Impact Analysis of Mega Vessels on Container Terminal Operations

Transcription

1 Available online at ScienceDirect Transportation Research Procedia 25C (2017) World Conference on Transport Research - WCTR 2016 Shanghai July 2016 Impact Analysis of Mega s on Container Terminal Operations Qiang Meng a*, Jinxian Weng b, Suyi L c a Department of Civil and Environmental Engineering, National University of Singapore, Singapore b College of Transport and Communications, Shanghai Maritime University, Shanghai, , China c Centre for Maritime Studies, National University of Singapore, Singapore Abstract Mega vessels currently play a vital role in maritime transportation and their deployment may have significant impacts on container terminal operations. This study is concerned with the impact analysis of mega vessels on container terminal operations. First, the container operation process at a container terminal is formulated as a queuing network. Based on the queuing network, a simulation model is then developed. Because of the computational complexity of the simulation, the ARENA software tool is used to solve the developed model, based on a realistic case involving the Hong Kong port. The case analysis comprises ten scenarios that represent current and possible future situations regarding the utilization of more mega container vessels. The results suggest that the current port facilities may not be sufficient to accommodate more mega container vessels The Authors. Published by Elsevier B.V The Authors. Published by Elsevier B.V. Peer-review under responsibility of WORLD CONFERENCE ON TRANSPORT RESEARCH SOCIETY. Keywords: Simulation; Container Terminal; Queuing Network; Ship Size 1. Introduction The container shipping industry has witnessed fast growth in the last three decades due to the increase in the demand for containerized cargo delivery. As reported by Ebeling (2009), more than 90% of the world s trade goods are transported in containers. As stated by Cullinane and Khanna (2000), a larger container vessel may produce lower unit costs, owing to the existence of scale economies, thus making shipping liners disposed to use more largesized container vessels. Although there are no global standards, container vessels can roughly be classified based on their dimensions and carrying capacities. As reported by MAN Diesel (2008), the Panama Canal is often used as a standard to classify container vessels. Due to the restrictions of the Panama Canal, the capacity of a container vessel cannot exceed 5,000 TEUs (twenty-foot equivalent units) before Since the 1990s, the Post-Panamax vessels, with capacities of 5,000-10,000 TEUs, have made up the majority of fleets serving the Asia-Europe and Trans-Pacific lines. The Panama Canal has planned to build a new lane to handle the larger container vessels and the project is expected to The Authors. Published by Elsevier B.V. Peer-review under responsibility of WORLD CONFERENCE ON TRANSPORT RESEARCH SOCIETY /j.trpro

2 188 Qiang Meng et al. / Transportation Research Procedia 25C (2017) be completed in The New-Panamax vessels, with capacities of 12,000-14,500 TEUs, will be able to travel through the canal after the project is accomplished. The major ports, such as the port of Singapore, Hong Kong port, and the port of Rotterdam, could be greatly affected by this increase in the use of large-sized container vessels. The main issue is that, when a large container vessel arrives, the terminal needs to handle thousands of containers within a short time window. From the viewpoint of port operators, examining the impacts of the increasing calls from the mega vessels at their ports is of great interest, as it could help them make better-informed decisions, such as whether or how much to expand their existing port infrastructure and facilities, and how to enhance their productivity to attract more business. On the other hand, shipping companies are also interested in this impact analysis. Although mega vessels are cost-efficient in terms of dollars per TEU mileage, their turnaround time may cause serious concerns, especially in light of the high capital investments and maintenance costs. This study aims to investigate the productivities and efficiencies of container terminals under an assumption that calls from mega vessels will continue to increase in the future. The results of this study will shed some light on the impacts brought by the increase of mega container vessels visits. Bottlenecks at current container terminals for accommodating more mega vessels, and the turnaround time of mega vessels will be investigated in this study Literature review Research topics related to container terminal operations can be divided into three categories. The first category focuses mainly on the planning and development of container terminals, including the studies of container terminal equipment, layout planning, implementation of information systems, and the automation of transport trucks (e.g., van Hee and Wijbrands, 1988; Gademann and Van De Velde, 2000; Nam and Ha, 2001; Alicke, 2002; Lokuge and Alahakoon, 2007; Roy and De Koster, 2012). The second category is concerned with solving problems associated with container terminal operations, such as the berth allocation problem, the quay crane allocation problem, human resource scheduling problems, strategies for stacking containers, strategies for using yard cranes, and rules for dispatching transport trucks using optimization models (e.g.,daganzo,1989; Peterkofsky and Daganzo,1990; Steenken et al., 1993; Kozan and Preston, 1999; Imai et al., 2001; Bish, 2003; Chen et al., 2007; Imai et al., 2007; Said et al., 2014). The third category includes various case studies which provide links between the theories and their practical applications (e.g., Nam et al., 2002; Shabayek and Yeung, 2002; Steenken et al., 2004; Cordeau et al., 2007; Stahlbock and Voß, 2008). The queuing network model has been used to investigate container terminal operations for decades. Edmond and Maggs (1978) applied queuing models to address berth assignment and investment decision problems. Legato and Mazza (2001) presented a queuing network model of logistic activities related to the container terminal operations, and used it to solve berth planning and optimization problems by considering the case of the Gioia Tauro terminal in Italy. A simulation model was also developed based on the queuing network model, in which the number of quay cranes was considered a decision factor and the waiting time of a container vessel was viewed as a performance measurement. Lagana et al. (2006) further simplified this queuing model, and solved the problem using a distributed computation approach. More recently, Canonaco et al. (2008) constructed a queuing network model to represent loading and unloading procedures in a container terminal. They aimed to provide decision aids for quay crane and transport truck assignments. A simulation model was developed in the Delphi software. Roy and De Koster (2014) used new integrated queuing network models to rapid design evaluation of container terminals with Automated Lift Vehicles (ALVs) and Automated Guided Vehicles (AGVs). Shabayek and Yeung (2000) used a queuing model to analyze a Hong Kong container terminal. However, the model only took into account the uncertainty associated with vessel arrivals and the quay crane service rate and is therefore over-simplified. Shabayek and Yeung (2002) later took into account more uncertainties and used a simulation approach to analyze their model. However, the model still did not capture the inter-correlation between yard side and quay side operations. Since it is well accepted that the queuing network can appropriately represent container terminal operations, we will also apply this idea in our study. Simulation has been used extensively to analyze container terminal operation processes, since analytical methods are hardly feasible for solving detailed large-scale problems (e.g., Khatiashvili et al. 2006; Gambardella et al., 1998; Kia et al., 2002; Kozan, 1997). Liu et al. (2002) evaluated the performances of four different automated terminal

