291 71 18MB
English Pages 312 [337] Year 2020
Soft Computing with NeuroFuzzy Systems
Soft Computing with NeuroFuzzy Systems
Edited by: Jovan Pehcevski
ARCLER
P
r
e
s
s
www.arclerpress.com
Soft Computing with NeuroFuzzy Systems Jovan Pehcevski
Arcler Press 224 Shoreacres Road Burlington, ON L7L 2H2 Canada www.arclerpress.com Email: [email protected]
e-book Edition 2021 ISBN: 978-1-77407-982-9 (e-book) This book contains information obtained from highly regarded resources. Reprinted material sources are indicated. Copyright for individual articles remains with the authors as indicated and published under Creative Commons License. A Wide variety of references are listed. Reasonable efforts have been made to publish reliable data and views articulated in the chapters are those of the individual contributors, and not necessarily those of the editors or publishers. Editors or publishers are not responsible for the accuracy of the information in the published chapters or consequences of their use. The publisher assumes no responsibility for any damage or grievance to the persons or property arising out of the use of any materials, instructions, methods or thoughts in the book. The editors and the publisher have attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders if permission has not been obtained. If any copyright holder has not been acknowledged, please write to us so we may rectify. Notice: Registered trademark of products or corporate names are used only for explanation and identification without intent of infringement. © 2021 Arcler Press ISBN: 978-1-77407-779-5 (Hardcover) Arcler Press publishes wide variety of books and eBooks. For more information about Arcler Press and its products, visit our website at www.arclerpress.com
DECLARATION Some content or chapters in this book are open access copyright free published research work, which is published under Creative Commons License and are indicated with the citation. We are thankful to the publishers and authors of the content and chapters as without them this book wouldn’t have been possible.
ABOUT THE EDITOR
Jovan obtained his PhD in Computer Science from RMIT University in Melbourne, Australia in 2007. His research interests include big data, business intelligence and predictive analytics, data and information science, information retrieval, XML, web services and service-oriented architectures, and relational and NoSQL database systems. He has published over 30 journal and conference papers and he also serves as a journal and conference reviewer. He is currently working as a Dean and Associate Professor at European University in Skopje, Macedonia.
TABLE OF CONTENTS
List of Contributors........................................................................................xv
List of Abbreviations..................................................................................... xix
Preface..................................................................................................... ....xxi Section 1 Intelligent Neuro-fuzzy Methods Chapter 1
Fuzzy-Neuro Model for Intelligent Credit Risk Management..................... 3 Abstract...................................................................................................... 3 Introduction................................................................................................ 4 The Fuzzy Logic And Neural Networks Algorithms..................................... 5 Hybrid Fuzzy Logic-Neural Network (HFNN) Model For Intelligent Credit Risk Management................................................................... 6 Experiment Results.................................................................................... 12 Discussion And Analysis Of Results.......................................................... 16 Conclusions And Recommendations......................................................... 17 References................................................................................................ 18
Chapter 2
Method to Improve Airborne Pollution Forecasting by Using Ant Colony Optimization and Neuro-Fuzzy Algorithms........................... 21 Abstract.................................................................................................... 21 Introduction.............................................................................................. 22 Methodology............................................................................................ 30 Results And Discussion............................................................................. 31 Conclusions.............................................................................................. 33 Acknowledgements.................................................................................. 33 References................................................................................................ 34
Chapter 3
A TSK-Type Recurrent Neuro-Fuzzy Systems for Fault Prognosis............. 37 Abstract.................................................................................................... 37 Introduction.............................................................................................. 38
Temporal Neuro-Fuzzy Systems................................................................ 38 Prognostics Process................................................................................... 44 Experimental Results................................................................................. 44 Conclusion............................................................................................... 46 References................................................................................................ 47 Chapter 4
A Neuro-Fuzzy Model for QoS Based Selection of Web Service.............. 49 Abstract.................................................................................................... 49 Introduction.............................................................................................. 50 Qos Properties Of Web Service................................................................. 51 Related Work............................................................................................ 52 Refinement Of The Framework.................................................................. 54 Comments And Recommendations........................................................... 57 Conclusions.............................................................................................. 57 References................................................................................................ 58 Section 2 Adaptive Neuro-Fuzzy Systems
Chapter 5
Adaptive Neuro-Fuzzy Logic System for Heavy Metal Sorption in Aquatic Environments.......................................................................... 63 Abstract.................................................................................................... 63 Introduction.............................................................................................. 64 Methodology............................................................................................ 66 Adaptive Neuro-Fuzzy Inference System (Anfis)........................................ 67 Simulation And Forecasting...................................................................... 70 Conclusion............................................................................................... 72 Acknowledgements.................................................................................. 72 References................................................................................................ 73
Chapter 6
Automatic Heart Disease Diagnosis System Based on Artificial Neural Network (ANN) and Adaptive Neuro-Fuzzy Inference Systems (ANFIS) Approaches................................................................................. 77 Abstract.................................................................................................... 77 Introduction.............................................................................................. 78 Related Works.......................................................................................... 79 Research Algorithms And Concepts........................................................... 80 Automatic Heart Disease Diagnosis System’s Design And Implementation.83
x
Experimental Results And Analysis............................................................ 85 Conclusion And Future Work.................................................................... 88 References................................................................................................ 89 Chapter 7
Reliability Estimation of Services Oriented Systems Using Adaptive Neuro Fuzzy Inference System................................................................. 91 Abstract.................................................................................................... 91 Introduction.............................................................................................. 92 Related Work............................................................................................ 95 Discussion And Research Approach.......................................................... 96 Result And Discussion.............................................................................. 99 Conclusion And Future Work.................................................................. 102 References.............................................................................................. 103
Chapter 8
Prediction of Soil Fractions (Sand, Silt and Clay) in Surface Layer Based on Natural Radionuclides Concentration in the Soil Using Adaptive Neuro Fuzzy Inference System................................................ 105 Abstract.................................................................................................. 106 Introduction............................................................................................ 106 Materials And Methods........................................................................... 108 Results And Discussion........................................................................... 113 Conclusion............................................................................................. 116 Acknowledgements................................................................................ 117 References.............................................................................................. 118 Section 3 Neuro-Fuzzy Inference Systems
Chapter 9
Adaptive Neuro-Fuzzy Inference System for Prediction of Effective Thermal Conductivity of Polymer-Matrix Composites............................ 125 Abstract.................................................................................................. 125 Introduction............................................................................................ 126 Adaptive Neuro-Fuzzy Inference System (Anfis)...................................... 129 Results And Discussion........................................................................... 132 Conclusion............................................................................................. 138 References.............................................................................................. 139
xi
Chapter 10 Application of Adaptive Neuro Fuzzy Inference System in Supply Chain Management Evaluation................................................... 143 Introduction............................................................................................ 143 Construction Supply Chain..................................................................... 146 Methodology.......................................................................................... 148 Neurofuzzy Model.................................................................................. 149 Conclusion............................................................................................. 154 References.............................................................................................. 155 Chapter 11 Application of the Adaptive Neuro-Fuzzy Inference System for Optimal Design of Reinforced Concrete Beams..................................... 159 Abstract.................................................................................................. 159 Introduction............................................................................................ 160 Genetic Algorithms................................................................................. 162 The Adaptive Neuro-Fuzzy Inference System.......................................... 163 Design Of Two-Span Continuous Reinforced Concrete Beams................ 165 Numerical Results................................................................................... 170 Conclusion............................................................................................. 175 References.............................................................................................. 176 Chapter 12 Comparison between Neural Network and Adaptive Neuro-Fuzzy Inference System for Forecasting Chaotic Traffic Volumes.................... 181 Abstract.................................................................................................. 181 Introduction............................................................................................ 182 Diagnosis Of Chaos................................................................................ 183 Forecasting Models................................................................................. 185 Numerical Results................................................................................... 187 Conclusion............................................................................................. 192 References.............................................................................................. 194 Chapter 13 The Development of an Alternative Method for the Sovereign Credit Rating System Based on Adaptive Neuro-Fuzzy Inference System......... 197 Abstract.................................................................................................. 197 Introduction............................................................................................ 198 Sovereign Credit Rating And Related Works............................................ 199 Description Of Selected Models............................................................. 200 Empirical Study....................................................................................... 203 xii
Empirical Results.................................................................................... 205 Conclusions............................................................................................ 208 References.............................................................................................. 211 Section 4 Neuro-Fuzzy Controllers Chapter 14 Implementation of Adaptive Neuro Fuzzy Inference System in Speed Control of Induction Motor Drives........................................................ 217 Abstract.................................................................................................. 217 Introduction............................................................................................ 218 State Space Approach............................................................................. 219 System Description................................................................................. 220 Anfis Based Speed Controller.................................................................. 224 Adaptive Neuro-Fuzzy Principle............................................................. 224 Simulation Results.................................................................................. 226 Conclusions............................................................................................ 227 References.............................................................................................. 229 Chapter 15 Neuro-Fuzzy Based Interline Power Flow Controller for Real Time Power Flow Control in Multiline Power System............................ 231 Abstract.................................................................................................. 231 Introduction............................................................................................ 232 Working Principle Of IPFC..................................................................... 234 Problem Formation................................................................................. 235 State Space Modeling Of IPFC................................................................ 237 Fuzzy Control Scheme For Master Converter Of IPFC............................. 240 Ann Control Scheme For Slave Converter Of IPFC.................................. 241 Simulation Results And Discussion......................................................... 242 Conclusion............................................................................................. 246 References.............................................................................................. 248 Chapter 16 Controlling Speed of DC Motor with Fuzzy Controller in Comparison with ANFIS Controller........................................................ 251 Abstract.................................................................................................. 251 Introduction............................................................................................ 252 Background............................................................................................ 254 Methodology.......................................................................................... 255
xiii
Conclusions............................................................................................ 262 References.............................................................................................. 263 Chapter 17 A Neuro-Fuzzy Controller for Collaborative Applications in Robotics Using LabVIEW........................................................................ 265 Abstract.................................................................................................. 265 Introduction............................................................................................ 266 Prototype Description............................................................................. 267 Neuro-Fuzzy Controller.......................................................................... 267 Collaborative Task................................................................................... 272 Results.................................................................................................... 272 Conclusions............................................................................................ 280 References.............................................................................................. 281 Chapter 18 An Adaptive Fuzzy Sliding Mode Control Scheme for Robotic Systems.283 Abstract.................................................................................................. 283 Introduction............................................................................................ 284 Sliding Mode Control (SMC)................................................................... 286 Decoupled Robot Tracking Control Design............................................. 289 Simulation Results.................................................................................. 295 Conclusions............................................................................................ 300 References.............................................................................................. 301 Index...................................................................................................... 305
xiv
LIST OF CONTRIBUTORS Elmer P. Dadios Department of Manufacturing Engineering and Management, De La Salle University, Manila, Philippines James Solis College of Computer Studies, De La Salle University, Manila, Philippines Elizabeth Martinez-Zeron Facultad de Informática, Universidad Autónoma de Querétaro, Querétaro, México Marco A. Aceves-Fernandez Facultad de Informática, Universidad Autónoma de Querétaro, Querétaro, México Efren Gorrostieta-Hurtado Facultad de Informática, Universidad Autónoma de Querétaro, Querétaro, México Artemio Sotomayor-Olmedo Laboratory of Artificial Intelligence, Faculty of Engineering, Universidad Autónoma de Querétaro, Querétaro, México Juan Manuel Ramos-Arreguín Facultad de Informática, Universidad Autónoma de Querétaro, Querétaro, México Rafik Mahdaoui Laboratoire d’Automatique et Productique (LAP), Université de Batna, Batna, Algérie; Leila Hayet Mouss Centre Universitaire Khenchela, Khenchela, Algérie Abdallah Missaoui LSTS-ENIT, Tunis, Tunisia; Kamel Barkaoui CEDRIC-CNAM, Paris, France. Ahmad Qasaimeh Department of Civil Engineering, Jerash University, Jerash, Jordan xv
Mohammad Abdallah Chemical Engineering Department, AlHuson University College, Al-Balqa Applied University, Salt, Jordan Falah Bani Hani Chemical Engineering Department, AlHuson University College, Al-Balqa Applied University, Salt, Jordan Mohammad A. M. Abushariah Computer Information Systems Department, King Abdullah II School for Information Technology, The University of Jordan, Amman, Jordan Assal A. M. Alqudah Computer Information Systems Department, King Abdullah II School for Information Technology, The University of Jordan, Amman, Jordan Omar Y. Adwan Computer Information Systems Department, King Abdullah II School for Information Technology, The University of Jordan, Amman, Jordan Rana M. M. Yousef Computer Information Systems Department, King Abdullah II School for Information Technology, The University of Jordan, Amman, Jordan Ashish Seth Department of Computer Science, Punjabi University, Patiala, India Himanshu Agarwal University College of Engineering, Punjabi University, Patiala, India Ashim Raj Singla Department of Information Technology, Indian Institute of Foreign Trade, New Delhi, India Saad Al-Hamed Department of Agricultural Engineering, College of Food and Agriculture Sciences, King Saud University, Riyadh, KSA Mohamed Wahby Department of Agricultural Engineering, College of Food and Agriculture Sciences, King Saud University, Riyadh, KSA Mohamed Al-Sulaiman Community College, Huraimla, Shaqra University, Huraimla, KSA xvi
Abdulwahed Aboukarima Community College, Huraimla, Shaqra University, Huraimla, KSA Agricultural Engineering Research Institute, Agricultural Research Centre, Cairo, Egypt Rajpal Singh Bhoopal Thermal Physics Laboratory, Department of Physics, University of Rajasthan, Jaipur, India Ramvir Singh Thermal Physics Laboratory, Department of Physics, University of Rajasthan, Jaipur, India Pradeep Kumar Sharma Thermal Physics Laboratory, Department of Physics, University of Rajasthan, Jaipur, India Thoedtida Thipparat Faculty of Management Sciences, Prince of Songkla Universit, Kohong Hatyai, Songkhla, Thailand Jiin-Po Yeh Department of Civil and Ecological Engineering, I-Shou University, Kaohsiung, Taiwan Ren-Pei Yang Department of Civil and Ecological Engineering, I-Shou University, Kaohsiung, Taiwan Jiin-Po Yeh Department of Civil and Ecological Engineering, I-Shou University, Kaohsiung City, Taiwan; Yu-Chen Chang Institute of Civil Engineering Technology, National Kaohsiung University of Applied Sciences, Kaohsiung City, Taiwan. Hakan Pabuçcu Department of Business Administration, Bayburt University, Bayburt, Turkey Tuba Yakıcı Ayan Department of Econometrics, Karadeniz Technical University, Trabzon, Turkey K. Naga Sujatha Department of Electrical Engineering, AU College of Engineering, Andhra University, Visakhapatnam, India.
