of the agricultural holdings. The labor force employed in these farms is also an input parameter, as well as the productivity factor (α) that express the ratio of small farms to large farms productivity. Finally, the total population of a given region (P), and the target population, i.e. the total market size/sum of potential consumers, are also critical inputs for the diffusion model. The data regarding the aforementioned input parameters were obtained through Eurostat, FAO, and similar systems. The outcome variables of the model, which stand as the system s key performance and monitoring indicators, are the total employees in small farms, and the total profit of small farms across the region under study. The model was applied in the case of Greece. The preliminary results (Figure 2) indicate the positive future potential of small-scale farming, in terms of employment (total SFs employees) and profits (total SFs profits). Indicatively, given that no intervention is applied to accelerate the SFs products market diffusion, the total number of SFs employees and the total SFs profits, are estimated to rise by 45%, and 30% respectively within 20 years. This auspicious prospect could be further accelerated through appropriate interventions on the part of governments and smallholders in order to influence consumers behavior regarding SFs products. Small Farm Total Employees + Small Farms Total Profit + Potential Small Farm Products Consumers Small Farm Products Consumers + Small Farm Products Production B Market Saturation Word-ofmouth R Environmental Awareness + Adoption as Innovators + Social Awareness External Interventions + + Economic Incentives Adoption as Imitators + + Health and Nutrition Figure 18. Conceptual model of the system under study. 196

198 Towards this direction and motivated by the EU s recognition of the need to enhance the role of SFs towards safeguarding sustainable development and food security across Europe, the proposed model allows for the simulated implementation, monitoring, and evaluation of a set of real-world policy-making interventions on the SFs products diffusion rate, through the general intervention function x(t). The scope of these policy-making actions is to increase the rate of consumption of agricultural products cultivated in SFs and consequently increase the SFs market share in the developed countries. Specifically, four external factors that affect the adoption rate of SFs products in the Greek food market were identified: (i) environmental awareness, (ii) social awareness, (iii) economic incentives, and (iv) health and nutrition awareness. Indicative policymaking actions towards the direction of developing SFs products production and consumption include the increase of advertising expenditure in promoting the sustainability aspect of smallscale farming (products and farming practices) (environmental awareness), the market promotion of local products, related training and/or educational interventions, the promotion of healthy food as a critical component of wellbeing (social awareness health and nutrition), and the subsidy of investments in agriculture (new farmers) (economic incentives). 4. Conclusions Figure 2. Total employees and total profit evolution (small farms). Food security has a multidisciplinary nature and it is an important issue not only for the developing world, but also in the developed countries. The prevailing farming practices are unsustainable. Small Farms could play a critical role in Food Supply Chains ensuring Food Security and Sustainable development at the same time. The operation of food markets is quite complex and perhaps, the (mainly social and environmental) merit of a small-scale agricultural production is not enough to build momentum for increase the number of small farms. Thus, (national and international) governance with selected interventions may have a key role. To this end, this paper provides a new quantitative policy-making support tool based on System Dynamics methodology, which has been used for the Greek food market. This is a first time effort and there is space for improvements. Future research directions include further validation and verification of the System Dynamics model based upon data from FAO and Eurostat, the examination of alternative diffusion models, the development of a multi-level model incorporating all the EU-27 countries and different agricultural products on a single-product basis (i.e. cereals, nuts, wheat etc.). 197

199 Acknowledgement This research has received funding from the European Union's Seventh Framework Programme (FP7-REGPOT ) under Grant Agreement No , Project Acronym: GREEN- AgriChains, Project Full Title: Innovation Capacity Building by Strengthening Expertise and Research in the Design, Planning and Operations of Green Agrifood Supply Chains, Project Duration: All the above reflect only the authors views; the European Union is not liable for any use that may be made of the information contained herein. References Altieri M.A. and Koohafkan P. Enduring farms: Climate change, smallholders and traditional farming communities. Environment and Development Series 6. Third World Network, Penang, Borras S.M., Kay C. and Akram-Lodhi A.H. Agrarian reform and rural development: Historical overview and current issues. In A.H. Akram-Lodhi, S.M. Borras Jr and C. Kay, Land, poverty and livelihoods in an era of globalization. Routledge, London and New York, 2007, pp Christiaensen L., Demery L. and Kuhl J. The (evolving) role of agriculture in poverty reduction an empirical perspective. Journal of Development Economics. Vol. 96, 2011, pp Diao X., Hazell P., Resnick D. and Thurlow J. The role of agriculture in development: Implications for sub-saharan Africa IFPRI Research Report No International Food Policy Research Institute. Washington, DC, Dimitri C., Effland A. and Conklin N. The 20 th Century Transformation of U.S. Agriculture and Farm Policy. United States Department of Agriculture. Washington, D.C., EAFRD. Short Food Supply Chains and Local Food Systems in the EU. A State of Play of their Socio-Economic Characteristics. European Agricultural Fund for Rural Development. Seville, ETH. Food Security in Europe in the Age of Global Climatic Change. Available at: < Accessed on June 1st, EU. EU farm economics European Commission Directorate-General for Agriculture and Rural Development. Brussels, Eurobarometer. Special Eurobarometer 378 Active Ageing. European Commission. Luxembourg, European Commission. Sustainable Food Security H2020-SFS Small farms but global markets: the role of small and family farms in food and nutrition security. European Commission. Brussels,

200 Eurovia. Land concentration, land grabbing and people s struggles in Europe. European Coordination Via Campesina. Brussels, Faber M. and Wenhold F. Nutrition in contemporary South Africa. Water SA. Vol. 33, No. 3, 2007, pp FAO. The State of Food Insecurity in the World Eradicating world hunger: taking stock ten years after the World Food Summit. Food and Agriculture Organization of the United Nations. Rome, FAO. How to Feed the World in Food and Agriculture Organization of the United Nations. Rome, FAO. The State of Food Insecurity in the World How does international price volatility affect domestic economies and food security?. Food and Agriculture Organization of the United Nations. Rome, FAO. The State of Food Insecurity in the World Undernourishment around the world. Food and Agriculture Organization of the United Nations. Rome, FAO. The State of Food Insecurity in the World The multiple dimensions of food security. Food and Agriculture Organization of the United Nations. Rome, Foodsecure. Foodsecure briefing for key European commission stakeholders. FOODSECURE. Brussels, Hubbard L.J. and Hubbard C. Food security in the United Kingdom: External supply risks. Food Policy. Vol. 43, 2013, pp Lin B.-H., Ver Ploeg M., Kasteridis P. and Yen S.T. The roles of food prices and food access in determining food purchases of low-income households. Journal of Policy Modeling. In Press, Noleppa S. and Cartsburg M. Agricultural self-sufficiency of the European Union: Statistical evidence. Agripol network for policy advice GbR. Berlin, NRDC. Wasted: How America Is Losing Up to 40 Percent of Its Food from Farm to Fork to Landfill. Natural Resources Defense Council. New York, UN. World Economic and Social Survey 2013: Sustainable Development Challenges. Department of Economic and Social Affairs of the United Nations Secretariat. New York, UNEP/IFAD. Smallholders, food security, and the environment. United Nations Environment Programme/International Fund for Agricultural Development. Kenya/Rome,

201 UNWFP. Emergency Food Security Assessment Handbook. United Nations World Food Programme. Rome, WHO. World health statistics World Health Organization. Geneva, WHO. Global status report on noncommunicable diseases World Health Organization. Geneva,

202 Towards the implementation of optimal train loading plan in the Athens Thessaloniki freight services A. Ballis, Assoc. Professor Department of Transportation Planning and Engineering, National Technical University of Athens, 5, Iroon Polytechniou, GR-15773, Athens, Greece, F. Karapetis, Senior Software Eng. TRAINOSE, Karolou 1 Athens , Greece Th. Ballis Civil/Transportation Eng. National Technical University of Athens, Greece Abstract An effective train loading plan contributes positively to the profitability of the railway services, to train safety, to energy consumption and to the efficiency of rail terminal operations. The main goal of an optimized train loading is the proper assignment of loading units to the wagons of a train so that the utilization of the train is maximized while taking under consideration the maximum axle load restrictions imposed by the design or the condition of the railway infrastructure, operating conditions and safety regulations. The problem typically is expressed in two ways: (a) given a predefined commodity load, which is the minimum number of wagons required to perform the transportation task or (b) given a standard set of wagons (e.g. in the case of a shuttle train) which is the maximum commodity weight or volume that can be transported. The current work describes the way that the train loading plan has been analyzed, solved and integrated in the information system that supports the new railway service of TRAINOSE for container transport in the Athens Thessaloniki line. This new service, named ics, was launched in December 2013 and since then operates on a daily basis. The work includes the literature review on the train loading plan, the pragmatic aspects of the wagon loading problem and the heuristic implemented in the information system of ics service. It also includes the results of the validation process of the above heuristic against an accurate (but more demanding in computational time) mixed integer optimization model. The presentation concludes with the proposed solution for the ics wagon loading problem. Keywords: Train loading plan, wagon loading heuristic, mixed-integer programming. 201

203 1. Introduction An effective train loading plan contributes positively to the profitability of the railway services, to train safety, to energy consumption and to the efficiency of rail terminal operations. The main goal of an optimized train loading plan is the proper allocation of loading units to the wagons so that the utilization of the train is maximized while satisfying a number of restrictions. For the narrow view (optimization of train utilization) typical restrictions are the length of the wagon, the maximum wagon axle load due to railway infrastructure conditions or due to safety regulations, etc. For the wider view of the problem (optimization of the whole system namely, train, terminal and trucks) typical restrictions are the performance of terminal handling equipment, the acceptable average truck waiting time etc. In the current work, the algorithm that performs the container train loading plan for TRAINOSE s intermodal Cargo Shuttle (ics) is presented and validated. The remainder of the paper includes: the literature review (Section 2) where studies addressing the narrow and the wider view of train loading problem are presented. Section 3 that presents the pragmatic aspects of a wagon loading process as well as the algorithm/heuristic developed and used by ics. Next in Section 4 the process for the evaluation of the above heuristic is analyzed. The last part, Section 5, hosts the conclusions. 2. State of the Art The problem of train loading is a subcategory of the known "bin packing" problem. In this problem, objects of different volumes (containers) must be packed into a finite number of bins (wagons) in a way that minimizes the number of bins used. Solving techniques based on mathematical models using integer linear programming or metaheuristics are the most well-known for this class of problems. Examples of the constraints taken into account are (narrow view) the maximum weight attached to wagons, the maximum number of containers per wagon, the number of wagons attached to a train and its total weight as well as (wider view) terminal equipment resources, maximum truck waiting time allowed, etc. One of the first works concerning the narrow view of the problem is the one of Feo and Gonzáles- Velarde (1995) who treated a problem where trailers had to be assigned to wagons. Their models and solutions were based on the assumption that no more than two trailers can fit on one wagon. Nearly a decade later, Corry and Kozan (2006) optimize the load planning with respect to the service time of the train and the weight distribution along the train. Only one type of container and no weight restrictions for the wagons have been taken into account. In a subsequent paper Corry and Kozan (2008) aimed at minimizing the train length and the train service time. Neither weight restrictions for the wagons nor for the whole train have been considered. The model is formulated (but not solved) as an integer linear program and real-world problem instances have been treated by use of heuristics (local search). A recent (2011) work on the subject was this of Aggoun et al (2011) that incorporated into the problem the aspects of business constraints, like handling of dangerous goods and incompatibilities between families of containers. 202

204 As far as the wider view of the train loading problem, Powell and Carvalho (1998) aimed to optimize the circulation of intermodal wagons. The same year, in Bostel and Dejax (1998) the positioning of containers on incoming and outgoing trains is optimized with the aim of reducing transport distances of containers that have to be handled by cranes. Bruns et al. (2010) proposed three different integer linear programming formulations in which the real weight restrictions related to wagon configurations are considered. The resulting train loading plan is calculated so the terminal s operating costs are minimized. Anghinolfi et al. (2012) studied a case with multiple trains of different destinations and took into account the minimization of distances between the stocking area and the train. In 2012, Basetas investigated the spectrum of parameters involved in the problem and performed a comparison among the allocation methods concerning their applicability to the train loading problem. 3. Pragmatic aspects in container wagon loading The problem of train loading plan (in its narrow view) is typically expressed in two ways: (a) given a predefined commodity load, which is the minimum number of wagons required to perform the transportation task or (b) given a standard set of wagons (e.g. in the case of a shuttle train) which is the maximum commodity weight or volume that can be transported. In any case and in order to take into account pragmatic aspects of train loading, the understanding of the mechanics of wagon loading must be acquired. Figure 1 presents the (typically uneven) distribution of forces on the bogies of the wagon. The effect of uneven allocation of forces is explained through an example of good wagon utilization versus bad wagon utilization shown in Figure 2. In this example a 20 tons per axle limit has been assumed, resulting to maximum of 40 tons per bogie. On the left side of Figure 2 the loaded containers exhausts the loading capacity of the left bogie (F1 = 39 tn ~ 40 tn) while the loading capacity of the right bogie remains at 18 tn. This leads to lower wagon utilization by 21 tn. On the contrary, the right side of Figure 2 both forces F3 and F4 are reaching the loading limits of 40 tn, leading to a 79 tn wagon weight. The identification of suitable container combinations (among the set of containers to be transported by the train) can maximize the overall transported commodity weight and therefore the profitability of the transportation service. 203

205 Figure 1: Distribution of the weight of a loaded container into wagon bodies Figure 2: Examples of good and bad wagon utilization 204

206 The Current Train Loading System of TRAINOSE In December 2013 TRAINOSE launched an intermodal Cargo Shuttle (ics) which connects Athens to Thessaloniki on a daily basis. Customers have to enter their orders in the web site of ics (Figure 3) which must respond immediately to inform them if the order can be accepted or not. In order to do this, the program must simulate the train loading process in a way that maximizes the utilization factor of each wagon and provide the answer (to accept or refuse the customer s order) instantly. It must be noted that the above train loading problem is re-solved each time a new order enters the system and for this reason the program code must be lightweight to keep the computational effort low. For this reason the software has incorporated a heuristic based on the Best Fit allocation method (see Figure 4). According to Best Fir rule, each new container is allocated onto the wagon which, after the container loading, has the least remaining capacity. The ics heuristic is used to load 45, 40 and 20 feet containers on a 60 feet wagon used for the ics container transport service. In terms of length, the following cases are investigated: Case (three 20 feet containers loaded on the same wagon) Case 2040 (one 20 feet container along with a 40 feet container on the same wagon) Case 45 (only one 45 feet container on the wagon as the remaining length of 15 feet cannot accommodate any other container) Case x2020 (two 20 feet containers loaded on the center of the wagon. This is technically feasible as the wagons used allow for a container placement starting 10 feet away from the edge of the wagon. The mark {x} indicates a 10 feet space) Case x20x20 (ten feet space, one 20 feet container, ten feet space, 20 feet container) As it is shown in Section 5, the results of the heuristic are not always optimal but the computational effort is always very low. Figure 3: The web based interface of the program First fit: container loaded on the first available slot (leftmost or rightmost side) Next fit: container loaded on the next available slot relatively to the reach stacker s position Best fit: container loaded on the slot which minimizes the wagon s remaining capacity Worst fit: container loaded on the slot which maximizes the wagon s remaining capacity Figure 4: Illustration of Best/Worst/First/Next fit allocation [Source: Basetas (2012)] 205

207 4. Evaluating the efficiency of the loading algorithm The train loading plan, in its narrow view, can be solved to optimality if modeled as a mixed integer optimization problem. The disadvantage of this approach is the high computational power that demands as well as for the need for an (expensive commercial) solver. In the current work, the accuracy of the currently used heuristic was measured by comparing its results with these of the mixed integer programming model using the solver of NEOS (Network- Enabled Optimization System) server (see, Czyzyk et.al. (1998), Dolan (2001) & Gropp et.al. (1997)). The model takes into account wagon loading weight restrictions, the overall train weight and the train length restriction. One aspect that differentiates this model from the work of other researches is that the sequence of servicing the customer orders is retained. Customer orders are served according to the first-in sequence (e.g. an order which was accepted cannot be replaced by a posterior order, even if this increases the overall train utilization). In order to measure the efficiency of the heuristic a series of tests were executed. Each scenario assumes a train with 20 wagons (having 60 feet useful platform length) to be loaded by 35 container orders. Each order concerns a container transport with specific length and weight characteristics. The container length is generated randomly from an empirical {45, 40 and 20 feet container} distribution. The container weight is defined randomly from an empirical {container weight distributions per container type} (see Tournaki (2014)). Chart 1 presents the results after the completion of 800 scenarios, each one solved with ics heuristic and with the mixed integer model. In 257 cases (32%) the heuristic produced the same (optimal) solution with the optimization model. In 424 cases (53%) the heuristic used one extra wagon and in 105 cases (13%) used two extra wagons compared to the mixed-integer model. Finally, for the last 14 cases (2%) the ics heuristic used three more wagons in order to allocate all containers onto wagons. The best solution that combines low computational effort and optimal solution is a suitable combination of both solvers: e.g. in the case of a 20 wagon train, the ics heuristic can be used to allocate containers to the first 16 wagons (20 3 worst difference 1 safety margin) while the mixed integer model will be used to finalize the loading of the latest customer orders (last 4 wagons) rearranging all the available containers to the appropriate wagon positions. Next the wagons are sorted from the heaviest to the lightest to satisfy train breaking rules that require that heavy loaded wagons must be placed near to the locomotive of the train. 206

208 Scenarios Book of Proceedings Additional wagons required by the heuristic Chart 1: Comparison of the best fit heuristic against mixed integer optimization 5. Conclusions The current work presents the wagon loading heuristic of the information system of ics service that performs the simulation of train loading each time a customer enters a new order in the system. The performance of this algorithm was evaluated against the solutions produced by an integer programming model. The analysis concluded that the current algorithm although fast, produces results which in most cases are near optimal. Precisely, the tests concluded that in 32% of the cases the ics heuristic provides optimal results while in 53% of the cases the ics heuristic uses one more wagon to accomplish the allocation of all containers. For the 13% of the cases the gap of the result between the heuristic and the optimization model is two wagons and in some rare cases (2%) the optimization model uses three wagons less in comparison to the heuristic. In order to improve the results while retaining small execution time, a combined approach could probably be applied: The first container orders will be allocated to wagons by using the ics heuristic (in order to benefit from the low computational effort required) while the few latest container orders will be allocated by using the mixed-integer optimization model that provides the optimal solution (rearranging all the available containers to the appropriate wagon positions). Next the wagons are sorted from the heaviest to the lightest to satisfy train breaking rules. References Feo TA, Gonzáles-Velarde JL. The intermodal trailer assignment problem. Transportation Science 29(4), 1995, pp Corry P, Kozan E. An assignment model for dynamic load planning of intermodal trains. Comput. Oper Res 33, 2006, pp.:

209 Corry P, Kozan E. Optimised loading patterns for intermodal trains. OR Spectr 30(4), 2008, pp: Aggoun A., Rhiat A. & Grassien J.P. Online Assignments of Containers to Trains Using Constraint Programming. Proceedings of the 5th International Workshop on Multi-disciplinary Trends in Artificial Intelligence Hyderabad, India, December 7-9, 2011, pp Powell, B. & Carvalho, A. Real-time optimization of containers and flatcars for intermodal operations. Transportation Science, 32, 1998, pp: Bostel N, Dejax P. Models and algorithms for the container allocation problem on trains in a rapid transshipment yard. Transp Sci, 32(4), 1998, pp Bruns F., Knust S. Optimized load planning of trains in intermodal transportation. OR Spectrum, 2010, Published online. Anghinolfi D., Foti L., Maratea M., Paolucci M., Siri S. Optimal loading plan for multiple trains in container terminals. 5th International Workshop on Freight Transportation and Logistics. 2012, Mykonos, Greece. Basetas E. Railway freight transport: Algorithms for the loading of containers onto railway freight wagons, Diploma Thesis, NTUA, Czyzyk J., Mesnier M. P., and Moré J. J. The NEOS Server. IEEE Journal on Computational Science and Engineering 5(3), 1998, pp This paper discusses the design and implementation of the NEOS Server. Dolan E. The NEOS Server 4.0 Administrative Guide. Technical Memorandum ANL/MCS-TM- 250, Mathematics and Computer Science Division, Argonne National Laboratory, (This technical report, which discusses the implementation of the server and its use in detail, is available for download in PDF format). Gropp W. and Moré J. J. Optimization Environments and the NEOS Server. Approximation Theory and Optimization, M. D. Buhmann and A. Iserles, eds., Cambridge University Press, 1997, pp (This paper discusses the NEOS Server as a problem-solving environment that simplifies the formulation of optimization problems and the access to computational resources). Tournaki E. The carbon footprint of rail intermodal freight transport: Case study of Athens- Thessaloniki line, Diploma Thesis, NTUA,