3 Qiang Meng et al. / Transportation Research Procedia 25C (2017) design concepts by using the simulation approach. The costs incurred in handling one container were considered for each of the four design concepts. The study revealed that the Automated Guidance Vehicles (AGV) system performed the best among the other terminal design concepts, and that automated terminal equipment can dramatically increase a port s throughput and reduce handling costs. Due to the complexity and computational burden, Bielli et al. (2006) developed a simulator for the container terminal and proposed a distributed simulation approach to evaluate different port policies and improve management proficiencies. The model comprised several parallel computational processes on different machines that are interacted with each other by sending and receiving messages. Huang et al. (2008) presented a simulation tool which integrates all the activities of a container terminal to analyze its capacity. Petering and Murty (2009) also utilized a simulation approach, in this case, to analyze the impacts of the layout of container stacks on the quay crane rate in order to obtain the optimal stack length. This paper proposes a simulation model for container terminal system analysis. Veloqui et al. (2014) developed a simulating solution to analyse the congestion problem in the Port of Naples. The results of this study showed that the solution should consider the decreasing simultaneously the time of service in the access gate and in the yard. It can be seen from the existing literature that little work has been done to examine the impacts of mega vessels on container terminal operations using an efficient simulation approach. However, more and more shipping liners want to use large-sized container vessels because of their low unit costs. Therefore, this study makes an effort to examine the impacts of mega vessels on container operations, and whether the current port facilities are able to accommodate them using a simulation approach Objectives and contributions In this study, we develop a queuing network model to capture all key operation processes in a container terminal and propose a simulation tool to implement the model. Moreover, a real case is used to test the applicability of the proposed model and the simulation tool. The results give insightful perspectives on the impact of using more mega vessels at major container ports. This information will benefit both the ports and the shipping companies. The contributions are twofold. First, we independently develop a flexible simulation model with a reasonable level of details. Industry users can easily modify the model for their own purposes and the model could also be used for many other purposes, such as to assess new rules for allocating port equipment, and thus provide useful information to the port managers. Second, the proposed simulation model can be considered the first attempt to quantitatively analyze the impact of mega vessels on container terminal operations. The remainder of the paper is organized as follows. Section 2 describes the queuing network that we use in the study. A simulation model, as well as the input and output parameters for the model, are presented in section 3. In section 4, the simulation model is used to examine the productivity and efficiency of a real container terminal. Conclusions are drawn in the last section. 2. Assumptions, Problem Statement and Queuing Network Model Formulation Two main tasks need to be accomplished when a container vessel calls at a port. One is to unload containers from the vessel to the port and the other is to load containers from the port to the vessel. In general, the unloading process will be executed firstly, followed by the loading process. Vis and Koster (2003) explored in detail the procedure of loading, and unloading and transshipping containers at a container terminal. More details on loading and unloading operations are described below. In the unloading process, the vessel will first wait for an available berth. When a berth that fits the vessel s size (length, width, depth, etc.) is available and the port permits the vessel to enter, it begins berthing. Once the berthing process is complete, the quay cranes (QCs) that have been assigned to the vessel will unload containers from the vessel onto the internal trucks. After picking up a container, the truck will transport it to a certain storage stack and wait to be served by a yard crane (YC) which is capable of lifting the container from the truck onto the stack. When the YC finishes the unloading process, the internal truck may return to the quay side to pick up another container or transport another container to the quay side ready for the loading process. Containers that are destined for the port s hinterland will wait for external trucks to pick them up. When the external truck arrives, YCs are needed again to lift the containers onto the trucks, and the trucks will then approach the yard gate to wait for clearance.