xvii
K. Vaisakh Department of Electrical Engineering, AU College of Engineering, Andhra University, Visakhapatnam, India. A. Saraswathi Department of Electrical and Electronics Engineering, University College of Engineering Villupuram, Anna University, Chennai, India S. Sutha Department of Electrical and Electronics Engineering, University College of Engineering Dindigul, Anna University, Chennai, India Aisha Jilani Electrical Engineering Department, Lahore College for Women University, Lahore, Pakistan Sadia Murawwat Electrical Engineering Department, Lahore College for Women University, Lahore, Pakistan Syed Omar Jilani Electrical Engineering Department, University of Lahore, Lahore, Pakistan Hiram E. Ponce Escuela de Graduados en Ingeniería y Arquitectura, División de Ingeniería y Arquitectura , Instituto Tecnológico y de Estudios Superiores de Monterrey, Campus Ciudad de México, Mexico City 14380, Mexico Dejanira Araiza Escuela de Graduados en Ingeniería y Arquitectura, División de Ingeniería y Arquitectura , Instituto Tecnológico y de Estudios Superiores de Monterrey, Campus Ciudad de México, Mexico City 14380, Mexico Pedro Ponce Escuela de Graduados en Ingeniería y Arquitectura, División de Ingeniería y Arquitectura , Instituto Tecnológico y de Estudios Superiores de Monterrey, Campus Ciudad de México, Mexico City 14380, Mexico Abdel Badie Sharkawy Shaaban Ali Salman Mechanical Engineering Department, Faculty of Engineering, Assiut University, Assiut, Egypt
xviii
LIST OF ABBREVIATIONS FLS
Fuzzy logic system
GMI
Gross monthly income
TR
Training samples
ACO
Ant colony optimization
ANFIS
Neuro-fuzzy inference system
FCM
Fuzzy C-Means
FIS
Fuzzy inference system
FMECA
Failure mode and effects analysis
FNN
Fuzzy neural network
MS
Manufacturing systems
UDDI
Universal description, discovery, and integration
AIS
Artificial immune system
MDSS
Medical decision support systems
MLPNN
Multilayer perceptron neural network
SOSs
Service oriented systems
SOA
Service-oriented architecture
SMEs
Small and medium enterprises
MAD
Mean absolute deviation
RMSE
Root mean square error
ETC
Effective thermal conductivity
FAM
Fuzzy associative memories
HDPE
High-density polyethylene
LDPE
Low-density polyethylene
PVDF
Polyvinylidene fluoride
ASC
Agile supply chain
FIS
Fuzzy inference system
GDP
Gross domestic product
GNP
Gross national product
MAE
Mean absolute error
OLS
Ordinary least squares
VIF
Variance inflation factor
IM
Induction motor
FLC
Fuzzy logic controller
GSA
Gravitational search algorithm
PIM
Power injection model
RPC
Reactive power compensation
SCUC
Security constrained unit commitment
VSC
Voltage source converter
PID
Proportional integral derivative controller
FCM
Fuzzy cluster means
AFSMC
Adaptive fuzzy sliding mode control
SMC
Sliding mode control
xx
PREFACE
The computing with Fuzzy Inference System (FIS) is the most important modeling tool based on the theory of fuzzy sets. The basic structure of FIS is a model that maps the following parameters: input characteristics to input membership function (input MF – Membership Function), the input function of membership to rules, the rules to the set of output characteristics, the output characteristics to the output membership functions (output MF), and the output membership function to a single value of the output, or decision according to the output. The structure of the neural network maps the inputs through the input membership functions and their parameters, and then through the output membership functions and their parameters. These systems are used in the field of automatic control, decision-making processes, etc. The most common fuzzy inference methodology is that of the Mamdani type; however, besides it there is also a methodology known as Sugeno type inference. Advantages of the Sugeno method is its computational efficiency, as it works well with linear techniques, works well with optimization and adaptive techniques, and it’s also suitable for mathematical analysis. Since it is more compact and computationally efficient than a Mamdani system, the Sugeno system is also used for adaptive fuzzy model construction techniques. The afore-mentioned adaptive techniques can be used to define affiliation functions in order to adjust the system to model the data in the best way. An adaptive neuro-fuzzy inference system (ANFIS) can learn to make humanlike decisions and uses fuzzy membership functions for the soft constraints (input variables). The benefit is that ANFIS generates the most appropriate decisions resulting from fuzzy rules, for matches between different alternatives. Further benefit is that it extracts a small number of fuzzy rules, which are still very effective without giving up robustness. This edition covers different topics from soft computing and neuro-fuzzy systems, including intelligent neuro-fuzzy models, adaptive neuro-fuzzy systems, neuro-fuzzy inference systems, and neuro-fuzzy control systems. Section 1 focuses on intelligent neuro-fuzzy models, describing fuzzy-neuro
model for intelligent credit risk management; a method to improve airborne pollution forecasting by using ant colony optimization; TSK-type recurrent neuro-fuzzy systems for fault prognosis; and neuro-fuzzy model for QoS based selection of web service. Section 2 focuses on adaptive neuro-fuzzy systems, describing adaptive neuro-fuzzy logic system for heavy metal sorption in aquatic environments; an automatic heart disease diagnosis system based on artificial neural network (ANN); reliability estimation of services oriented systems using adaptive neuro fuzzy inference system; and prediction of soil fractions (sand, silt and clay) in surface layer based on natural radionuclides concentration. Section 3 focuses on neuro-fuzzy inference systems, describing an adaptive neuro-fuzzy inference system for prediction of effective thermal conductivity of polymer-matrix composites; application of adaptive neuro-fuzzy inference system in supply chain management evaluation; application of the adaptive neuro-fuzzy inference system for optimal design of reinforced concrete beams; comparison between neural network and adaptive neuro-fuzzy inference system for forecasting chaotic traffic volumes; and development of an alternative method for the sovereign credit rating system based on adaptive neuro-fuzzy inference system. Section 4 focuses on neuro-fuzzy control, describing implementation of an adaptive neuro fuzzy inference system in speed control of induction motor drives; a neuro-fuzzy based interline power flow controller for real time power flow control in multiline power system; controlling speed of DC motor with fuzzy controller in comparison with ANFIS controller; a neuro-fuzzy controller for collaborative applications in robotics using LabVIEW; and an adaptive fuzzy sliding mode control scheme for robotic systems.
xxii
SECTION 1
INTELLIGENT NEURO-FUZZY METHODS
CHAPTER
1
Fuzzy-Neuro Model for Intelligent Credit Risk Management
Elmer P. Dadios1, and James Solis2 Department of Manufacturing Engineering and Management, De La Salle University, Manila, Philippines 1
2
College of Computer Studies, De La Salle University, Manila, Philippines
ABSTRACT This paper presents hybrid fuzzy logic and neural network algorithm to solve credit risk management problem. Credit risk is the risk of loss due to a debtor’s non-payment of a loan or other line of credit. A method of evaluating the credit worthiness of a customer is complex and non-linear
Citation: E. Dadios and J. Solis, “Fuzzy-Neuro Model for Intelligent Credit Risk Management,” Intelligent Information Management, Vol. 4 No. 5A, 2012, pp. 251-260. doi: 10.4236/iim.2012.425036. Copyright: © 2012 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY). http:// creativecommons.org/licenses/by/4.0
4
Soft Computing with NeuroFuzzy Systems
due to the diverse combinations of risk involve. To address this problem a credit scoring method is proposed in this paper using hybrid fuzzy logicneural network (HFNN) model. The model will be implemented, tested, and validated for individual auto loans using real life bank data. The neural network is used as the learner and the fuzzy logic is used as the implementer. The neural network will fine tune the fuzzy sets, remove redundant input variables, and extract fuzzy rules. The extracted fuzzy rules are evaluated to retain the best k number of rules that will give final and intelligent decisions. The experiment results show that the performance of the proposed HFNN model is very accurate, robust, and reliable. Comparison of these results to other previous published works is also presented in this paper. Keywords:- Fuzzy Logic; Neural Networks; Fuzzy-Neuro Model; Credit Risk Management
INTRODUCTION Credit scoring is a method of evaluating the credit worthiness of a customer. A credit scoring model is built to assist credit analysts to decide whether a new loan should or should not be granted [1,2]. The model is used as a gauge for every applicant’s profile. If a profile is equal or better than the model, the account is predicted to be “good”. Otherwise, the account is predicted to be “bad”. There have been several automated approaches presented before to solve credit scoring problem. Among them are: 1) Rule-based, 2) Statisticalbased, 3) Genetic Algorithm (GA), 4) Neural Networks (NN). The rulebased (or knowledge based) approach is believed to be the easiest for a credit professional to formalize and the least expensive to implement [3]. It uses a set of rules derived from past credit experiences. It provides consistency to the account review process since it is an automation of the traditional risk assessment process [2]. But one major problem with rule-based scoring is the difficulty of determining the source of error if the system makes a series of bad decisions [3], hence, also difficult to improve. The statisticalbased credit scoring method uses linear discriminant analysis and logistic regression. Thus, it requires specialized education, training and experience. Also, the traditional regression techniques cannot be fully automated. It is labor-intensive and time consuming to design and updates the model [4]. Limited in its effecttiveness as a long-term decision tool, the credit scoring models have to be updated and improved as trends in customer behavior changes by which the performance of the system falls below the acceptable level of prediction rate.
Fuzzy-Neuro Model for Intelligent Credit Risk Management
5
One of the successful techniques used to solve the credit scoring problem is the neural networks (NN) [5,6]. It is believed that NN provide an essential technology for a faster and more effective tool for credit scoring [7]. NN are capable of modeling very complex mathematical and logical relationships that are unknown to the credit analyst and NN are able to model linear and non-linear relationships. In terms of accuracy, in most cases, the rulebased and statistical-based credit scoring systems correctly classifies at 74% [8], commercially available NNbased at 75% - 80% [8,9] and Genetic Algrithm between 72% to 74% [9].
THE FUZZY LOGIC AND NEURAL NETWORKS ALGORITHMS Fuzzy Logic is a system that imitates the way a human being thinks [1013]. Unlike conventional logical systems, fuzzy logic does not need actual theoretical data and input-output relations to solve a problem. Rather, it defines complex systems with linguistic descriptions. The fuzzy logic system (FLS) contains four components: the fuzzifier, rules, inference engine and defuzzifier. The fuzzifier maps crisp input numbers into fuzzy sets. Data obtained from the outside world is converted into data understandable by the system. Rules are linguistic variables expressed in IF-THEN statements. The inference engine maps fuzzy sets into fuzzy sets. It also handles situations wherein two or more rules are combined. The defuzzifier maps the fuzzy output sets into crisp outputs. The crisp output is an output that is easily understood by the outside environment. Neural Network (NN) is an information processing paradigm that is inspired by the way biological nervous systems, such as the brain, process information [14-16]. The key element of this paradigm is the novel structure of the information processing system. It is composed of a large number of highly interconnected processing elements (neurons) working in unison to solve specific problems. NN, like people, learn by example and solve problems for a specific application, such as pattern recognition or data classification [16]. It has been suggested that, since all the other methods are known to possess definite advantages and disadvantages [2,17], the best approach is to combine the methods, taking advantage of the strengths and, thus, creating a hybrid approach to the overall credit scoring process.
6
Soft Computing with NeuroFuzzy Systems
HYBRID FUZZY LOGIC-NEURAL NETWORK (HFNN) MODEL FOR INTELLIGENT CREDIT RISK MANAGEMENT The hybrid fuzzy logic neural network (HFNN) model combines the desirable properties of both neural network and fuzzy logic to form a system that supercedes the limitations of neural network and fuzzy logic algorithms. The HFNN model system learns inductively from the data using the neural network. Fuzzy rules are extracted from the trained neural network and with the extracted rules, previously unseen data can be discriminated whether “good” or “bad” accounts.The HFNN model developed in this paper undergoes two stages, namely: learning (or training) and implementation (or testing). Figure 1 shows the learning stage of the HFNN model. The different phases that took placed in the learning stage are: Fuzzification of the Data, Neural Network Learning, Fuzzy Sets Tuning, Pruning Input Variables, and Fuzzy Rule Extraction & Evaluation.
Fuzzification of Data The fuzzification of data is conducted and configured manually. All fuzzy sets are initialized before the training data are inputted into the neural net. The input variables are partitioned by overlapping fuzzy sets and the membership functions are initialized based on the recommendation of a loan evaluation expert [18]. Note that the specification of membership function is subjective and may vary from different experts. However, membership functions cannot be assigned arbitrarily. The number of membership functions is chosen such that the resulting fuzzy rules will be easily readable and accurate enough to classify data. Figure 2 shows sample typical fuzzy sets for an input variable Total Income.
Figure 1. Architecture of the learning stage of the HFNN model.
Fuzzy-Neuro Model for Intelligent Credit Risk Management
7
The fuzzification and distribution of data are automated processes. Through the fuzzy sets, the crisp input from 1000 samples is fuzzified one at a time and serves as a fuzzy input to the NN [19]. To fuzzify a crisp input, the degree of membership of the crisp input in each of the affected fuzzy members is computed. The fuzzy member that gets the bigger degree of member gets the value 1, the rest of members get the value 0. In case, the degree of member is the same for 2 adjacent fuzzy members, the leftmost member gets the value 1. The rightmost member gets the value 0. For example, as shown in Figure 2, when Income = P45,000 and based in the membership functions shown, the degree of membership of P45,000 is 7/8 in Low and 1/8 in Medium. Hence, μ(45,000) = max(7/8, 1/8) = 7/8 So, μ(45,000) is Low, which means that the Total Income is Low. The data then are distributed after fuzzifying the entire 1000 samples. The first 630 samples becomes the training set, the next 70 samples becomes the evaluating set, and the last 300 samples becomes the testing set.
Neural Network Learning for HFNN Model In this research, the neural network (NN) learns purely from the training data presented to the model. The NN is initialized with input neurons equal to the number of fuzzy input members. This means that for every fuzzy input member there should be a corresponding neuron in the NN. Hence, for the input variable Total Income shown in Figure 2, there must be also 4 input neurons in the NN shown in Figure 3. The output layer has two neurons one for good and the other one for bad [19,20]. The number of hidden neurons is 2/3 of the sum of the total number of input neurons plus the total number of the output neurons [21].
Figure 2. Sample fuzzy sets for an input variable total income.
Soft Computing with NeuroFuzzy Systems
8
The neural network is trained using backpropagation method to map the fuzzified inputs to the desired output [14]. The optimum accuracy is achieved when the classification error is minimized for the training data and at the same time giving the best accuracy performance for the validation data. When the neural network attained optimum classification accuracy, fuzzy rules are extracted from the NN. The extracted rules are used by the fuzzy system during the implementation for the classification of the previously unseen samples.
Fuzzy Input Set Tuning The tuning process of fuzzy input sets is conducted automatically [19]. Only inputs that are continuous variables are tuned. Adjustment on the boundaries of the fuzzy members of a given fuzzy set follows specific restraints. In this paper, the following restrictions are adapted: 1)
The fuzzy sets are kept overlapped with the adjacent fuzzy sets, as shown in Figure 2. 2) When updating the parameters, the parameters a, b, and c should remain valid; that is l ≤ a ≤ b ≤ c ≤ u must always hold, where [l, u] is the domain of the corresponding variable. 3) When updating the position parameters of the current fuzzy member, the parameters of current member should not become smaller than the corresponding parameters of the left neighbor or larger than the corresponding parameters of the right neighbor. 4) The parameter a of the leftmost member and the parameter c of the rightmost member remain fixed. 5) The parameter b of the current fuzzy member being updated should not become greater than the parameter c of the right member or smaller than the parameter c of the left member. For every error back propagated from the output to input fuzzy sets, the input fuzzy sets that are continuous variables are adjusted so as to reduce the error. After satisfactory numbers of adjustment were made with the input fuzzy sets and yet the error does not go down, the tuning had reached its optimum. Once the fuzzy input sets are tuned, the NN is re-trained to an optimum accuracy for the newly tuned fuzzy input sets [18,19].
Fuzzy-Neuro Model for Intelligent Credit Risk Management
9
Figure 3. Architecture of fuzzy input to the NN during learning stage.
