Neuro fuzzy modeling of rice husk combustion in Fluidised bed

Transcription

1 Neuro fuzzy modeling of rice husk combustion in Fluidised bed Srinath. Suranani, Venkat Reddy. Goli Abstract This paper presents an adaptive-network based fuzzy system (ANFIS) for modeling the performance of fluidized bed combustor firing rice husk. Different ANFIS models have been constructed and tested in order to find the best ANFIS for modeling the performance of fluidized bed combustor (i.e combustion efficiency). Two parameters have been considered in the construction and plausible ANFIS models. The type of membership function and the number of linguistic variables are two mentioned parameters. Based on the experimental data, the proposed model consists of three input variables such as feed rate, fuel particle size and air to fuel ratio were used to predict the combustion efficiency of the rice husk fired combustor. Six different models based on these inputs are defined. All of the trained ANFIS are then compared with respect to the Absolute percentage error. To meet the best performance of the intelligent based approaches, data are pre processed (scaled) and finally out puts are post processed (returned to its original scale). The ANFIS model is capable of dealing with both complexity and uncertainty in the data set. Keywords Adaptive neuro fuzzy inference system (ANFIS), Neural networks, Fuzzy-logic, Combustion efficiency. E I. INTRODUCTION NERGY is vital for the social and economic development of any nation. The world energy demand is increasing very rapidly with development of civilization and growing industrialization. The energy requirement of India is expected to grow at % per year in coming years which means around four fold increases in energy requirement over the next 25 years. Coal is most important and abundant fossil fuel in India and accounts for 55% of India s energy need, whereas 30% of the requirement is met by petroleum products. A large population of India in the rural areas depends on the traditional sources of energy such as firewood, animal dung and agricultural residues consisting mainly of rice husk, saw dust, groundnut shell, coconut coir, cotton flower shell, etc. Srinath Suanani is with the National Institute of Technology Warangal,, INDIA (phone: ; srinathnit@yahoo.co. in). Venkat Reddy. Goli is with National Institute of Technology Warangal, INDIA. ( v.reddy@nitw.ac.in).. Biomass is an important source of energy in tropical countries, particularly in South and Southeast Asia, accounting for some 40% of the total regional energy consumption. Worldwide, biomass is the fourth largest energy resource after coal, oil, and natural gas [11]. Energy derived from the biomass is called bioenergy. These biomass feed stocks can be used for power generation applications. In view of this, a variety of processes exists for biomass conversions. The most used of these are thermal conversions, bio-chemical and chemical conversions and direct combustion. The thermal conversion processes consist of fast and slow pyrolysis, biomass gasification and fluidized bed combustion. Fuel flexibility, excellent solid gas mixing, temperature homogeneity and effective emission control makes the fluidized-bed combustion technology [5] the most efficient and environmentally friendly technology for conversion of energy from various biomass fuels, particularly, from agricultural residues sustainably produced on a large scale. The performance of fluidized bed combustor i.e. the efficiency of carbon conversion may greatly depend on the movement of solids and gas in bed and freeboard. In addition to mixing there are several other factors influencing conversion, such as the residence time in the bed. The conversion in gas solids fluidized bed combustors has been observed to vary from plug flow, to well below mixed flow. In fluidized bed combustion process internally lot of changes in terms of composition, temperature, reactions and flow rate occur. The process of developing model equations is very complex, and difficult to visualize. Multiple differential equations with strong cross influences are the norm. Hence an attempt has been made to model the FBC process using blackbox approach. An often used approach for black-box models are Artificial Neural Networks [8][2]. Neuro-fuzzy modeling refers to the way of applying various learning techniques developed in the neural network literature to fuzzy modeling or a fuzzy inference system (FIS)[1]. Neuro-fuzzy systems, which combine neural networks and fuzzy logic, have recently garnered a lot of interest in research and application. The neuro-fuzzy approach [7] has added the advantage of reduced training time, not only due to its smaller dimensions but also because the network can be initialized with parameters relating to the problem domain. Such results emphasize the benefits of the fusion of fuzzy [4]and neural network technologies as it facilitates an accurate initialization of the network in terms of the parameters of the fuzzy reasoning system. Various types of 193