4 190 Qiang Meng et al. / Transportation Research Procedia 25C (2017) The containers that need to be transshipped will wait for their vessels. The loading process can be considered as the inverse of the unloading process. An internal truck first goes to a storage stack and waits for the YC to lift a container onto it. The truck then proceeds to a preassigned QC and waits for the QC to lift the container onto the vessel. 6, Container from hinterland 1, Berthing 2, QC unloading 3, YC stacking 4, External YC unstacking 5, Container to hinterland 9, departure 8, QC loading 7, Internal YC unstacking Fig. 1. Diagram of the proposed queuing network The unloading and loading processes described above,using a network of queues, is shown in Fig. 1. The network consists of several interconnected queues. In queuing theory, a queue has some basic elements, namely, customers, servers, arrival rate of customers, service rate of servers, and a priority rule (e.g., first-come-first-served, first-come-last-served). We define such elements for each queue and make the following major assumptions. Queue 1 - berthing. Upon a vessel s arrival, it waits for an available berth, QCs, and tugs. In this queue, the visiting vessels are the customers, and the tugs together with the pilot service are the servers. Furthermore, the availabilities of berths and QCs also impose restrictions on the queue. Many studies have already shown that the Poisson process is an appropriate means of modeling the vessel arrivals (e.g., Kozan, 1997; Pachakis and Kiremidjian, 2003; Huang et al., 2008). This is because inter-arrival times become random due to unpredictable weather conditions and delays in service caused by other ports. The situation also fits well with the exponential distribution when the vessel traffic to the whole port is considered. For the Hong Kong port, the assumption of Poisson arrivals of vessels has already been justified. For example, Shabayek and Yeung (2011) collected 11 months of data from the Hong Kong port and found that the historical inter-arrival times agreed reasonably well with the exponential distribution. According to this finding, they concluded that the Poisson distribution was appropriate to represent the arrival process at the Hong Kong port. Hence, this study also assumes that the arrival process of customers follows the Poisson distribution. The service process depends on the vessel s size and follows a triangular distribution. Queue 2 - QC unloading. After the berthing process, QCs start to unload containers onto the internal trucks (ITs). Several QCs may work on one vessel simultaneously; the number of QCs allocated to a vessel depends on the vessel s size and the terminal operator s rule. It is reasonable to assume that a QC only lifts containers when a truck is waiting under it, although some buffer capacity may exist. In this queue, the containers to be unloaded are considered as customers and quay cranes are regarded as servers. For every berthed vessel, many customers (containers) arrive in the same time. The service rate of a quay crane follows the triangular distribution. The availability of internal trucks is another restriction. Queue 3 - YC stacking. Three categories of containers are temporarily stored in yard stacks: (i) the containers that are unloaded from a vessel and are going to be carried to the port s hinterland, (ii) the containers that are unloaded from a vessel and are going to be loaded onto another vessel, and (iii) the containers originating from the

5 Qiang Meng et al. / Transportation Research Procedia 25C (2017) hinterland and that are going to be exported. The first two categories of containers are transported by internal trucks, and those in the third category are transported by external trucks. Once a truck arrives at its designated stack, it waits for an available YC to lift containers onto the stack. In this queue, the trucks are the customers and the YCs are the servers. The arrival rate depends on the truck s travelling time and the outputs of Queue 2 and Queue 6. The service process is assumed to follow a triangular distribution. Queue 4 - external YC unstacking. The containers destined for the hinterland wait for external trucks to pick them up. Once an external truck arrives at its designated stack, it waits for an YC to become n available to lift the containers onto it. In this queue, the external trucks are the customers and the YCs are servers. The arrival process of customers is assumed to be in accordance with a Poisson distribution. The service process is assumed to follow a triangular distribution. Queue 5 transporting containers to the hinterland. After picking up containers, external trucks approach the yard gate for clearances. In this queue, external trucks are customers and gate lanes are servers. The arrival process depends on the outputs of Queue 4 and the service process follows the triangular distribution. Queue 6 - containers transported from the hinterland. The containers from the hinterland are transported to the port terminal by external trucks. In this queue, the external trucks with their loaded containers are the customers and the gate lanes are viewed as servers. The arrival process is assumed to follow a Poisson distribution and the service process follows a triangular distribution. Queue 7 - internal YC unstacking. Internal trucks are sent to pick up the containers that are going to be loaded onto vessels from the internal stacks. In this queue, the containers are the customers and the YCs are the servers, moving the containers from the stacks to the internal trucks. The YCs will only begin to serve the containers after the internal trucks arrive. The service process is assumed to be triangularly distributed. Queue 8 - QC loading. After picking up the containers, the internal trucks travel to the quay side and wait for QCs to lift the containers from the trucks onto these vessels. In this queue, the trucks with the containers are the customers, and the QCs are the servers. The arrival rate depends on the outputs of Queue 7. The service process follows a triangular distribution. Queue 9 - vessel departing. Once the unloading and loading tasks have been completed for a vessel, it will depart from the port. The vessel has to wait for tugs and the pilot service and to obtain permission to leave from the port. In this queue, the departing vessels are the customers and the relative services on the port side are the servers. The arrival rate depends on the outputs of previous queues. The service process is assumed to follow a triangular distribution. 3. Simulation Model We consider N types of container vessels, labeled by Type 1,,, Type j,, Type N. The j th type of min max min max vessel has an overall length range of ( lj, l j ), and a capacity range denoted by ( Cj, C j ). Furthermore, let Q j denoted the number of QCs assigned to the j th vessel type. We denote the proportion of all container vessels visiting a terminal that are of type j to be P j %. The total quay length of the terminal is considered to be divided into M separate berth areas. The length of the i th berth area is denoted by LB i. We further assume that QCs do not move from one berth area to another, so QCs can be divided into M groups as well, with each group associated with a specific berth area. The number of QCs in the i th group is denoted by N BQi. We assume that one QC can handle one FEU (forty-foot equivalent unit) or two TEUs at a time. The number of internal trucks is N IT, and we assume that one truck can carry one FEU or two TEUs at a time. The number of QCs is N YC. The number of gate lanes is denoted by N G and it is assumed that incoming and the outgoing trucks can use any of the lanes. The arrivals of vessels are assumed to follow a Poisson process with parameter l v. The annual container throughput is represented by C TEUs, the number of inward containers by I TEUs, and the number of outward containers are denoted by (C-I) TEUs. The transshipment percentage of inward containers is represented by C I %, and the transshipment percentage of outward containers is denoted by C O %. Several factors may affect the service rate of a QC, such as the port operator s experience, the arrangement of the containers on the vessel s deck, the discharging sequence, and the percentage share of TEUs. Under certain