Pruning the Input Variables or Fuzzy Sets Pruning or removal of redundant input variables or fuzzy sets will improve the readability of a fuzzy rule base extracted during the learning stage. Removal of the redundant input fuzzy sets will simplify the model. Pruning techniques are adapted from NN wherein, tests are made for its parameters. i.e. either weights or neurons to determine how the error would change if the parameter is removed [20]. In this paper, fuzzy sets that represent various input variables in all possible combinations are removed and the network classification performance is evaluated. The process in determining redundant variables is conducted by systematically enabling or disabling the inputs. When an input is enabled, its contribution is accepted into the NN. But when an input is disabled, its value is blocked in the NN. It is as if the input does not exist. With all the possible combination of enabled and disabled inputs, the prediction accuracy of each combination is recorded. In this study, the 16 input variables shown in Table 1 is presented in binary and equivalent to 216 − 1 or 65,535 combinations. For a given sample, the HFNN model may take in the values of 2 input variables, e.g., Age and Total Income, and ignoring the values of the remaining 14 input variables. This is one unique combination of the inputs. The ignored values of input variables are changed to zero before being inputted to the hybrid network. This is to neutralize the effect of the input variables that are not included in that particular combination. The 65,535 unique combinations will be ranked according to their accuracy performance. Among the combinations with the same accuracy performance, say 100%, the combination with the least number of fuzzy inputs and maximum number of rules is the most desirable combination.
10
Soft Computing with NeuroFuzzy Systems
Fuzzy Rule Extraction and Evaluation From the pruned system, all possible fuzzy rules are extracted. Figure 4 shows how to extract fuzzy rules in the HFNN model. The combination pattern 0 3 0 2 is the exact representation of the fuzzy inputs combination both enabled and disabled. The zero value represents disabled input. With the output of G for Good, 0 3 0 2 G is one fuzzy rule. Every possible combination of the fuzzy inputs is considered a rule. This rule is being evaluated by NN whether the combination results is “good” or “bad”. Note that rules are derived from unique combinations of the enabled fuzzy inputs. There are no two rules that have exactly the same combination of the enabled fuzzy inputs. Hence, there are no conflicting rules that results from these combinations. The final rules for the rule base are selected from the extracted rules by computing the performance of each rule. Rules that are only responsible for a fixed numbers of classifications may be deleted. Only a number of k best rules are kept [19]. In here, k is a number determined by the number of rules derived from specific combination of input variables. k is directly proportional with the number of input variable and fuzzy member per input variable. Given these factors, k is typically determined by the number of rules that have “hits” at least twice in the training set. Each extracted fuzzy rule is rated by counting how many times it is used or “hit” by the training samples. All rules that were not used were eliminated from the list of best k rules. Rules that were used only once are also eliminated from the list. The best k rules define the credit scoring model. Figure 5 shows the sample diagram of the process for getting the best k rules. Table 1. Snapshot of a single record in the sample data.
Fuzzy-Neuro Model for Intelligent Credit Risk Management
11
Figure 4. Extracting rules from the HFNN model.
Fuzzy Logic Implementation Stage of HFNN Model The optimized fuzzy sets achieved during the learning stage are used in the implementation stage. The test samples are first fuzzified through these optimized fuzzy sets before these input are inferred with the fuzzy rule base. Once the rules are evaluated and compiled, each fuzzified input is compared with the compiled rules and are classified, as shown in Figure 6. If corresponding rule is not found, the input is classified as unclassified. Otherwise, input is classified either as correctly classified or misclassified. A sample miniature compilation of the fuzzy rules learned and evaluated from the learning data are as enumerated below: 1)
IF Total Income is Low AND Civil Status is Single AND Age is Young THEN Account is BAD.
Figure 5. Illustration of retaining the best k rules.
Soft Computing with NeuroFuzzy Systems
12
Figure 6. Architecture of the fuzzy logic implementation stage of the HFNN model.
2)
IF Total Income is High AND Civil Status is Married AND Age is Middle THEN Account is GOOD. 3) IF Total Income is High AND Civil Status is Single AND Age is Old THEN Account is BAD. The Classification of the data is the end result of the process. A sample classification of the systems is shown below: 1)
IF Total Income is High AND Civil Status is Married AND Age is Middle. A fuzzified account that is actually BAD but is inferred by the fuzzy rule base as GOOD is a misclassification. 2) IF Total Income is Low AND Civil Status is Single AND Age is Young. This account is actually GOOD and classified as GOOD is correctly classified. 3) IF Total Income is Low AND Civil Status is Married AND Age is Young If this account cannot be found in the fuzzy rule base then it is unclassified. Table 1 presents sample data record, from its original form to its fuzzified form. Just refer to section 4.1 for more details.
EXPERIMENT RESULTS Description of Experiment Data In this research a total of one thousand records from the bank were selected uniformly to be used for experiments. There are 630 “good” and 370 “bad” of the thousand accounts selected. The data contains 16 variables, shown in Table 1. These 16 variables were used by statistical tool of the bank in
Fuzzy-Neuro Model for Intelligent Credit Risk Management
13
assisting the human credit evaluators in reviewing the loan applications. Also, the same input variables were considered in earlier credit scoring systems developed and published as described in section 1. Some variables of the data have to be transformed so as to reduce the complexity of the computation of the traditional NN for the benchmark. The variables Total Income, Equity Ratio and Gross Monthly Income (GMI) Ratio, and Court Case were transformed. Total Income values were transformed to the nearest million, e.g. from 100,000 to 0.10. Equity Ratio and GMI Ratio values were transformed from percentage to their decimal values equivalent, e.g. from 78.45% to 0.7845. Out of 12 possible values for the Court Case variable, these values were grouped into 2 simple values, each account have either “With” court case or “Without” court case. This is because only 14 out of the 1000 accounts selected have “With” court case value and 986 accounts have “Without” court case value.
Traditional Neural Network (MLP) Experiments Results In this research, experiments using traditional NN (Multi-Layer Perceptron) is conducted and compared to the proposed HFNN model developed. Figures 7 and 8 shows the results of the NN training and samples classification performance. Figure 7 presented a typical behavior of a neural network that is being trained. It just continuously learned and cross-validation was used to determine when the training stopped. When compared to the performance of the proposed HFNN model shown in Figure 9, the traditional NN performance does not indicate any spike because there was no fuzzy input tuning process that happened. It can be observe from Figure 8 that the patterns of training (TR), evaluation (EV), and testing (TE) samples are similar. When one group of samples, say TR, are decreasing in classification accuracy, the other groups are also decreasing. Moreover, Figure 8 shows that as classification rate for the TR samples is still increasing, the classification rates for EV and TE samples are already declining. This is an indication that the network is getting over-fitted for TR samples. Hence, the training is stopped. The final classification accuracy on this experiment is 94.67%. However, it is important to note that the training of the traditional MLP to get optimum accuracy took longer time compared to the proposed HFNN Model. It recorded and average of 48 hours to train the MLP for each set using Pentium 4 - 2.8 GHz Single Core Processor with 512 Mb RAM.
14
Soft Computing with NeuroFuzzy Systems
Figure 7. Neural network training performance.
Figure 8. Neural network classification accuracy comparison.
Figure 9. HFNN model classification performances.
The Proposed HFNN Model Experiments Results Figure 9 shows the performance of the proposed HFNN model developed in this research. It can be seen from the graph that each group of the samples behave similarly as the system is being trained. The classification accuracies of Training (TR), Evaluation (EV), and Testing (TE) samples were already dropping after the 25th epoch. However, when the tuning process for input
Fuzzy-Neuro Model for Intelligent Credit Risk Management
15
fuzzy set is activated all classification accuracies started to improve particularly the EV samples, which reached 100% accuracy from 77th epoch to 197th epoch. The classification accuracy of the system for the training samples (TR) started to slow down at 117th epoch. However, at 347th epoch the system achieved 100% classification accuracy. The evaluation samples (EV) had reached 100% classifycation accuracy at 77th epoch. However it was overfitted after the 197th epoch. Hence, a drop in the classification accuracy for EV begun until it settled at 98% classification accuracy at 247th epoch. The reason why EV had reached 100% prediction accuracy ahead of TR is because the system looks EV patterns as just part of the TR patterns. But as the system was further trained for TR, the system started to see the minor differences in the patterns between EV and TR. The continuous drop in EV classification accuracy continued until it reached a plateau staring at the 217th epoch. The test samples, TE, on the other hand, were classified by the system as having patterns much similar with EV. But the graph in Figure 9 shows that the TE samples have more patterns that are diverse than that of EV samples. That is why, TE classification rate dropped earlier and lower than EV. TE classification rate reached 96%. The tuning of the fuzzy sets served its purpose—the classification performance of the HFNN Model had improved so much. As can be seen in Figure 9, all three graphs were falling continuously until at their lowest at 47th epoch.
Performance Comparison of the Proposed HFNN Model against Traditional Neural Network (MLP) Figure 10 shows the behavior of the proposed HFNN model against the traditional NN developed in this research. Both of these two models showed superior performance against the previous published works mentioned in Section 1. It can be seen from this figure that the highest TE classification accuracy equal to 95.33% obtained by the proposed HFNN model using 2560 rules. The neural network classification accuracy is lower than this at 94.67%. Furthermore, it can be observed that using HFNN model with 95 rules resulted to 86% classification accuracy. It even dropped to 83.67% when using 60 rules. This is a trade-off of the system developed which the human experts have to decide. It should be noted that minimizing the number of rules and input variables will make easy and simple for the human credit evaluators to read and decide. The simplification of the rules may result the expense of the
16
Soft Computing with NeuroFuzzy Systems
classification accuracy. However, in this proposed HFNN model, the 83.67% classification accuracy using 60 rules is still way above acceptable than that of classical method reported in Section 1 mentioned earlier in this paper. The training time of the proposed HFNN model to get optimum accuracy performance took only 3 hours compared to 48 hours of the traditional Multilayer Perceptron (MLP) Neural Network. The fuzzification increased the granularity of the continuous input values. The complexity of the various patterns found in the samples was reduced, and so with the training time as a consequence.
DISCUSSION AND ANALYSIS OF RESULTS In this research, the developed HFNN model tuning process improved the classification accuracy significantly. Without tuning process, the classification accuracies of the 3 sets; TR, EV, and TE could have been peg to 94%. But employing the tuning process, TR samples had reached and set at 100%, EV samples at 98.57% and TE samples at 96%. This can be seen in Figure 9. The original HFNN model with complete 16 input variables got a classification accuracy of 98.57% only when tested with the evaluating samples (EV). When the redundant input variables were removed, the performance improved to 100%. The improvement signified that some input variables or their particular combinations contributed to noise. Hence, the removal of the redundant input variables not only improved the readability of the rules but it also improved the accuracy performance of the HFNN model developed.
Figure 10. Traditional NN and proposed HFNN model classification accuracy comparison for TE samples.
The improvement of the rule base can be defined in terms of performance (i.e. reduction of error) and in terms of complexity or simplicity (i.e. number of variables or parameters). There is a trade-off between performance and simplicity. To obtain high accuracy, a large number of free parameters are needed, which again resulted in a very complex and thus less comprehensible or readable model. However, often the performance of a model can actually
Fuzzy-Neuro Model for Intelligent Credit Risk Management
17
increase with the reduction of the number of parameters because the generalization capabilities of the model may increase. If the model has too many parameters, it tends to over-fit the training samples, TR, and displays poor generalization on test samples, TE. The removal of 12 input variables found to be redundant resulted to 896 extracted rules only. Eliminating the 836 noise rules, the ones with single or no “hits” in the training set, further simplified the readability of the set of rules down to 60 rules. These 60 rules define the credit scoring model with a classification performance of 83.67% when tested with the test set, way above the industry standard of 74% classification performance. Finally, The HFNN model developed in this research trains faster than the traditional NN by 16 times and has better classification accuracy of 95.33% compared to 94.67% of traditional NN.
CONCLUSIONS AND RECOMMENDATIONS The HFNN model developed in this research to solve credit risk management problem is capable of selflearning similar to the traditional neural network. Subsequently, once trained, it is capable of discriminating the “good” and the “bad” accounts with better accuracy compared to the traditional NN. Unlike the neural network’s “black box” configuration, which is an undesirable feature for credit evaluation, the HFNN model is capable of generating the rules behind the discrimination of each account subjected to it. The system behaves much like a traditional fuzzy logic system in this aspect. However, the HFNN model is better than the traditional fuzzy logic system because of its learning capability. The fuzzy logic system does not have this capability. Although, this research was done for auto loan, the Hybrid Fuzzy-Neuro Network is easily transferable to similar loan products like mortgage loan, salary loan, and even for credit card grants. These types of loans are the same because they have similar input and output variables required. In this research, the extracted rules were just listed in the order of their importance, i.e. the most relevant rules were listed first in the list. For future works, it is worth to investigate some of these rules that can be fussed together to further simplify the list of rules. The output of the developed HFNN model is limited to 2 possible values; either good or bad. By providing the data with more than 2 outputs, say 2 additional outputs, namely: marginally good and marginally bad. Marginal accounts can be taken for a closer look before a decision is granted. This can be considered for future study.
18
Soft Computing with NeuroFuzzy Systems
REFERENCES 1. 2. 3. 4. 5. 6.
7. 8. 9. 10.
11. 12.
13.
14.
M. Bonilla, I. Olmeda and R. Puertas-An, “Application of Genetic Algorithms in Credit Scoring,” WASL Journal, 2000. L. P. Wallis, “Credit Scoring,” Business Credit Magazine, March, 2001. A. Fensterstock, “Credit Scoring Basics,” Business Credit, March, 2003. A. Jost, “Neural Networks,” Credit World, Vol. 81, No. 4, 1993, p. 26. A. Fensterstock, “The Application of Neural Networks to Credit Scoring,” Business Credit, March, 2001. G. Vasconcelos, P. Adeodato and D. S. M. P. Monterio, “A Neural Network Based Solution for the Credit Risk Assessment Problem,” Proceedings of the IV Brazilian Conference on Neural Networks, Washington, 20-22 July 1999, pp. 269-274. S. A. DeLurgio and F. Hays, “Understanding the Financial Interests in Neural Networks,” WASL Journal, 2001. BrainMaker, “Credit Scoring with Brainmaker Neural Network Software,” 2003. http://www.calsi.com/CreditScoring.html J. Whitehouse, “Human Capabilities + Computer Automation = Knowledge,” IT Briefing, Pacific Lutheran University, Parkland, 2000. E. P. Dadios and D. J. Willaims, “A Fuzzy-Genetic Controller for the Flexible Pole-Cart Balancing Problem,” Proceedings of 1996 IEEE International Conference on Evolutionary Computation, Nagoya, 2022 May 1996, pp. 223-228. L. A. Zadeh, “A Theory of Approximate Reasoning,” In: Machine Intelligence, John Wiley & Sons, New York, 1979, pp. 149-194. I. Iancu, “A Mamdani Type Fuzzy Logic Controller,” In: E. P. Dadios, Ed., Fuzzy Logic: Controls, Concepts, Theories and Applications, InTech Croatia, Rijeka, 2012, pp. 55-54. A. Achs, “From Fuzzy Datalog to Multivalued Knowledge-Base,” In: E. P. Dadios, Ed., Fuzzy Logic: Algorithms, Techniques and Implementations, InTech Croatia, Rijeka, 2012, pp. 25-54. E. P. Dadios, K. Hirota, M. Catigum, A. Gutierrez, D. Rodrigo, C, San Juan and J. Tan, “Neural Network Vision Guided Mobile Robot for Driving Range Golf Ball Retriever,” Journal of Advanced Computational Intelligence and Intelligent Informatics, Vol. 10, No. 4, 2006, pp. 181- 185.