2 FIS are reported in literature and each is characterized by their consequent parameters only [3]. A specific approach in neuro-fuzzy development is the adaptive neuro-fuzzy inference system (ANFIS)[1][10], which has shown significant results in modeling nonlinear functions. ANFIS uses a feed forward network to search for fuzzy decision rules that perform well on a given task. Using a given input-output data set, ANFIS creates a FIS whose membership function parameters are adjusted using a back propagation algorithm alone or a combination of a back propagation algorithm with a least squares method. This allows the fuzzy systems to learn from the data being modelled. II. EXPERIMENTAL PROCEDURE A. Experimental Set-up Fig 1 shows schematics of an experimental set-up with the rectangular FBC used in this study. The FBC was made of 6-mm thick stain less steel covered internally with 25-mm thick refractory. The combustor consists of three parts, (1) a rectangular furnace of 450 X 440 X 480 mm at the bottom base, (2) a cylindrical section with 1500mm height and 200mm inner diameter at the middle and (3) a cylindrical section of 500 mm height and 300mm inner diameter at the top. The whole vessel is insulated with ceramic wool of 120mm thickness. Induced draft was used to maintain a sufficient vacuum in the furnace. A heating coil of stainless steel pipe of one inch diameter is provided inside main vessel in the form of helical shape through which water is circulated. Fuel was fed pneumatically into the bed surface from a sealed hopper through an inclined feeding pipe. A cyclone was fitted to the combustor exit, and the carryover from the bed was collected for analysis. Flue gases were measured using Quintox KM9106 continuous emission analyser. Fig 1. A schematic diagram of the Fluidized bed Combustor III. METHODOLOGY In fluidized bed combustion process internally lot of changes in terms of composition, temperature, reactions and flow rate occur. Because of changeable nature of complex combustion process, the use of conventional methods may not give us accurate results. Hence this paper presents an adaptive network based fuzzy inference system (ANFIS)[1] to model the process. The main structure of the intelligent approach is explained in the following. The ANFIS algorithm has the following basic steps[9]. Step 1: Selection of important input variables of the model. Based on the process, the most important variables which have considerable impact on the performance of fluidsed bed combustion are considered. Step 2: Collection of data sets from the experiments for each of the input variables and output variable. In addition, all of inputs and output data are scaled and normalised using a normalisation method. Step 3: Divide the data into three sets, one for estimating the model, called the train data set, the second one for testing the model, called test data set and third one for checking the validity of the estimated model. Step 4 : This step is concerned with executing and estimating all of the plausible ANFIS models regarding two main parameters i.e the type of membership function and number of linguistic variables. Step 5: The models prediction capability is evaluated in this step through estimation of relative errors. Considering the value of percentage relative error, the best ANFIS is selected in each case in order to predict the performance of fluidised bed combustor i.e combustion efficiency. The significance of the proposed ANFIS for the prediction of combustion efficiency is four fold. First, the pre processing and post processing approaches of ANFIS eliminate possible noise. Second, it identifies the best ANFIS model based on minimum relative percentage error. Fourth, it provides a more reliable accurate solution than conventional approaches i.e regression because it uses adaptive neural modelling and fuzzy logic. ANFIS efficiently handles uncertainty, noise, and non-linearity in the given data set and provides the optimum solution. a) Adaptive neuro-fuzzy Inference System (ANFIS) Neural networks (NNs) and adaptive neurofuzzy inference system (ANFIS) models are universal approximators with the capability of identification and approximation of nonlinear relationships between independent (inputs) and the dependent (targets) variables. [1]- [2]. Optimising the values of the adaptive parameters is of vital importance for the performance of the adaptive system. Hybrid learning algorithm [3] used for ANFIS is faster than the classical back propagation method to approximate 194