6 192 Qiang Meng et al. / Transportation Research Procedia 25C (2017) container arrangements on the deck, a QC may not be able to lift two TEUs at a time. In our model, we assume that min mode max the QC service process follow a triangular distribution with parameters ( T, QC T, QC T ). This means that most QC mode min probably a QC will take T minutes to make a lift, at the least it will take QC T minutes to make a lift, and at the QC most it will take T max minutes to make a lift. Similarly, we assume that the service process of a YC follows a QC triangular distribution with parameters ( T, T, T ). The transportation time of an internal truck depends on min YC mode YC max YC many factors, such as the distance it has to travel, the route, the port s layout, the congestion level, and whether it is loaded with containers. For the sake of presentation, the transportation time for an internal truck loaded with containers and for an empty truck are considered separately and follow two different triangular distributions. More specifically, when an internal truck is empty, its transportation time follows the triangular distribution with min mode max parameters ( T, IT1 T, IT1 T ), while a truck loaded with containers has a transportation time that follows the IT1 min mode max triangular distribution with parameters ( T, IT 2 T, IT 2 T ). The transportation time of an external truck without IT 2 min mode max containers is assumed to follow the triangular distribution with parameters ( T, ET1 T, ET1 T ), while that of a truck ET1 min mode max loaded with containers is assumed to follow the triangular distribution with parameters ( T, ET 2 T, ET 2 T ). For ET 2 min mode max each gate lane, we assume that the service process follows the triangular ( T, G T, G T ) distribution. The G output parameters are listed in Table 1. Table 1. Output parameters of the proposed simulation model R Bi R qi R YC R IT T Wj Utilization rate of every berth area i Utilization rate of every QC group j Utilization rate of YCs Utilization rate of internal trucks Wait time of vessel type j T Sj Total stay time (turnaround time) of vessel type j In the model, we try to represent realistic container terminal operations, and avoid making strong assumptions. The model consists of many logical constraints, policies and rules, thus presenting a large-scale problem. Hence, an analytical approach is hardly feasible for solving the queuing network model and a simulation approach would be a more useful tool (Shabayek and Yeung 2002). The ARENA is a suitable software tool to conduct queuing network simulation and it has been employed in a recent study (Shyshou et al., 2010) to simulate a network of queues for offshore anchor handling fleet sizing problem. In ARENA software, the model can be easily adjusted with the built-in process module, and not much programming technique is needed. In Fig. 2, we present the flow chart of our ARENA simulation model s main level, using the default flow chart shapes from the ARENA software. In this model, nine types of container vessels (the choice of vessel types will be discussed later) are considered. The first type is a river vessel and the other eight types ( to Type8) are ocean vessels. We classify the ocean vessels by their sizes and carrying capacities and assume that a different number of QCs is assigned to each vessel type. For instance, a vessel needs one QC, a vessel needs two QCs and the biggest vessel (Type8) needs eight QCs.

7 Qiang Meng et al. / Transportation Research Procedia 25C (2017) Arrive Particular Unload and External YC Unstacking Pass Gate Record container to land Arrive Type 1 Unload and Dispose to Land Type Type 3 Type 4 Type 5 Unload and Unload and Unload and Unload and Assign Container Particular Record Transship Container Balance Transshipm ent Percentage Hold Container Pool Dispose Exported Container Record Exported Container Type 6 Unload and Initial Stock Record Hinterland Container Type 7 Type 8 Unload and Type8 Unload and Container from hinterland Container from hinterland process Fig. 2. The ARENA simulation model, main level The vessel arrive module creates a flow of river vessels visiting the container terminal, the vessel arrive module creates a flow of ocean vessels, which is further divided into eight vessel types through the Type module. The Particular and Type X modules assign various particulars to the generated vessels. For every vessel, we randomly assign its length, maximum TEU capacity, the percentage of containers that need to be discharged and the percentage of containers that need to be loaded. Every vessel will go through the unload and load sub-module, and then leave the system. The unloaded containers proceed to be stacked in the yard. Containers from the port s hinterland enter the system and are also stacked in the yard. Transshipped containers are loaded onto vessels and leave the system. Containers destined for the hinterland leave the system on external trucks. In Fig.3, we present the expanded sub-module of the Unload and load module for illustration purpose; we will not go through all of the details of the model in this paper. It should be pointed out that the proposed simulation model is flexible and any part of the model can be modified easily.