Fuzzy-Neuro Model for Intelligent Credit Risk Management
19
15. R. Gustilo and E. P. Dadios, “Optimal Control of Aquaculture Prawn Water Quality Index Using Artificial Neural Networks,” Proceedings of the 5th IEEE International Conference on Cybernetics and Intelligent Systems and the 5th IEEE International Conference on Robotics, Automation and Mechatronics, Qingdao, 17-19 September 2011, pp. 266-271. 16. S. Hilado and E. P. Dadios, “Face Detection Using Neural Networks with Skin Segmentation,” Proceedings of the 5th IEEE International Conference on Cybernetics and Intelligent Systems and the 5th IEEE International Conference on Robotics, Automation and Mechatronics, Qingdao, 17-19 September 2011, pp. 261-265. 17. N. Marcos, “Belief-Evidence Fusion through Successive Rule Refinement in a Hybrid Intelligent System”, Ph.D. Thesis, De La Salle University, Manila, 2002. 18. D. Nauck, “Combining Neural Networks and Fuzzy Controllers,” FLAI, Linz, 28 June-2 July 1993. 19. D. Nauck, “Data Analysis with Neuro-Fuzzy Methods,” Habilitation Thesis, University of Magdeburg, Magdeburg, 2000. 20. S. Haykin, “Neural Networks: A Comprehensive Foundation,” Prentice Hall, Upper Saddle River, 1999. 21. S. Piramuthu, “Financial Credit-Risk Evalution with Neural and Neurofuzzy Systems,” European Journal of Operation Research, Vol. 112, No. 1, 1999, pp. 310-321. doi:10.1016/S0377-2217(97)00398-6.
CHAPTER
2
Method to Improve Airborne Pollution Forecasting by Using Ant Colony Optimization and Neuro-Fuzzy Algorithms
Elizabeth Martinez-Zeron1, Marco A. Aceves-Fernandez1, Efren Gorrostieta-Hurtado1, Artemio Sotomayor-Olmedo2, Juan Manuel Ramos-Arreguín1 Facultad de Informática, Universidad Autónoma de Querétaro, Querétaro, México 1
Laboratory of Artificial Intelligence, Faculty of Engineering, Universidad Autónoma de Querétaro, Querétaro, México 2
ABSTRACT This contribution shows the feasibility of improving the modeling of the non-linear behavior of airborne pollution in large cities. In previous works, Citation: Martinez-Zeron, E., Aceves-Fernandez, M., Gorrostieta-Hurtado, E., Sotomayor-Olmedo, A. and Ramos-Arreguín, J. (2014), “Method to Improve Airborne Pollution Forecasting by Using Ant Colony Optimization and Neuro-Fuzzy Algorithms”. International Journal of Intelligence Science, 4, 81-90. doi: 10.4236/ijis.2014.44010. Copyright: © 2014 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY). http:// creativecommons.org/licenses/by/4.0
22
Soft Computing with NeuroFuzzy Systems
models have been constructed using many machine learning algorithms. However, many of them do not work for all the pollutants, or are not consistent or robust for all cities. In this paper, an improved algorithm is proposed using Ant Colony Optimization (ACO) employing models created by a neuro-fuzzy system. This method results in a reduction of prediction error, which results in a more reliable prediction models obtained. Keywords:- Neuro-Fuzzy models, Ant Colony Optimization, Airborne Pollution
INTRODUCTION In recent years, the environment has been affected by the presence of particulate pollutants such as the Ozone (O3), NO2 nitrogen oxide, carbon monoxide CO, sulfur dioxide (SO2) and particulate matter less than 10 microns PM10 (≤10 microns) and Particles less than 2.5 micrometers PM2.5 (≤2.5 microns) [1] . For this reason, pollution monitoring has been necessary for large cities with high concentration of population and industries.
During several years the air quality in Mexico City for prevention of toxicity levels in health and the environment have been observed and evaluated [1] . Measures were performed to obtain information efficiency and reliability of the air quality. Some of these contributions consists in predicting pollution levels such as the work of Cortina [2] , who forecast levels of ozone pollution in the city of Guanajuato in Mexico by using neural networks, Aceves [3] takes modeling variables between relative humidity and temperature level of contamination. Sotomayor [4] uses support vector machines and kernel functions particulate ozone (O3), (PM10) and nitrogen dioxide in Mexico City. Patterns of pollution levels do not show linear behavior [3] [4] whereby a pattern is generated in clusters can be represented by several linear functions, this for ease of interpretation. In this case study, Fuzzy C means is used to generate clusters which have similar characteristics, subsequently establish the cluster centers as membership functions in a fuzzy system [3] , Also, neuro-fuzzy inference system (ANFIS) with multiple inputs and one output (“Multiple-Inputs-Single-Output” or MISO) is used to approximate nonlinear functions [5] [6] . Subsequently, three models are generated by the above steps after obtaining improved by using the algorithm of ant colony
Method to Improve Airborne Pollution Forecasting by Using Ant ...
23
optimization (ACO) prediction. This methodology shows that it is possible to improve existing algorithms to predict the levels of particulate pollutants, in this case study in Mexico City.
Fuzzy inference methods A Fuzzy Inference System (FIS) is basically a logic that allows intermediate values defined between conventional evaluations as Yes/No, True/False, etc. A FIS is the main unit of a fuzzy logic system, the part where the decision is made, by constructing appropriate rules based on the theory of fuzzy sets. The rules are constructed linguistically (IF-THEN-) having the general form of “If A Then B” where A and B are (collections of) propositions containing linguistic variables linked by connectors (AND, OR) to make the correct decision-making [7] . A fuzzy inference system consists of 5 blocks as described in Figure 1. Fuzzy inference system operates as follows: The input data are treated by a method of fuzzification then the Fuzzy rules are formed and the already fuzzyfied input data are analyzed. Defuzzification method is used to convert the value obtained after analysis with fuzzy rules to obtain an output value applicable to the real world. The steps taken by fuzzy reasoning, i.e. inference operations on rules IF - THEN performing the FIS, are: •
Compare the input variables with the membership functions for the linguistic value of each part, this process is called fuzzyfication. • Combine by an operator (t-norm, usually multiplication or minimum) the membership values to obtain the weight of each rule. • Add the resulting output to produce a crisp, this step is called defuzzification. A fuzzy rule has a general structure in a fuzzy model as shown on equation (1): if x is A and y is THEN z = f (x, y) (1)
where A,B fuzzy sets in the previous data sets are, z=f(x, y)is a consequent function. Usually f(x, y) is polynomial in the input variables x and y, the output of the system with a fuzzy region specified by the predecessors of the rule.
Soft Computing with NeuroFuzzy Systems
24
Figure 1. Block Diagram of a Generic Fuzzy inference system.
A typical rule in FIS model has the form as shown in equation (2) [8] : (2) if Input1 = x AND Input 2 = y, THEN output is z = ax + by + c
(2)
The final output of the FIS system is the weighted average of all the outputs of the rules:
∑ Final Output = ∑
N
i =1 N
w i zi
i =1
w i (3)
The Takagi-Sugeno model or TSK model proposed in 1985 [8] [9], is widely used in the theoretical analysis, application control and Fuzzy Modeling. A Fuzzy system needs a precedent and consequent to express a logical connection between the input and output data used as a basis to produce the desired behavior of the system.
Clustering Clustering is the grouping of data based on certain criteria that clusters have similar properties and characteristics, clusters are based mainly on the distance interpreted as similarity [10] .
Fuzzy C-means Fuzzy C-Means (FCM) is a clustering method which allows a data to belong in one or more clusters, this method developed by Dunn [11] and improved
Method to Improve Airborne Pollution Forecasting by Using Ant ...
25
by Bezdek [10] , is often used for pattern recognition, it is an iterative optimization algorithm that minimizes the cost function given by:
∑ ∑ n
= J
c
= k 1 =i 1
2
µikm x k − vi (4)
where n is the number of data, c is the number of clusters, xk is the kth data point, vi is the center of the ith cluster, mik is the degree of membership of the kth data in the ith cluster, m is any real number greater than 1, (typically m=2) [10] . The degree of membership mik is defined by: 1
µik =
x −v ∑ j=1 x k − vi j k c
2/ (m −1)
(5) Starting with the desired number of cluster c and an initial randomly chosen center for each cluster vi, vi=1,2,3,L,c, FCM will converge in a solution to vi which represents both a local minimum or a point function cost [10] . This clustering method uses fuzzy partition such that each point can belong to several clusters with different membership values between 0 and 1. Fuzzy c-means, has predefined parameters such the weight of the exponent m and the number of clusters. To measure the effect of clustering algorithm, validity and accuracy are required. Validity: If the algorithm can find all internal type in the data collected. Truth: If the algorithm can set the same kind of data to the same group, different types of data at different clusters. We define the rate of decline of Clustering (Cr) to measure the effect of clustering [7] : = Cr
m best × 100 m result
(6)
mbest is best clustering number and mresult is the actual number of cluster after clustering.
∑ ( µ (x) ) ∑ ( µ (X) )
m
vi
x∈X
C
n
x∈x
C
m
, 1≤ i ≤ k
(7)
After determining the number of clusters, the value of m, and the stopping criterion the FCM algorithm performs two steps, first calculates the membership functions using equation (5), as a second step updates the proto
Soft Computing with NeuroFuzzy Systems
26
types using equation (7) the two steps are repeated iteratively to achieve the stability criterion, in which the change of the output values by the previous equations is minimal. The Figure 2 shows an example FCM function.
Figure 2. Example of Fuzzy C-Means showing three different clusters with a set of data points each, and their centroids.
The Adaptive Neuro-Fuzzy Inference System (ANFIS) This method is a type of neural network that is based on a method for the process of fuzzy modeling to learn information about a data set by generating fuzzy rules, this is called neuro-fuzzy, this architecture was proposed by Jang in 1993 [12] . Has a network structure with directional links connecting a number of nodes. Each node has a function with adjustable or fixed parameters. The classical ANFIS consists of 5 layers with specific tasks as show in Figure 3 [13] . First Layer (Layer 1). Layer one is performed the input fuzzification. That is, each entry will set a value of belonging that only covers the total of the input variables to be treated [13]. This is expressed in mathematical terms as follows: Out ij(1) = µ j (ln i(1) ) (8)
While: Out ij(1) is the layer 1 node’s output which corresponds to the j-th linguistic
term of the i-th input variable. The membership is: = µj
1 = , i 1,...., Numln = Vars, j 1,....., Nu m ln tTer bo x i − cij 1+ a ij
(9)
While the parameters (aij, bij, cij) are referred to as premise parameters or non-linear parameters and they adjust the shape and the location of the membership function.
Method to Improve Airborne Pollution Forecasting by Using Ant ...
27
Second Layer (Layer 2 from Figure 3). Executes the fuzzy AND of the antecedent part of the fuzzy rules. This results to each node’s output being the product of all of its inputs (every input term node that is connected to that rule node) [14] : Inputs Out (2) Out ij(1) , (10) k = w k = ∏ i =1
N
For all the terms nodes j connected to the k-th the rule node, k=1, ….., NumRules Third layer (Layer 3 from figure 3). Normalizes the membership functions (MFs). The output of the k-th node is the firing strength of each rule divided by the total sum of the activation values of all the fuzzy rules. This results in the normalization of the activation value for each fuzzy rule. This operation is simply written as: (3) Out= w= k k
∑
Out (2) k
N Rules m =1
Out (2) m
,= k 1,..., NumRules
(11)
Fourth layer (Layer 4 from figure 3) executes the consequent part of the fuzzy rules. Each node k in this layer is accompanied by a set of adjustable parameters a1k , a 2k ,......., a N
Inputs
k
, a ok and implements the linear function:
(1) (1) (1) Out (4) k = w k f k= w k (a1k ln1 + a 2k ln 2 + ..... + a N Inputs k ln N
Inputs
k
+ a 0k ),
k= 1,...., NumR
(12)
Figure 3. ANFIS architecture.
The weight wk is the normalized activation value of the k-th rule, calculated by from equation (7). Those parameters are called consequent parameters or linear parameters of the ANFIS system and are adjusted by the RLS
Soft Computing with NeuroFuzzy Systems
28
algorithm. Layer 5 computes the output of fuzzy system by summing up the outputs of layer fourth. Rules Rules Out (5) ∑= Out (4) w k fk = = ∑ k = k 1= k 1
N
N
∑ ∑
N Rules
k =1 N Rules k =1
w k fk wkf
(13)
Optimization The optimization algorithm is a numerical method which finds a value xi ∈ Rn , where Rn is an n-dimen- sional search space, which minimizes or maximizes the function optimization J(x); through the systematic selection of values of the variable xi using some restrictions. Here, J(x) is called the objective function. A feasible solution that minimizes or maximizes the objective function is called an optimal solution [15] . As mentioned in [15] the techniques used for providing solution to complex combinatorial optimization problems have evolved from constructive methods to local search methods and finally to population-based algorithms. These techniques are classified as: Extensive and not exhaustive. These non-analytical computational techniques are based, and the classification is performed based on the solution space.Exhaustive optimization techniques are those that guarantee always find the optimal (maximum or minimum) walking in the worst case the whole solution space (which can be considerably large) [16] . Non-exhaustive techniques rely on getting good enough solutions without exceeding the time constraints established or memory. Solutions are based on steps taken to find new solutions that approximate the optimal solution more, increasing quality. Some of the best known are: • • • • • •
Genetic Algorithms Taboo search Ant Colony Optimization Greedy Randomized Adaptive Search Procedure (GRASP). Scatter Search Simulated Annealing
Ants Colony Optimization The ant colony optimization (ACO) is a bio-inspired technique of classification based on the external behavior of a real ant colony when
Method to Improve Airborne Pollution Forecasting by Using Ant ...
29
foraging. This algorithm was created in 1992 by Dorigo [17] . The ACO algorithm solves complex problems of combinatorial optimization in various fields of engineering, commerce, industry, among others. When a problem does not have a polynomial equation that describes it belongs to the NP-hard problems. The results provided by ACO can be approximated to address the combinatorial optimization problemsas shown in the Figure 4. [18] . The probability of an ant k to choose the next node, where x is the current state and y is the next state we have:
Figure 4. Example of ACO where ants choose the shortest path from node 1 to node 2. p
k xy
( τ )( η ) = ∑ ( τ )( η ) (14) α xy
β xy
α xy
β xy
where: ηxy
1
is the level of convenience that has a transition state, d xy d xy is the distance. τij
is the trace levels, or that the amount of pheromone which will be deposited in a state of transition. β is a parameter that controls the influence of ηxy α is a parameter that controls the influence of τxy
Updated Trail Pheromone must be updated because this changes which followed the iterations by offering increased likelihood based on the accumulated experience [19] . In this case the level of pheromone raises based on the best solution path according to equation (14) τij= ü − ρ τij + ∆τij (15)
where: ρ ∈ (0,1] It is the persistence factor trail
Soft Computing with NeuroFuzzy Systems
30 = ∆τij
∑
NrosAnts k =1
∆τij (16)
Accumulating trace is proportional to the quality of solutions. 1 , if ant k used e deg pi , p j ; ∆τij = L k 0, otherwise.
(17)
Lk represents the objective value of the solution k (in the denominator for minimization problems). Every time an ant travels a node applies the rule: τij ← (1 − ϕ) × t ij + t 0 (18)
ϕ ∈ (0,1] Decrement pheromone parameter.
The stop condition This process is iterative until it stops for compliance with established criteria
METHODOLOGY In Figure 5, the environmental monitoring map shown in Mexico City, in this study we only evidence in the northwest area of the spring season. The list of the sites used in this contribution is shown on figure 5, whilst the proposed methodology for this contribution is shown in the figure 6. The monitoring stations used for this case study were chosen due to data availability and specific industrial or polluted areas. Since obtaining data from such stations is filtered data, i.e., data values are complete and validated readings are sought. The data are arranged in time series along the day; hence the clustering is performed to generate the characteristics of groups depending on pollution levels. These are grouped according through the fuzzy c means algorithm, which generates fuzzy clusters, based on a center that iteratively changes to accommodate more data groups. This provides greater robustness for modeling. Data modeling is performed using a fuzzy logic system, wherein each generated center is interpreted as fuzzy rules language.
Method to Improve Airborne Pollution Forecasting by Using Ant ...
31
Figure 5. Monitored areas.