3 the precise value of the model parameters. The hybrid learning algorithm [10] of ANFIS consists of two alternating phases: (1) gradient descend, which computes error signals recursively from the output layer backward to the nodes, and (2) the least square method, which finds feasible set of consequent parameters. IV. RESULTS AND DISCUSSION. The proposed algorithm [9]is applied to 80 set of data which are obtained from the rice husk combustion experiments carried out in a rectangular fluidized bed combustor. Step 1 Feed rate, fuel particle size and air to fuel ratio are considered as the input variables of the ANFIS in the present study. Step 2 Experiments were conducted at four different feed rates, based on the emission data combustion efficiency was calculated at different excess air factors at each feed rate for three fuel particle sizes. Step 3 The 80 sets of data are divided into 65 training data sets and 15 test data sets. Step 4 Two parameters have been considered in the construction of plausible ANFIS models. The type of membership function and the number of linguistic variables are two mentioned parameters. Five different member ship functions are considered in building the ANFIS, as follows: the Gaussian combination membership function(gaus), the generalized bell shaped built in membership function(gbell), The trapezoidal shaped built-in member ship function(trap),the triangular membership function(tri) and (pimf) Step 5 The architectures which are shown below have minimum Absolute Percentage Error among all of the architectures. Table presents structure and training information of the ANFIS model. Fig 2 Comprison of experimenatl and neuro fuzzy output Table 1 : Training information of the ANFIS model ANFIS PARAMETER VALUE Type of input membership Triangular shaped function(mf) Number of MFs 3 for each input Type of output MF Linear Optimization method hybrid Number of training data pairs 65 Number of checking data pairs 12 Number of linear parameters 108 Number of non linear parameters 27 Number of parameters 135 Number of fuzzy rules 27 Maximum epochs 1000 Error tolerance TABLE 2 : AVERAGE TESTING ERROR RELATED TO DIFFERENT MEMBERSHIP S.No. FUNCTIONS USING HYBRID OPTIMIZATION METHOD FOR RICE HUSK Membership function Output membership type Avg. testing error (training) 1 Trimf Constant Trimf Linear Trapmf Constant Trapmf Linear Gbellmf Constant Gbellmf Linear Gaussmf Constant Gausmf Linear pimf Linear Table 2 shows the various membership functions applied for modeling the FBC process. Out of five types of membership functions, the average absolute error is found to be lowest for triangular type member ship function. Fig 2 presents the comparison of experimental and neuro fuzzy output for the fuel rice husk at the feed rate of 30 kg/hr for the smallest fuel particle size of 0.345mm. It can be observed that the results obtained by neuro fuzzy model with triangular membership function are found to be in good agreement with the experimental values. V. CONCLUSIONS Modelling problem that was studied in the paper originates from the fluidised bed combustion of rice husk which is highly nonlinear and complex process is thus making conventional modelling difficult. Estimative models for the performance of fluidised bed combustion were identified using computational intelligence. Applied ANFIS networks were capable of 195

4 capturing the nonlinearities in process data, the training was efficient and prediction accuracy of the obtained models was good. A neuro fuzzy (ANFIS) trained with triangular membership functions for each input and linear membership function for output with hybrid optimization method gave satisfactory results for the fuel rice husk. 27 decision rules were framed in human understandable terms and combining with the numerical data gave an accurate ANFIS output related to experimental values. Based on the experimental results reported in this paper it could be concluded that application of computational intelligence for the modelling of the performance of fluidised bed combustion process has proven its potential and opened interesting directions for future research. Above all, hybrid computational intelligence methodologies could be further explored to provide more efficient and precise estimation by integration of available expert knowledge with other sources of information. ACKNOWLEDGMENT The authors would like to acknowledge the financial support from the MHRD (Ministry of Human Resource and Development) under R&D scheme.. REFERENCES [1] J.S. jang, ANFIS: Adaptive-network-based fuzzy inference system, IEEE Trans. Syst., Man, Cybern, vol. 23, no. 3, pp , May/jun,1993. [2] K.Hornik, M, Stinchcombe, and H, White, Multilayer feed forward networks are universal approximators, Neural nets., vol 2, no.5, pp , 1989 [3] Jang, R., Sun, C., Mizutani, E. Neuro fuzzy and soft computation, New jersy: Prentice hall [4] Babuska, R. Fuzzy modeling and identification. Delft, The Netherlands. (1997). [5] Oka S., Fluidized bed combustion processes and application, Yugoslav thermal association, Beograd,1994 [6] Ćojbašić Ž., Ćojbašić Lj., Nikolić V., Fuzzy and neuro-fuzzy systems in problems of process control and modeling: possibilities and some aspects of application, "Journal of Process Technology", Beograd, pp , [7] Chiu S., Fuzzy Model Identification Based on Cluster Estimation, Journal of Intelligent & Fuzzy Systems, Vol. 2, No. 3, 1994 [8] David M. Himmelblau, Applications of Artificial Neural Networks in Chemical Engineering. Korean J. Chem. Engg., 17(4), (2000). [9] Ali Azadeh et al An Integrated Intelligent Neuro fuzzy algorithm for long term electricity consumption: cases of Selected EU countries, Acta Polytechnica Hungarica,Vol. 7, No. 4, [10] Neural Networks, fuzzy Logic, and Genetic Algorithms systhesis and applications by S. Rajasekaran, G. A. Vijayalaxmi Pai [11] Werther J, Saenger M, Hartge EU, Ogada T, Siagi Z, Combustion of agricultural residues. Progress in Energy and Combustion Science, 26, 1 27, (2000).. 196