210 Evangelia Chrysochoou An exact method for the inventory routing problem University of Thessaly / Department of Mechanical Engineering Pedion Areos, 38334, Volos Greece Prof. Athanasios Ziliaskopoulos University of Thessaly / Department of Mechanical Engineering Pedion Areos, 38334, Volos Greece of corresponding author: echryso@certh.gr Abstract Vendor inventory management is a concept which is adapted highly nowadays where the decision maker in the process is the supplier. The combination of the inventory management with the vehicle routing problem constitutes one of the latest trends of logistics and supply chain management and constitutes the backbone of vendor managed inventory systems. As new emerging technologies are introduced in the context of freight transportation systems, research requires the development of new models and algorithms that can incorporate their advantages. In this context, this paper aims to discuss all significant elements of inventory routing problem. New valid inequalities are proposed to stronger the formulation of the transported quantities and enhance the Maximum Level (ML) policy. This approach was motivated by the fact that, nowadays where infrastructures were manufactured for much higher consumption rates of goods, retailers are opposed to the Order Up to level (OU) policy and look for more economic and competitive inventory plans. A branch and cut algorithm was developed to solve the problem exactly. In order to evaluate the performance of the algorithm the benchmark instances set for the single vehicle case created by Arhetti et al. (2007) was used. Computational results have shown that this approach improves the optimal solution on an average at least 20%. Keywords: Vendor Managed Inventory, Inventory Routing, Transshipment, Branch and Cut 1. Introduction Vendor Managed Inventory (VMI) systems seem to be one of the most tractable business model nowadays in global logistics and supply chain operations. This is increasingly the case for electronics and automotive parts manufactured in China and assembled in the European Union countries. Most of these parts are assembled in five (5) major plants in Central Europe, operating with Just In Time production procedures, using the VMI principles. The general concept behind this model is that replenishments and distribution making process is centralized at the supplier level. It is characterized as a win win situation for both supplier and manufacturers, or in general retailers due to the fact that it provides the ability to the supplier to combine and coordinate the demand and shipments of a network of retailers (or more generally stock holding entities, such as manufacturers, wholesalers, retailers or 3rd party logistics providers). On the other hand these retailers secure the shortage of their inventories without allocating resources to control and manage them. Backbone of the VMI system is the solution of inventory routing problem (IRP) 209

211 which is one of the most interesting extensions of routing problems. IRP combines the decision process of inventory management and distribution transportation of goods. The decision maker in such a model has to make three decisions: the amount to be transported, the frequency of shipments as well as the distribution plan. However, the IRP in practice becomes meaningful when customers demand is considered to be stochastic instead of assuming a fixed usage rate. The basic difference behind the SIRP and the deterministic IRP is the level of realism and the difficulty of solving instances given the data in a probabilistic sense. In a two stage stochastic program a long term anticipatory decision must be made prior to the full information of the random parameter of the problem and short terms decisions are available as recourse actions once the uncertainty has been revealed. The overall aim is to make here and now a decision which minimizes the total expected cost associated with both the long term and the short term decisions (Carøe and Tind.,1998). IRP was introduced 30 years ago by the seminal paper of Bell et al., (1983) which studied the case with stochastic demand accounting only for transportation costs. They proposed a linear programming model to solve the deterministic version of the problem. To the best of our knowledge there are two very recent literature reviews on the subject. We refer to the work of Andersson et al. (2010) which was focused mostly on industrial aspects and Coelho et al. (2014) which provides the most up to date overview of the problems and methodologies of the VMI problem. Bertazzi, Palettas and Speranza (2002) introduced a practical VMI policy, called deterministic order up to level (OU) policy for the IRP. Based on the proposed policy Arhetti et al. (2007) developed the fist exact algorithm using a branch and cut scheme for the single vehicle. Based on their work very recently Coelho and Laporte (2013) and Adulyasak et al. (2014) have solved multivehicle version of IRP in a branch and cut fashion under OU and maximum level (ML) policies. Solyali and Sural (2011) also based on the work of Arhetti et al. (2007) proposed a strong formulation for the inventory replenishment part of the IRP. In this paper new valid inequalities are introduced to enhance the computational process of the optimal transported quantities under the ML policy. This approach was motivated by the fact that in the context of a deterministic model all parameters are known at the beginning of the process; thus a vendor can take advantage of the fact that the he knows the total demand of each stock keeping venue in advance and can transport quantities in an early stage in order to fulfill the future known demand. However, the amounts that he is able to transport are bounded by the amounts that are made available to him at each stage. This seems to be an important issue for major multinationals that ship parts from China to Europe to be assembled in a number of locations in Central Europe, but also keep inventory either in 3rd party facilities or at the production sites. Transshipment is in fact a recourse action they use in practice in case that there is shortage at a particular venue. These important realizations gave us the motivation to introduce new inequalities in order to enhance the ML policy. The key deference of ML policy in contrast with the OU policy is that the supplier is free to decide about any quantity to be transported to the inventories of his retailers (in fact stock keeping venues) bounded only by their capacity or maximum level defined by them. On the other hand OU policy restricts the amount to be such that fills the inventory to its capacity. However nowadays where infrastructures were manufactured for much higher consumption rates of goods retailers are opposed to the OU policy and look for more economic and competitive inventory plans. A convenient approach to address these particularities is Coelho and Laporte s (2013) proposed new tactical policy, called optimized target level that yields lower cost and inventory levels than the OU policy. Reviewing their approach in comparison to the strong formulation of Solyali and Sural (2011) this paper was motivated to introduce new bound in order to determine optimal quantities to be transported. Therefore, we introduce a modification for the mixed integer 210

212 programming model of the IRP. To the best of our knowledge this assumption was not proposed before. The remainder of this paper is organized as followed. In 2, we give the formal description of the deterministic IPR model and the branch and cut. Computational results are provided in 3. Significant remarks as well as conclusions are given in Inventory routing problem deterministic model We consider an inventory routing problem where a supplier denoted by node 1 is distributed to N- 1 retailers over a finite discrete time T, using a single vehicle of capacity C. Traditionally the problem is defined on an undirected graph G=(V,E) where {1} is the vertex representing the supplier and vertices V = {2,3,,N-1} represent the set of stock keeping venues (will be called retailers from thereafter as it is commonly called in the literature). A = {(i, j): i j, i, jϵv} is the set of arcs. Inventory holding cost occurs for both supplier and set of retailers and is denoted as h i iεv per period and each vertex has and inventory capacityc i iεv. The length of the discrete planning horizon is H where tεt = {1,, H}. At the beginning of the planning horizon the decision maker knows that (1) each period the quantities r t is made available to the supplier in order to fulfill the request of his retailers; (2)the initial inventory levels of both supplier and t retailers are known { I 1 0, I i 0, iεv } (3) of each retailer at each period is denoted with d i t, iεv. A single vehicle can perform the route one at each period with capacity C, and a routing cost c i,j is associated with arc(i, j)ϵa. Throughout the paper we assume that since the supplier has the information of the demand of his retailers in advance he can transport the quantities q i t to meet the demand of period t and subsequent periods as well. However the available quantities r t shall be added to the total available quantities at period t can be used for deliveries to retailer in the same period t and subsequent periods. The objective function is defined in a way to minimize the total transportation and inventory cost of the whole planning horizon while meeting the demand of each retailer. minimize h i I i t tεt iεv t + c ij x i,j tεt iεv iεv i<j (1) Subject to the following constrains: I 1 t 0 t T (2) I t 1 = I t r t t q i t T (3) i V I t i 0 t T, i V (4) I t i = I t 1 i + q t t i d i t T, i V (5) I t i C i t T, i V (6) t q i C t T (7) i V q t t i = d i I 0 i, i V (8) t T t T 211

213 H q t i y t j i d i j=t, i V, t T (9) t t j q i I r j, t T, i V (10) j=1 j=1 t x 1j t T j V H (11) t x 1j j V y 1 t t T (12) t x ij t = x ji t T, i V (13) j V j V t x ij + x t t ji = 2y i t T, i V (14) j V j V t x i,j y t i t T, i, j V (15) t x i,j y t j t T, i, j V (16) C(1 x t i,j ) + u t i u t t j + q j i, j V : i j, t T (17) u i t q i t i, j V, t T (18) u i t y i t C i V, t T (19) y i t y 1 t t T, i V (20) q i t, u i t 0 i V, t T (21) t x i,j {0,1} i, j V : i j, t T (22) y i t {0,1} i V, t T (23) Constraints (2) and (3) are related to the inventory level at the supplier s site. The first one expresses the fact that inventory level at the supplier level cannot be negative in any period, thus avoiding a stock out situation. The second one defines the inventory level at the supplier at the end of period t by the inventory level at the end of period t-1, minus the total quantities to be transported at period t, plus the quantities r t that are made available at time t. Constraint (4) secures the stock out avoidance of each retailer as well. Constraint (5) defines the inventory level at each retailer at the end of period t by the inventory level at the end of period t-1, plus the quantities that is made available at period t the demand at period t as well. Constraint (6) secures that the inventory level of each retailer cannot exceed its capacity. Constraint (7) (10) defined the quantities delivered. These set of constrains are opposed to the OU policy instead they aim to secure the ML policy. More precisely constraint (7) secures that for each period the quantities to distribute cannot exceed the capacity of the vehicle. Constraint (8) declares that the total quantities to be transported to each retailer are equal to the total demand over the whole planning horizon minus the starting inventory 212

214 level. Constraint (9) expresses the fact that the quantities to be transported to each retailer at period t can be less or equal to the demand requested at period t and subsequent periods when the retailer is served at period t. Constraint (10) ensures that the transported quantities at period t cannot exceed the suppliers staring inventory level plus the product made available since period t. Constraints (11) (20) serves the routing counterpart of the problem. More specifically, constraint (11) secures that the total number of routes cannot exceed the number of periods of the planning horizon, however it is not necessary to perform a route for each period. Constraint (12) ensures that if a route is performed at time t it will start from the supplier and will visit only one retailer. Constraints (13) and (14) secure the flow of the route among intermediate retailers. Constraints (15) and (16) define the relationship of the two indexed of the three indexed variables of the routing constrains and stated that when a retailer is served at time t he will be an origin or a destination of a valid path. Constraints (17) (19) is the well known sub tour elimination constrains based on the Miller-Tucker-Zemlin (MTZ) constraint formulation also suggested by Anken et. al. (2012); this is achieved by introducing extra variables u t i that express the quantities that are in the vehicle until retailer i. Constraint (20) secures that if a route is performed at period t, then there will be intermediate points in the route. Constraints (21) (23) enforce integrality and non - negativity conditions. The IRP is NP hard since it contains the VRP as a special case. If the problem size is relatively small the formulation can be solved by the framework of a branch and cut algorithm as follows: Initially at a generic node of the search tree the relaxed linear program defined by the (1) (16) and (20) to (23) is solved. Next a search of violated sub tour elimination constraints (17) (19) is made and sequentially those constraints are generated and introduced to the current problem which is then re - optimized. The process is repeated until a feasible or dominated solution is reached, or until there are no more cuts to added and then branching on fractional variables is performed. 3. Computational results The algorithm described above was coded in C++ using IBM Concert Technology and CPEX 12.4 with 2 threads. All computations were executed in an Intel Atom 1.83 GHz and 2 GB RAM personal laptop with maximum time of 2 hours. To evaluate the performance of the algorithm, we have used the benchmark instances set for the single vehicle case created by Arhetti et al. (2007). Those instances was used to evaluate the performance of the proposed valid inequalities for the ML policy in coherent to the OU policy. The small instances up to 20 customers were used for both high and low level of inventory holding cost. The small number of experiments is indicative in order to present proposed approach potential solutions that yield almost 20% less IRP cost. The computational results are shown in table 1-2. Table 1 provides optimal solution of each of the 5 instances with 5, 10, 15 and 20 retailers. Table 1 contains the results of instances with time horizon H = 3 and high inventory cost ( h i [0.1, 0.5] and h 1 = 0.3) and results with low inventory cost (h i [0.01, 0.05] and h 1 = 0.03). Column 1 shows the corresponding name of the data set, columns 2 3 contain the CPU time (in sec) and the optimal value of the objective function as it was found by Arhetti et al (2007). Columns 4 5 contain the CPU time (in sec) and the optimal value of the objective function of our model, and columns 6 7 contain the difference of the optimal solutions and the percentage of it as well. Analogues remain columns contain the results on low inventory cost. 213

215 High Inventory cost, Horizon = 3 Low Inventory cost, Horizon = 3 Arhetti et.al. Chrysochoou& Ziliaskopoulos z* Arhetti et.al. Chrysochoou & Ziliaskopoulos z* Instances CPU z* CPU z* Diff %Diff CPU z* CPU z* Diff %Diff abs1n5.dat , ,83 13% , ,1 71,58 6% abs2n5.dat , ,7 375,39 19% , ,76 208,9 18% abs3n5.dat , ,3 732,14 22% , ,4 387,3 19% abs4n5.dat , ,8 356,65 18% , ,89 203,5 14% abs5n5.dat , ,4 542,74 23% , , % abs1n10.dat , ,9 1291,7 26% , ,44 40,93 2% abs2n10.dat , ,4 960,78 20% , ,79 367,3 15% abs3n10.dat , ,8 864,03 20% , ,02 297,7 14% abs4n10.dat , % ,31 485,7 22% abs5n10.dat , ,2 1206,5 24% , ,55 437,6 20% abs1n15.dat , ,9 20% , ,18 256,4 12% abs2n15.dat ,4 1227,6 21% , ,03 406,2 16% abs3n15.dat , ,3 22% , ,44 496,6 18% abs4n15.dat , ,3 1144,2 22% , ,16 449,9 19% abs5n15.dat , ,2 1258,7 24% , ,1 539,4 22% abs1n20.dat , ,4 1768,4 24% , ,09 669,2 24% abs2n20.dat ,8 1563,2 21% , ,76 458,1 16% abs3n20.dat ,8 1897,2 24% , ,14 670,5 22% abs4n20.dat , ,2 1480,7 21% , ,13 659,2 20% abs5n20.dat , , % ,41 599,6 18% Table 19 Computational Results on instances with time horizon H = 3 and high and low inventory cost In the set of instances with high inventory cost the average percentage is 21. 3% and yields within the interval ( ) %. However in the second case with low inventory cost the average percentage of improvement on the optimal solution found is 16.7% and yields within the interval ( ) %. This is due to the fact that the transportation cost is higher. Thus our approach can perform significant saving in the cases where the inventory cost is high and competitive to the transportation cost. 214

216 4. Conclusions In this paper the IRP problem was analyzed which constitutes the backbone of the well known VMI systems. New valid inequalities were introduced in order to enhance the performance of the ML policy in contrast to the OU policy which is used in most recent research papers. This approach was motivated by the fact that nowadays retailers are opposed to the OU to level policy and seek for more economic and competitive inventory plans. In the context of a deterministic model all parameters are known at the beginning of the process; thus a vendor can take advantage of the fact that the he knows the total demand of each stock keeping venue in advance and can transport quantities in an early stage in order to fulfill the future known demand. However, the amounts that he is able to transport are bounded by the amounts that are made available to him at each stage. A branch and cut algorithm was developed to solve the problem exactly. In order to o evaluate the performance of the algorithm the benchmark instances set for the single vehicle case created by Arhetti et al. (2007) was used. Computational results have shown that this approach improves the optimal solution on an average at least 20%. Acknowledgement This research has been co-financed by the European Union (European Social Fund-NSF) & Greek national funds through the Operational Program "Education and Lifelong Learning" of the National Strategic Reference Framework (NSRF)-Research Funding Program: Heracleitus II. Investing in knowledge society through the European Social Fund. References Adulyasak,Y., J.-F. Cordeau, R.Jans. Formulations and Branch-and-Cut Algorithms for Multivehicle Production and Inventory Routing Problems. INFORMS Journal on Computing Vol. 26 (1), 2014,pp Aksen D., O. Kaya, F. Salman, and Y. Akça. Selective and periodic inventory routing problem for waste vegetable oil collection. Optimization Letters, Vol. 6(6), 2012, pp Andersson H., A. Hoff, M. Christiansen, G. Hasle, and A. Løkketangen. Industrial aspects and literature survey: Combined inventory management and routing. Computers & Operations Research, Vol. 37(9), 2010,pp Archetti C., L. Bertazzi, G. Laporte, and M. G. Speranza). A branch-and-cut algorithm for a vendor-managed inventory-routing problem. Transportation Science, Vol. 41(3), 2007,pp Bell W. J., L. M. Dalberto, M. L. Fisher, A. J. Greenfield, R. Jaikumar, P. Kedia, R. G. Mack, and P. J. Prutzman. Improving the distribution of industrial gases with an on-line computerized routing and scheduling optimizer. Interfaces, Vol. 13(6), 1983, pp Bertazzi L., G. Paletta, and M. G. Speranza,. Deterministic order-up-to level policies in an inventory routing problem. Transportation Science, Vol. 36(1), 2002, pp Carøe C. C., and J.Tind,. L- shaped decomposition of two stage stochastic programs with integer recourse. Mathematical Programming, Vol. 83, (1-3), 1998, pp Coelho L. C. and G. Laporte. The exact solution of several classes of inventory routing problems. Computers & Operations Research, Vol. 40(2), 2013, pp Coelho L. C. and G. Laporte. An Optimized Target Level Inventory Replenishment Policy for the VMI Systems. 2013, Technical Report CIRRELT , Montreal, Canada. 215

217 Coelho L. C., J.-F. Cordeau, and G. Laporte. Thirty years of Inventory Routing. Transportation Science, Vol. 48(1), 2014, pp Solyalı O.and H. Süral. A branch-and-cut algorithm using a strong formulation and an a priori tour based heuristic for an inventory-routing problem. Transportation Science, Vol. 45(3), 2011, pp

218 Open Governmental data sources in Europe: A comparative evaluation of semantic and technical characteristics Olga Vasileiou National Technical luniversity of Athens Heroon Polytechniou 9, 15780, Zografou, Greece Charilaos Georgis London School of Economics and Political Science Michael Petychakis Decision Support Systems Laboratory, NTUA Heroon Polytechniou 9, 15780, Zografou, Greece Spiros Mouzakitis Decision Support Systems Laboratory, NTUA Heroon Polytechniou 9, 15780, Zografou, Greece, smouzakitis@epu.ntua.gr Dimitris Askounis Decision Support Systems Laboratory, NTUA Heroon Polytechniou 9, 15780, Zografou, Greece Abstract This paper aims to summarize a comparative study of end-user services and technical characteristics of current Open Governmental data sources in Europe. The analysis has been performed in the context of the ENGAGE FP7 e-infrastructures Project. The research began with a snapshot of the current situation of open data sources in Europe. Thereafter, we managed to collect, categorize, statistically analyze and comparatively assess the open government data throughout the European Union. The results of our study show that there is still no uniform policy regarding the provision of public sector information across data sources in the countries of the European Union. The quality of the government data sources varies significantly depending on the country and the data provider. In general, the majority of the datasets is not completely open, as it has been published under restricted or non-specified licenses. Nevertheless, in the recent years there is an increasing effort in the adoption of open licenses; especially in the newly launched national open data portals. The ENGAGE project will constitute the initial test-bed of the results of the work performed. Keywords: Open Data, Infrastructures, governmental data 217