The fuzzy system is improved with training data, which in this case, the training data are data from a previous year validated dataset. The training is through the use of neuro-fuzzy system AN-FIS which generates a more robust data modeling system. The training is carried out using of neuro-fuzzy system ANFIS which generates a more robust data modeling system. Three models for the prediction of air quality are generated; these models are subject to the ant colony optimization algorithm to improve the prediction closer to an outcome more accurately. Such three models along with the real measured data and its corresponding prediction are shown on Figure 7. The mean square error for each model generated with neuro-fuzzy and neurofuzzy with ACO to measure the difference between the estimated and the model.
RESULTS AND DISCUSSION The models are created using Fuzzy Logic where the rules are formed by clustering, the ANFIS to improve each model and make each one more accurate and the ACO algorithm use these three models to improve the forecast having different options base on the three previous models.
32
Soft Computing with NeuroFuzzy Systems
Figure 6. Proposed methodology.
Figure 7. Model selection by ACO. The models are named Mod 2 Mod 1 and Mod 3, respectively. These models are generated based on the characteristics of the stations, i.e. stations show greater similarity between their measurements generate a model. The results are shown on table 1 and table 2. CO is sensed to the northwest, with three models generated based on the area in 2009, and data validation are the stations belonging to the same area but in 2010. Table 1. Results of spring in northwestern federal district with CO pollutant.
Method to Improve Airborne Pollution Forecasting by Using Ant ...
33
Table 2. Results of spring in northwestern federal district with O3 pollutant.
Likewise the table 2 is showing the models of Ozone (O3). In the following tables the mean square error, it can be seen that is closer to the real value, the prediction of stations Station 1, Station 2, Station 3 and Station 4 based Mod 1, Mod 2 and Mod 3 models. The station in which there was no good approximation was at Station 1.Figure 7 shows how the algorithm is based on the nearest location from the current node to the next node following the ACO algorithm as shown on equations (14)-(16) and the methodology shown on figure 6. The influence of the pheromone that can choose the model that is closest to the actual data, likewise considering also the shortest distance.The figure 7 shows the comparison between the models created by the trained logic system with ANFIS and the ACO model. This to appreciate the difference between the accuracy for each model generated with neuro- fuzzy and neuro-fuzzy improved with ACO.
CONCLUSIONS The algorithms based on swarm intelligence are feasible for solving problems different than the typical machine learning methods. It has shown that a combination of modeling and optimization methods may be used to improve the prediction for this type of non-linear problem. Such combination of methods showed that systems working cooperatively is well complemented and integrated to generate a task system. For these reasons ACO is an algorithm that provides an improvement in the prediction of contamination if used properly. ACO prediction has a point of view on the combinatorial type issue, and thus, the use of this algorithm improves the contaminants prediction by making the shortest search and the use of pheromone provides a best approximation to real data.
ACKNOWLEDGEMENTS Elizabeth Martinez-Zeron, Marco A. Aceves-Fernandez, Efren GorrostietaHurtado, Artemio Sotomayor-Olmedo, Juan Manuel Ramos-Arreguín This work was made possible through funding provided by the National Council of Science and Technology CONACYT.
34
Soft Computing with NeuroFuzzy Systems
REFERENCES 1.
http://www.calidadaire.df.gob.mx/calidadaire/index.php?opcion=4&o pcionrecursostecnicos=3 2. Cortina. M. (2012) Aplicación de inteligencia artificial a la predicción de contaminantes atmosféricos. Tesis Doctoral, E.T.S.I. Telecomunicación (UPM). 3. Aceves-Fernández, M.A., Sotomayor-Olmedo,A., Gorrostieta-Hurtado, E., Pedraza-Ortega, J.C., Tovar-Arriaga, S. and Ramos-Arreguin, J.M. (2011) Performance Assessment of Fuzzy Clustering Models Applied to Urban Airborne Pollution. 2011 21st International Conference on Electrical Communications and Computers (CONIELECOMP), San Andres Cholula, 28 February-2 March 2011, 212-216. 4. Sotomayor-Olmedo, A., Aceves-Fernández, M.A., GorrostietaHurtado, E., Pedraza-Ortega, C., Ramos-Arreguín, J.M. and VargasSoto, J.E. (2013) Forecast Urban Air Pollution in Mexico City by Using Support Vector Machines: A Kernel Performance Approach. International Journal of Intelligence Science, 3, 126-135. http://dx.doi. org/10.4236/ijis.2013.33014 5. (2005) Adaptive Neuro-Fuzzy Inference Systems (ANFIS). In: Nedjah, N. and de Macedo Mourelle, L., Eds., Fuzzy Systems Engineering: Theory and Practice. Vol. 181, Springer, Rio de Janeiro, 53-83. 6. Adaptive Neuro-Fuzzy Inference Systems (ANFIS) Library for Simulink Ilias S. Konsoulas M.Sc., CEng, MIET. 7. Aceves-Fernandez, M.A., Sotomayor-Olmedo, A., GorrostietaHurtado, E., Pedraza-Ortega, J.C., Ramos-Arreguín, J.M, CancholaMagdaleno, S. and Vargas-Soto, E. (2011) Advances in Airborne Pollution Forecasting Using Soft Computing Techniques. In: Popovic, D., Ed., Air Quality: Models and Applications, INTECH Publisher, Querétaro, 1-14. http://dx.doi.org/10.5772/16273 8. Sugeno, M. and Kang, G.T. (1988) Structure Identification of Fuzzy Model. Fuzzy Sets and Systems, 28, 15-33. http://dx.doi. org/10.1016/0165-0114(88)90113-3 9. Takagi, T. and Sugeno, M. (1985) Fuzzy Identification of Systems and Its Application to Modeling and Control. IEEE Transactions on Systems, Man and Cybernetics, SMC-15, 116-132. 10. Bezdek, J.C. (1981) Pattern Recognition with Fuzzy Objective Function Algoritms. Plenum Press, New York. http://dx.doi.org/10.1007/978-1-
Method to Improve Airborne Pollution Forecasting by Using Ant ...
11.
12.
13.
14. 15.
16. 17. 18.
19.
35
4757-0450-1 Dunn, J.C. (1973) A Fuzzy Relative of the ISODATA Process and Its Use in Detecting Compact Well-Separated Clusters. Journal of Cybernetics, 3, 32-57. http://dx.doi.org/10.1080/01969727308546046 Jang, J.S.R. (1993) ANFIS: Adaptive-Network-Based Fuzzy Inference System. IEEE Transactions on Systems, Man and Cybernetics, 23, 665-685. Ghanbari, A., Abbasian-Naghneh, S. and Hadavandi, E. (2011) An Intelligent Load Forecasting Expert System by Integration of Ant Colony Optimization, Genetic Algorithms and Fuzzy Logic. 2011 IEEE Symposium on Computational Intelligence and Data Mining (CIDM), Paris, 11-15 April 2011, 246-251. Adaptive Neuro-Fuzzy Inference Systems (ANFIS) Library for Simulink Muñoz, M., López, J. and Caicedos, E. (2008) Inteligencia de enjambres: Sociedades para la solución de problemas (una revisión). Ingeniería e Investigación, 28, 119-130. Carretero, J.C.O., Cánovas, D.G. and Pereira, F.D.Q. (2008) Técnicas Heurísticas para Problemas de Diseño en Telecomunicaciones. Dorigo, M. (1992) Optimization, Learning and Natural Algorithms. Ph.D. Thesis, Politecnico di Milano, Italian. Calixto, C.A.C. (2005) Implementacion En Hidroinformática De Un Método De Optimización Matemática Basado En La Colonia De Hormigas. Tesis no publicada, Pontificia Universidad Javeriana, Bogota, Colombia. Kole, A., Chakrabarti, P. and Bhattacharyya, S. (2013) An Ant Colony Optimization Algorithm for Uncapacitated Facility Location Problem.
CHAPTER
3
A TSK-Type Recurrent Neuro-Fuzzy Systems for Fault Prognosis
Rafik Mahdaoui1,2, Leila Hayet Mouss2 Laboratoire d’Automatique et Productique (LAP), Université de Batna, Batna, Algérie; 1
2
Centre Universitaire Khenchela, Khenchela, Algérie
ABSTRACT As a result from the demanding of process safety, reliability and environmental constraints, a called of fault detection and diagnosis system become more and more important. In this article some basic aspects of TSK (Takigi Sugeno Kang) neuro-fuzzy techniques for the prognosis and diagnosis of manufacturing systems are presented. In particular, a neuroCitation: R. Mahdaoui and L. Mouss, “A TSK-Type Recurrent Neuro-Fuzzy Systems for Fault Prognosis,” Journal of Software Engineering and Applications, Vol. 5 No. 7, 2012, pp. 477-482. doi: 10.4236/jsea.2012.57055. Copyright: © 2012 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY). http:// creativecommons.org/licenses/by/4.0
38
Soft Computing with NeuroFuzzy Systems
fuzzy model that can be used for the identification and the simulation of faults prognosis models is described. The presented model is motivated by a cooperative neuro-fuzzy approach based on a vectorized recurrent neural network architecture. The neuro-fuzzy architecture maps the residuals into two classes: a one of fixed direction residuals and another one of faults belonging to rotary kiln. Keywords:- TSK Neuro-Fuzzy Systems; Faults Diagnosis; Fault Prognosis
INTRODUCTION Failure may cause large amount of loss. Therefore, fault diagnosis and prognosis system is very important for safe operation and preventing rescue. Recent progress in the field of diagnostics of manufacturing systems (MS) drives is a result of broadly conceived basic research carried out over many years. According to [1], there is no single method that could accommodate the entire system fault. Thus, the combination of ANN and fuzzy logic is considerably practical because it combined both of the advantages and makes the entire system more robust. ANN can operate simultaneously on qualitative and quantitative data and very useful when no mathematical model of the system is available whereas fuzzy logic has an ability to mimic the sensing, generalizing, processing, operating and learning ability of human operator [2]. In order to achieve this goal we organize this article into three parts. The first part presents principal architectures of TSK Temporal Neuro-Fuzzy systems operation and their applications. The second part is dedicated to the workshop of clinker of cement factory. Lastly, in the third part we propose a Neuro-Fuzzy system for system of production diagnosis.
TEMPORAL NEURO-FUZZY SYSTEMS Fuzzy neural network (FNN) approach has become a powerful tool for solving real-world problems in the area of forecasting, identification, control, image recognition and others that are associated with high level of uncertainty [2]. The Neuro-fuzzy model combines, in a single framework, both numerical and symbolic knowledge about the process. Automatic linguistic rule extraction is a useful aspect of NF especially when little or no prior knowledge about the process is available [1,3]. For example, a NF model of a non-linear dynamical system can be identified from the empirical data.
A TSK-Type Recurrent Neuro-Fuzzy Systems for Fault Prognosis
39
This model can give us some insight about the on linearity and dynamical properties of the system. The most common NF systems are based on two types of fuzzy models TSK [4,5] combined with NN learning algorithms. TSK models use local linear models in the consequents, which are easier to interpret and can be used for control and fault diagnosis [6]. Mamdani models use fuzzy sets as consequents and therefore give a more qualitative description. Many Neuro-fuzzy structures have been successfully applied to a wide range of applications from industrial processes to financial systems, because of the ease of rule base design, linguistic modeling, and application to complex and uncertain systems, inherent non-linear nature, learning abilities, parallel processing and fault-tolerance abilities. However, successful implementation depends heavily on prior knowledge of the system and the empirical data [7]. Neuro-fuzzy networks by intrinsic nature can handle limited number of inputs. When the system to be identified is complex and has large number of inputs, the fuzzy rule base becomes large. NF models usually identified from empirical data are not very transparent. Transparency accounts a more meaningful description of the process i.e. less rules with appropriate membership functions. In ANFIS [2] a fixed structure with grid partition is used. Antecedent and consequent parameters are identified by a combination of least squares estimate and gradient based method, called hybrid learning rule. This method is fast and easy to implement for low dimension input spaces. It is more prone to lose the transparency and the local model accuracy because of the use of error back propagation that is a global and not locally nonlinear optimization procedure. One possible method to overcome this problem can be to find the antecedents & rules separately e.g. clustering and constrain the antecedents, and then apply optimization. Hierarchical NF networks can be used to overcome the dimensionality problem by decomposing the system into a series of MISO and/or SISO systems called hierarchical systems [6]. The local rules use subsets of input spaces and are activated by higher level rules [7]. The criteria on which to build a NF model are based on the requirements for faults diagnosis and the system characteristics. The function of the NF model in the FDI scheme is also important i.e. Preprocessing data, Identification (Residual generation) or classification (Decision Making/ Fault Isolation).
Soft Computing with NeuroFuzzy Systems
40
For example a NF model with high approximation capability and disturbance rejection is needed for identification so that the residuals are more accurate. Whereas in the classification stage, a NF network with more transparency is required. The following characteristics of NF models are important: • Approximation/Generalisation capabilities; • Transparency: Reasoning/use of prior knowledge/rules; • Training Speed/Processing speed; • Complexity; Transformability: To be able to convert in other forms of NF models in order to provide different levels of transparency and approximation power. Adaptive learning Two most important characteristics are the generalising and reasoning capabilities. Depending on the application requirement, usually a compromise is made between the above two. In order to implement this type of Neuro-Fuzzy Systems for Fault Diagnosis and Prognosis and exploited to diagnose of dedicated production system we have to propose data-processing software NEFDIAG (NeuroFuzzy Diagnosis). The Takagi-Sugeno type fuzzy rules are discussed in detail in Subsection A. In Subsection B, the network structure of FENN is presented.
Temporal Fuzzy Rules Recently, more and more attention has paid to the Takagi-Sugeno type rules [8] in studies of fuzzy neural networks. This significant inference rule provides an analytic way of analyzing the stability of fuzzy control systems. If we combine the Takagi-Sugeno controllers together with the controlled system and use state-space equations to describe the whole system [9], we can get another type of rules to describe nonlinear systems as below:
Rule r
A TSK-Type Recurrent Neuro-Fuzzy Systems for Fault Prognosis
where nonlinear system; dimensions;
41
is the inner is the inner state vector of the is the input vector to the system, and N, M are the
are linguistic terms (fuzzy sets) defining the conditions for xi and uj respectively, according to Rule r;
When considered in discrete time, such as modeling using a digital computer, we often use the discrete statespace equations instead of the continuous version. Concretely, the fuzzy rules become:
Rule r
where is the discrete sample of state vector at discrete time t. In following discussion we shall use the latter form of rules. In both forms, the output of the system is always defined as: (1) where vector Y
a matrix of P × N, and P is the dimension of output
The fuzzy inference procedure is specified as below. First, we use multiplication as operation AND to get the firing strength of Rule r: (2) where are the membership functions of respectively? After normalization of the firing strengths, we get (assuming R is the total number of rules) (3)
42
Soft Computing with NeuroFuzzy Systems
where S is the summation of firing strengths of all the rules, and hr is the normalized firing strength of Rule r. When the defuzzification is employed, we have
(4) Where, Using Equation (4), the system state transient equation, we can calculate the next state of system by current state and input.
The Structure of Temporal Neuro-Fuzzy System The main idea of this model is to combine simple feed forward fussy systems to arbitrary hierarchical models. The structure of recurrent Neuro-fuzzy systems is presented in Figure 1. In this network, input nodes which accept the environment inputs and context nodes which copy the value of the state-space vector from layer 3 are all at layer 1 (the Input Layer). They represent the linguistic variables known as uj and xi in the fuzzy rules. Nodes at layer 2 act as the membership functions, translating the linguistic variables from layer 1 into their membership degrees. Since there may exist several terms for one linguistic variable, one node in layer 1 may have links to several nodes in layer 2, which is accordingly named as the term nodes. The number of nodes in the Rule Layer (layer 3) and the one of the fuzzy rules are the same—each node represents one fuzzy rule and calculates the firing strength of the rule using membership degrees from layer 2. The connections between layer 2 and layer 3 correspond with the antecedent of each fuzzy rule. Layer 4, as the Normalization Layer, simply does the normalization of the firing strengths. Then with the normalized firing strengths hr, rules are combined at layer 5, the Parameter Layer, where A and B become available. In the Linear System Layer, the 6th layer, current state vector X(t) and input vector U(t) are used to get the next state X(t + 1), which is also fed back to the context nodes for fuzzy inference at time (t + 1). The last layer is the Output Layer, multiplying X(t + 1) with C to get Y(t + 1) and outputting it.