8 194 Qiang Meng et al. / Transportation Research Procedia 25C (2017) Wait Berth Record Wait Berth Time Berthing Wait QC Record Wait QC Time unload container number QC Unload container finish unload Start load change entity type mark as finish unload assign container particular Record unload Time Dispose 1 container load number Check container pool YC Unstacking decrease container pool IT move to Yard Depart Start Depart mark as finish load QC Release Berth and QC Record Total Stay Time Record Depart count YC Stacking Dispose 2 Fig. 3. Expanded sub-module of Unload and load module 4. Assumptions, Problem Statement and Queuing Network Model Formulation As one of the five container terminal operators at the Hong Kong port, MTL is located in the Kwai Tsing container terminal, as shown in Fig. 4. In the figure, the four separate berth areas of MTL are denoted by the letters A, B, C, and D Data collection regarding port facilities The data regarding the geometric configurations of each berth area and the number of QCs used was visually collected from satellite pictures provided by GoogleTM Maps. In addition, the total container throughput of MTL is 5.9 million TEUs and the number of yard cranes is 106 (Hong Kong Marine Department, 2009c). The number of internal trucks is not available from the existing literature or any reports. According to Shanghai Pudong International Container Terminals Ltd. (2004), they were running 36 YCs and 73 trucks. Based on our preliminary simulation experiments, we found it appropriate to assume that 200 internal trucks are used in MTL, because this number will roughly balance the utilization rates of the various items of port equipment Data collection regarding parameters used in queuing network The service rate of aqc is an important index for measuring the performance of a container terminal. According to Petering and Murty (2009), a quay crane s peak service rate could be 40 lifts/hour. Most terminals will only achieve 25 lifts/hour on average. According to the Hong Kong Marine Department (2006), the Hong Kong container terminal s QC service rate is 36 lifts per hour on average. It should also be noted that one crane can lift one or two TEUs. Based on our preliminary simulation experiment results, we set the MTL quay crane service rate to 50 TEUs/hour on average, with the highest and lowest service rates set to 60 and 40 TEUs/hour, respectively. The Hong Kong Marine Department (2009a) provides the daily schedule of vessels visiting the Hong Kong port.

9 Qiang Meng et al. / Transportation Research Procedia 25C (2017) A total of 582 records of container vessels were collected between October 14 and October 23, A list of the characteristics of each container vessel (e.g., size, capacity in terms of TEUs) was compiled from the collected records. This information provided us with the container vessel fleet mix currently visiting the Hong Kong port. It is assumed that the smallest type of ocean container vessels (400-1,000 TEUs) need one QC and the largest vessels (20,000 TEUs and above) need eight QCs. MTL also serves many river container vessels. We consider all river vessels to be the same type, and further assume that each river vessel is served by one QC.Avoid hyphenation at the end of a line. Symbols denoting vectors and matrices should be indicated in bold type. Scalar variable names should normally be expressed using italics. Weights and measures should be expressed in SI units. All non-standard abbreviations or symbols must be defined when first mentioned, or a glossary provided. MTL DP World HIT COSCO- HIT ACT Berth Area B: 660m, 9 QCs Berth Area A: 500m, 6 QCs Berth Area C: 900m, 12 QCs Berth Area D: 340m, 4QCs HIT MTL Quay Crane Yard Stack & Yard Crane Fig. 4. Bird s-eye view of Kwai Tsing container terminal and the berth areas operated by MTL Table 2. Container vessel types and the percentage share of each type of vessels under each fleet mix Types Capacity Range (1000 TEUs) Length Range (m) No. of QCs Fleet Mix 1 Fleet Mix 2 Fleet Mix 3 RV [0.1, 0.3] N.A. N.A. N.A. OT 1 [0.4, 1.0] % 6% 3% OT 2 [1.0, 3.0] % 35% 7% OT 3 [3.0, 5.0] % 20% 10% OT 4 [5.0, 7.0] % 15% 15% OT 5 [7.0, 10.0] % 13% 20% OT 6 [10.0, 12.0] % 6% 25% OT 7 [12.0, 20.0] % 5% 15% OT 8 [20.0, 22.0] % 0% 5%

10 196 Qiang Meng et al. / Transportation Research Procedia 25C (2017) Note: RV= vessel; OT= Type; N.A.=Not available In this study, the following three different fleet mixes are considered: (i) Fleet Mix 1: the current fleet mix based on the records we compiled; (ii) Fleet Mix 2: the fleet mix for the near future (e.g., the next three years) (iii) Fleet Mix 3: the fleet mix for the distant future (e.g., ten years in the future). Based on the data reported in MAN Diesel (2008), we can estimate the fleet mix in the near future (i.e., Fleet Mix 2), based on the difference between the existing fleet mix of world container vessels in 2008 and the fleet mix of container vessels under construction in 2008 (expected to be delivered 2-3 years later). For the third fleet mix (i.e., Fleet Mix 3), we assume there are more ultra large container vessels and the estimation on the percentage shares of the various types of vessels is somehow subjective. The classifications of the vessel types together with their percentage shares are listed in Table 2. Various other key data were collected from the Hong Kong Marine Department (2009b), as shown in Table 3. Based on these data, we can further calculate or estimate the remaining input data for the simulation. Table 3. Data relevant to the Hong Kong port A1. MTL throughput (1,000 TEUs) 5900 A2. Number of ocean vessels arriving at the container terminal (with containers to be handled) A3. Number of river vessels arriving at the container terminal (with containers to be handled) A4. Total throughput of the container terminals (1,000 TEUs) A5. Throughput of ocean vessels at the container terminal (1,000 TEUs) A6. Throughput of river vessels at the container terminal (1,000 TEUs) 2413 A7. Total containers unloaded (TEUs) 9840 A8. Total containers loaded (TEUs) A9. Total containers unloaded directly (1,000 TEUs) 3531 A10. Total containers unloaded by transshipment (1,000 TEUs) 6309 A11. Total containers loaded directly (1,000 TEUs) 3922 A12. Total containers loaded by transshipment (1,000 TEUs) Scenarios in the simulation Based on the data from the Hong Kong Marine Department (2009b), the average annual growth rate in the container throughput of the Hong Kong port between 2002 and 2008 was 4.67%. It is thus reasonably to assume a 5% annual growth rate in container throughput in the simulation. In this study, we denote the container throughputs in 2008, 2010 and 2018 as throughputs 1, 2, and 3, respectively. Moreover, a straight berth area instead of several separate berth areas is also considered in the simulation. The simulation scenarios are given in Table 4, of which Scenarios 1, 2, and 3 are the situations that are most likely to happen. In Scenarios 4 and 5, we fix the fleet mix to the current situation and only change the throughput. In Scenarios 6 and 7, we fix the throughput and only change the fleet mix. In addition to the separate berth scenarios, we set Scenarios 8-10 to have a straight berth area, and compare them to Scenarios 1-3, respectively.