219 1. Introduction Public sector information (PSI) is the single largest source of information in Europe. It is produced and collected by public bodies and includes digital maps, meteorological, legal, traffic, financial, economic and other data. Most of this raw data could be re-used to new products and services, such as car navigation systems, weather forecasts, financial and insurance services. Since the last decade there has been an ever-increasing number of open data initiatives and portals. The European Commission (EC) and national governments all over the world have established of national open data portals in order to increase public access to high value, machine readable datasets generated by public agencies and organizations. In 2003, the EU adopted the Directive on the re-use of public sector information (PSI Directive) that introduced a common legislative framework regulating how public sector bodies should make their information available for re-use in order to remove barriers such as discriminatory practices, monopoly markets and a lack of transparency. During the last years, a vast number of PSI portals and open data communities that develop new ideas and apps have emerged. For the first time in decades the importance of government openness has received attention. Previously, government openness was regarded as a passive provision of information to citizens, while nowadays more proactive approaches of data handling in open formats are preferred (Zuiderwijk, Helbig, Gil-García, & Janssen, 2014; Zuiderwijk & Janssen, 2014). More and more governmental organizations are releasing their data, as can be seen for instance in the United States of America and the United Kingdom. Open government data are expected to bring many advantages. The literature shows that in general potential advantages of big open data can be, for instance, political and social, economic, technical and operational (M. Janssen, Charalabidis, & Zuiderwijk, 2012). Open data provide the potential to unlock business innovation and financial performance (European_Commission, 2011a; K. Janssen, 2011; Jetzek, Avital, & Bjørn-Andersen, 2012; Kundra, 2012; Yang & Kankanhalli, 2013). As confirmed by various studies, proactive release of public and private data may create considerable benefits for businesses, researchers, citizens and other stakeholders. The proliferation of such Open Government Data initiatives and particularly Open Government Data portals during the recent years, however, has raised significant questions in terms of the variety in the interoperability, standards, services, openness and vision of each portal. Both systematic analysis of these portals and comprehensive analysis of the capabilities and potential of these initiatives are hardly available or generally missing from the recent research literature. In order to address this gap, current paper presents our initial findings from the analysis and potential of the European Open Data portals. 2. Methodology More specifically, we focused on the countries of the European Union, where we intended to identify irregularities and challenges with regard to the provision of Open Data across the studied countries. The research began with a snapshot of the current situation of open data sources in 218

220 Europe. Thereafter, we managed to collect, categorize, statistically analyze and comparatively assess the open government data throughout the European Union. Within the current study, a representative number of diverse and distributed open government data sources from all the countries of the European Union were analyzed. Data was collected by means of an online research for organizations and public offices of each country that provide open public data in electronic form, as well as data aggregators. The collected datasets were classified and analyzed by genre and country in view of enabling a qualitative comparison between the EU member states. Our research mainly focused on a central government level, meaning that for each of the 27 EU countries we reviewed every working ministry website with publicly available datasets. Our research was also extended to other central government-related websites, such as national and regional open data portals, public services, national statistical offices, central banks, national geodata-related websites as well as each country s official police, fire service and army website. Furthermore, besides national public datasets, we also included official European Union portals and websites. In addition to this research, which targeted at covering an in-breadth analysis, we also conducted in-depth research for 3 representative countries of the European Union that have exhibited a proliferation of open data initiatives during the last years. For that purpose, we chose the United Kingdom (being a pioneer about open data amongst European countries), France (being a representative country of mainland Western Europe) and Greece (being a country of Eastern Europe). With regard to this research: The United Kingdom demonstrates great differentiation among local administration structures. Unlike other European countries, there is a great number of local administrative sections, a tradition holding since the Middle-Ages. In terms of this research, counties, boroughs and unitary authorities, were regarded as equivalent to the municipalities in the rest of Europe. In summary, we proceeded to the investigation of 98 counties, 61 boroughs, 15 unitary authorities and 60 urban areas (cities/towns). In France, we investigated 95 departments and 11 urban areas (cities/towns) data sources). In Greece, 13 Regions and 325 Municipalities were investigated, as predefined by the Kallikrates program of the Greek Ministry of Internal Affairs (Site 4). We subsequently proceeded to the collection and categorization of the datasets from each data source researched. The process was initiated on March 2012 and was concluded on January In total, 3,466 data portals were analyzed. 219

221 3. Results In this section, the results of the analysis per key attributes are presented. Licenses One of the most important issues of public data is the definition of the license related to it and whether it is open or not. The results showed that the vast majority of the datasets (3,045 or 87.88% of all of the datasets) are published without a clearly defined or open license (License Not Specified or Restricted) while in the case of 64 of them (1.85%) the original providers held the license. On the contrary, only 356 datasets (10.27%) were published with an open license, with most of them under either the UK Open Government License or the Creative Commons Attribution License. It is evident that one of the most crucial steps into making data truly open apply an open license that will support the openness of data- is now lacking in Europe, with the bright exception of the emerging national open data portals. Interface and Data Languages Another point of interest in an ever-diverse and multilingual set of countries such as the European Union is the existence of multilingual support in the data that each provider publishes. More specifically, it was deemed important to separate the case of languages that the user interface, through which the data is provided, is available in from the case of the languages of the data itself. The results of the analysis showed that nearly half of the web interfaces (47.39%) support only the native language of the country they belong to. Additionally, a large percentage of the interfaces (38.18%) is accessible by an additional language, (mainly in English), while 7.07% of them support 3 languages and 7.36% 4 languages or more. The percentages are quite lower with regard to the language of the actual provided datasets. The vast amount (77.80%) of the datasets is available only in their native language. As a result, only 22.19% are available in 2 languages or more, contrary to the UIs where 52.50% of them provide multilingual support. Thus, it is now obvious that the separation of UI and data languages was the only way to showcase the differences between them and to prove that while the former has now reached an acceptable level of multilingual support, the latter remains lacking in that aspect. The most common language used is English. Data Provision This attribute refers to utilities that serve the purpose of providing the data / information to users and applications, either in a human readable format or a machine-processable format. For instance, this attribute indicates whether the data is available for online view only, via a downloadable file or both, as well as the existence of value-added services (charts, maps, APIs). Our analysis 220

222 revealed in absolute numbers the different ways the data is provided in EU websites. According to the analysis, 2542 datasets are provided in downloadable form (73.34%), 1751 datasets are provided through an online view (50.52%), 328 datasets through a map service (9.46%), 293 datasets through charting capabilities (8.54%) and only 37 through an API (1.07%). The above analysis was repeated specifically for the national open data portals in order to highlight the difference in data provision between them and the rest of the public data providers. According to this analysis, in 5 out of the 14 at the time available open government data portals, the user can access datasets through an API; a percentage (35.71%) much higher than the average 1.08%. This fact clearly demonstrates the technical maturity of the new open data portals in contrast to the legacy websites of ministries, municipalities and public agencies. Data Formats The available data representation formats of the published information are one of the key features of open government portals, as this defines the inherent properties of the datasets, their usability and interoperability. In our analysis we portrayed the most prominent data formats encountered in all of the 27 European countries: The largest percentage of the datasets is stored in the PDF (38%) and in the HTML format (28%). In addition, 14% of the datasets are available through RSS feeds, 8% in the XLS/XLSX formats, 7% in the DOC/DOCX formats, 4% in the CSV format and 1% in the KML format. According to the results we clearly conclude that the majority of the data sources do not provide datasets in machine processable-formats that can be directly consumed through applications, thus limiting their usability significantly. The analysis was repeated across the EU countries so as to reach a conclusion about the status of each country in each category under investigation. The countries which have made the greatest progress are the Netherlands (56% of the data sources contain machine processable datasets), Bulgaria (51.61%), Spain (51.45%), United Kingdom (48.95%) and Hungary (41.15%). On the contrary, the countries that are lagging the most are Slovenia (4.18%), Luxembourg (7.46%), Malta (11.95%), Slovakia (15.85%), Germany (23.71%) and Cyprus (24%). Our analysis also illustrates the findings particularly for local administration data sources in the UK, France and Greece where notably few differences were found compared to their national average. In the UK, 43.56% of the data sources contain machine processable datasets (compared to 48.95% of the UK average) and 56.44% contain non machine processable datasets. Meanwhile, in France 31.06% of the data sources provide machine processable datasets (compared to 32.67% of the French average) while 68.94% do not. Finally, in Greece 25.67% of the datasets are machine processable (compared to 27.33% of the Greek average) and 74.33% are not. Similarly to open licenses and multilingual support, it is apparent that local administration data infrastructure in each country is lacking compared to that of the central administration. 221

223 4. Conclusions The aim of this paper was to investigate and provide an insight into the current public data infrastructures in the European Union. For this purpose, our research included both central and local government data sources as well as official EU portals and websites in order to ensure that the data sources investigated complement each other and provide the full picture of the current public data landscape. The results of our study show that there is still no uniform policy regarding the provision of public sector information across data sources in the countries of the European Union. The quality of the government data sources varies significantly depending on the country and the data provider. In general, the majority of the datasets is not completely open, as it has been published under restricted or non-specified licenses. Nevertheless, in the recent years there is an increasing effort in the adoption of open licenses, especially in the newly launched national open data portals. In terms of multilingual support, only 22% of the actual datasets are available in more than one language, whereas in the case of user-interfaces of the data portals (static website text), 52% of them support multiple languages. The discrepancy between the two cases is expected, given the fact that the task of translating the rapidly growing volume of information published by each data provider in more languages other than the original is challenging. Hence, there is a notable difficulty faced by researchers or citizens to access and utilize foreign datasets. Moreover, most data portals provide the ability to search data only through browsing of categories and simple text search, rarely supporting semantic search with the bright exception of open government data portals. Thus, the low percentage of SPARQL and CKAN searches as well as the small number of cases where data is provided through an API, clearly indicates that currently there is low usage of Linked Data and Semantic Web technologies. Furthermore, most datasets are published in nonmachine processable formats, rendering their technical re- use demanding. This is evident in the case of ministry and other public administration websites, where a simple publication of the data is sufficient, as opposed to the national portals where the availability of technically re-usable formats and semantic interoperability is also a concern. Despite these shortcomings, it should be noted that the quality of open government infrastructures is steadily improving. Particularly, throughout the EU there is an ever-growing trend of countries, cities and regions towards launching official open data portals where data is published under universally open standards. The United Kingdom was found to be the leading country in that trend, but also France, Austria, Italy, Spain, Germany, Belgium, Portugal, Estonia and the Netherlands have launched their own national and regional open data portals. It is expected that even more countries are going to adopt open government policies bringing together considerable advancements in the following years. 222

224 Acknowledgement This work has been partly funded by the European Commission through the Project ENGAGE (An Infrastructure for Open, Linked Governmental Data Provision towards Research Communities and Citizens). References Book ENGAGE, Deliverable Analysis Report of Public Sector Data and Knowledge Sources United Nations Data, UN statistical databases, Papers J. C. Bertot, P. T. Jaeger, S. Munson and, T. Glaisyer, Social media technology and government transparency, Computer, vol. 43, no 11, pp , A. Burton, D. Groenewegen, C. Love, A. Treloar and, R. Wilkinson, Making research data available in Australia, Intelligent Systems, IEEE, vol. 27, no. 3, pp , Carte de France [online], Available at: A. Cordella and F. Iannacci, Information systems in the public sector: The e-government enactment framework, The Journal of Strategic Information Systems, vol. 19, no. 1, pp , L. Ding, T. Lebo, J. S. Erickson, D. DiFranzo, G. T. Williams, X. Li, J. Michaelis, A. Graves, J. G. Zheng, Z. Shangguan, J. Flores and, J. A. H. Deborah L. McGuinness, TWC LOGD: A portal for linked open government data ecosystems, Web Semantics: Science, Services and Agents on the World Wide Web, vol. 9, no. 3, p , European Commission. (2012, October) Directive 2003/9 8/EC of parliament and council on the re-use of public sector information. European Commision. [Online]. Available: ctive_en.pdf M. B. Gurstein, Open data: Empowering the empowered or effective data use for everyone?, First Monday, vol. 16, no. 2, pp. 2-7, B. Hogge. (2010, May) Transparency accountability initiative, Open data study, [Online]. Available: ions/publications/open-data-study

225 A branch and price solution algorithm for the tail assignment problem George Kozanidis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece, Elina Gioti Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Abstract We consider the tail assignment problem, i.e., the problem of assigning a set of passenger flights, which must be carried out by a commercial fleet, to specific aircraft. Each flight has a fixed departure time, as well as an origin and a destination airport, which, combined, determine the duration of the associated trip. The aim is to schedule all the flights, while also minimizing the number of aircraft utilized for this purpose. Motivated from theory that has been developed in the past, we develop an integer programming formulation and a branch and price solution algorithm for this problem. Keywords: aircraft scheduling, integer programming, column generation, branch & price. 1. Introduction We consider the tail assignment problem, i.e., the problem of assigning a set of passenger flights, which must be carried out by a commercial fleet, to specific aircraft. Each flight has a fixed departure time, as well as an origin and a destination airport, which, combined, determine the duration of the associated trip. The aim is to schedule all the flights, while also minimizing the number of aircraft utilized for this purpose. Motivated from theory that has been developed in the past (Grönkvist, 2005), we develop an integer programming formulation and a branch and price solution algorithm for this problem. The proposed methodology utilizes a master problem that tries to schedule the maximum possible number of flights using a set of aircraft-routes, and a column generation (colgen) subproblem that generates cost-effective aircraft-routes which are fed into the master problem. Due to the huge number of alternative aircraft-routes, the master problem minimizes the number of aircraft utilized to operate the flights, using only a small subset of these routes. At each iteration, the column generation sub-problem uses dual information obtained from the optimal solution to the master problem s linear relaxation, in order to generate the most cost-effective (the one with the minimum reduced cost) aircraft-route, out of those that have not been considered yet. This route is then added to the master problem. The optimal solution of the column generation subproblem is obtained with an efficient network optimization solution algorithm, which outperforms 224

226 existing commercial optimization software packages that can be utilized for the same purpose alternatively. The procedure continues similarly, until the optimal solution to the linear relaxation of the master problem is obtained. This happens when no other aircraft-route with negative reduced cost can be identified. In order to eliminate the non-integralities, the algorithm creates next a branch and price tree by branching on the fractional decision variables of this solution. Additional decision variables representing aircraft-routes are generated during this phase, due to the integrality restrictions that are gradually introduced. The remainder of this work is structured as follows. In Section 2, we introduce the two optimization models that the proposed methodology utilizes. In Section 3, we develop the proposed solution algorithm, and finally, in Section 4 we summarize the present work and we point to promising directions for future research. 2. Problem Formulation In this section, we present the model formulation of the master problem and that of its column generation counterpart. Both these models utilize the following two common sets: I : set of aircraft, S : set of flights. Additional notation specific to each of the two formulations is defined in each corresponding part. 2.1 Master problem formulation For the formulation of the master problem, we introduce the following mathematical notation: Sets: Ri : set of routes of aircraft i, Parameters: f : cost for each aircraft utilized, h : cost for each flight that remains uncovered, aijs : binary parameter that takes the value 1 if route j of aircraft i covers flight s, and 0 otherwise, ii, jri, ss, Decision Variables: xij : binary decision variable that takes the value 1 if route j of aircraft i is scheduled, and 0 otherwise, ii, jri, ys : binary decision variable that takes the value 1 if flight s remains uncovered, and 0 otherwise, ss. Utilizing this notation, the master problem is formulated as follows: ij hys (19) ii jr i ss Min fx + s.t. xij 1, i I (20) jri 225

227 ys aijs xij 1, s S (21) ii jr i x, y binary, i, j, s (22) ij s The objective function (1) minimizes the total cost, which comprises of the aircraft utilization cost and the cost of the uncovered flights. Cost coefficient h is always much larger than cost coefficient f, imposing the relative priority between the two objectives. Constraint set (2) ensures that at most one route is scheduled for each aircraft. In conjunction with the objective coefficient of variables xij, a cost equal to f is imposed this way, for each aircraft utilized. We call these constraints the aircraft-rows. Constraint set (3) states that each shipment, s, is either covered by exactly one aircraft-route, or remains uncovered (variable ys is equal to 1 instead), in which case the corresponding penalty h is imposed in the objective. We call these constraints the flight-rows. Finally, constraint set (4) restricts the decision variables of the problem to binary values. Clearly, any two distinct flights covered by the same aircraft-route must be temporally nonoverlapping. Additionally, economic efficiency reasons dictate that an aircraft is not allowed to travel empty. This implies that for any pair of consecutive flights covered by the same aircraft, the arrival airport of the preceding one must coincide with the departure airport of the succeeding one. These, as well as several other rules that the generated aircraft-routes must abide by, are incorporated into the colgen sub-problem formulation which is presented next. 2.2 Column generation sub-problem formulation The routes that are candidate to enter the master problem are ranked in terms of their reduced-cost with respect to the optimal solution of the current master LP relaxation. The aim of the colgen subproblem is to identify the aircraft-route with the minimum reduced cost. If this reduced-cost is negative, this is an indication that the associated aircraft-route has the potential to improve this solution; therefore, it is added to the master problem, and the new optimal master LP dual solution is updated. If not, this implies that no other aircraft-route can improve the optimal solution of the current master LP relaxation; therefore, the column generation procedure terminates. Let dls, als, dts and ats be the departure airport, the arrival airport, the departure time, and the arrival time of flight s, respectively. For each flight ss, we define two sets, as explained next. Ns is the set of flights which are next-compatible with shipment s, while Ps is the set of flights which are previous-compatible with flight s. A flight s is next-compatible with flight s if als = dls and ats < dts. A flight s is previous-compatible with flight s, if flight s is next-compatible with flight s. Let also li be the airport at which aircraft i is currently located. We also define two additional sets. For each aircraft ii, Fi is the set of flights which are next-compatible with aircraft i, while for each flight ss, Vs is the set of aircraft which are previous-compatible with flight s. A flight s is next-compatible with aircraft i if li = dls. An aircraft i is previous-compatible with flight s if s is next-compatible with i. For the formulation of the colgen sub-problem, we consider a network N = {A,V}. The set of vertices, V, includes one node for each aircraft, one node for each flight, as well as one node E, which acts as the terminal node. This latter node is fictitious because the duty of each aircraft stops upon completion of the last flight. The set of arcs, A, comprises of the edges which connect aircraft 226

228 nodes with nodes that correspond to next-compatible flights for the associated aircraft, the edges which connect pairs of nodes that correspond to compatible flights, as well as one edge for each flight that connects the node that corresponds to this flight with the terminal node. The aim of the colgen sub-problem is to identify the longest (minimum negative-distance, to be precise) path in this network that begins in one of the aircraft nodes, visits at least one flight node, and ends in the terminal node. The length of any path is equal to f+ci d, where i is the index of the aircraft associated with the node this path begins from, C Sis the set of flight nodes that this path visits, and ci/ds is the dual variable of the corresponding aircraft/flight row in the current master LP optimal solution. Since the cost of any aircraft route in the master problem is equal to f, this is equivalent to finding the aircraft-route with the minimum reduced-cost. With these in mind, we introduce the following additional mathematical notation for the colgen sub-problem: Sets: Fi : set of flights which are next-compatible with aircraft i, ii, Vs : set of aircraft which are previous-compatible with flight s, ss, Ns : set of flights which are next-compatible with flight s, ss, Ps : set of flights which are previous-compatible with flight s, ss, Parameters: ci : dual value of aircraft row i in current master LP optimal solution, ii, ds : dual value of flight row s in current master LP optimal solution, ss, Decision Variables: zi : binary decision variable that takes the value 1 if the generated route utilizes aircraft i, and 0 otherwise, ii, xis : binary decision variable that takes the value 1 if the generated route includes a direct travel from aircraft-node i to flight-node s, and 0 otherwise, ii, sfi, xst : binary decision variable that takes the value 1 if the generated route includes a direct travel from flight-node s to node t, and 0 otherwise, where t is either a flight node, or the terminal node, ss, t N { E}, s ys : binary decision variable that takes the value 1 if the generated route covers flight s, and 0 otherwise, ss. Utilizing the above notation, the colgen sub-problem is formulated as follows: sc Min f ci zi ds ys ii ss s (23) s.t. zi 1 (24) ii zi xis, i I (25) sfi x x x, s S (26) is rs st iv s rp s tn s{ E} (27) y x x, s S s is rs iv s rp s 227