A TSK-Type Recurrent Neuro-Fuzzy Systems for Fault Prognosis
43
Next we shall describe the feed forward procedure of TNFS by giving the detailed node functions of each layer, taking one node per layer as example. We shall use notations like to denote the ith input to the node in layer k, and o[k] the output of the node in layer k. Another issue to mention here is the initial values of the context nodes. Since TNFS is a recurrent network, the initial values are essential to the temporal output of the network. Usually they are preset to 0, as zero-state, but non-zero initial state is also needed for some particular case.
Figure 1. The structure of a simple TNFS.
•
Layer 1. There is only one input to each node at layer 2. The Gaussian function is adopted here as the membership function:
(5) where c and s give the center (mean) and width (variation) of the corresponding m[1] linguistic term of input µ [2] in Rule r. r
•
r
Layer 2. This layer has several nodes, one for figuring matrix A and the other for B. Though we can use many nodes to represent the components of A and B separately, it is more convenient to use matrices. So with a little specialty, its weights of links from layer 4 are matrices Ar (to node for A) and Br (to node for B). It is also fully connected with the previous layer. The functions of nodes for A and B are respectively. (6)
Soft Computing with NeuroFuzzy Systems
44
•
Layer 3. The Linear System Layer has only one node, which has all the outputs of layer 1 and layer 2 connected to it as inputs. Using matrix form of inputs and output, we have
PROGNOSTICS PROCESS The first step in building a prognostics system, as published in the ISO standard, is the identification of the set of failure modes (FM), their influence factors on each other and the detection measures (descriptors) that allow to track the evolution of the degradation. The international standard IEC 60812 [10] has presented a procedure named “Procedure for failure mode and effects analysis (FMECA)”, which helps the identification of all the failure modes for a specific system, by the analysis of its subsystem s and components. Also, the FMECA method classifies the FMs using risk priority numbers (RPN) that are calculated with three failure mode parameters: occurrence (Occ), detection (Det) and severity (Sev). So, the FMECA [11] allows the definition of the appropriate detection method and measures to be used in the diagnostics as well as in the prognostics of the failure modes. The network structure is build in three steps: Step 1. The determination of fuzzy subsets for every input variable. The initial values of the centres and variances characterising the membership functions of the first layer down, can be arbitrarily established (equidistant on the domain of definition of the linguistic variable) or applying a clustering algorithm of the type Fuzzy C-Means. Step 2. Obtain the minimal dimension of the rule base. The extraction of most significant rule that determines the number of the nodes in the second layer. Step 3. Optimization of the parameters of rules determined at Step 2. The objective is to alternate the parameter values (c,w) of the network in order to improve the rule base minimizing the quadratic criteria of performance,
EXPERIMENTAL RESULTS To test the quality of the model, several actions were generated, and fixed goals were defined. The goals were defined in a way that the results were
A TSK-Type Recurrent Neuro-Fuzzy Systems for Fault Prognosis
45
understood without ambiguity by human knowledge, the Figure 2 illustrate the fuzzy base rules with 3 parametres to classify the defaults modes In order to illustrate the learning effect of the proposed immune based FNN (IM-FNN), we use One of the most important types of systems present in the process industry is workshop of SCIMAT clinker. A fault in a workshop of SCIMAT clinker may lead to a halt in production for long periods of time. Apart from these economic considerations faults may also have security implications. A fault in an actuator may endanger human lives, as in the case of a fault in an elevator’s emergency brakes or in the stems position control system of a nuclear power plant [9,12]. The design and performance testing of fault diagnosis systems for industrial process often requires a simulation model since the actual system is not available to generate normal and faulty operational. In Figure 3 the detection of fault mode in the rotary kiln is observed with the classification after training the neuro-fuzzy system.
Figure 2. The generated rules base.
Data needed for design and testing, due to the economic and security reasons that they would imply. According to this Figure 4 we can say the prediction of a faults is a complex problem and need the correction of inverse problem.
46
Soft Computing with NeuroFuzzy Systems
Figure 3. Detection of fault mode.
Figure 4. The failure mode of rotary kiln.
CONCLUSION The successful of implementing neuron-fuzzy is heavily depends on prior knowledge of the system and the training data. In the intrinsic nature, the neuro-fuzzy only can handle a limited number of inputs and can usually be identified in a not very transparent way from the empirical data [2]. The transparency is the determination of the process with a less amount of fuzzy rules with appropriate membership function. For the complex system, a large architecture is needed to represent a model.
A TSK-Type Recurrent Neuro-Fuzzy Systems for Fault Prognosis
47
REFERENCES 1.
R. J. Patton, P. M. Frank and R. N. Clark, “Issues of Fault Diagnosis for Dynamic Systems,” Springer, London, 2000. 2. L. Marinai, “Gas Path Diagnostics and Prognostics for Aero-Engines Using Fuzzy Logic and Time Series Analysis,” Ph.D. Thesis, School of Engineering, Canfield University, Canfield, 2004. 3. J. M. Koscielny and M. Syfert, “Fuzzy Logic Applications to Diagnostics of Industrial Processes,” Preprints of the 5th IFAC Symposium on Fault Detection, Supervision and Safety for Technical Processes, Washington, 9-11 June 2003, pp. 771-776. 4. C.-F. Juang, “A TSK-Type Recurrent Fuzzy Network for Dynamic Systems Processing by Neural Network and Genetic Algorithms,” IEEE Transactions on Fuzzy Systems, Vol. 10, No. 2, 2002, pp. 155-170. doi:10.1109/91.995118 5. C. D. Bocaniala and J. Sa da Costa, “Tuning the Parameters of a Fuzzy Classifier for Fault Diagnosis. HillClimbing vs Genetic Algorithms,” Proceedings of the 6th Portuguese Conference on Automatic Control, Faro, 7-9 June 2004, pp. 349-354. 6. J. He, “Neuro-Fuzzy Based Fault Diagnosisf or Nonlinear Processes,” Ph.D. Thesis, The University of New Brunswick, New Brunswick, 2006. 7. F. J. Uppal and R. J Patton, “Fault Diagnosis of an Electro-pneumatic Valve Actuator Using Neural Networks with Fuzzy Capabilities,” 2002. 8. C. D. Bocaniala, J. Sa da Costa and V. Palade, “A Novel Fuzzy Classification Solution for Fault Diagnosis,” International Journal of Fuzzy and Intelligent Systems, Vol. 15, No. 3-4, 2004, pp. 195-206. 9. K. B. Ariffin, “On Neuro-Fuzzy Applications for Automatic Control, Supervision, and Fault Diagnosis for Water Treatment Plant,” Ph.D. Thesis, Faculty of Electrical Engineering Universiti, Teknologi, 2007. 10. D. Henry, X. Olive and E. Bornschlegl, “A Model-Based Solution for Fault Diagnosis of Thruster Faults: Application to the Rendezvous Phase of the Mars Sample Return Mission,” European Conference for Aero-Space Sciences, St. Petersburg, 4-8 July 2011. 11. F. Xi, Q. Sun and G. Krishnappa, “Bearing Diagnostics Based on Pattern Recognition of Statistical Parameters. Journal of Vibration and Control, Vol. 6, No. 3, 2000, pp. 375-392. doi:10.1177/107754630000600303 12. J. Biteus, “Distributed Diagnosisand Simulation Based Residualgenerators,” Ph.D. Thesis, Vehicular SystemsDepartment of Electrical Engineering, Linkopings Universitet, Linkoping, 2005.
CHAPTER
4
A Neuro-Fuzzy Model for QoS Based Selection of Web Service
Abdallah Missaoui1 and Kamel Barkaoui2 1
LSTS-ENIT, Tunis, Tunisia;
2
CEDRIC-CNAM, Paris, France.
ABSTRACT The automatic selection and composition of Web services rely strongly on the manner to deal with ambiguity inherent to the description of functionalities of these services and the client’s requests. Quality of Service (QoS) criteria become crucial in Web services selection and the problem of checking that a web service satisfies a given level of QOS is considered in recent research works. This paper presents a QoS based automatic classification method of Citation: A. Missaoui and K. Barkaoui, “A Neuro-Fuzzy Model for QoS Based Selection of Web Service,” Journal of Software Engineering and Applications, Vol. 3 No. 6, 2010, pp. 588-592. doi: 10.4236/jsea.2010.36068. Copyright: © 2010 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY). http://creativecommons.org/licenses/by/4.0
50
Soft Computing with NeuroFuzzy Systems
web services. These services give generally similar functionalities and are offered by different providers. The main feature of our Web service selection model is to take advantage of the neuro-fuzzy logic for coping with the imprecision of QoS constraints values. Keywords:-Web Service, Selection, Neuro-Fuzzy, QoS, Constraint
INTRODUCTION Web services are modular, self-contained, self-describing software components which are distributed over the Web. They can be readily located and checked-out online and dynamically, using a new directory and corresponding search mechanism known as Universal Description, Discovery, and Integration (UDDI). The requester accesses the description using a UDDI or other types of registry, and requests the execution of the provider’s service by sending a SOAP message to it (see Figure 1). SOAP and HTTP provide exactly what they were designed for a simple, lightweight mechanism for interoperability and distributed communication. However, SOAP and HTTP do not provide the traditional enterprise qualities of service typically needed for an enterprise.
Figure 1. Basic web services architecture.
Furthermore, SOAP was designed to be extensible, and it can be extended to support any desired QoS feature by adding SOAP headers to the SOAP messages and adding QoS features to the basic SOAP run-time facilities. In recent years, several service providers offer QoS features to their customers. Then, multiple providers may provide similar functionalities with different values of non-functional properties.
A Neuro-Fuzzy Model for QoS Based Selection of Web Service
51
Their non-functional properties need to be considered during service selection. There are characterised as quality of service (QoS). In many practice cases of business applications, it is recommended to be taken into account during the provider selection. The human faculty of cognition and perception is very complex, but it possesses an efficient mechanism for information processing and expression [1,2]. This paper applies the neuro-fuzzy decision making approach in the process of selection and choice of the most appropriate web service with respect to quality of service criteria. This paper is organized as follows: Section 2 presents web service QoS generic description. In Section 3, we discuss and evaluate related works on web service selection adopting a common fuzzy logic approach. Section 4, we enlighten our QoS requirement description model exploiting neurofuzzy logic in order to deal with the imprecision of QoS constraints values. Comments and recommendations for our model are explicitly presented in Section 5. Finally, Section 6 draws a conclusion.
QOS PROPERTIES OF WEB SERVICE Many services are appearing on the Web, several requesters are presented to a group of service providers offering similar services. Different service providers may have different qualities of service. QoS is one of the most important factors for user’s choice of Web service. This will require sophisticated patterns of selection process. It is necessary to provide an appropriate negotiation mechanism between clients and service providers to reach mutually-agreed QoS goals. QoS management in Web service architecture includes the definition of QoS attributes and the specifications of the following processes: QoS publication, discovery, validation, and monitoring. Many works have studied QoS management on web service. Several QoS languages and architectures are proposed. The proposed approaches for QoS management can be classified into two groups: one based on extending web service technologies including SOAP, WSDL and UDDI to support QoS [3-5]. The second group use independent entities to perform QoS management [6]. Quality of service is defined by the ability to provide different priorities to different applications, users, or data flow, or to guarantee a certain level of performance to a data flow. A QoS property may include several
Soft Computing with NeuroFuzzy Systems
52
sub-properties representing different evaluation criteria, e.g. availability, performance, accessibility. In addition, a QoS property can be evaluated by many metrics and therefore it is necessary to define the units of measurements. QoS in web service architecture is a combination of several qualities or properties of a service, such as: • • •
• •
•
•
Response time: the interval between a usercommand and the reception of an action, a result or a feedback from the service. Availability: availability is the percentage of time that a service is available for use; Accessibility: Accessibility represents the degree that a system is normatively operated to counteract request messages without delay. Throughput: It means the max number of services that a platform providing Web services can process for a unit time. Reliability: Reliability is the quality aspect of a Web service that represents the degree of being capable of maintaining the service and service quality. The number of failures per month or year represents a measure of reliability of a Web service. Price: represents the money that the customer should pay for this service. It is always associated with the value of the service’s functionality, i.e. the more a service costs, the more complicated functions it provides. Security Level: represents the security level of a service. It includes the existence and type of authentication mechanisms the service offers, confidentiality and data integrity of messages exchanged, non-repudiation of requests or messages, and resilience to denialof-service attacks [7].
RELATED WORK With the strong popularity of the development of service oriented application, quality of service becomes a central interest of more and more researchers and enterprises. QoS values are proportional to the reliability degree and performance of service and thus play a very important role in the provider choice. A large number of services are exposed constraint information’s for comparison providers. Many researches [5,8-10] have studied QoS issues to improve two processes of discovery and selection of services. Several QoS-aware web
A Neuro-Fuzzy Model for QoS Based Selection of Web Service
53
service selection mechanisms have been developed in recent years in order to perform the web service composition and to improve performance applications based on services. This mechanisms’ main objective is how to how properly select a set of providers that most satisfy the constraints defined by the user in his business processes. Menascé studies the problem of finding services that minimize the total execution time. It presents an optimized heuristic algorithm that finds the optimal solution without exploring the entire solution space. The solution provided in [11] covers only the important case of execution constraints but not all QoS properties. Pfeffer proposed a fuzzy logic based model for representing any kind of non-functional service properties. This representation of user preferences enables the fast evaluation of the requested service composition by a fuzzy multiplication of the service composition properties. Thus service composition’ properties are measured during or after execution [12]. Other works have been done in fuzzy logic based web service selection. In [12-17], various methods have been proposed for specifying fuzzy QoS constraints and for ranking Web services based on their fuzzy representation. There is a more suitable technique to quantify functional properties: Linear Programming. These properties are not fitting well for measuring the nonfunctional attributes, because the majority of them are not easy to be quantified in numerical form. In the meantime, user’s QoS constraints regularly remain imprecise or ambiguous due to various human mental states, and it is very difficult to distinguish the priority order among QoS criteria. Furthermore, in web services selection, the applied QoS constraints are not explicitly defined. It is necessary to relax the constraints to make an optimal solution. The use of fuzzy logic offers improvements in the overall satisfaction level. The QoS information’s represented at abstract level such that it could efficiently select the best services However they are still initial efforts which need further investigation for more complete solutions. In the following, we specify several open issues that can be solved: • •
When we use some kinds of fuzzy numbers like triangular fuzzy they may not be easy to be defined by end users. It is very important to correctly define the QoS properties that we use in the selection process. These criteria’s QoS have important effects on ranking methods.
Soft Computing with NeuroFuzzy Systems
54
• •
How to improve fuzzy based web service discovery and the representation of QoS to achieve effective web service selection? How to automatically set the weights of service providers attributes?