11 Qiang Meng et al. / Transportation Research Procedia 25C (2017) Table 4. Ten scenario settings for the MTL case Fleet Mix 1 Fleet Mix 2 Fleet Mix 3 Throughput 1 Throughput 2 Throughput 3 Separate Berth Straight Berth Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 6 Scenario 7 Scenario 8 Scenario 9 Scenario Footnotes The proposed model is stochastic and as a result the simulated results are not always the same. To minimize stochastic errors and obtain relative stable results, it is necessary to perform the simulation more than once. However, the more simulation runs, the simulation time required is longer. Therefore, it is preferable to find the particular number of simulation runs, which not only provides sufficiently precise results but does not increase the simulation time too much. Based on the theory of probability and statistics, the following equation can be used to estimate the required number of runs if the mean with a specified confidence interval and an error range are provided. N 2 æza /2s ö run = ç è E ø (1) where N represents the required number of simulation runs, s is the sample standard deviation, run z a is the /2 threshold value for a 100(1 - a) percentile confidence interval and E represents the allowed error range. For this study, we initially ran the simulation model 30 times. For each scenario, the simulation required a warm-up period of 10 days, and another 60 days for collecting statistics (e.g., the sample mean value and standard deviation). Based on the preliminary statistics, we were then able to calculate the required number of simulation runs can be obtained using Eq.(1). Taking the total stay time in Scenario 1 as an indicator, and targeting to estimate the expected total stay time with a relative error of 0.05 at an confidence level of 95%, 30 replications (with different random seeds) were conducted in the preliminary testing. It was found that the mean total stay time of OT7 is hr and the corresponding z0.025s =3.76 hr. Hence, the required number of simulation runs for OT7 in terms of the total stay stay time time, denoted by N, can be estimated to be OT7 N stay time æ 3.76 OT7 ö = ç = 20 è ø 2 (2)

12 198 Qiang Meng et al. / Transportation Research Procedia 25C (2017) Therefore, the number of simulation runs for all vessel types in terms of the total stay time, denoted by can be estimated by stay time stay time stay time stay time RV OT1 OT7 stay time N, N = max{ N, N,, N } = 20 (3) As the utilization rate is also an important indicator in this study, we also calculated the this gives the number of replications required to give a stable result for this output value, as follows: N = max{ N,, N, N,, N, N, N } = 9 (4) utilization utilization utlization utilization utlization utilization utilization Berth A Berth D QC A QC D IT YC Obviously, based on this, 30 replications will be adequate to generate a relative error of not more than 0.05 with a confidence level of 95% in Scenario 1, for both the total stay time and the utilization rate. Similarly, 30 replications were also found to be appropriate for the other nine scenarios. Table 5 gives the container terminal equipment utilization rates under each scenario. Table 6 gives the wait time and total stay times of each vessel type, where the wait time refers to the time from a vessel s arrival until the quay cranes start working on the vessel, and the total stay time means the time between a vessel s arrival and its departure. 7 6 Utilization S2-S1 7 6 Utilization S3-S Throughput Berth QC YC Internal Truck Throughput Berth QC YC Internal Truck 25% Total Stay Time S2-S1 1200% Total Stay Time S3-S1 20% 1000% 15% 10% 800% 600% 400% 5% 200% 0% Throughput 0% Throughput Fig. 5. Comparison of utilization rates and total stay times between Scenarios 2 and 1, and between Scenarios 3 and 1 Figs. 5-8 compare the results for the various scenarios. The effects of using more mega vessels are measured as the marginal increase in equipment utilization rates and the marginal increase in the vessels stay time. For instance, the berth utilization increase rate of S2-S1 (Scenario 2 compared to Scenario 1) is calculated as the difference between the berth utilization rate of Scenario 2 and that of Scenario 1 divided by the berth utilization rate of Scenario 1. Based on these results, we draw the following conclusions: In the near future (i.e., Scenario 2), the MTL will still be able to handle the increased throughput and changed fleet mix because the utilization rates of the various pieces of equipment will not increase significantly. The total stay time does not increase significantly either. Larger vessels tend to cause a larger marginal increase in the total stay time, but their total stay time is still acceptable. However, if more mega vessels visit the terminal in the distant