229 z, x, x, x, y binary, i, s, r, t (28) i is rs st s The objective function (5) minimizes the reduced-cost of the aircraft-route that will be identified, which is equal to the fixed cost, f, of each aircraft-route, minus the dual of the aircraft this route pertains to, minus the sum of the duals of the flights that this route covers. Constraint (6) ensures that this route utilizes exactly one aircraft. Constraint set (7) states that the selected aircraft should visit a node that corresponds to one of its next-compatible flights first. Constraint set (8) ensures the flow balance at each flight-node. Incoming flow can originate either at an aircraft-node or at a flight-node, while outgoing flow can be directed either to a flight-node or to the terminal node. Constraint (9) states that a flight is covered if and only if there is incoming flow in the corresponding flight-node. Finally, constraint set (10) imposes integrality on the decision variables. 3. Solution Methodology 3.1 Solving the master LP relaxation using column generation Each node of the branch and price tree is associated with a distinct master problem and its companion colgen sub-problem, which are based upon the two fundamental formulations (1)-(4) and (5)-(10), respectively. Two distinct master or colgen problems differ from each other only with respect to the additional constraints that have been added as a result of branching. The optimal solution of each master LP relaxation is obtained with column generation, according to the logicflow shown in Figure 1. Solve restricted master LP problem Update dual values in objective function of colgen sub-problem Add this vehicle-route to the master problem Solve colgen sub-problem to find vehicle-route with minimum reduced-cost Is the reduced-cost of the identified vehicle-route negative? YES 3.2 Solving the colgen sub-problem NO DONE Figure 1: Column generation logic-flow. Instead of using commercial optimization software for solving each colgen sub-problem, it is far more efficient to utilize a modification of the shortest path algorithm of Desrochers and Soumis (1988) in order to find the aircraft-route with the minimum reduced-cost, taking advantage of the 228

230 fact that the associated network is acyclic. This algorithm scans the network nodes in topological order finding possible path extensions for each node through its next-compatible nodes, and updates their corresponding distances accordingly. For each node, the algorithm stores a label denoting the best path distance of this node from the source, which is updated accordingly each time that an improved path length is discovered. Naturally, since the complexity of the algorithm is linear in the number of network arcs, its performance is significantly superior to that of commercial optimization software packages that can be alternatively utilized for solving the colgen sub-problem, which is an integer program. Moreover, the additional constraints that are added as a result of branching can be incorporated rather easily into this algorithm, making it possible to find the optimal integer solution of the problem without resorting to generic optimization software for the solution of colgen. 3.3 Branching When the optimal solution to the currently explored master LP relaxation is fractional, the algorithm performs branching in order to continue the search for the optimal integer solution. As in the case of a typical branch and bound solution algorithm, new sub-problems are created through the addition of constraints that eliminate fractional solutions. A typical design involves the selection of one such fractional decision variable for branching, and the partition of the solution space by setting this variable equal to 0 and 1. Rather than appending the associated branching constraints to the master problem, we incorporate them directly into the existing formulation instead. The benefit of doing this is that it retains the same number of master problem constraints, and thus we do not have to deal with extra dual variables. 4. Summary and Future Work In this work, we have presented a solution algorithm for the tail assignment problem, i.e., the problem of assigning a set of passenger flights to specific aircraft. The proposed methodology utilizes a branch and price tree, at each node of which a linear problem comprising a suitable modification of the original formulation is solved with column generation. When the problem is too large and cannot be handled efficiently, the user must inevitably compromise for a near-optimal solution instead of the exactly optimal one. The most common technique for achieving this is the incorporation of tolerances on the optimal objective. When such tolerances are present, the algorithm does not backtrack to nodes created earlier in the search tree, unless these tolerances are violated. In any other case, the algorithm continues its dive deeper, which makes it easier to obtain faster an integer solution. Through the choice of the tolerance values and the relaxation bounds on the optimal objective, the user can control how close this solution will be to the truly optimal one, and may select to interrupt the execution of the algorithm before its termination if the quality of the best integer solution that has been found so far is acceptable. Future research should be directed towards the development of similar enhancements that can be incorporated into the design of the proposed solution algorithm, in order to enable the more efficient treatment of large scale problems. 229

231 References Desrochers, M., and F. Soumis. A Generalized Permanent Labelling Algorithm for the Shortest Path Problem with Time Windows. INFOR, Vol. 26(3), 1988, pp M. Grönkvist. The Tail Assignment Problem. PhD Dissertation, Göteborg University, Department of Computer Science and Engineering,

232 A multi-stage column generation solution approach for the bidline aircrew scheduling problem Panagiotis Andrianesis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. George Kozanidis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece, Abstract We consider the bidline scheduling problem that typically arises in the commercial airline industry, i.e., the problem of generating anonymous duty lines, which will be subsequently matched to specific aircrew members, according to their seniority and preferences. Each duty line consists of duty and rest periods beginning and ending at the same crew-base, typically called pairings. It must abide by certain safety and collective agreement rules, and the credited hours (i.e., the hours for which aircrew members will be paid) it contains must fall within a certain interval. The generated bidlines must collectively cover a given set of pairings; at the same time, they should also satisfy, to the greatest extent possible, a given set of quality criteria. Such criteria include purity (the degree to which the bidline contains routes of the same type), regularity (the degree to which the duties/offs are repeated in a specific pattern), hour balance (the degree to which the number of credited hours approaches a desired target), as well as several other criteria related to particular characteristics (e.g., number of single duties/offs, etc.). For this problem, we develop an integer programming formulation, and a multi-stage branch and price solution methodology. The first stage aims at generating a large number of high quality duty lines that satisfy a certain quality threshold. The second stage aims at generating additional duty lines so as to cover the flights that remain uncovered at the end of the first stage, without any special concern for their quality. Finally, the last stage aims at further improving the quality of the combined solution obtained by the first two stages. The proposed solution methodology utilizes a master problem that tries to optimize some appropriate measure of performance using a given set of duty lines, and a column generation subproblem that generates cost-effective duty lines which are fed into the master problem. We describe the steps of the proposed solution algorithm, focusing on novel formulations for the master problem. We supplement our analysis with numerical results demonstrating the performance of the algorithm on a medium-sized airline company. Keywords: aircrew scheduling, bidline, column generation. 231

233 1. Introduction The airline crew scheduling problem is one of the most important problems in the airline industry, due to the fact that the involved cost is the second largest following the fuel cost. The objective of the problem is to determine schedules (typically over a monthly period) for the crewmembers, given the flight schedule of one or more aircraft types. Traditionally, the crew scheduling problem is solved in two steps. Firstly, the crew pairing problem is solved, which creates sequences of duties and rest periods that start and end at the crewmember s base (pairings). Secondly, the crew assignment problem is solved to assign crewmembers to given pairings. Recently, integrated approaches have appeared in the literature (Saddoune et al. 2012). The main approaches that airlines employ for solving the crew assignment problem can be classified in three categories: (a) Bidline scheduling, where anonymous monthly schedules are produced, and the crewmembers bid on them (Boubaker et al., 2010). (b) Rostering, where personalized schedules are produced that take into account preassignments and preferences. (c) Preferential bidding, which is similar to rostering, but accounts for seniority, too. In this paper, we consider the bidline scheduling problem, which is a typical problem encountered by many US airline companies. The remainder of the paper is structured as follows. In Section 2, we formulate the bidline problem, and in Section 3, we present the proposed methodology. Lastly, in Section 4, we present numerical results for a medium-sized airline company, and provide directions for further research. 2. The bidline problem Given a set of pairings produced by the solution of the crew pairing problem, the bidline scheduling problem aims at finding bidlines, which must collectively cover this set of pairings and have the following characteristics: (a) The credited hours that the bidline contains must fall within a certain interval (minimum and maximum credited hours). (b) The bidline must abide by certain safety and collective agreement rules, e.g., required rest periods between pairings, minimum number of days off in a monthly or weekly period, etc. (c) The bidline should satisfy, to the greatest extent possible, a given set of quality criteria, such as: (1) Purity, i.e., the degree to which the bidline contains pairings of the same type, e.g., pairings that contain morning daily flights. (2) Regularity, i.e., the degree to which the duties/offs are repeated in a specific pattern, e.g., duties from Monday to Thursday, and offs from Friday to Sunday. (3) Hour balance, i.e., the degree to which the number of the credited hours approaches a desired target. (4) Avoid single duties/offs, and/or include a certain number of consecutive duties/offs. The bidline problem is formulated as a set partitioning problem, as follows: 232

234 minimize cb yb M x b p p (1) subject to: ap, b yb xp 1 p (2) yb b {0,1} b (3) xp 0 p (4) where c b is the cost of bidline b, y b is a binary variable, which takes the value 1 if bidline b participates in the solution, and 0 otherwise, M is a large number which represents the cost of not covering a pairing p, x is a slack variable which allows for a pairing not being covered subject p to the large cost M, and a pb is a coefficient which indicates if bidline b covers pairing p, i.e.,, a pb, = 1 if bidline b covers pairing p, and 0 otherwise. 3. Methodology For the bidline problem, as formulated in the previous section, we develop an integer programming formulation, and a multi-stage branch and price solution methodology. The first stage aims at generating a large number of high quality duty lines that satisfy a certain quality threshold. The second stage aims at generating additional duty lines so as to cover the flights that remain uncovered at the end of the first stage, without any special concern for their quality. Finally, the last stage aims at further improving the quality of the combined solution obtained by the first two stages. The proposed solution methodology utilizes a master problem that tries to optimize some appropriate measure of performance using a given set of duty lines, and a column generation pricing sub-problem that generates cost-effective duty lines, which are fed into the master problem. The column generation methodology is presented in the following figure. Solve Restricted Master Problem Add new column Duals Update and solve pricing subproblem Bidline with negative reduced cost? NO OPTIMAL YES Figure 1. Column generation procedure The above procedure solves the linear relaxation of the master problem. To obtain an integer solution, branching is applied (see Barnhart et al., 1998), including special branching rules and tree-searching strategies. In what follows, we discuss the formulations of the master problem. The pricing sub-problem is not presented, due to space consideration. At the first stage, the objective is to generate a large number of high quality duty lines that satisfy a certain quality threshold; measured in terms of cost, this can be expressed as generating 233

235 bidlines with cost c b C (C is the threshold). This objective is achieved by modifying the objective function in (1) as follows: minimize cb C y b b (5) Note that (5) does not include the second term of (1), i.e. M = 0, and that the coefficient c b C is negative. Hence, (5) can be viewed as a maximization of C c b y b b, where C c b represents a measure of the merit of bidline b. In other words, the objective is to maximize the cumulative merit of the bidlines. By controlling the value of C compared to the quality penalties that determine the cost of a bidline, we can ensure a minimum quality threshold for the generated bidlines. At the second stage, the objective is to cover as many pairings as possible. To meet this objective, we solve the original problem (1)-(4) with c b = 0. Since this stage may sacrifice quality in order to achieve maximum coverage, we choose to fix a certain number of high quality bidlines obtained at the first stage. Finally, at the last stage, we revert c b to its actual value, in order to fine tune the quality of the solution. 4. Numerical Results and Further Research We applied the proposed methodology on actual instances of a medium-sized US airline company (~100 cockpit crewmembers, ~700 pairings; sub-problem: ~ 1000 discrete variables, ~10,000 constraints). We used CPLEX 12.4 to solve both the master problem and the pricing sub-problem, and AIMS proprietary software for modeling and supporting the bidline problem on an Intel Core2 2.4 GHz, with 3 GB RAM, and no parallel processing. The results are shown in the following Table. Month April 2014 May 2014 June 2014 July 2014 No. of Pairings Pre-processing 17 min 15 min 16 min 17 min 1 st stage 30 min 24 min 20 min 28 min 2 nd stage 15 min 13 min 24 min 13 min 3 rd stage 48 min 33 min 10 min 28 min Total time 110 min 85 min 70 min 86 min Table 1. Numerical Results for a Medium-Sized US Airline Company In all four instances, all pairings were covered, and with suitable user-acceptable tolerances the optimality gap was less than 5%. An initial integer solution, with all pairings covered was obtained within an hour, at the end of the second stage. 234

236 In our future research, we intend to link the three stages and pass information of the later stages to the earlier ones. Some preliminary results indicated about 20% reduction in computational time and about 10% improvement in the quality of the first integer solution this way. We also recognize that there is significant potential in improving the methodology for producing the bidlines (subproblem), as well as improving the tree-search strategies and branching rules. References Barnhart, C., Johnson, E.L., Nemhauser, G.L., Savelsbergh, M.W.P., and P.H Vance. Branchand-price: Column generation for solving huge integer programs. Operations Research, Vol. 46 No. 3, 1998, pp Boubaker, K., Desaulniers, G., and I. Elhallaoui. Bidline scheduling with equity by heuristic dynamic constraint aggregation. Transportations Research Part B, Vol. 44, 2010, pp Saddoune M., Desaulniers G., Elhallaoui, I., and F. Soumis. Integrated Airline Crew Pairing and Crew Assignment by Dynamic Constraint Aggregation. Transportation Science, Vol. 46, No. 1, 2012, pp Acknowledgement The work of the authors is sponsored by the research grant Development of operations research tools for the optimization of functional procedures of commercial airlines, which is funded by AIMS Corporation. 235

237 A Calibration Tool for Macroscopic Traffic Flow Models Anastasia Spiliopoulou* Dynamic Systems and Simulation Laboratory Technical University of Crete, Chania, Greece Ioannis Papamichail Dynamic Systems and Simulation Laboratory Technical University of Crete, Chania, Greece Markos Papageorgiou Dynamic Systems and Simulation Laboratory Technical University of Crete, Chania, Greece John Chrysoulakis Department of Civil and Infrastructure Engineering TEI of Athens,12210 Egaleo-Athens, Greece Abstract This paper presents a software tool that has been recently developed for the calibration and validation of macroscopic traffic flow models using real traffic data and appropriate optimization methods. The software has a user-friendly graphical interface which makes the calibration procedure an easy task. Apart from the description of the software tool, an application example is presented as well. Keywords: Traffic flow models, calibration, validation, calibration tool. *Corresponding author, natasa@dssl.tuc.gr 236

238 1. Introduction During the last decades, several mathematical road traffic flow models have been proposed (see Hoogendoorn and Bovy (2001) for an overview on macroscopic traffic flow models). These models can be used for planning of new or upgraded road infrastructures, for development and testing of traffic estimation algorithms, for designing and testing of traffic control strategies, as well as, for other traffic engineering tasks. The models include a number of physical or nonphysical parameters whose values may differ for different freeway sites. Thus, before employing a traffic flow model in practice, it is important to first calibrate it against real traffic data. The calibration procedure aims to appropriately specify the model parameter values, so that the representation of the network and traffic flow characteristics is as accurate as the model structure allows. The most common approach is to minimize the discrepancy between the model s estimations and the real traffic data, by use of appropriate optimization tools. The nonlinear, nonconvex least-squares optimization problem of parameter estimation is known to have multiple local minima and hence only derivative-free optimization algorithms should be utilized (see Kontorinaki et al. (2014) for an overview on suitable optimization methods). Within the literature there are only few works on the calibration of macroscopic traffic flow models. The main reasons being: first, it is quite difficult to have access to real traffic data, and second, there is no available tool that can be easily employed to solve the parameter estimation problem. Within this work, a software tool has been developed for the calibration and validation of macroscopic traffic flow models with a user-friendly graphical interface. In the following sections, first the software tool is shortly described followed by a short application example using real traffic data from a freeway stretch in Athens, Greece. Finally the last section concludes with the main remarks of this study. 2. CALISTO graphical user interface CALISTO (CALIbrationS Tool) is a software tool that enables the calibration and validation of macroscopic traffic flow models for various freeway sites using real traffic data. Figure 1 presents the application window of the software which contains the following basic elements: Freeway network description: this feature includes all the required information needed so that a freeway site is described adequately, such as the number of freeway links, the number of freeway on-ramps and off-ramps and their location, the number of detector stations and their location etc. See Figure 2 for an example of the Freeway network description editor. Simulated traffic data: this feature contains information about the simulated data, such as the simulation step, the traffic measurements interval and the simulation duration, as well as the specification of the traffic data input file. 237

239 Other settings: it consists of some extra simulation features regarding the utilized performance index and the simulation outputs. Figure 11 CALISTO application window. 238

240 Figure 12 Example of Freeway network description editor. 239

241 Figure 13 Example of METANET parameters editor. Selection of the traffic flow model: the user may select one of the available macroscopic traffic flow models. Two models are available in the current version of the software. In particular, the first-order Cell Transmission Model (CTM) (Daganzo, 1995a, 1995b) and the second-order model METANET (Messmer and Papageorgiou, 1990). Both models are discrete-time state-space models and they are the most commonly used models for the freeway traffic flow representation. The structure of the program is modular enabling the addition of more discrete-time state-space models in the future. Depending on the model that is selected, some model parameter values need to be specified, by clicking on the Model parameters button (see Figure 1 and Figure 3). Note, that in case of calibration these values are actually the initial model parameter values while in case of validation these are the values of the resulted model. Selection of the optimization algorithm: the user may select one of the available optimization methods. Three methods are available in the current version of the software, namely, the Nelder-Mead method, a genetic algorithm and the cross-entropy method. All three optimization methods are derivative-free methods and are suitable for the calibration of macroscopic traffic flow models (see Spiliopoulou et al. (2014) for an illustration). Again, the structure of the program is modular enabling the addition of more optimization algorithms in the future. Depending on the algorithm that is selected, some parameters need to be specified, by clicking on the Algorithm parameters button (see Figure 1). 240

242 Selection of the operation: two operations are available, either Calibration or Validation. The calibration aims at estimating the optimal model parameter values so that the model may represent the traffic conditions of a particular freeway site with the highest achievable accuracy. The validation, is usually carried out after the model calibration, and aims to test the validity of the produced model, thus the resulting model is applied to the same freeway site using different traffic data than the data used for its calibration. Execution: the selected operation is executed, by clicking on the Run button (see Figure 1), taking into account all the introduced information. The output of the calibration procedure includes graphs of the calibration progress, the optimal model parameter values, the performance index (PI) value and plots of the real traffic data and the corresponding model estimations at various network locations; while the output of the validation procedure includes the obtained PI value and plots of the of the real traffic data and the corresponding model estimations. 3. Application example This section presents the application of the developed software tool for the calibration of the METANET model using real traffic data from a part of the Attiki Odos freeway (34 th to 28 th km, direction from the Airport to Elefsina) in Athens, Greece. This freeway stretch includes three onramps and three off-ramps, as shown in Figure 4. In order to model the network by use of METANET, the freeway stretch is represented through 9 nodes (N0 N8) and 8 links (L1 L8), where each node corresponds to a bifurcation or a junction or any location marking a change of the network geometry; whereas the homogeneous road stretches between these locations are represented by links. Each network link is subdivided in model sections of equal length; see for example link L1 which is divided in 3 sections, with the vertical short lines denoting the section borders. Figure 4 displays the length, number of sections and number of lanes for each link; the exact location of the on-ramps and off-ramps; as well as the location of 19 available detector stations which are depicted by bullets. The METANET model was calibrated using real traffic data from the morning peak hours, 6 12 am, of 16/06/2009. In this calibration exercise, the Nelder-Mead algorithm is employed to solve the parameter estimation problem, aiming at minimizing the RMSE (Root Mean Squared Error) of the real-speed measurements and the model s estimation of speed. 241