REFINEMENT OF THE FRAMEWORK Neuro-fuzzy technique is the combination of two artificial intelligence (AI) methods: fuzzy logic techniques and neural networks. Neuro-fuzzy system has the ability to handle the nonlinear and complex systems. It is constructed based on the learning algorithm of neural networks technique to adjust the appropriate parameters for fuzzy logic system [18]. In this paper, we aim to solve the selection of web services in a global and flexible manner by introducing a neuro-fuzzy way. For this purpose, we have developed a neural-fuzzy system based on the Sugeno Approach [19]. This is known as the ANFIS (i.e., Adaptive Neuro-Fuzzy Inference Systems). We assume that semantic matchmaking has taken place to identify functionally equivalent services. When several of them are available to perform the same task, their quality aspects become important and useful in the selection process. An ANFIS is a multi-layered feed forward network, in which each layer performs a particular task. The layers are characterized by the fuzzy operations they perform. Figure 2 describes the global architecture of this neural-fuzzy system. It shows a n-input, type-5 ANFIS. Three membership functions are associated with each input. We assume that the fuzzy inference system under consideration has n inputs Q1, Q2,…,Qn (which are one service attributes). Each input has five linguistic terms, for example, the input Q1 possesses the terms {M11, M12,…, M15}. For each input Qi, we have defined linguistic expressions Li = {Very Poor(vp), Poor(p), Medium(m), Good(g), Very Good(vg)}
The common fuzzy if-then rule has the following type: Rule 1: If (Q1 is M11) and (Q2 is M21) and … and (Qn is Mn1) then f1 (Q1, Q2,…, Qn)
We denote the output of the i th node in layer k as Ok,i. Figure 2 shows the schematic diagram of the ANFIS structure, which consists of five layers. one
Layer 1: Every node i in this layer transform the crisp values to a fuzzy
A Neuro-Fuzzy Model for QoS Based Selection of Web Service
55
where QK is the input to node K and and i ∈{1,...,5} ) is a linguistic label (very poor, poor, fair, good, very good) associated with this node. In other words, O1,i is the membership grade of a fuzzy set and it specifies the degree which the given input QK ( k ∈{1,...,n} ) satisfies the quantifier M.
We use the following generalized Bell function as the membership function (MF)
where ai, bi and ci are the parameters set of MF. The bell-shaped function varies depending on the values of these parameters. Where the parameters a and b vary the width of the curve and the parameter c locates the center of the curve. The parameter b should be positive. The parameters in this layer are referred to as premise parameters. The generalized Bell-shaped function is shown in Figure 3.
Figure 2. The structure of the neural fuzzy selector.
Layer 2: Every node in this layer is a fixed node labeled ∏. The weighting factor, wk, of each rule is calculated by evaluating the membership expressions in the antecedent of the rule. This is accomplished by first converting the input values to fuzzy membership values by utilizing the input membership functions and then applying the and operator to these membership values.
56
Soft Computing with NeuroFuzzy Systems
The and operator corresponds to the multiplication of input membership values. Each node output represents the firing strength of a rule. Layer 3: Every node in this layer is a fixed node labeled N. The function of the fixed node is used to normalize the input firing strengths.
Layer 4: Every node in Layer 4 is a parameterized function, and the adaptive parameters are called “consequent parameters”. The node function is given by:
Layer 5: The single node in this layer is a fixed node labeled ∑, which computes the overall output as the summation of all inputs:
Thus, the ANFIS network is constructed according to the TSK fuzzy model. This ANFIS architecture can then update its parameters according to the backpropagation algorithm [20]. This algorithm minimizes the error between the output of the ANFIS and the desired output.
Figure 3. Generalized bell-shaped (a = 2, b = 4, c = 6).
Our neuro-fuzzy system allows classifying service providers in several categories: very poor, poor, fair, good, very good. It allows automating the selection process in the dynamic composition of services. According to the
A Neuro-Fuzzy Model for QoS Based Selection of Web Service
57
QoS requirements of web service providers and the functions of Neuro-fuzzy system, we believe that each service invoked is appropriate candidate to increase the composition ability of web services and to decrease the burden of composition cognition and the minimal development cost.
COMMENTS AND RECOMMENDATIONS In fuzzy inference system (FIS), The MF of the consequent of each rule is a constant of a fuzzy MF. There are two steps to construct this system: the specification of an appropriate number of input/output and the specification of the shape of MFs. The main problem is that structure identification requires human expertise to solve the parameter estimation. In our selector system we used a different approach, which take advantage of adaptive neural networks algorithms during fitting procedures. MF parameters are fitted to a dataset through a learning algorithm.A significant number of samples of service providers are needed in order to have better result and to avoid having too many defect values during selection process. The database must be as complete as possible, including samples of providers attributes over a broad range. The number of samples depends on the context and on the runtime environment.On the other hand, fuzzy logic sets are based on transparence, linguistic rules and establish a framework to include human expertise into modelling. The number of rules is decided by an expert who is familiar with the system to be modeled. In our work, however, no expert is available and the number of membership functions assigned to each input qualities is chosen empirically by examining the desired input-output data. We merged the fuzzy logic approach with the ability of learning algorithms from neural networks to adjust the model.
CONCLUSIONS Web service composition is an emerging area involving important technological challenges. Some of the main challenges are to correctly describe QoS of Web services, to compose them adequately and automatically, and to discover suitable providers and QoS composition issues. Neuro-fuzzy logic can be seen as a promising formal technique for representing imprecise QoS constraints. In this paper, we have presented a solution to use neurofuzzy approach in Web service discovery and selection. We have proposed methods for ranking and selecting web services based on a neuro-fuzzy specification of fuzzy QoS constraints. The user’s constraints are formalized as fuzzy sets and the Qos criteria’s are expressed as fuzzy expressions. This model can be seen as a contribution towards a more complete solution for web service composition integrating fully QoS features.
58
Soft Computing with NeuroFuzzy Systems
REFERENCES 1.
J. S. Jang, “ANFIS: Adaptive-Network-Based Fuzzy Inference System,” IEEE Transactions on Systems, Man, and Cybernetics, Vol. 23, No. 3, 1993, pp. 665-684. 2. J. R. Jang and C. T. Sun, “Neuro-Fuzzy and Soft Computing: A Computational Approach to Learning and Machine Intelligence,” Prentice-Hall, Inc., Upper Saddle River, New Jersy, 1997. 3. V. Diamadopoulou, C. Makris, Y. Panagis and E. Sakkopoulos, “Techniques to Support Web Service Selection and Consumption with QoS Characteristics,” Journal of Network and Computer Applications, Vol. 31, No. 2, 2008, pp. 108-130. 4. A. F. M. Huang, C. W. Lan and S. J. H. Yang, “An Optimal QoS-Based Web Service Selection Scheme,” Information Sciences, Vol. 179, No. 19, 2009, pp. 3309-3322. 5. L. Zeng, B. Benatallah, A. H. H. Ngu, M. Dumas, J. Kalagnanam and H. Chang, “QoS-Aware Middleware for Web Services Composition,” IEEE Transactions on Software Engineering, 2004, pp. 311-327. 6. D. A. Menascé, H. Ruan and H. Gomaa, “QoS Management in ServiceOriented Architectures,” Journal of Performance Evaluation, Vol. 64, No. 7-8, 2007, pp. 646-663. 7. D. A. Menasce, “QoS Issues in Web Services,” IEEE Internet Computing, Vol. 6, No. 6, 2002, pp. 72-75. 8. M. Sultana, M. M. Akbar and M. Rouf, “Network Flow Heuristic Algorithm for a Distributed Web Service Selection Problem,” IEEE Conference on Communications, Computers and Signal Processing, 2009, pp. 465-470. 9. D. Tsesmetzis, I. Roussaki and E. Sykas, “QoS-Aware Service Evaluation and Selection,” European Journal of Operational Research, Vol. 191, No. 3, 2008, pp. 1101- 1112. 10. S. Chaari, Y. Badr and F. Biennier, “Enhancing Web Service Selection by QOS-Based Ontology and WS-Policy,” Proceeding of the 23rd ACM Symposium on Applied Computing, Ceará, 2008, pp. 2426-2431. 11. D. A. Menascé, E. Casalicchio and V. Dubey, “On Optimal Service Selection in Service Oriented Architectures,” Performance Evaluation Journal, Vol. 67, No. 8, 2010, pp. 659-675. 12. H. Pfeffer, S. Krüssel and S. Steglich, “A Fuzzy Logic based Model for Representing and Evaluating Service Composition Properties,”
A Neuro-Fuzzy Model for QoS Based Selection of Web Service
13.
14.
15.
16.
17.
18.
19.
20.
59
The Third International Conference on Systems and Networks Communications, Bangalore, 2009. M. Lin, J. Xie, H. Guo and H. Wang, “Solving Qos-Driven Web Service Dynamic Composition as Fuzzy Constraint Satisfaction,” IEEE International Conference on e-Technology, e-Commerce and e-Service, Hong Kong, 2005. P. Wang, K. Chao, C. Lo, C. Huang and Y. Li, “A Fuzzy Model for Selection of QoS-Aware Web Services,” IEEE International Conference on e-Business Engineering, IEEE Computer Society, Shanghai, 2006, pp. 585-593. K. M. Chao, M. Younas, C. C. Lo and T. H. Tan, “Fuzzy Atchmaking for Web Services,” The 19th International Conference on Advanced Information Networking and Applications, Taipei, 2005. L. Zhuang, Y. F. Huang, W. G. Jian, J. B. Zhou and H. Q. Guo, “Solving Fuzzy QoS Constraint Satisfaction Technique for Web Service Selection,” International Conference on Computational Intelligence and Security Workshops, Harbin, 2007. H. Tong and S. Zhang, “A Fuzzy Multi-Attribute Decision Making Algorithm for Web Services Selection Based on QoS,” The IEEE AsiaPacific Conference on Services Computing, Guangzhou, 2006. M. A. Denai, F. Palis and A. Zeghbib, “ANFIS Based Modelling and Control of Non-Linear Systems: A Tutorial,” IEEE International Conference on Systems, Man and Cybernetics, Vol. 4, 2004, pp. 34333438. O. Nelles, A. Fink, R. Babuka and M. Setnes, “Comparison of Two Construction Algorithms for TakagiSugeno Fuzzy Models,” International Journal of Applied Mathematics and Computer Science, 2000, pp. 835-855. P. Werbos, “The Toots of the Back Propagation: From Ordered Derivatives to Neural Networks and Political Forecasting,” John Wiley and Sons, Inc, New York, 1993.
SECTION 2
ADAPTIVE NEURO-FUZZY SYSTEMS
CHAPTER
5
Adaptive Neuro-Fuzzy Logic System for Heavy Metal Sorption in Aquatic Environments
Ahmad Qasaimeh1, Mohammad Abdallah2, and Falah Bani Hani2 1
Department of Civil Engineering, Jerash University, Jerash, Jordan
Chemical Engineering Department, AlHuson University College, Al-Balqa Applied University, Salt, Jordan 2
ABSTRACT In this paper, adaptive neuro-fuzzy inference system ANFIS is used to assess conditions required for aquatic systems to serve as a sink for metal removal; it is used to generate information on the behavior of heavy metals (mercury) in water in relation to its uptake by bio-species (e.g. bacteria, fungi, algae, etc.)
Citation: A. Qasaimeh, M. Abdallah and F. Bani Hani, “Adaptive Neuro-Fuzzy Logic System for Heavy Metal Sorption in Aquatic Environments,” Journal of Water Resource and Protection, Vol. 4 No. 5, 2012, pp. 277-284. doi: 10.4236/jwarp.2012.45030. Copyright: © 2012 by authors and Scientific Research Publishing Inc. This
work is licensed under the Creative Commons Attribution International License (CC BY). http://creativecommons.org/licenses/by/4.0
64
Soft Computing with NeuroFuzzy Systems
and adsorption to sediments. The approach of this research entails training fuzzy inference system by neural networks. The process is useful when there is interrelation between variables and no enough experience about mercury behavior, furthermore it is easy and fast process. Experimental work on mercury removal in wetlands for specific environmental conditions was previously conducted in bench scale at Concordia University laboratories. Fuzzy inference system FIS is constructed comprising knowledge base (i.e. premises and conclusions), fuzzy sets, and fuzzy rules. Knowledge base and rules are adapted and trained by neural networks, and then tested. ANFIS simulates and predicts mercury speciation for biological uptake and mercury adsorption to sediments. Modeling of mercury bioavailability for biospecies and adsorption to sediments shows strong correlation of more than 98% between simulation results and experimental data. The fuzzy models obtained are used to simulate and forecast further information on mercury partitioning to species and sediments. The findings of this research give information about metal removal by aquatic systems and their efficiency. Keywords:-Adaptive Neuro-Fuzzy; Simulation, Heavy Metal; Sorption; Aquatic Systems; Forecast
INTRODUCTION The release of heavy metals from industries into the environment has resulted in many problems for both human health and aquatic ecosystems [1,2]. Heavy metals released into the environment by technological activities tend to persist indefinitely, circulating and eventually accumulating throughout the food chain, becoming a serious threat to the environment [3]. The presence of heavy metals in the environment is of major concern because of their toxicity, bio-accumulating tendency, threat to human life and the environment [4,5]. Lead, cadmium and mercury are examples of heavy metals that have been classified as priority pollutants by the U.S Environmental protection Agency (US EPA) [6]. Various biomaterials have been examined for their biosorptive properties and different types of biomass have shown levels of metal uptake [7]. Tables 1 and 2 show examples of biomass ability to sorb heavy metals. In recent years, applying biotechnology in controlling and removing metal pollution has been paid much attention, and gradually becomes hot topic in the field of metal pollution control because of its potential application. Alternative process is biosorption, which utilizes various certain natural materials of biological origin, including bacteria, fungi, yeast, algae, plant,
Adaptive Neuro-Fuzzy Logic System for Heavy Metal Sorption in ...
65
etc. These biosorbents possess metal-buffering property and can be used to decrease the concentration of heavy metal ions in solution [8]. A large quantity of materials has been investigated as biosorbents for the removal of metals extensively. The tested biosorbents can be basically classified into the following categories: bacteria (e.g. Bacillus subtillis), fungi (e.g. Rhizopus arrhizus) (Table 3), yeast (e.g., Saccharomyces cerevisiae), algae, industrial wastes (e.g., S. cerevisiae waste biomass from fermentation and food industry), water plants (e.g. Water Hyacinths and Reeds), agricultural wastes (e.g. corn core), and other polysaccharide materials [9]. The importance of metallic ions to fungal and yeast metabolism has been known for a long time [10]. The yeast biomass has been successfully used as biosorbent for removal of Ag, Au, Cd, Co, Cr, Cu, Ni, Pb, U, Th and Zn from aqueous solution. Yeasts of genera Saccharomyces, Candida, Pichia are efficient biosorbents for heavy metal ions [11]. Algae are of special interest in search for and the development of new biosorbents materials due to their high sorption capacity and their ready availability in practically unlimited quantities in the aquatic systems as seas and oceans (Table 4) [24,25]. Table 1. Biomass and their biosorbent capacity [12].
Table 2. Metal biosorption capacity by different biosorbents.
66
Soft Computing with NeuroFuzzy Systems
Table 3. The value of sorption capacity (mmol/g) for different sorbents [23].
Aquatic systems are considered as natural ecosystems that are designed to take advantage of the natural processes to provide efficient and low-cost wastewater treatment. The removal of metals from the water column within an aquatic system is performed generally by biological species uptake and by adsorption to sediments. Therefore, the aquatic systems such as streams, rivers, reservoirs and lakes serve as sink for heavy metal in aqueous solutions. The metal ion bioavailability for sorption to the biotic surface is pH dependent, as well for metal ion adsorption to sediments. The binding of a metal ion to the biotic surface of an organism decreases with increasing pH, whereas the binding behavior of metal ion to sediments increases with increasing pH.
METHODOLOGY The research methodology implies Neuro-Fuzzy system to model and assess mercury removal from aquatic natural systems. Investigational data and information for mercury bioavailability in water and adsorption to sediments were adopted from previous research work-literature review of Prof. Elektorowicz research team attained in Concordia University-Canada [26-32]. Table 4. Algae sorption capacity (mmol/g) for metal ions in aqueous solution.
Neuro-Fuzzy system simulates mercury sorption and evaluates the efficiency of removal by verifying the effects of pH and mercury concentration in water. The computational tools used in this research are those in MATLAB; fuzzy toolbox and simulink.