13 Qiang Meng et al. / Transportation Research Procedia 25C (2017) future (Scenario 3), the MTL will not be able to handle them using the current equipments. This is because the utilization rates of the equipment will increase too much and the total stay times for the majority of the fleet will be unacceptable. For example, the average utilization rates of four berths increase to 92.58% in Scenario 3, as shown in Table 5. In addition, the total stay times for OT5, OT6, OT7 and OT8 range from 118 hours to 274 hours, shown in Table 6 and Fig.5. The high utilization rates and big total stay times for these mega vessels suggest that the MTL terminal as it is at present will not be able to meet the higher demand in the distant future. Table 5. Container terminal equipment utilization rates, MTL case Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 6 Scenario 7 Scenario 8 Scenario 9 Scenario 10 Berth A 50.72% 54.28% 83.14% 62.88% 89.69% 43.83% 31.95% NA NA NA Berth B 60.01% 64.38% 90.93% 72.11% 93.09% 54.14% 44.25% NA NA NA Berth C 64.05% 70.96% 94.23% 75.68% 94.67% 60.24% 56.93% NA NA NA Berth D 62.15% 71.95% 95.85% 75.85% 95.99% 59.65% 69.17% NA NA NA All Berth 60.61% 67.60% 92.58% 73.19% 94.13% 56.42% 55.34% 59.76% 65.52% 95.73% QC A 42.96% 46.40% 73.44% 53.56% 78.24% 37.35% 27.96% NA NA NA QC B 53.98% 59.20% 88.37% 64.78% 82.96% 49.78% 44.23% NA NA NA QC C 52.92% 61.61% 73.12% 62.79% 77.77% 52.83% 55.08% NA NA NA QC D 54.00% 65.41% 70.76% 65.79% 81.39% 54.88% 60.55% NA NA NA All QC 52.26% 60.65% 75.20% 63.15% 80.24% 51.04% 51.60% 52.16% 60.50% 92.26% YC 51.69% 59.78% 72.60% 69.21% 82.60% 51.16% 51.55% 51.39% 59.27% 80.46% IT 69.91% 78.12% 90.10% 78.87% 85.97% 70.14% 73.35% 68.67% 75.80% 91.78% By fixing the throughput, we can observe the effects of using different fleet mixes under the same level of throughput. By comparing Scenarios 4 to 2 and 5 to 3, we can see, as shown in Fig.6, that the situation will become even worse if we continue to employ the current fleet mix (Fleet Mix 1) instead of Fleet Mixes 2 and 3 (comprising more mega vessels) in the near and distant future. This is because, with the larger throughput in the future, the terminal will become busier and the vessels will have to wait longer. By comparing Scenarios 6 to 1 and 7 to 1, we can conclude that, if the current throughput were carried by more mega vessels (i.e., Fleet Mixes 2 and 3), the terminal would be less busier, as shown in Fig. 7. However, some types of mega vessels (e.g., OT7) would have to wait longer in this situation. If there were a single straight berth area (Scenarios 8-10) instead of several separate berth areas, the equipment utilization rates and total stay times of the majority of vessels would decrease dramatically, as shown in Fig. 8. In Scenarios 3 and 10, the arrival rates of containers exceed the handling capacity of the terminal. One possible reason is that the actual container throughput for the straight berth area is higher than that for several separate areas. For example, according to the simulation results, the actual container throughput in Scenario 10 is about 12% higher than the throughput in Scenario 3 during a given simulation time window. To summarize, the simulation results indicate that using more mega container vessels will be more efficient if the throughput keeps increasing. There will be a bottleneck at the MTL terminal because the current equipment and infrastructure may not be able to accommodate the larger number of mega vessels in the distant future (e.g., 10 years from now). This suggests that the MTL terminal must be expanded and improved. In addition, the simulation results show that a longer continuous berth area would be helpful for accommodating more large vessels.

14 200 Qiang Meng et al. / Transportation Research Procedia 25C (2017) Table 6. Total stay time (hr) and wait time (hr) of each vessel type, MTL case type Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 6 Scenario 7 Scenario 8 Scenario 9 Scenario 10 RV Total Stay Time Wait Time OT1 Total Stay Time Wait Time OT2 Total Stay Time Wait Time OT3 Total Stay Time Wait Time OT4 Total Stay Time Wait Time OT5 Total Stay Time Wait Time OT6 Total Stay Time Wait Time OT7 Total Stay Time Wait Time OT8 Total Stay Time NA NA NA NA NA NA NA Wait Time NA NA NA NA NA NA NA 84.60

15 Qiang Meng et al. / Transportation Research Procedia 25C (2017) % Utilization S4-S % Utilization S5-S % 13.00% 8.00% 3.00% 8.00% 3.00% -2.00% Berth QC YC Internal Truck -2.00% Berth QC YC Internal Truck -7.00% -7.00% 30% Total Stay Time S4-S2 350% Total Stay Time S5-S3 25% 300% 20% 15% 10% 250% 200% 150% 100% 5% 50% 0% 0% Fig. 6. Comparison of utilization rates and total stay times between Scenarios 4 and 2, and between Scenarios 5 and % Utilization S6-S1 6.00% Utilization S7-S1-1.00% Berth QC YC Internal Truck 4.00% 2.00% -2.00% -3.00% -4.00% -5.00% -2.00% -4.00% Berth QC YC Internal Truck -6.00% -6.00% -7.00% -8.00% -8.00% % 2.00% Total Stay Time S6-S % 4 Total Stay Time S7-S1 1.50% 35.00% 1.00% % -0.50% -1.00% 25.00% % % 5.00% -2.00% -2.50% -3.00% -5.00% -1 Fig. 7. Comparison of utilization rates and total stay times between Scenarios 6 and 1, and between Scenarios 7 and 1