243 Figure 14 Representation of the considered freeway stretch. Figure 15 Convergence of the Nelder-Mead algorithm over iterations. 242

244 Figure 16 Optimal model parameter values. 243

245 Figure 17 Time-series of the real speed measurements and the model s estimations of speed at various detector locations. Figure 5 presents the PI value over iterations. It is seen that the algorithm converges after around 140 iterations achieving a PI value equal to At the end of the calibration procedure, a window with the optimal model parameters appears, as shown in Figure 6. The user may utilize these values to validate the resulted model using real traffic data from other dates, different than the data used for its calibration. Moreover, Figure 7 presents the time-series of the real speed measurements and the corresponding model s estimations at various network locations. It is observed that the resulted model is able to reproduce the real traffic conditions of this particular freeway stretch with sufficient accuracy. As presented above, this tool enables the calibration of various traffic flow models for different freeway sites. The software is very easy to use and is expected to be very useful to both researchers and practitioners. 244

246 4. Conclusions This paper presents a software tool that has been recently developed for the calibration and validation of macroscopic traffic flow models and has a user-friendly graphical interface. An application example illustrates the easiness of use and the effectiveness of the software enabling the calibration and validation of various traffic flow models at different freeway sites. Acknowledgements This research was co-financed by the European Union (European Social Fund - ESF) and by national funds through the Operational Program "Education and Lifelong Learning" of the National Strategic Reference Framework (NSRF) - Research Funded Project: ARCHIMEDES III. Investing in society s knowledge through the European Social Fund. The authors would like to thank ATTIKES DIADROMES S.A. for providing the utilised traffic data from Attiki Odos motorway in Athens, Greece. References Hoogendoorn, S. P., and P. H. Bovy, State-of-the-art of vehicular traffic flow modelling. Proceedings of the Institution of Mechanical Engineers, Part I. Journal of Systems and Control Engineering, Vol. 215, No. 4, pp Daganzo, C. F., 1995a. A Finite Difference Approximation of the Kinematic Wave Model of Traffic Flow. Transportation Research Part Β, Vol. 29, No. 4, pp Daganzo, C. F., 1995b. The Cell Transmission Model, Part II: Network Traffic. Transportation Research Part Β, Vol. 29, No. 2, pp Messmer, A., and M. Papageorgiou, METANET: A macroscopic simulation program for motorway networks. Traffic Engineering & Control, Vol. 31, No. 8-9, pp

247 Kontorinaki, M., Spiliopoulou, A., Papamichail, I., Papageorgiou, M., Tyrinopoulos, Y., Chrysoulakis, J., Overview of nonlinear programming methods suitable for calibration of traffic flow models. Operational Research: An International Journal. In press. Spiliopoulou, A., Papamichail, I., Papageorgiou, M., Tyrinopoulos, Y., Chrysoulakis, J., Macroscopic traffic flow calibration using different optimization algorithms. Proc. of the International Symposium of Transport Simulation, Ajaccio, Corsica, France, June

248 Air Traffic Management: The free flight concept. Coletsos John National Technical University of Athens, School of Applied Mathematical and Physical Sciences Zografou campus, 15780, Athens, Greece, Ntakolia Charis National Technical University of Athens, School of Applied Mathematical and Physical Sciences Zografou campus, 15780, Athens, Greece Abstract The insufficient air routes combined with the adverse weather and congestion to air sectors lead to economic, environmental and safety problems to political aviation in Europe. This situation creates negative aspects to airlines and airports, as well. Furthermore, according to recent studies over 40,000 daily flights are predicted for 2020, and therefore the current ATM system will not be able to handle this volume of traffic in an efficient manner. A new and promising approach of solving these problems in the future consists of transforming the ATM system from an airport-centered to an airplane-centered system so it can: (i) prioritize the airline preferences, (ii) support the free flight concept, (iii) distribute fairly ground holding and air delays among the flights, (iv) minimize the volume of work of ATCs as an observer, (v) relax the existing distance limits between airplane since the human factor has been annihilated, and therefore, (vi) increase the air sectors capacity avoiding congestions and (vii) increase safety and efficiency. Our attempt will be to develop a mathematical model for a support system for the free flight concept. We divide the problem into two sub problems (upper and lower level) in order to decrease the computational efforts and the complexity of the air traffic flow management problem and to allow flexibility between the decision maker levels enforcing in the same time free flight scenario Keywords: Air-Traffic Management, Integer Programming, Operations Research. 247

249 1. Introduction Information technology has already had a major influence on air travel. And with the number of global travelers expected to double by 2030 according to the Federal Aviation Administration (FAA), it will continue to lead the way for the industry. However, every day over 100,000 flights take off at airports across the world. Some are short hops to nearby destinations; some flights cross the oceans, but all have to fly in the same sky. Until recently, air traffic has been managed by routing airplane into narrow, predetermined routes much like highways in the sky originally developed to meet the domestic airspace requirements of countries and often defined by the location of ground-based navigational aids. The above can cause congestion, ground holding and airborne delays or even worse cancellation. Indeed, Congestion phenomena are persistent and arise almost on a daily basis as a consequence of bad weather conditions which cause sudden capacity reductions. The Air Transport Association has estimated that system delays drove an estimated $5.9 billion in direct operating costs for United States airlines in As a result, air traffic flow management (ATFM) has become increasingly crucial. Free flight is a developing air traffic control method that uses no centralized control (e.g. air traffic controllers). Instead, parts of airspace are reserved dynamically and automatically in a distributed way using computer communication to ensure the required separation between aircraft. Free flight is a new concept being developed to take the place of the current air traffic management methods through the use of technology. True free flight eliminates the need for Air Traffic Control (ATC) operators by giving the responsibility to the pilot in command. This gives the pilot the ability to change trajectory in mid-flight. With the aid of computer systems and/or ATC, pilots will be able to make more flight path decisions independently. As in most complex systems, distributed yet cooperative decision making is believed to be more efficient than the centralized control characterized by the current mode of air traffic management Therefore, research should be turn to free flight in order to solve these issues and increase network s capacity, safety and efficiency. Research should also focus on European airspace, since en-route airspace, and not only airports, is highly congested. This is due to both the airway system, built up by a fixed track system connecting airports, and to the existing air navigation and air traffic control rules. Nowadays, the minimum safe separation between aircraft is assured only by means of altitude and/or longitudinal separations. This type of structure represents a bottle-neck for air traffic flow with the increase of flight volume. Though some measures have been taken to reduce traffic congestion, much more is needed before air traffic can once again flow smoothly and efficiently. We are interested in the development of all decision methodologies which can make maximum use of airspace without violating safety constraints. 2. Mathematical model overview Our philosophy is based on Dell Olmo and Lulli s (2002) model, where we divide the problem into two sub problems (upper and lower level) in order to decrease the computational efforts and the complexity of the air traffic flow management problem and to allow flexibility between the decision maker levels enforcing in the same time free flight scenario (Fig.1). 248

250 Figure 1 - Mathematical model's structure. In the upper level, the inputs, such as airspace structure, schedule timetables and the airplane s characteristics are integrated in our model. In this way, we represent a fixed topology network, and it is formulated as a special dynamic multicommodity network flow problem with side constraints. The upper level s output is the airplane sequences for each way point and it is the input of the lower level. The lower level represents some free flight aspects in a constrained scenario, as it is analytically mentioned in (Dell'Olmo and Lulli 2002). It is formulated as an optimization model. The output of this level is the 4D trajectories for each airplane. The 4D trajectory of an aircraft consists of the three spatial dimensions plus time as a fourth dimension. The novelty of this research is the enhancement to the above model the ability: (i) to segregate the airborne delay cost from the ground holding cost, (ii) to impose cancellation policies, (iii) to take into account the arrival and departure capacity of each airport, (iv) to ensure connectivity between airports for continued flights and (v) to minimize the cost of flight due to airborne delays, ground holding delays, flight speed and cancellation. The proposed architecture supports the free flight concept by guaranteeing freedom of movement to all the airplanes. 3. Free flight mathematical model (Lower Level) In this section, based on the (Dell'Olmo and Lulli 2002), the lower level is presented as a decision maker that represents the flight operation of the airplane (mainly the navigation operation) according to the new possible routes, guaranteeing at the same time freedom of movement to all airplanes in the network (free flight scenario) with respect to conflict measures and safety assurance. After the process, a 4 D trajectory is obtained for each airplane. Sets: K is the set of airplanes, T = {1,, t N } the subset of time periods, V k the set of speeds for airplane k and Z k the set of admissible flight levels for airplane k. k k Input data: t s is the starting time of airplane k, c z,v the fuel consumption cost at level z (z Z k ) and speed v (v V k k ) for airplane k, c trn the fuel consumption cost due to turn for airplane k, u k the initial flight level of airplane k, S min lon the minimum longitudinal separation, S min lat the minimum lateral separation, D max the maximum route deviation, B the airway breadth and τ the length of 249

251 the time interval. We consider also η k (t), η + k (t) the negative and the positive deviation variable, respectively, δ k,l (t), δ + k,l (t) the negative and the positive ordinate deviation variable of airplane k and l, respectively and θ k,l (t), θ + k,l (t) the negative and the positive abscissa deviation variable of airplane k and l, respectively. In addition we define the decision variables x k (t) the abscissa of airplane k at time t and y k (t) the ordinate of airplane k at time t. φ k z,v 1, if airplane k is at flight level z with velocity v at time t (t) = { 0, otherwise, k K, z Z k, v V k σ k,l 1, if airplanes k, l have to respect lateral separation at time t z (t) = { 0, if airplanes k, l have to respect longitudinal separation at time t, k, l K, z Z k Z l Objective function: The objective function minimizes the total cost of the traffic through the considered airway. It includes the fuel consumption cost term k z Z k,v V k c z,v φ z,v k (t) and the increased fuel consumption cost term due to turn c k trn y k (t + 1) y k (t). In order to linearize this term, we introduce two non-negative variables η k (t), η k + (t), for the positive and the negative deviation of ordinate of y k (t) of airplane k at time t. Therefore, we set y k (t + 1) y k (t) = η k + (t) η k (t), k K, t [t k s, t N 1]. In case of positive lateral deviation we have η k + (t) > 0 and η k (t) = 0. In case of negative lateral deviation: η k + (t) = 0 and η k (t) > 0. So, we substitute y k (t + 1) y k (t) = η k + (t) + η k (t) and the term becomes c k trn (η k + (t) + η k (t)). So the objective function is the following: min { [c k z,v k K,t T z Z k,v V k φ k z,v (t)] Subject to the following constraints + c k trn (η k + (t) + η k (t))} k We have initial conditions v V k φ u (t k k,v s ) = 1, k K (10), z( u k ) Z k,v V k φ z,v k (t k s ) = 0, k K (11), x k (t k s ) = 0, k K (12) and y k (t k s ) = 0, k K (13). We introduce representation constraints in order to impose that only one of the variables φ k z,v (t) is equal to one, for each airplane k and for each time period t t k s, so z Z k,v V k φ k z,v (t) = 1, k K, t k t s (14). The continuity constraints link the positions of each airplane at consecutive time periods. z+1 w=z 1 v V k φ k w,v (t + 1) φ k z,v (t) 0, v V k x k (t + 1) x k (t) = τ vφ k z,v z Z k,v V k (t) k K, z Z k, t [t s k, t N 1] (15), k K, t [t k s, t N 1] (16) 250

252 y k (t + 1) y k (t) D max τ vφ k z,v linearization z Z k,v V k (t) η k + (t) + η k (t) D max τ vφ k z,v (t) z Z k,v V k, k K, t [t k s, t N 1] (17) linearization, k K, t [t k s, t N 1] (17l) The separation constraints must be satisfied in order to assure the minimal separation among airplanes. Note that for each couple of airplanes which are at the same flight level, at least one of either longitudinal or lateral separation constraints must be assured. y k (t) y l (t) S min lat [ φ k z,v (t) v V k l + φ z,v (t) v V l k, l(k l) K, z (Z k Z l ), t max(t s k, t s l ) 2 + σ k,l z (t)], (18) x k (t) x l (t) S min lon [ φ k z,v (t) v V k l + φ z,v (t) v V l k, l(k l) K, z (Z k Z l ), t max(t s k, t s l ) 1 σ k,l z (t)], (19) In order to linearize these constraints, we introduce two non-negative variables δ k,l (t), δ + k,l (t), the positive and the negative deviation, respectively, of ordinate of y k (t) of airplane k and y l (t) of airplane l at time t. Therefore, we set: y k (t) y l (t) = δ k,l (t) δ + k,l (t), k, l(k l) K, z (Z k Z l ), t max(t s k, t s l ). In case of positive lateral deviation we have δ + k,l (t) > 0 and δ k,l (t) = 0. On the other hand in case of negative lateral deviation δ + k,l (t) = 0 and δ k,l (t) > 0. So, we substitute y k (t) y l (t) = δ + k,l (t) + δ k,l (t) in the constraint (18). Similarly, for the constraint (19), we introduce two non-negative variables θ k,l (t), θ + k,l (t), the positive and the negative deviation, respectively, of abscissa of x k (t) of airplane k and x l (t) of airplane l at time t. Therefore, we set: x k (t) x l (t) = θ k,l (t) θ + k,l (t), k, l(k l) K, z (Z k Z l ), t max(t s k, t s l ). In case of positive lateral deviation we have θ + k,l (t) > 0 and θ(t) = 0. In case of negative lateral deviation θ + k,l (t) = 0 and θ k,l (t) > 0. So, we substitute x k (t) x l (t) = θ + k,l (t) + θ k,l (t) in the constraint (19) δ k,l + (t) + δ k,l (t) S min lat [ φ k z,v (t) v V k l + φ z,v (t) v V l k, l(k l) K, z (Z k Z l ), t max(t s k, t s l ) θ k,l + (t) + θ k,l (t) S min lon [ φ k z,v (t) v V k l + φ z,v (t) v V l k, l(k l) K, z (Z k Z l ), t max(t s k, t s l ) 2 + σ k,l z (t)], (20) 1 σ k,l z (t)], In the sequence there are track constraints x k (t) 0, k K, t > t s k (22), B y k (t) B, k K, t > t s k (23), non negativity constraints: η k (t), η + k (t) 0, k K, t (21) 251

253 [t k s, t N 1] (24) and δ k,l (t), δ k,l + (t), θ k,l (t), θ k,l + (t) 0, k, l(k l) K, z (Z k Z l ), t max(t k s, t l s ) (25). Also, we have the binary constraints φ k z,v (t) {0,1}, k K, t [t k s, t N 1], z Z k, v V k (26), σ k,l z,v (t) {0,1}, k, l(k l) K, t T, z (Z k Z l ), t max(t k s, t l s ) (27) and linearity constraints y k (t + 1) = y k (t) + η + k (t) η k (t), k K, t [t s k, t N 1] (28) y k (t) y l (t) = δ k,l (t) δ + k,l (t), k, l(k l) K, z (Z k Z l ), t max(t s k, t s l ) (29) x k (t) x l (t) = θ k,l (t) θ + k,l (t), k, l(k l) K, z (Z k Z l ), t max(t s k, t s l ) (30) This mathematical model, is a contribution to the free flight concept, and must be considered as part of an integrated decision system, where a batch of models cooperates under one or more global strategies. 4. Conclusions Our research s goal is to develop a dynamic integer optimization mathematical model with two level hierarchical architecture that support free flight in a constrained scenario that will: (i) Reassure safety between airplanes, (ii) Minimize airborne and ground holding delays, (iii) Cancellation costs, (iv) Minimize speed deviations, (v) Distribute fairly the delay among the flight path, (vi) Take into account airport s arrival and departure capacity and arc s capacity as well. References Agustín A., Alonso-Ayuso A., Escudero L.F., Pizarro C., On air traffic flow management with rerouting. Part I: Deterministic case, European Journal of Operational Research, Vol. 219, pp Betsimas D., Stock S., The air traffic flow management problem with en route capacities, Operations Research, Vol. 46, pp Betsimas D., Lulli G., Odoni A., The air traffic flow management problem: an integer optimization approach. 13th International Conference, IPCO. 2008, pp Betsimas D., Lulli G., Odoni A., An integer optimization approach to large-scale air traffic optimization flow management, Operations Research, Vol. 59, pp D'aspremont A., Sohier D., Nilim A., El Ghaoui L., Duong V., Optimal path planning for air traffic flow management under stochastic weather and capacity constraints, Proceedings of 4th IEEE International Conference on Research, Innovation and Vision for the Future, RIVF' , pp

254 Dell'Olmo P, Lulli G., A new hierarchical architecture for air traffic management : Optimization of airway capacity in a free flight scenario. European Journal of Operational Research, Vol. 144, pp Grignon L., Analyses of Delay in an Air Traffic System with Weather Uncertainty. PhD thesis. s.l., University of Washington. Leal de Matos P., Powell P.L., Decision support for flight rerouting in Europe, Decision Support Systems, Vol. 34, pp Lulli G., Odoni A., The european air traffic flow management problem, Transportation Science, Vol. 41, No. 4, pp Richetta O., Odoni A.R., Dynamic solution to the ground-holding policy problem in air traffic control, Transportation Research, Vol. 28A, No. 3, pp Sheridan, T.B Next Generation air transportation systems: human automation interaction and organizational risks, [ s.l, Paper Presented at the Second Symposium on Resilience Engineering. Soomer M.J., Franx G.J., Scheduling aircraft landings using airline's preferences, European Journal of Operational Research, Vol. 190, No. 1, pp Vranas P.B., Bertsimas D.J., Odoni A.R., Dynamic ground-holding policies for a network of airports, Transportation Science, Vol. 28, pp Waslander S.L., Raffard R.L., Tomlin C.J., Market-based air traffic flow control with competing airlines. Journal of Guidance Control and Dynamics, Vol. 31, No. 1, pp

255 Comparison of pricing mechanisms in markets with non-convexities Panagiotis Andrianesis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. George Liberopoulos Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece, Abstract We consider markets that are characterized by non-convexities or indivisibilities, due to the presence of avoidable costs and minimum supply requirements. The motivation for our work has been the area of electricity markets, which allow the submission of multi-part bids and take into account the technical characteristics of the generation units. Such market designs, when operated under marginal pricing, may lead to market outcomes where truthful bidding results in losses for some participants. To deal with this highly undesirable prospect, some approaches provide makewhole payments, or uplifts, as they are often called, whereas others modify the market-clearing prices to ensure sufficient revenues to the suppliers. In this work, we present and compare revenueadequate pricing approaches. These include the Semi-Lagrangean Relaxation and the so-called Primal-Dual approaches for generating efficient revenue-adequate prices. We supplement these schemes with a newly proposed scheme, which we refer to as Minimum Zero-Sum Uplift (MZU). To facilitate the comparisons, we apply these schemes on a stylized example that appears in the literature. Keywords: non-convexities, electricity market, revenue-adequate pricing. 254