5 3rd International Conference on Medical, Biological and Pharmaceutical Sciences (ICMBPS'2013) January 4-5, 2013 Bali (Indonesia) Prevalence of Health-Related Risk Behaviors and Its Psychological Correlates YoungHo Kim, Ph.D. Abstract - The purpose of the study was to investigate the prevalence of health risk behaviors among a random sample of Korean adolescents and the relationship of psychological variables with health risk behaviors. 885 students ranged from 7 th to 9 th grade were randomly selected from 3 junior high schools in Dobong-gu district, Seoul. Four Korean-version measures were used to assess the health risk behavior and psychological variables of adolescents. Korean adolescents showed high prevalence of physical inactivity (n = 67%), smoking (n = 54%), drinking alcohol (n = 69%), eating problem (n = 49%), mental health problem (n = 57%), and viewing pornography (n = 47%). In addition to this, this study revealed that the three psychological variables were significantly correlated with health risk behaviors, and had significant effect to account for health risk behaviors. This study has the potential to influence the development of better health education and promotion programs for adolescents. Keywords-Health risk behaviors, Health locus of control, Selfesteem, Self-efficacy, Adolescent I. INTRODUCTION It has been well documented that many health risk behaviors are often initiated during the adolescent years and the initiation of risk behaviors is occurring at progressively younger age. A large volume of study indicated that the rates of smoking, drinking alcohol and drug use during adolescence have remarkably increased since 1980 s, and many adolescents experienced health risk behaviors at markedly earlier ages [1]. Traditionally, in many areas of public health a number of studies aimed at understanding why the majority of adolescents initiate health risk behaviors have focused on providing information, education and counseling programs without fully considering the psychological factors associated with adolescents risk behaviors [2]. In this regard many studies across a wide range of populations and settings have demonstrated the existence of a relationship between health risk behaviors and psychological variables [3,4]. YoungHo Kim, Ph.D. is with Department of Sport Science, Seoul National University of Science and Technology 172 Gongneung-dong, Nowon-gu, Seoul, , Korea. ) yk01@seoultech.ac.kr However, most of previous studies have been conducted in Western countries. The same level of research has not been focused on the Korean adolescent populations. The study attempted to identify the prevalence of health risk behaviors among a random sample of Korean adolescents. Specifically, the study investigated the relationship between psychological variables and health risk behaviors. II. METHODS A. Participants A total of 885 students ranged from 7 th to 9 th grade were randomly selected from 3 junior high schools in Dobong-gu district, Seoul. All participants in the age cohort were years old (M = 15.1 years). B. Measures Korean Health Survey Kit was applied to evaluate adolescents health risk behaviors. Coefficient alpha was.92, indicating high internal consistency and test-retest r was.83, indicating stability. The Multidimensional Health Locus of Control Scale (MHLC) was translated into Korean, and used in the study [5]. Test-retest Cronbach s α reliability coefficients of each sub-scale were as below:.87 for internal health locus of control (IHLC);.84 for powerful other health locus of control (PHLC);.81 for chance health locus of control (CHLC). The Self-efficacy Scale was also revised for the Korean version, and adopted in the study [6]. A Cronbach alpha coefficient of.90 was found for the questionnaire. The Korean version of the Self-esteem Scale was applied in the study [7]. A two week test-retest was performed and obtained a reliability of.87. III. RESULTS Please see the results in Table I and II on the next page. IV. DISCUSSION In practical terms the findings reinforced the argument for consideration of the psychological aspects in the development of the health risk reduction strategy. In addition, if further studies were to be undertaken to look at relationships between other psychological variables and other specific dimensions of adolescent health, then such findings of the existence of significant relationships could increase the understanding. 197