16 202 Qiang Meng et al. / Transportation Research Procedia 25C (2017) % -0.40% Berth QC YC Internal Truck -2.00% -4.00% -0.60% -6.00% -0.80% -1.00% -1.20% -8.00% % -1.40% % -1.60% % -1.80% -2.00% Utilization S8-S % -2 Total Stay Time S8-S1-0.50% Berth QC YC Internal Truck -5.00% -1.00% -1.50% -2.00% -2.50% % % % -3.50% Utilization S9-S2-3 Total Stay Time S9-S % Utilization S10-S Total Stay Time S10-S % % Type8 Berth QC YC Internal Truck -6-8 Fig. 8. Comparison of utilization rates and total stay times between Scenarios 8 and 1, Scenarios 9 and 2, and between Scenarios 10 and 3 5. Conclusions In this study, we propose a queuing network model to represent the container operation process at a container terminal. A simulation model is developed and implemented by using ARENA software. With the flexible simulation model, we are able to investigate large-scale problems and obtain reliable results efficiently. An illustrative case is employed to evaluate the impacts of mega vessels on the Kwai Tsing container terminal in Hong Kong, using the proposed queuing network model and the developed simulation model. Ten scenarios with different container fleet mix and different throughputs were simulated in the illustrative case. Based on the results, the impacts of mega vessels on the container terminal operations were analyzed. The simulation results showed that the current container terminal will not be able to meet the larger container demand in the distant future because of high utilization rates, the big total stay times and wait times for the majority of vessels. It is also found that using more mega vessels would be more efficient for the container terminal because it would reduce the vessels total stay times and wait times to an extent. However, even doing this, the vessels total stay times and wait times are still unacceptable in the distant future. These results suggest that there will be a bottleneck at the container terminal if it stays as it is now, and the terminal may need to expand its capacity so as to

17 Qiang Meng et al. / Transportation Research Procedia 25C (2017) cater for mega vessels. The higher throughput that can be accommodated in a single berth area indicates that building a longer continuous berth would help the terminal to meet the container handling demand in the future. References Alicke, K., Modeling and optimization of the intermodal terminal mega hub. OR Spectrum, 24(1), Bielli, M., Boulmakoul, A., Rida, M., Object oriented model for container terminal distributed simulation. European Journal of Operational Research, 175(3), Bish, E. K., A multiple-crane-constrained scheduling problem in a container terminal. European Journal of Operational Research, 144(1), Canonaco, P., Legato, P., Mazza, R. M., Musmanno, R., A queuing network model for the management of berth crane operations. Computer and Operations Research, 35(8), Chen, L., Bostel, N., Dejax, P., Cai, J., Xi, L., A tabu search algorithm for the integrated scheduling problem of container handling systems in a maritime terminal. European Journal of Operational Research, 181(1), Cordeau, J. F., Gaudioso, M., Laporte, G., Moccia, L., The service allocation problem at the gioia tauro maritime terminal. European Journal of Operational Research, 176(2), Cullinance, K., Khanna, M., Economies of scale in large containerships: optimal size and geographical implications. Journal of Transport Geography, 8 (3), Daganzo, C. F., The crane scheduling problem. Transportation Research Part B, 23(3), Debjit Roy and Rene de Koster., Optimal design of container terminal layout. Proceedings of International Material Handling Research Colloquium. Debjit Roy and Rene de Koster., Modeling and design of container terminal operations. ERIM Report Series Reference No. ERS LIS. Available at SSRN: or Ebeling, C. E., Evolution of a Box. Invention and Technology, 23(4), 8 9. Edmond, E. D., Maggs, R. P., How useful are queue models in port investment decisions for container berths? Journal of the Operational Research Society, 29 (8), Gademann, A. J. R. M., Van De Velde, S. L., Positioning automated guided vehicles in a loop layout. European Journal of Operational Research, 127(3), Gambardella, L. M., Rizzoli, A. E., Zaffalon, M., Simulation and planning of an intermodal container terminal. Simulation, 71(2), Gamal Abd El-Nasser A. Said, Abeer M. Mahmoud, El-Sayed M. El-Horbaty., Simulation and optimization of container terminal operations: a case study, 4(4), Hong Kong Marine Department, Port benchmarking for assessing Hong Kong s maritime services and associated costs with other major international ports. Available online: last accessed 4th August Hong Kong Marine Department, 2009a. Arrivals and Departures. Available online: accessed 12th-23rd August Hong Kong Marine Department, 2009b. Port and Maritime Statistics. Available online: last accessed 4th August Hong Kong Marine Department, 2009c. Port of Hong Kong Handbook 2009 Available online: last accessed 4th August Huang, S.Y., Hsu, W-J., Chen, C. C., Nautiyal, S., Capacity analysis of container terminals using simulation techniques. International Journal of Computer Applications in Technology, 32(4), Imai, A., Nishimura, E., Hattori, M., Papadimitriou, S., Berth allocation at indented berths for mega-containerships. European Journal of Operational Research, 179(2), Imai, A., Nishimura, E., Papadimitriou, S., The dynamic berth allocation problem for a container port. Transportation Research Part B: Methodological, 35(4), Khatiashvili, S., Bakeev, C., Fidler, M., Application of simulation modeling to harbour operations. Maritime Engineering, 159(3), Kia M., Shayan, E., Ghotb, F., Investigation of port capacity under a new approach by computer simulation. Computers and Industrial Engineering, 42 (2-4), Kozan, E., Comparison of analytical and simulation planning models of seaport container terminals. Transportation Planning and Technology, 20(3), Kozan, E., Preston, P., Genetic algorithms to schedule container transfers at multimodal terminals. International Transactions in Operational Research, 6(3), Lagana, D., Legato, P., Pisacane, O., Vocaturo, F., Solving simulation optimization problems on grid computing systems. Parallel Computing, 32(9), Legato, P., Mazza, R. M., Berth planning and resources optimisation at a container terminal via discrete event simulation. European Journal of Operational Research, 133(3), Liu, C-I., Jula, H., Ioannou, P. A., Design, simulation, and evaluation of automated container terminals. IEEE Transactions on Intelligent Transportation Systems, 3(1), Lokuge, P., Alahakoon, D., Improving the adaptability in automated vessel scheduling in container ports using intelligent software agents. European Journal of Operational Research, 177(3),