256 1. Introduction Electricity markets in which generation units are allowed to submit multi-part bids and which take into account the technical characteristics of these units are characterized by non-convexities. Such market designs, when operated under marginal pricing, may result in market outcomes where truthful bidding results in losses for some participants. To deal with this undesirable prospect, some approaches provide external payments, or uplifts, as they are often called, to ensure sufficient revenues to the market participants (O Neill et al., 2005; Hogan and Ring, 2003; Bjørndal and Jörnsten, 2008; Gribik et al., 2007; Andrianesis et al. 2013), whereas others ensure sufficient revenues without the provision of external uplifts (Motto and Galiana, 2002; Galiana et al., 2003; Araoz and Jörnsten 2011; Ruiz et al., 2012; Van Vyve, 2011). In this paper, we focus on the latter approaches, which are pure revenue-adequate in that the prices that they generate guarantee that no supplier incurs losses without the need for additional external/internal uplifts. We also discuss a new mechanism, referred to as Minimum-Zero Sum Uplift. The remainder of the paper is structured as follows. Section 2 presents the market model we use for our study and various pricing approaches. Section 3 illustrates the application of the approaches on a numerical example, and discusses some interesting findings. 2. Market Model and Pricing Approaches We consider a single-commodity, single-period stylized Unit Commitment and Economic Dispatch (UCED) problem, where supplier i submits a bid for its marginal cost bi and its startup cost fi, to an auctioneer. The auctioneer solves a bid/cost minimization problem to obtain the optimal commitment and dispatch, represented for supplier i by variables zi and qi respectively, that satisfy a deterministic and inelastic demand d. Supplier i is subject to technical maximum and minimum constraints denoted by parameters ki for the capacity and mi for the minimum output. The formulation of the Mixed Integer Linear Programming problem is presented below. min L bi qi zi fi (1) zi, q i i subject to: d i (2) i q qi ziki i (3) qi zimi i (4) zi qi 0 i (5) {0,1} i (6) Problem (1)-(6) is characterized by non-convexities due to the presence of the fixed costs and the minimum output requirements. We mark with an asterisk the optimal solution, and we denote by λ * the marginal cost price, which is equal to the dual variable associated with constraint (2), if the commitment variables are fixed to their optimal value, so that problem (1)-(6) is transformed into a Linear Programming problem. In what follows, we present the basic elements of the Minimum Zero-Sum Uplift, the Semi-Lagrangean Relaxation and the Primal-Dual approaches. 255

257 2.1 Minimum Zero-Sum Uplift (MZU) The MZU scheme is based on the idea of maintaining the optimal solution and increasing the commodity price so that eventually all suppliers who would incur losses under marginal pricing, break even. The profitable suppliers are allowed to keep the profits that they would make under marginal cost pricing but are not allowed to gain any additional profits beyond that. This can be achieved if the extra commodity payments that the profitable suppliers receive as a result of the price increase are transferred as side-payments to the non-profitable suppliers, in addition to the extra commodity payments that the latter suppliers also receive as a result of the price increase. The smallest price at which the non-profitable suppliers break even is such that the total additional payments that they receive are just enough (hence the term minimum zero-sum ) to cover their losses. The MZU price λ is given as follows: 2.2 Semi-Lagrangean Relaxation * * * min 0, bi qi zi fi * i (7) The Semi-Lagrangean Relaxation (SLR) approach computes a uniform price that produces the same solution as the original UCED problem while ensuring that no supplier incurs losses. The formulation of the SLR problem is presented below. zi, qi d (8) min L ( ) b q z f q d SLR i i i i i i i subject to: q d i i (9) and primal constraints (3) (6). The SLR approach consists of solving the dual problem: * max ( ) (8) L SLR To find λ, Araoz and Jörnsten (2011) suggested using an iterative algorithm that increases λ in each iteration and solves the relaxed problem until the objective function reaches the optimal value of the objective function of the original UCED problem. 2.3 Primal Dual Approach Ruiz et al. (2012) proposed a so-called primal-dual (PD) approach for deriving efficient uniform revenue-adequate prices. This approach consists of: (a) relaxing the integrality constraints of the MILP problem so that it becomes a (primal) LP problem, (b) deriving the dual LP problem associated with the primal LP problem, (c) formulating a new LP problem that seeks to minimize the duality gap of the primal and dual LP problems, subject to both primal and dual constraints, and (d) adding the integrality constraints back to the problem as well as additional constraints to ensure that no participant incurs losses. This procedure yields a new Mixed Integer Non-Linear Programming (MINLP) problem, which is not presented due to space considerations. 256

258 Book of Proceedings 3. Numerical Results and Discussion Α common test-bed for evaluating different pricing schemes that deal with non-convexities has been an example introduced by Scarf (1994). In this paper, we use a modification of this example, introduced by Gribik et al. (2007). We modeled the pricing approaches using GAMS and solved the SLR and MZU schemes with the CPLEX solver and the PD scheme with BARON, on an Intel Core i5 at 2.67GHz, with 6GB RAM. Diagram 1 shows the price vs. the demand level for the aforementioned pricing schemes for a demand granularity of 0.5 units. Note that all schemes except PD actually use the optimal UCED solution. PD is the only scheme that allows for different allocations. Diagram 2 shows the percent increase of the total cost under PD compared to the optimal (minimum) total cost. Price PD SLR MZU Demand Diagram 1. Price vs demand under PD, SLR, MZU schemes (modified Scarf Example) % % Cost Increase [PD] Diagram 2. Cost increase (%) under the PD scheme vs demand (modified Scarf example) Diagram 1 indicates that the prices under all pricing schemes are not monotonically increasing in demand. This is the main effect of the non-convexities. Diagram 2 indicates that the PD scheme may result in inefficient commitment and dispatch quantities; the cost increase reaches up to about 7%. This effect is due to the fact that the PD scheme exchanges price for cost efficiency, by reallocating the quantities, so that the average costs are actually lower than the ones of the optimal allocation. Diagram 1 also shows that the SLR scheme exhibits price spikes. The SLR prices yield competitive prices that are high enough to make the market participants willing to generate the amounts of electricity scheduled by auctioneer. To achieve this, the SLR scheme may result in prices that are higher than the ones required to cover the losses. Demand 257

259 Lastly, we note that the prices of the PD and MZU schemes are comparable. The MZU scheme allows for internal transfers between the suppliers, and the uplifts are zero-sum. Hence, the profitable suppliers may transfer part of their revenues to the non-profitable ones, which in general keeps prices low. The PD scheme may yield lower prices than the MZU price, exchanging price for cost efficiency. In all cases where the PD price is lower than the MZU price, we observe that the dispatching is less efficient than the optimal one. This is the tradeoff for seeking price efficiency. References Andrianesis, P., Liberopoulos, G., Kozanidis, G., and A. Papalexopoulos. Recovery mechanisms in day-ahead electricity markets with non-convexities Part I: Design and evaluation methodology. IEEE Transactions on Power Systems, Vol. 28, No. 2, 2013, pp Araoz, V. and K. Jörnsten, Semi-Lagrangean approach for price discovery in markets with nonconvexities. European Journal of Operational Research, Vol. 214, No. 2, 2011, pp Bjørndal, M. and K. Jörnsten, Equilibrium prices supported by dual price functions in markets with non-convexities. European Journal of Operational Research, Vol. 190, 2008, pp Galiana, F.D., Motto, A. L., and F. Bouffard, Reconciling social welfare, agent profits, and consumer payments in electricity pools. IEEE Transactions on Power Systems, Vol. 18, No. 2, 2003, pp Gribik, P.R., Hogan, W.W. and S.L. Pope. (2007, Dec.). Market-clearing electricity prices and energy uplift. Working Paper, John F. Kennedy School of Government, Harvard University. Available: Price _Uplift_ pdf Hogan W. W. and B. J. Ring. On minimum-uplift pricing for electricity markets. 2003, unpublished. Available: Motto, A. L. and F. D. Galiana. Equilibrium of auction markets with unit commitment: The need for augmented pricing. IEEE Transactions on Power Systems, Vol. 17, No. 3, 2002, pp O Neill, R.P., Sotkiewicz, P.M., Hobbs, B.F., Rothkopf, M.H. and W. R. Stewart Jr. Efficient market-clearing prices in markets with nonconvexities. European Journal of Operational Research, Vol. 164, 2005, pp Ruiz, C., Conejo, A. J., and S. A. Gabriel. Pricing non-convexities in an electricity pool. IEEE Transactions on Power Systems, Vol. 27, No. 3, 2012, pp Scarf, H.E.. The allocation of resources in the presence of indivisibilities. Journal of Economic Perspectives, Vol. 8, No. 4, 1994, pp

260 Van Vyve, M., Linear prices for non-convex electricity markets: models and algorithms, CORE Discussion Paper 2011/50, Université Catholique de Louvain, Louvain-la-Neuve, Belgium, Oct

261 Development of Optimization Models for Addressing Various Decision and Information Related Issues in Supply Chain Planning George Liberopoulos Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece, Dimitrios G Pandelis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece, d_pantelis@mie.uth.gr George Kozanidis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece, gkoz@mie.uth.gr George K.D. Saharidis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece, saharidis@gmail.com Abstract We consider a model of a two-stage serial supply chain that processes a single part type. Each stage has an infinite-capacity raw-parts (RP) buffer, a finite-capacity production facility (PF) with deterministic production lead time (PLT), and an infinite-capacity finished-parts (FP) buffer. Stage 2 receives orders from end customers and places orders to stage 1. Stage 1 receives orders from stage 2 and places orders to an initial supplier with inexhaustible supply of initial raw parts. Upon receipt of an order, a stage immediately ships the order quantity from its FP buffer to its customer. The order arrives after a deterministic order lead time (OLT). If there are not enough parts in the FP buffer to meet the order, an expensive external inexhaustible-supply subcontractor (S) immediately complements the missing parts of the order. Each stage has revenue from the parts it sells and incurs inventory holding costs in its RP and FP buffers, as well as fixed and variable production and order costs. In case it cannot meet all the demand, it either pays the cost of complementing the order to the subcontractor, or it passes this cost to its customer. For this model, we formulate several variants of a finite-horizon production-and-order planning problem. The variants differ with respect to the level of collaboration and information 260

262 sharing between the two stages. First, we distinguish between the cases where the decisions are made in a centralized/decentralized way. In the letter case, we further distinguish between the cases where the decisions are made sequentially/simultaneously and use local/global information. In a follow up work, we plan to numerically experiment with these variants in order to quantify the effect of the problem parameters, the type of collaboration, and the level of information sharing on order and production variability and supply chain profitability. Keywords: supply chain planning; centralized vs. decentralized decision making; local vs. global information. Nomenclature Facilities R i : stage- i raw-parts (RP) buffer, i 1,2 ; R 3 : customer demand source; P i : stage- i production facility (PF), i 1,2 ; F i : stage- i finished-parts (FP) buffer, i 1,2 ; F 0 : (inexhaustible-supply) initial raw-parts buffer; S i : (inexhaustible-supply) stage- i subcontractor, i 1,2 ; Indices i : stage index, i 1,2 ; t : period index, t 1,, T; Decision variables ( i 1,2, t 1,, T); P it, : quantity produced by P i in period t ; X it, : indicator (binary) variable of it, R it, : inventory in R i at the end of period t ; F it, : inventory in F i at the end of period t ; P equal to 0 if P, 0, and 1 if P, 0 ; D it, : quantity of order placed by R i to Fi 1 at the end of period t ; Y it, : indicator (binary) variable of it, it D equal to 0 if D, 0, and 1 if D, 0; S it, : quantity of order placed by F i to S i at the end of period t ; it it it 261

263 Parameters ( i 1,2, t 1,, T); P max it, : production capacity of PF P i in period t ; p L i : production lead time (number of periods) of PF P i ; L : order lead time (number of periods) from Fi 1 to d i D 3,t : (external) final customer orders placed by R 3 to F 2 at the end of period t ; R i ; I i : M : interest rate used by stage i to compute inventory holding cost rates; a very large number; Costs ( i 1,2 ) p i : x i : (variable) unit production cost at P i ; fixed setup cost at P i ; r i : unit inventory holding cost per period in R i ; f i : unit inventory holding cost per period in F i ; I i : interest rate used by stage i to compute inventory holding cost rates; d i : (variable) unit order cost from R i to Fi 1, i 1,2,3 ; s i : (variable) unit order cost from F i to S i ; y i : fixed order cost from R i to Fi 1 ; 1. Introduction The work presented in this paper is part of a project supported by grant MIS ODYSSEUS: A holistic approach for managing variability in contemporary global supply chain networks, which is co-financed by the EU-ESF and Greek national funds through NSRF Operational Program Education and Lifelong Learning THALES: Reinforcement of the Interdisciplinary and/or Inter-Institutional Research and Innovation. The main goal of ODYSSEUS is to study the phenomenon of supply chain demand variability, identify the physical points of its creation, analyze its causes, and evaluate its negative impact on supply chain performance. One of the requirements of ODYSSEUS is to develop quantitative models to support decisions related to demand variability and in particular the bullwhip effect (the phenomenon that demand variability increases as one moves upstream in the supply chain). The 262

264 literature on the bullwhip effect is vast. Much of it involves the development and analysis of stochastic dynamic models of supply chains. Representative examples are Chen et al. (2000a,b), Cachon and Lariviere (2001), Lee et al. (1997a,b), Alwan et al. (2003), and Zhang (2004). In this paper, we formulate a deterministic dynamic capacitated lot-sizing planning problem (Buschkühl, et al. 2010) and variants of it for a simple two-stage serial supply chain model, in order to study the bullwhip effect. Such problems are simple and fit the practical MRPframework (Tempelmeier, 1997). They are also solvable with readily available mathematical programming software and heuristic approaches (Tempelmeier and Destroff, 1996). In a follow up work, we plan to use these variants to quantify the effect of the problem parameters, the type of collaboration, and the level of information sharing on order and production variability and supply chain profitability. In this respect, our models are related to Saharidis et al. (2006, 2009). 2. Basic Supply Chain Model Diagram 1 shows a graphical representation of the basic model described in the Abstract. Triangles represent buffers, and circles represent production facilities. Solid black arrows indicate the material flow and dashed grey arrows indicate the order flow. The decision variables of the model are shown in blue color, while its parameters are shown in red color. F 0 d L1 Stage 1 Stage 2 p L1 S 1,t R P 1 1 F1 R1,t 1, t F1,t R2,t 2, t F2,t 1, t X, max Y1, t, D1, t P P Y2, t, D2, t P P D3,t max 1, t S1 2 S 2,t p d L2 L2 R P 2 2 F2 2, t 2, t Diagram 1. Basic supply chain model. X, S R 3 We make the following assumptions regarding the variable cost rates: d i 1 d i p, i 1,2 (29) i ri Ii di, i 1,2 (30) f I ( d p ), i 1,2 (31) i i i i s i, i 2,3 (32) di 1 Inequalities (29) and (32) are necessary for ensuring the profitability and competitiveness of stage i, respectively. Expressions (30) and (31) are the usual inventory holding cost assumptions. 263

265 We consider a finite-horizon planning problem for the basic model. The horizon is divided into T discrete time periods, and decisions are made at the end of each period. The final customer orders of each period are known in advance. The PLTs and OLTs are constant. In each period, the order of events and decisions is as follows. For i 1,2 : 1) R i receives D parts from Fi 1 ; 2) P i starts processing P it, parts which it takes from i, t 1 L d i P p i, t L i parts from P i. 4) and Fi 1 immediately sends these parts to R i. 3) F i receives F i also receives S it, parts from S i. 5) R i orders D it, parts from Fi 1, R i. Next, we formulate several variants of the finite-horizon planning problem. The variants differ in terms of the level of collaboration and information sharing between the two stages. 3. Variants of the Planning Problem A) Centralized Decision Making: The two stages maximize their total profits jointly and simultaneously subject to customer order requirements and other constraints. This problem can be formulated as the following MILP problem: max T 2 di 1Di 1, t yiyi, t didi, t xi Xi, t pipi, t sisi, t rr i i, t fifi, t (33) t1 i1 Subject to R i, t R D P i, t1 i, t, i 1,2 d i, t 1 L i t 1,, T (34) F F P S D, i 1,2 i, t i, t 1 p i, t L i, t i 1, t i t 1,, T (35) P P X, i 1,2 max i, t i, t i, t D t 1,, T (36) M Y, i 1,2 i, t i, t t 1,, T (37) R, F, P, S, D 0, i 1,2 i, t i, t i, t i, t i, t t 1,, T (38) X, Y {0,1}, i 1,2 i, t i, t t 1,, T (39) B) Decentralized Sequential Decision Making: Stage 2: Leader; Stage 1: Follower. The stages maximize their individual profits separately and sequentially, starting with stage

266 B.1) Local Information: Stage 1 pays the cost of S1. Stage 2 solves a local-information self-profit maximization problem and decides, among others, the values of D2,t. Stage 1 takes these values as given and solves its own local-information self-profit maximization problem. Stage-2 problem: max T d3d3, t y2y2, t d2d2, t x2 X 2, t p2p2, t s2s2, t r2 R2, t f2f2, t (40) t 1 Stage 1 problem: max Subject to (34)-(39) for i 2 only. T d2d2, t y1y 1, t d1d1, t x1 X1, t p1p 1, t s1s1, t r1 R1, t f1f1, t (41) t 1 Subject to (34)-(39) for i 1 only. B1.2) Global information: Stage 2 pays the cost of S1. It solves a global-information self-profit maximization problem and decides, among others, the values of D2,t and S1,t. Stage 1 takes these values as given and solves its own local-information self-profit maximization problem. Stage-2 problem: T max ( ) d3d3, t y2y2, t d2d2, t s1 d2 S1, t x2x 2, t p2p2, t s2s2, t r2 R2, t f2f2, t (42) t 1 Stage-1 problem: T max ( ) Subject to (34)-(39). d2 D2, t S1, t y1y 1, t d1d1, t x1 X1, t p1p 1, t r1 R1, t f1f1, t (43) t 1 Subject to (34)-(39) for i 1 only. C) Decentralized Sequential Decision Making: Stage 1: Leader; Stage 2: Follower. The stages maximize their individual profits separately and sequentially, starting with stage 1. Stage 1 decides Y2,t, D2,t, and ΣS2,t and pays y2,ty2,t. Stage 2 plans only its production and detailed supply from S2, given that ΣS2,t has been decided by stage 1. Stage-1 problem: max T d2d2, t y1y 1, t d1d1, t y2y2, t x1 X1, t p1p 1, t s1s1, t r1 R1, t f1f1, t (44) t 1 Subject to (34)-(39) and T T D D S 2, t 3, t 2, t t1 t1 (45) 265

267 Stage-2 problem: max T d3d3, t d2d2, t x2 X 2, t p2p2, t s2s2, t r2 R2, t f2f2, t (46) t 1 Subject to (34)-(36) and (38)-(39) D) Decentralized simultaneous Decision Making: The stages maximize their individual profits separately and simultaneously. Stage 2 solves the same problem as in variant B.1 and decides, among others, the values of D2,t. Stage 1 solves a local-information self-profit maximization problem and decides the selling price d 2 that allows it to achieve a desired profit margin β. The two problems comprise the components of an equilibrium problem. Stage-2 problem: max T d3d3, t y2y2, t d2d2, t x2 X 2, t p2p2, t s2s2, t r2 R2, t f2f2, t (47) t 1 Stage-1 problem: Subject to (34)-(39) for i 2 only min d 2 (48) Subject to (34)-(39) for i 1 only, and T d2d2, t (1 ) y1y 1, t d1d1, t x1 X1, t p1p 1, t s1s1, t r1 R1, t f1f1, t (49) t 1 d2 0 (50) References Alwan, L.C., J.J. Liu, and D. Q. Yao,. (2003). Stochastic characterization of upstream demand processes in a supply chain. IIE Transactions, Vol. 35, No. 3, pp Buschkühl, L., F. Sahling, S. Helber, and H. Tempelmeier (2010). Dynamic capacitated lotsizing problems: a classification and review of solution approaches. OR Spectrum, Vol. 32, pp Cachon, G.P. and M. Lariviere, M. (2001). Contracting to assure supply: how to share demand forecasts in a supply chain. Management Science, Vol. 47, No. 5, pp

268 Chen, F., Z. Drezner, J.K. Ryan, and D. Simchi-Levi (2000a). Quantifying the bullwhip effect in a simple supply chain: The impact of forecasting, lead times, and information. Management Science Vol. 46, No. 3, pp Chen, F., J.K. Ryan, and D. Simchi-Levi (2000b). The impact of exponential smoothing forecasts on the bullwhip effect. Naval Research Logistics, Vol. 47, No. 4, pp Lee, H., P. Padmanabhan, and S. Whang (1997a). The bullwhip effect in supply chains. Sloan Management Review, Vol. 38, No. 3, pp Lee, H., P. Padmanabhan, and S. Whang (1997b). Information distortion in a supply chain: The bullwhip effect. Management Science, Vol. 43, No. 4, pp Saharidis, G., Y. Dallery, and F. Karaesmen (2006). Centralized versus decentralized production planning. RAIRO Operations Research, Vol. 40, pp Saharidis, G., V. Kouikoglou, and Y. Dallery (2009). Centralized and decentralized control policies for a two-stage stochastic supply chain with subcontracting. International Journal of Production Economics, Vol. 117, No. 1, pp Tempelmeier H. (1997). Resource-constrained materials requirements planning MRP rc. Production Planning and Control, Vol. 8, No 5. pp Tempelmeier H. and M. Derstroff (1996). A Lagrangean-based heuristic for dynamic multilevel multiitem constrained lotsizing with setup times. Management Science, Vol. 42, No. 5 pp Zhang, Χ. (2004). The impact of forecasting methods on the bullwhip effect. International Journal of Production Economics, Vol. 88, No. 1, pp

269 Measuring employee satisfaction in a Greek academic environment Nancy Bouranta Department of Business Administration of Food and Agricultural Enterprises, University of Patras, 2 G. Seferi St, Agrinio Greece, bouranta@gmail.com Christian Hurson IAE, Université de Rouen, 3, Avenue Pasteur, Rouen Cedex, France. Yannis Siskos Department of Informatics, University of Piraeus, 80, Karaoli & Dimitriou Street, Piraeus, Greece. Abstract The cycle of organizational success proposed by Schlesinger and Heskett suggests that satisfied employees deliver high service quality. Employee satisfaction or dissatisfaction hinges on whether there is a productive, fulfilling relationship between staff and management. This paper focuses on administrative staff job satisfaction measurement in the academic environment. The survey conducted at a Greek Business University used the MUSA system of multi-criteria satisfaction analysis. Looking at the partial satisfaction, weight, demanding and impact indexes, it is observed that the criterion of co-workers is a leverage opportunity for the university authorities. Keywords: Job satisfaction, Multicriteria decision analysis, MUSA system. 268

270 1. Introduction It has been well established in the literature that job satisfaction exerts an important influence on customer satisfaction, as satisfied employees tend to be more productive, provide better services to customers, and hence, can significantly enhance business profitability. Motivated by the importance of employee satisfaction, the current paper focuses on administrative staff job satisfaction measurement. The survey was conducted at the same Business University, using the same multi-criteria method. The following part of the paper presents the theoretical background. In the next section, the research methodology is reported. A discussion of the findings and the managerial implications comprises the following section. The paper ends with survey conclusions. 2. Literature Review Job satisfaction has been one of the most germane issues for researchers, because it contains useful information for predicting employee work-related behaviours and attitudes (Zimmerman and Darnold, 2009). A significant trend toward overwork has also been observed. Specifically, job satisfaction is associated with increased productivity, customer satisfaction, less absenteeism, lower turnover and life satisfaction (Chen et al., 2006). Job satisfaction has also been described as a person s overall affective reaction to a set of work and work-related factors. According to this perception, job satisfaction contains a number of job characteristics that need to be obtained within a broad measure of employee beliefs and attitudes about the job. Some of these facets can be divided among five main groups of job characteristics and work environment as follows: organisational image, organisational vision, superiors, co-workers and conditions of work (Eskildsen et al., 2010). These characteristics may not be of equal importance to every individual (Boles et al., 2003). This means that the overall rating for job satisfaction is not a simple average of the employees satisfaction levels for the different facets of a job, but will be a more complex assessment (D Addio et al., 2003). In some research, multi-faceted questions are used as stand-alone questions regarding job satisfaction; in other cases, they are used in addition to single-item questions (Robbins and Judge, 2011). 3. Methodological frame 3.1 Objectives of research and Satisfaction Criteria The objective of the present research is to identify the level of reported job satisfaction among the participating employees by using Multicriteria Satisfaction Analysis, as well as to formulate proposals for improvement or modification of administrative practices. A pilot questionnaire was created to reflect the policy of the Greek universities. Employees completed the pilot questionnaire and indicated any ambiguity or other difficulty they experienced in responding to the questions, as 269

271 well as offering suggestions. Based on this feedback, some questions were eliminated, others were modified, and additional items were developed. The first measure of global job satisfaction is derived from responses by employees to the statement: Generally speaking, I am very satisfied with my job. Participants were asked to select a number from 1 to 5, where 1 = strongly disagree and 5 = strongly agree; 3 is interpreted as a neutral response. For the purposes of this analysis, it is assumed that the higher the number selected, the greater the level of job satisfaction. A second measure contained a set of criteria (5) and sub-criteria (19) about employee job satisfaction (Figure 1). The items of the job satisfaction scale were adapted from Brown and Peterson (1993). All the items were in the form of statements on Likert type scale. 3.2 The MUSA system The Multicriteria Satisfaction Analysis (MUSA) system of Grigoroudis and Siskos (2010) has been used in order to measure customer or employee satisfaction, assuming that their global satisfaction depends on a set of criteria representing service characteristic dimensions. Thus, the global satisfaction is denoted as a variable Y and the set of criteria is denoted as a vector Χ = (Χ1, Χ2,,Χn). MUSA system uses a preference disaggregation logic. In the traditional aggregation approach, the criteria aggregation model is known a priori, while the global preference is unknown. On the contrary, the philosophy of disaggregation involves the inference of preference models from given global preferences. This preference disaggregation methodology is implemented through an ordinal regression based approach in the field of multicriteria analysis used for the assessment of a set of a marginal satisfaction functions in such a way, that the global satisfaction becomes as consistent as possible with participants marginal judgements. According to the survey, each participant is asked to express his/her own judgements, namely his/her global satisfaction and his/her satisfaction with regard to a set of discrete criteria. Based on these assumptions, the problem is approached as a problem of qualitative regression and solved via special linear programming formulations where the sum of deviations between global satisfaction evaluation explicitly expressed by employees and the one resulting from their multicriteria satisfaction evaluations is minimized. The main results from the aforementioned preference disaggregation approach are focused on global and partial explanatory analysis. Global explanatory analysis lays emphasis on customers global satisfaction and its primary dimensions, while partial explanatory analysis focuses on each criterion and their relevant parameters separately. Satisfaction analysis results consist of: Global satisfaction index: it shows in a range of 0-100% the level of global satisfaction of the customers; it may be considered as the basic average performance indicator for the organisation. 270

272 Global demanding index: it shows in a range of -100%-100% the demanding level of customers according to the following: demanding index 100%: extremely demanding customers, demanding index 0%: "normal" customers, demanding index -100%: non-demanding customers. Criteria/sub-criteria satisfaction indices: they show in a range of 0-100% the level of partial satisfaction of the customers according to the specific criterion/sub-criterion, similarly to the global satisfaction index. Weights of criteria/sub-criteria: they show the relative importance within a set of criteria or sub-criteria. Demanding indices: they show in a range of -100%-100% the demanding level of customers according to the specific criterion/sub-criterion, similarly to the global demanding index. The above methodology has been successfully implemented in many customer satisfaction surveys. Recently, Gosse and Hurson (2014) applied MUSA methodology to measure job satisfaction of recent employees in a major French organisation. Conditions Leadership Rewards Co-workers Job security of work Physical work environment Leadership style Pay Relationship with coworkers Job security before crisis Working hours Interesting for employees needs Benefit package Sense of social belonging Job security after crisis Workload Opportunity for initiatives Free days Relationship with supervisor Learning opportunities Job content Employee appraisal Opportunity of promotion Communication 271

273 Figure 1. Hierarchical structure of employee s job satisfaction criteria 3.3. Sampling The questionnaire was sent via to all the participants, along with an explanation of the purpose of this academic study. A sample of 78 questionnaires from the contacted employees was collected, of which six were excluded because they provided answers that were uniformly positive or negative (skewed responses). The 72 usable questionnaires constitute 49.6% of the total employee population. The demographic information showed that 58.3% of the respondents were females and 41.7% male. The age groupings were years (11.1%), years (41.6%), years (38.9%), <55 years (8.4%). The majority of the respondent had a bachelor s degree (76.4%) or a high school degree (23.6%). Tenure length groupings were 2-10 years (22.2%), years (56.5%), 23+ (21.3%). 4. Survey Results Seven out of ten employees declared they were from moderately (31.94%) to very satisfied (44.44%) while only 17.9% of them declared they were very unsatisfied. Only a few (6.94%) adopted a neutral attitude by declaring they were neither satisfied/nor unsatisfied. Globally administrative staffers seemed to be satisfied with their jobs (global satisfaction index equal to 88.05%). The satisfaction levels of the criteria regarding co-workers and work conditions- are very high, as they exceed 80%. The criteria of leadership (64.41%), job security (53.0%) and rewards (24.49%), show lower satisfaction levels. The global demanding index is -65.9%, indicating that employees are not very demanding. According to the weights computed by the MUSA system, the criterion of co-workers seems to be the most important (63.25%) (Table 1). The weights of the other criteria are about 10% with little fluctuation. This shows that administrative staffers give approximately the same importance to the other aspects of their work and are not particularly demanding with regard to them. The criterion of co-workers is the most important for the employees and this criterion also presented a higher degree of satisfaction. At the level of sub-criteria it was observed that in regard to the criterion of work conditions, employees are very satisfied with the physical work environment (91.34%) and working hours (81.34%).They are also satisfied with their workload (73.79%) and the content of their work (79.21%). With regard to the dimension of rewards, the satisfaction of employee regarding free days oscillates at very satisfactory levels (82.33%), while a low degree of satisfaction is observed for the sub-criterion of pay and benefits package (satisfaction indices: 3.7% and 25.95% respectively). Regarding job security, the administrative staff felt safe before the implementation of a mobility and reallocation scheme. However, after the new evidence, employees felt insecure. In addition, the sub-criteria regarding learning opportunities (17.18%) and opportunity of promotion (45.69%) show very poor performance ratings. The partial satisfaction index regarding the leadership style has low value as well. Moreover, the sub-criteria regarding employees needs (66.58%) and opportunity for initiatives (62.35%) also performed 272

274 poorly. Finally, regarding co-workers, the indices representing the satisfaction level of employees concerning all the tree sub-criteria, had satisfactory values. Criteria Weights Average satisfaction index Average demanding index Impact Conditions work of 10.85% 81.14% % 2.05% Leadership 8.62% 64.41% -6.70% 3.07% Rewards 8.13% 24.49% -0.86% 6.14% Co-workers 63.25% 97.94% % 1.30% Job security 9.14% 53.90% % 4.21% Table 1. Main results for numerical example 5. Conclusions The current survey illustrates the implementation of a preference dissagregation methodology for measuring employee job satisfaction in a university. The global satisfaction index exceeds 88.05%, indicating that administrative staffers at the university are satisfied with their jobs. The average satisfaction indices regarding two of the criteria (conditions of work and co-workers) exceed 80%. The criterion of rewards has the lowest value (24.49%). This may coincide with the country s economic situation that has led, in the last few years, to drastic cuts in public sector workers wages. As far as the importance of the criteria, it is observed that good organizational relationships (63.25%) is considered as the most important, while the other criteria average about 10% considering as not so important. At the level of sub-criteria, it is observed that overall employees feelings about job security have dramatically changed after the crisis. Over the past decades, Greek civil employees had jobs for life, ensuring security and stability. Nowadays, the economic crisis is forcing the government to put public-sector employees on mobility and reallocation schemes. The criteria of rewards and job security should be the first priorities for improvement in the future, as they have the lowest partial satisfaction index. However, they are not in the university leadership s control, thus university authorities can only exert influence in the same extent as government decisions. Given that, university should take effort to improve the criteria of leadership and conditions of work. However, the criterion of condition of work is difficult to be improved because the partial satisfaction index is high, the weight is slightly high, where as the demanding index is slightly low. 273

275 Acknowledgments This research has been co-financed by the European Union (European Social Fund) and Greek national funds through the Operational Program "Education and Lifelong Learning". References Boles, J.S., Wood, J.A. and Johnson, J. Interrelationships of role conflict, role ambiguity, and work-family conflict with different facets of job satisfaction and the moderating effects of gender. Journal of Personal Selling & Sales Management, Vol. 23, 2003, pp Brown, S.P. and Peterson, R.A. Antecedents and consequences of salesperson job satisfaction: meta analysis and assessment of causal effects. Journal of Marketing Research, Vol. 30, 1993, pp Chen S.H., Yang C.C., Shiau J.Y., Wang, H.H. The development of an employee satisfaction model for higher education. The TQM Magazine, Vol. 18, 2006, pp D Addio, A.C., Eriksson, T. and Frijters, P. An analysis of the determinants of job satisfaction when individuals baseline satisfaction levels may differ, Centre for Applied Micro econometrics (CAM), Department of Economics, University of Copenhagen, 16, 2003, available at: Eskildsen, J., Kristensen, K. and Antvor, H. G. The relationship between job satisfaction and national culture, The TQM Journal, Vol. 22, 2010, pp Gosse, B. and Hurson, Ch. Assessment and improvement of employee job-satisfaction: a full-scale implementation of MUSA methodology on recent employees in a major French organisation, Working Paper, NIMEC 2014, Universitéde Rouen, France. Grigoroudis, E.and Siskos, Y. Customer Satisfaction Evaluation. Methods for Measuring and Implementing Service Quality. International Series in Operations Research & Management Science, Vol. 139, 2010, pp Robbins, S. και Judge, T. Organisational Behaviour, 2011, Prentice Hall. Schlesinger, L.A. and Heskett, J.L. Customer satisfaction is rooted in employee satisfaction, Harvard Business Review, November-December, 1991, pp Zimmerman, R.D. and Darnold, T.C The impact of job performance on employee turnover intentions and the voluntary turnover process: A meta-analysis and path model, Personnel Review, Vol. 38, 2009, pp

276 Identifying factors of bank service quality during economic crisis in Greece Nancy Bouranta Department of Business Administration of Food and Agricultural Enterprises, University of Patras, 2 G. Seferi Street, Agrinio Greece, bouranta@gmail.com Christian Hurson IAE, Université de Rouen, 3, Avenue Pasteur, Rouen Cedex, France. Yannis Siskos Department of Informatics, University of Piraeus, 80, Karaoli & Dimitriou Street, Piraeus, Greece. Abstract Greek banking industry has been facing strong pressures resulting mainly from the country s economic concurrence. This crisis has led to structural changes in the banking sector, including mergers and acquisitions. In addition, competition between banks for attracting new customers or keeping the existing ones has become more intense. The unstable economic environment has a negative impact on customer confidence and trust in the banking industry, leading customers to be more demanding and careful in their bank selection. The purpose of this paper is to identify factors that influence Greek customers evaluation of bank service quality during the economic crisis period. A satisfaction questionnaire with self-determined scales was used and distributed to a randomly selected sample of customers. The data was processed using the MUSA system of multicriteria satisfaction analysis. The results show that, generally, bank customers seem to be satisfied with the service provided, as global satisfaction index is high; even the dissatisfied customers feel that their expectations have been met to an extent. Keywords: Service quality, Multicriteria decision analysis, MUSA system, banks. 275

277 1. Introduction The claim that service quality is among the most important selection criteria in bank selection has been supported (Frango et al., 2012). Thus, delivering high-quality financial service that keeps customers satisfied is viewed as an indispensable practice for staying alive and gaining a sustainable advantage. Vasiliadis (2009) pointed out that every market and every customer differs. Thus, the study of the specificities of each market and client will be considered in this case. In terms of this concept, it is worth assessing how Greek customers evaluate bank service quality during the crisis period. The main objective of this paper is to determine the factors that affect customer satisfaction in the Greek bank industry during the economic crisis period, providing managers with a more complete view of which factors regulate customers satisfaction. Moreover, the comparison of the current results with the results of older surveys may lead to interesting conclusions about customer evaluation and how it is affected by external circumstances. To do so a satisfaction survey was conducted in Greece using the MUSA multicriteria satisfaction analysis methodology. MUSA system was tested in empirical surveys of the Greek bank sector (Mihelis et al., 2001; Bouranta et al., 2002) and showed reliable measurements. The remainder of the paper is organized as follows. Section 2 presents a brief literature review about service quality in the bank sector. A brief sketch of the multicriteria methodological frame is outlined in section 3. The research findings are presented in section 4. Finally section 5 concludes the paper. 2. Literature Review In measuring service quality many frameworks have been developed. The most widely acknowledged between academics and practitioners and applied within service industries models are those of Grönroos (1990) and Parasuraman et al. (1988). Both models are based on the concept that a customer judges the quality of provided services based on the discrepancy among expectations and perceptions. However, some researchers have questioned SERVQUAL measurement raising theoretical and operational criticisms. One of these criticisms refers to its applicability to different service industries has been questioned in terms of the number and the nature of its dimensions (Jabnoun and Khalifa, 2005). The discrimination of bank services has been pointed by Vasiliadis, (2009, p. 89) who supported that unlike many other products, a bank s products are characterized by low levels of standardisation, high need for adaptations, high customer involvement in providing the service, and a need for a high volume of customers. Although no standard instrument for measuring service quality exists in the banking sector, most studies in this field have adopted SERVQUAL or the alternative SERVPERF model as the fundamental measure of service quality (Chen et al., 2012). Specifically, a modified version of SERVQUAL is used by Fatina and Razzaque (2014) in their survey involving retail banking services in Bangladesh. Ladhari et al. (2011), using the five SERVQUAL dimensions, compare 276

278 Tunisian and Canadian customer perceptions in the banking sector. All of the SERVQUAL dimensions were found to be important predictors of customer satisfaction and loyalty, while the importance of each dimension differs according to consumers in two countries, mainly due to their different economic and cultural environments. Mihelis et al. (2001) based on the assumption that customers satisfaction is an assessment of a set of criteria and subcriteria, proposed four criteria and nineteen sub-criteria. Their survey, which is conducted in the private bank sector of Greece pointed that customer satisfaction is a dynamic parameter changes in the current market can affect customers preferences and expectations (p. 357). The reputation of this survey in a totally different environment (during economic crisis) is among the purposes of this paper. The comparison of the results may lead to interesting conclusions about customer evaluation and how it is affected by external circumstances. In terms of attrition, the recent country economic crisis has created an exceptional environment that may determine customer evaluation of bank service quality. For example, Keisidou et al. (2013) point out that customers care more than they did before the economic crisis about the pricequality ratio, and they are not willing to pay a premium for the products and services they receive. 3. Methodological frame 3.1. Criteria of bank customer satisfaction As it was mentioned the hierarchical structure of customers satisfaction dimensions, proposed by Mihelis et al. (2001) was used as a base for this survey. The respondents were required to point out their own judgements about global bank service quality with regard to the set of criteria. The customer evaluation of each criterion was also measured using a single measurement index. The self-administered questionnaire is contained also of a set of sub-criteria. All the item were in the form of statements on Likert type scale, where 1 refers to the statement strongly disagree and 5 to the statement strongly agree. The main satisfaction criteria and sub-criteria consist of: Personnel: concerning personnel skills and knowledge, communication and collaboration with customers, as well as first line employee responsiveness. Products: refers mainly to the variety and price of the products and service (cards, loans, bankassurance, etc) as well as to the special service (leasing, factoring, internet banking etc.) Image: bank credibility (name, reputation), technological excellence (troubles in the service system like strikes, damaged ATMs, etc.), along with stores appearance are included in this criterion. Service: refers to the service offered to the customers; it includes waiting time (queue, telephone, etc.) and information provided (informing customers in an understandable way, explaining the service and other relevant factors, informing for new products, etc.). 277

279 Access: network bank expansion, branches location, as well as convenience are contained within this criterion The MUSA system The Multicriteria Satisfaction Analysis (MUSA) system of Grigoroudis and Siskos (2010) has been used in order to measure customer or employee satisfaction, assuming that their global satisfaction depends on a set of criteria representing service characteristic dimensions. Thus, the global satisfaction is denoted as a variable Y and the set of criteria is denoted as a vector Χ = (Χ1, Χ2,,Χn). MUSA system uses a multicriteria preference disaggregation logic. In the traditional approach, the criteria aggregation model is known a priori, while the global preference is unknown. The required information is collected via a simple questionnaire in which the customers evaluate the provided product/service, i.e. they are asked to express their judgments, namely their global satisfaction and their satisfaction with regard to the set of discrete criteria. Based on these assumptions, the problem is approached as a problem of qualitative regression and solved via special linear programming formulations where the sum of deviations between global satisfaction evaluation explicitly expressed by customers and the one resulting from their multicriteria satisfaction evaluations is minimized. The main results from the aforementioned preference disaggregation approach are focused on global and partial explanatory analysis. Global explanatory analysis lays emphasis on customers global satisfaction and its primary dimensions, while partial explanatory analysis focuses on each criterion and their relevant parameters separately. Satisfaction analysis results, in more detail, consist of: Global satisfaction index: it shows in a range of 0-100% the level of global satisfaction of the customers; it may be considered as the basic average performance indicator for the organisation. Global demanding index: it shows in a range of -100%-100% the demanding level of customers according to the following: demanding index 100%: extremely demanding customers, demanding index 0%: "normal" customers, demanding index -100%: non-demanding customers. Criteria/sub-criteria satisfaction indices: they show in a range of 0-100% the level of partial satisfaction of the customers according to the specific criterion/sub-criterion, similarly to the global satisfaction index. Weights of criteria/sub-criteria: they show the relative importance within a set of criteria or sub-criteria. Demanding indices: they show in a range of -100%-100% the demanding level of customers according to the specific criterion/sub-criterion, similarly to the global demanding index. 278

280 3.3. Sampling The survey was conducted in Greece. Respondents were customers of a particular bank and were approached by specifically trained interviewers during selected times of the day throughout a month-long period. The sampling method was similar to that of a mall intercept interview (Malhotra, 2004). While more than 200 customers were approached, only 182 gave their consent to participate. Finally, 151 usable questionnaires used for data analysis as thirty-one were excluded because they provided answers that were uniformly positive or negative (skewed responses). The survey took place in a provincial Greek city, similar in size population and structure, to the comparative one. As far as the demographic characteristics of the sample are concerned, the respondents are split reasonably evenly between males (53.6%) and females (46.4%). The customer age groupings are years (13.9%), years (19.2%), years (15.2%), (32.5%), and over 56 years of age (19.2%). 4. Results The majority of the respondents declared that they were from moderately (49.7%) to very satisfied (3.9%), while only 0.6% of them declared that they were very unsatisfied. Several of the respondents (35.8%) adopted a neutral attitude by declaring that they were neither satisfied nor unsatisfied. Globally, bank customers seemed to be satisfied with the service provided, as their satisfaction index is high (84.18%). However, it should be noted that only 3.9% of the participants stated that they are very satisfy. The global demanding index is %, indicating that customers are not very demanding, and this is in accordance with the high global satisfaction index. Bank customers are not demanding regarding the criterion of image, as the partial index is lower (-87.9%) than the average global demanding index (62.28%); this result explains its very high partial satisfaction index (95,69%). The partial impact index moves to very low levels (2.86%). This criterion illustrates the bank competitive advantage, and it should not be concentrated on bank efforts for improvement. The second benefit of the bank seems to be the criterion of accessibility, as it presents a rather high satisfaction level (73.56%). The criterion of serviceability exhibits the lowest satisfaction (45.69%), a level of demanding nature equal to 2.23%, and its partial impact index is also higher than the others (4.44%). This means that there is room for improvement regarding this criterion. The criteria related to personnel (62.10%) and financial products (60.18%), show lower satisfaction levels. The weights of all of the criteria, except that of image, fluctuate from 8.18% to 8.91%. Comparing the results of this survey which is conducted during the economic crisis and a relative survey by Mihelis et al. (2001) which is completed before the economic crisis, the following observations can be made: The global satisfaction index was before crisis 89.6% very close to the vale observed during the crisis (84.18%). 279

281 Skills and knowledge criterion seems to have greater important for the customers nowadays (weight 83.47%) comparing with their evaluation before the crisis (weight 6.2%). The personnel competiveness became more important than before as customers ask information about more complicated bank products and demand personalized financial solutions. Thus, branch personnel should be more focused on advice and expertise. Customers want also more than before crisis a variety of financial products to choose the best suited for them. Specifically, the criterion weight was 26.5% before crisis and 81.50% during crisis. One of the negative consequences of the financial crisis is customer feeling of insecurity. Because of this, they seek more control over their finances and they are more conservative. As customers lives become busier, they want not to wait in teller lines, for a financial advisor or wander through employee desks. The waiting time criterion weight was 6.9% and inclines to 90.82%. The partial satisfaction index is declined from 66.2% to 4.45%. 5. Discussion and Conclusion This criterion of brand name can be implied as the bank s advantage over competition. During the economic crisis, banks attempted to boost their brand names and rewind their reliability, as customers began to be more cautious. The bank examined in this case has a well-known brand name and a good reputation. However, the bank should take action in order to increase customer satisfaction on the criteria of serviceability, financial product, and employees. Serviceability has the lowest satisfaction rate compared to the other criteria. The bank should communicate customers needs to their employees and train them in order to nurture their capability to customize their bank services. Moreover, the bank should inform its employees on an ongoing basis about new financial products or services so they can answer customer questions and solve their problems. Employees should be encouraged to learn new skills, be alert to any external changes, be empowered, and exercise the delegation to make decisions. Providing individual attention to each customer, keeping promises, having the ability to conduct problem solving and decision making will build long-term relationship with customers. Acknowledgments This research has been co-financed by the European Union (European Social Fund) and Greek national funds through the Operational Program "Education and Lifelong Learning". 280

282 References Bouranta, A., Kouremenos, A., and Siskos, Y. Comparative satisfaction measurement of atms vs tellers. in C. Zopounidis (ed), New Trends in Banking Management, Physica-Verlag Heidelberg, N.Y., 2002, pp Chen, H.-G., Liu, J. Y.-CH., Sheu, T.S. and Yang, M.-H. The impact of financial services quality and fairness on customer satisfaction. Managing Service Quality, Vol. 22, 2012, pp Fatima, J. K. and Razzaque, M.A. Service quality and satisfaction in the banking sector. International Journal of Quality & Reliability Management, Vol. 31, 2014, pp Frangos, C.C., Fragkos, K.C., Sotiropoulos, I., Manolopoulos, G. and Valvi, A.C. Journal of Marketing Research & Case Studies, ID , pages, DOI: / Grigoroudis, E. and Siskos, Y. Customer Satisfaction Evaluation. Methods for Measuring and Implementing Service Quality. International Series in Operations Research & Management Science, Vol. 139, 2010, pp Grönroos, C. Relationship approach to marketing in service contexts: the marketing and organizational behaviour interface. Journal of Business Research, Vol. 20, 1990, pp Jabnoun, N. and Khalifa, A. A customized measure of service quality in the UAE. Managing Service Quality, Vol. 15, 2005, pp Keisidou, E. Sarigiannidis, L., Maditinos D.I. and. Thalassinos E.I. Customer satisfaction, loyalty and financial performance: A holistic approach of the Greek banking sector. International Journal of Bank Marketing, Vol. 31, 2013, pp Ladhari, R., Ladhari, I. and Morales, M. Bank service quality: comparing Canadian and Tunisian customer perceptions. International Journal of Bank Marketing, Vol. 29, 2011, pp Malhotra, N.K. Marketing Research, 4th ed., Prentice-Hall, Upper Saddle River, NJ, 2004, pp Mihelis, G., E. Grigoroudis, Y. Siskos, Y. Politis and Y. Malandrakis, Customer satisfaction measurement in the private bank sector, European Journal of Operational Research, Vol. 130, 2001, pp Parasuraman, A., Zeithaml, V.A. and Berry, L.L. SERVQUAL: a multi-item scale for measuring consumer perceptions of the service quality. Journal of Retailing, Vol. 64, 1988, pp Vasiliadis, L. Greek banks internationalisation: a suggested modelling approach. EuroMed Journal of Business, Vol. 4, 2009, pp

283 Touristic Guide: A prototype software for touristic journey planning Lampros Mpizas Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Nestoras Tsoutsanis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Zoi Moza Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Olena Pechak School of Chemical Engineering, National Technical University of Athens, Iroon Polytechniou 9, Zografou 15780, Athens, Greece. Dimitris Pantelis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Georgios K.D. Saharidis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece, saharidis@gmail.com Abstract Given a user profile, a well-defined network of sites and activities, and the availability of (realtime) information, Touristic Guide software will return to the user a collection of sites and activities that maximizes their preferences while respecting budget and time constraints at real time. Real-time is practically translated to the possibility of the system to take into account changes in real-time that affect the proposed plan (e.g. more time or cost spent in a site, a site is closed etc.) and require re-calculation. The user profile will be processed to eventually define a set of weights/preferences associated to each candidate site/activity that the system will make available. For instance, a certain user may be a family man who loves ancient history and is interested in agritourism and tours offering cultural and heritage experience. The prototype software would then favor places and activities that authentically represent the stories and people of the past, would promote visits to regions famous for their history, art, architecture, religion(s), and assign high score of preference to touristic elements of peoples way of life and lifestyle. Keywords: Touristic guide, Journey planning, Software, Multi criteria optimization. 282

284 1. Introduction Given a user profile, a well-defined network of sites and activities, and the availability of (realtime) information, the intended system will return to the user a collection of sites and activities that maximizes their preferences while respecting budget and time constraints at real time. Realtime is practically translated to the possibility of the system to take into account changes in realtime that affect the proposed plan (such as more time or cost spent in a site, a site is closed etc.) and require re-calculation. An already constructed network of sites and activities will feed the system with static information (e.g. distances between the sites, expected travelling time from one site to the other) and real-time information (e.g. related to availability, expected duration and visiting cost). An underlying optimization model will be constructed that will take into account the user profile, the network and the real-time information to output suggestions for visiting sites that maximize user preferences in time and budget. 2. Modeling The input and decision variables of Touristic Guide software are divided into two categories following their nature: Variables and data related to costs Variables and data related to time/duration The input and decision variables will be divided into two categories following their scope: Variables and data related to visits Variables and data related to transfers Eventually, a mixed-integer mathematical programming model is devised that aims to: Select sites and activities, such that the sum of the preferences assigned to the selected sites and activities is maximum. The total time spent on visits and transfers should be no greater to the time available by the user following their input. The money spent on visits and transfers should be no greater to the budget available by the user following their input. Mathematically-wise, the underlying problem is a mixture of two notoriously difficult problems (NP-hard): the travelling salesman problem and the knapsack problem. Anytime the user invokes the software, the mathematical programming model will be solved to output the suggested set of site/sight to visit and/or activities to undertake accompanied by the expected cost, expected time to spend and driving directions on how to travel from one site to the next. Unexpected events might occur during the tour and will be recorded in the network s database. For instance, a site or activity may be unavailable on the given day/time or the entrance cost may have changed. Alternatively, real-time changes might occur because the user has decided to spend more time than expected on a given site or changed their mind on what they want to see next. The system will re-solve the model taking into account the current user status (sites and types of sites 283

285 visited, available remaining budget and available remaining time to spend) in order to output the next optimal set of suggestions. 3. Software interface A beta version of the user interface of the prototype software is presented in the following Figure 1 where the user has introduced two numbers: The available time (ex. 300 minutes) and the available budget (ex. 400 Euro). Figure 1: Touristic Guide - User interface The user has the possibility to setup his/her profile by using the button PROFILE. Using the button PROFILE a new windows is popped-up (see Figure 2). Figure 2: Profile setup In PROFILE the user has to answer a number of question in order to give to the system the possibility to construct his/her profile. Based on this answers the weights of the objective function of the mathematical problem will be defined. Additionally, the user has the possibility to use two other buttons: SITE and ACTIVITIES buttons. In both buttons the user could find additional information about the available, by the software, sites and activities. 284

286 4. Conclusion In the current manuscript the Touristic Guide software is presented. The software returns to the potential user (ex. a tourist) a collection of sites and activities that maximizes his/her preferences while respecting budget and time constraints at real time. The prototype of the user interface of the software is presented as well as the available options to maximize the satisfaction of the potential user. The next step of this work is to introduce a critical mass of touristic information and transport data for the region of Thessaly as well as to improve the user interface of the software. The final outcome of this work is to develop a software full operational and available for free to any potential user. Acknowledge The author gratefully acknowledges financial support from the Department of Mechanical Engineering, University of Thessaly and InnovEco Company. 285

287 FindMyWay: A prototype web-based platform for journey planning in Athens city, Volos city and Crete island George Konstatzos InnovECO, 76, Kourtidou Str., , Athens, Greece, gkonsta@gmail.com George Emmanouilidis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Lampros Mpizas Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Nestoras Tsoutsanis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Zoi Moza Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Olena Pechak School of Chemical Engineering, National Technical University of Athens, Iroon Polytechniou 9, Zografou 15780, Athens, Greece. Georgios K.D. Saharidis Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos, Volos, Greece. Abstract The objective of the proposed research is to develop a Decision Support System (DSS) for a web based platform which will help individuals to move in Greece using public transportation means. The developed platform includes mainly three prototype platforms corresponding to three different regions of Greece. The first platform provides journey planning using public transport means in the region of Athens as well as the connection between city of Volos and Athens. The second platform provides journey planning in the city of Volos using the local buses and the third platform provides journey planning between the cities of Crete island using intercity buses. The final outcome of this research will be the development of a journey planner for the entire Greece. 286

288 Keywords: Journey planner, transport, web platform 1. Introduction Transport, the fastest growing sector in terms of energy use and emissions production, plays a central role in the European economy and accounts for almost 20% of total gross energy consumption in Europe, while 98% of the energy consumed in this sector is fossil fuel. Along with nutrition and housing, transport is responsible for 70 80% of all environmental impacts in urban zones. Climate change is one of the biggest challenges that man will be facing in the coming years. For several years, the European Union (EU) has committed to tackling climate change both internally and internationally and is taking action to curb greenhouse gas emissions in all its areas of activity. The Kyoto protocol is the first step towards cutting greenhouse gas emissions. This required emissions in to drop 8% below the 1990 levels. Furthermore, the EU's objective is to ensure that the global average temperature will not increase more than 2 C above its pre-industrial levels. To underpin these commitments, the European Commission established in 2011 two roadmaps for a Resource Efficient Europe and a low carbon economy in 2050, in reference to the objectives which should be reached by 2020 and the technological and structural changes that needed up to 2050, respectively. According to them EU has set the most ambitious greenhouse gas emissions reduction targets in the world, with binding mechanisms already in place that guarantee a unilateral 20% reduction by 2020 compared to the 1990 levels. The EU is committed to increase this to a 30% in This commitment will be reinforced if other developed countries commit themselves to comparable reductions, and if economically more advanced developing countries contribute adequately according to their responsibilities and respective capabilities. Therefore, it is necessary for the EU Member States to promote policies that will result in the prevention of climate change. Aiming at this direction, various industries have been called upon to measure their carbon footprints usually reported as greenhouse gas (GHG) emissions in CO2 equivalents. As reported by the Roadmap for a Resource Efficient Europe (2011) a better implementation of the existing legislation in combination with new science-based standards for air quality and the transition to a low-carbon economy would ensure the air quality benefits. For this reason, in order the EU to meet the air quality standards and to limit the significant impacts in the health and environment sectors it decides to propose strategies for the air quality and emission standards, further emissions reduction and implementation of air pollution and quality policies. Transport accounts for around a third of all final energy consumption in the EEA member countries and for more than a fifth of GHG emissions. It is also responsible for a large share of urban air pollution as well as noise nuisance. The last 10 years (mainly since the 2001 White Paper on Transport), a lot has been achieved in road and partly in rail transport. Nevertheless, according to the White Paper (2011), if we stick to the business as usual approach, energy consumption and CO2 emissions from transport instead of decreasing, would remain one third higher than their 1990 level. 287

289 Moreover, mobility belongs to one of three key sectors (in combination with nutrition and housing) which are responsible for 70%-80% of the environmental impacts in industrialized countries. As a result in order someone to deal with the challenges in energy and climate change; he/she should address the above sectors in collaboration with long term strategies, in reference to the implementation of partnerships under the scopes of the Resource Efficiency. FindMyWay in Greece is a demonstration project which will put into practice, methodologies and approaches in reference to combination of co-modal transportation, GHG emissions reduction and air quality improvement in the geographical region of Greece. While the idea of co-modal transportation seems to be known in other European countries, its absence in Greece results in a rather high gap in the transportation sector. The developed platform will be freely available to the public in an attempt to achieve the following goals: promote GHG emission minimisation and energy efficiency in transport raise awareness on the environmentally-friendliest journey planning decisions for passengers circulating in the Greek public transport network; highlight the importance, in terms of impact and results for the environment, on preferring comodal transport options against using own car; introduce an innovative policy in the Greek transport system, based on the efficient comodality scheme and EU emission regulations; The aim of FindMyWay project is to enhance sustainable mobility in Greece and reduce pollution and minimize public transport emissions in urban environment by delivering an environmentally friendly co-modal transport planner for passengers. Additionally, FindMyWay project contribute to the implementation of EU greenhouse gas emission reduction commitments under UNFCCC Kyoto Protocol, and help establish by 2020 the framework for a European multimodal transport information, management and payment system. Following the aim of FindMyWay project, the specific project objectives are: To develop tools that will accurately measure the emissions of public vehicle circulating in the EU public transport networks under consideration. To develop tools that will provide the environmentally friendliest journey from any point of the network under consideration to another by using public transport and optimize passengers logistics To change the culture and the commuting habits of passengers by providing an easy-to use service while raising awareness on the environmental benefits. To decrease the CO2-eq per passenger and kilometre. To bring together business, scientists, local and national authorities to deal with impacts of transport on air pollution and climate change in urban zones. 2. Prototype platforms In the frame of FindMyWay in Greece project the research team has developed 3 prototype platforms for three specific regions of Greece. The first platform provides journey planning using 288

290 public transport means in the region of Athens as well as the connection between city of Volos and Athens. The second platform provides journey planning in the city of Volos using the local buses and the third platform provides journey planning between the cities of Crete island using intercity buses. All the platforms are available through the site: The first platform is the one providing multimodal and intermodal routing solution. In the following Figure 1 the user interface of the platform is presented. Figure 1: FindMyWay in Athens The potential user requests by the platform the optimal route starting from the city of Volos and ending in a specific address in the city of Athens. The user has three different options suggested by the platform. The total travel time as well as the total CO2 emission produced are given as itinerary information. The second platform is the one providing intermodal routing solution. In the following Figure 2 the user interface of the platform is presented. 289

291 Figure 2: FindMyWay in Crete The potential user requests by the platform the optimal route starting from one city or village of Crete island and ending to another city of the island. The user gets the optimal route as well as information about the waiting time to the transitional node if there is not a direct route connecting the selected origin and destination. Finally, the third platform is the one providing routing solution for the city of Volos. In the following Figure 3 the user interface of the platform is presented. 290

292 Figure 3: FindMyWay in Volos city The potential user requests by the platform the optimal route starting from one address in the center of the city of Volos and ending to another address. The user gets the optimal route represented in the map as well as information about the local buses that he/she should use in order to arrive at his/her destination. 3. Conclusion In this research work three prototype platforms for three regions of Greece are presented. The next step of this work is to collect transport information for additional regions of Greece and introduced them in the database developed in the frame of FindMyWay project. The final outcome of this research will be the development of a journey planner for the entire Greece connecting all cities and villages having a population greater or equal to Acknowledge The author gratefully acknowledges financial support from the Department of Mechanical Engineering, University of Thessaly and InnovEco Company. 291