Adaptive Neuro-Fuzzy Logic System for Heavy Metal Sorption in ...
67
ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM (ANFIS) In natural systems where variables are interrelated and data is large, it is difficult to determine the membership functions for input variables. Neuroadaptive learning technique works similarly to that of neural networks. It provides a method for the fuzzy modeling procedure to learn information about a data set. Fuzzy Logic computes the membership function parameters that best allow the associated fuzzy inference system to track the given input/output data. The Fuzzy Logic accomplishes this membership function parameter adjustment is called adaptive neuro-fuzzy inference system. Using a given input/ output data set, the fuzzy inference system uses either a backpropagation algorithm alone or in combination with a least squares type of method. This adjustment allows fuzzy systems to learn from the data they are modeling. Then testing the data to check the generalization capability of the resulting fuzzy inference system is needed. Checking the data is set for model validation. Model validation is the process by which the non-trained input variables are presented to the trained fuzzy inference system model to see how well the model predicts the corresponding output data.
Adaptive Neuro-Fuzzy Inference System for Mercury Speciation The fuzzy inference system consists of two components: the linguistic term base (database) and the rule base. The database is fuzzified in two parts: fuzzy premises (input) and fuzzy conclusions (output). The fuzzy production rule base infers input to output and then defuzzified.
Soft Computing with NeuroFuzzy Systems
68
Figure 1. Fuzzy inference system.
Figure 1 shows a scheme for fuzzy inference system. Adaptive NeuroFuzzy Inference System (ANFIS) is applied to estimate mercury removal in natural waters. ANFIS simulates and predicts mercury bioavailability that will be bio-sorbed by biological species. ANFIS model entails the following input variables to estimate output variable (bioavailable mercury concentration): •
Initial concentration of total mercury is in the range 0.3 × 10–6 - 1 × 10–3 moles/l;
• The pH value is situated in the range 5.36 - 8. ANFIS model is constructed into two inputs (Hgi and pH), one output (Bioavailable Hg), and nine rules. ANFIS model, training the data, and training error are illustrated in Figure 2. ANFIS model fits the experimental data for bioavailable mercury concentration. Subsequently, comparison is conducted between results obtained from the model using ANFIS and experimental results for different initial total mercury concentrations in water and pH. The comparison shows strong correlation (Figures 3 and 4).
Adaptive Neuro-Fuzzy Logic System for Heavy Metal Sorption in ...
69
Figure 2. ANFIS model and training for mercury bioavailability in water by varying initial Hg concentration and pH.
Figure 3. Simulation of bioavailable mercury concentration to be uptaken by bio-species verses pH for the range of initial mercury concentration.
Figure 4. Simulation of bioavailable mercury concentration to be uptaken by bio-species versus initial total mercury concentrations and pH of 5.36.
Adaptive Neuro-Fuzzy System for Mercury Adsorption In the second stage of work, ANFIS provides solution for soil adsorption of mercury for different conditions of initial mercury concentration and pH value. Neuro-Fuzzy system depends on fuzzy knowledge bases that satisfy the following parameters:
Soft Computing with NeuroFuzzy Systems
70 •
The initial concentration of total mercury is in the range between 1 × 10–7 to 1 × 10–3 mole/l;
• The pH value is located between 5.36 to 8; • The adsorbent concentration is 10 g/l. ANFIS model is consisted of two inputs (Hgi and pH), one output (adsorbed Hg), and nine rules. The model and training the data are shown in Figure 5. ANFIS model shows strong correlation between simulation and experimental data of mercury adsorption within different initial mercury concentrations and pH as shown in Figures 6 and 7. For certain initial mercury concentrations, the adsorbed mercury is decreasing from its upper value when pH equals 8 to its lower value and when pH equals 5.36 as shown in Figure 6. The benefit of neural training of the fuzzy inference system is vital especially when there is large data and no experience of the system behavior.
SIMULATION AND FORECASTING In previous sections the ANFIS model is constructed, trained, and checked. Now the model is ready for further range of simulation and forecasting. More information could be predicted for mercury removal by sorption in aquatic natural systems. In this section a simulink diagram is used for forecasting. Figure 8 shows an example of using fuzzy logic systems for mercury bioavailability and adsorption that were produced in previous sections to expand more information about Hg removal.
Figure 5. (a) ANFIS model and (b) training data: for mercury adsorption to sediments by varying initial Hg concentration and pH.
Adaptive Neuro-Fuzzy Logic System for Heavy Metal Sorption in ...
71
The run of simulink model describes the total mercury removal performance by components of an aquatic natural system; this performance can be depicted within different pH at certain initial mercury concentration. Figure 9 shows the performance of an aquatic system where it is optimal at pH equals 6.5. The fitting equation in the figure provides forecasting for total removal of mercury by natural waters components (bio-species and sediments) at any value of pH. The analysis in this section supply more information for model performance and forecasting. It also gives information about removal efficiency of the overall system.
Figure 6. Comparison between ANFIS simulation and experimental data for adsorbed Hg within different initial Hg concentration, when pH is varying as 8, 6.5, and 5.36 respectively at each concentration.
Figure 7. Comparison between ANFIS simulation and experimental data for Adsorbed Hg concentration within variable pH in Solution.
Figure 8. Simulink diagram forecast for total mercury removal for different Hgi and pH variables.
72
Soft Computing with NeuroFuzzy Systems
Figure 9. Total mercury removed by natural water within different pH, and the forecast equation.
CONCLUSION Fuzzy logic proved to be useful for assessing ambiguous natural processes. Modeling of mercury bioavailability for bio-species and adsorption to sediments shows strong correlation of more than 98% between simulation results and experimental data. Using adaptive neuro-fuzzy system is important for hazy system and no experience about data behavior. The findings of this research provide information, simulation, and forecasting about heavy metal removal efficiency in natural systems.
ACKNOWLEDGEMENTS The financial support from the Natural Sciences and Engineering Research Council of Canada under grant RGPIN-18948 is gratefully acknowledged.
Adaptive Neuro-Fuzzy Logic System for Heavy Metal Sorption in ...
73
REFERENCES 1.
D. Inthorn, H. Nagase, Y. Isaji, K. Hirata and K. Miyamoto, “Removal of Cadmium from Aqueous Solution by the Filamentous Cyanobacterium Tolypothrix tenuis,” Journal of Fermentation and Bioengineering, Vol. 82, No. 6, 1996, pp. 580-584. doi:10.1016/S0922-338X(97)81256-1 2. L. C. Rai, J. P. Gaur and H. D. Kumar, “Phycology and Heavymetal Pollution,” Biological Reviews, Vol. 56, No. 2, 1981, pp. 99-103. doi:10.1111/j.1469-185X.1981.tb00345.x 3. V. Kuppusamy, J. R. Jegan, K. Palanivelu and M. Velan, “Copper Removal from Aqueous Solution by Marine Green Alga Ulva Reticulate,” Electronic Journal of Biotechnology, Vol. 7, No. 1, 2004, pp. 61-71. 4. M. Horsfall Jr. and A. I. Spiff, “Effects of Temperature on the Sorption of Pb2+ and Cd2+ from Aqueous Solution by Caladium Bicolour (wild Cocoyam) Biomass,” Electronic Journal of Biotechnology, Vol. 8, No. 2, 2005. 5. J. C. Igwe and A. A. Abia, “Maize Cob and Husk as Adsorbents for Removal of Cd, Pb and Zn Ions from Wastewater,” The Physical Science, Vol. 2, 2003, pp. 83- 94. 6. L. H. Keith and W. A. Telliard, “Priority Pollutants,” Environmental Science & Technology, Vol. 13, No. 4, 1979, pp. 416-424. doi:10.1021/ es60152a601 7. B. Volesky and Z. R. Holan, “Biosorption of Heavy Metals,” Biotechnology Progress, Vol. 11, No. 3, 1995, pp. 235-250. doi:10.1021/ bp00033a001 8. J. L. Wang and C. Chen, “Biosorption of Heavy Metals by Saccharomyces Cerevisiae: A Review,” Biotechnology Advances, Vol. 24, No. 5, 2006, pp. 427-451. doi:10.1016/j.biotechadv.2006.03.001 9. K. Vijayaraghavan and Y. S. Yun, “Bacterial Biosorbents and Biosorption,” Biotechnology Advances, Vol. 26, No. 2, 2008, pp. 266291. doi:10.1016/j.biotechadv.2008.02.002 10. G. M. Gadd, “Interactions of Fungi with Toxic Metals,” New Phytologist, Vol. 124, No. 1, 1993, pp. 25-60. doi:10.1111/j.1469-8137.1993. tb03796.x 11. V. S. Podgorskii, T. P. Kasatkina and O. G. Lozovaia, “Yeasts— Biosorbents of Heavy Metals,” Mikrobiol Z, Vol. 66, 2004, pp. 91-103. 12. D. Kratochvil and B. Volesky, “Advances in the Biosorption of Heavy
74
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
Soft Computing with NeuroFuzzy Systems
Metals,” Trends in Biotechnology, Vol. 16, No. 7, 1998, pp. 291-300. doi:10.1016/S0167-7799(98)01218-9 S. Tunali, A. Cabuk and T. Akar, “Removal of Lead and Copper Ions from Aqueous Solutions by Bacterial Strain Isolated from Soil,” Chemical Engineering Journal, Vol. 115, No. 3, 2006, pp. 203-211. doi:10.1016/j.cej.2005.09.023 M. X. Loukidou, T. D. Karapantsios, A. I. Zouboulis and K. A. Matis, “Diffusion Kinetic Study of Cadmiurn (II) Biosorption by Aeromonas caviae,” Journal of Chemical Technology and Biotechnology, Vol. 79, No. 7, 2004, pp. 711-719. doi:10.1002/jctb.1043 J. A. Davis, B. Volesky and R. H. S. F. Vierra, “Sargassum Seaweed as Biosorbent for Heavy Metals,” Water Research, Vol. 34, No. 17, 2000, pp. 4270-4278. doi:10.1016/S0043-1354(00)00177-9 A. Selatnia, A. Boukazoula, N. Kechid, M. Z. Bakhti and A. Chergui, “Biosorption of Fe3+ from Aqueous Solution by a Bacterial Dead Streptomyces Rimosus Biomass,” Process Biochemistry, Vol. 39, No. 11, 2004, pp. 1643- 1651. doi:10.1016/S0032-9592(03)00305-4 T. Srinath, T. Verma, P. W. Ramteke and S. K. Garg, “Chromium (VI) Biosorption and Bioaccumulation by Chromate Resistant Bacteria,” Chemosphere, Vol. 48, No. 4, 2002, pp. 427-435. doi:10.1016/S00456535(02)00089-9 K. Vijayaraghavan, J. R. Jegan, K. Palanivelu and M. Velan, “Copper Removal from Aqueous Solution by Marine Green Alga Ulva Reticulate,” Electronic Journal of Biotecnology, Vol. 7, No 1, 2004. doi:10.2225/vol7-issue1-fulltext-4 A. Ozturk, “Removal of Nickel from Aqueous Solution by the Bacterium Bacillus Thuringiensis,” Journal of Hazardous Materials, Vol. 147, No. 1-2, 2007, pp. 518- 523. doi:10.1016/j.jhazmat.2007.01.047 T. R. Muraleadharan, L. Iyengar and C. Venkobachar, “Screening of Tropical Wood-Rotting Mushrooms for Copper Biosoption,” Applied and Environmental Microbiology, Vol. 61, No. 9, 1995, pp. 3507-3508. A. Nakajima and T. Tsuruta, “Competitive biosorption of thorium and uranium by Micrococcus luteus,” Journal of Radioanalytical and Nuclear Chemistry, Vol. 260, No. 1, 2004, pp. 13-18. doi:10.1023/ B:JRNC.0000027055.16768.1e S. Gang and S. Weixing, “Sunflower Stalks as Adsorbents for the Removal of Metal Ions from Wastewater,” Industrial & Engineering
Adaptive Neuro-Fuzzy Logic System for Heavy Metal Sorption in ...
23.
24.
25.
26.
27.
28.
29.
30. 31. 32.
75
Chemistry Research, Vol. 37, No. 4, 1998, pp. 1324-1328. doi:10.1021/ ie970468j J. Rincon, F. Gonzalez, A. Ballester, M. L. Blazquez and J. A. Munoz, “Biosorption of Heavy Metals by Chemically-Activated Alga Fucus Vesiculosus,” Journal of Chemical Technology and Biotechnology, Vol. 80, No. 12, 2005, pp. 1403-1407. doi:10.1002/jctb.1342 N. Kuyicak and B. Volesky, “Biosorption by Fungal Biomass,” In: B. Volesky, Ed., Biosorption of Heavy Metals, CRC Press, Florida, 1990, pp. 173-198. E. Romera, F. Gonzalez, A. Ballester, M. L. Blazquez and J. A. Munoz, “Biosorption with Algae: A Statistical Review,” Critical Reviews in Biotechnology, Vol. 26, No. 4, 2006, pp. 223-235. doi:10.1080/07388550600972153 J. D. Allison, D. S. Brown and K. J. Novo-Gradac, “MINTEQA2/ PRODEFA2, Geochemical Assessments Model for Environmental Systems: Version 3.0 User’s Manual,” Computer Sciences Corporation, Environmental Research Laboratory, Athens, GA, 1991. A. El-Agroudy and M. Elektorowicz, “Kinetics of Inorganic Mercury Removal from Surface Water by Water Hyacinths and Reeds,” 34 CCSWPR, Burlington, 1999. A. El-Agroudy, “Investigation of Constructed Wetland Capability to Remove Mercury from Contaminated Waters,” Ph.D. Thesis, Concordia University, 1999. E. Maria, B. Marek and Q. Ahmad, “Application of the AI to Estimate the Constructed Wetland Response to Heavy Metal Removal,” ASCE/ CSCE Conference on Environmental Engineering, Niagara Falls, July 2002. R. J. Hunter, “Introduction to Modern Colloid Science,” Oxford University Press Inc., New York, 1993. K. H. Tan, “Principles of Soil Chemistry,” Marcel Dekker Inc., New York, 1982. R. N. Yong, A. M. O. Mohamed and B. P. Warketin, “Principles of Contaminant Transport in Soils,” Development in Geotechnical Engineering, Elsevier, 1992.
CHAPTER
6
Automatic Heart Disease Diagnosis System Based on Artificial Neural Network (ANN) and Adaptive Neuro-Fuzzy Inference Systems (ANFIS) Approaches
Mohammad A. M. Abushariah, Assal A. M. Alqudah, Omar Y. Adwan, Rana M. M. Yousef Computer Information Systems Department, King Abdullah II School for Information Technology, The University of Jordan, Amman, Jordan
ABSTRACT This paper aims to design and implement an automatic heart disease diagnosis system using MATLAB. The Cleveland data set for heart diseases was used as the main database for training and testing the developed system. In order to train and test the Cleveland data set, two systems were developed. The Citation: Abushariah, M., Alqudah, A., Adwan, O. and Yousef, R. (2014), “Automatic Heart Disease Diagnosis System Based on Artificial Neural Network (ANN) and Adaptive Neuro-Fuzzy Inference Systems (ANFIS) Approaches”. Journal of Software Engineering and Applications, 7, 1055-1064. doi: 10.4236/jsea.2014.712093. Copyright: © 2014 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY). http://creativecommons.org/licenses/by/4.0
78
Soft Computing with NeuroFuzzy Systems
first system is based on the Multilayer Perceptron (MLP) structure on the Artificial Neural Network (ANN), whereas the second system is based on the Adaptive Neuro-Fuzzy Inference Systems (ANFIS) approach. Each system has two main modules, namely, training and testing, where 80% and 20% of the Cleveland data set were randomly selected for training and testing purposes respectively. Each system also has an additional module known as case-based module, where the user has to input values for 13 required attributes as specified by the Cleveland data set, in order to test the status of the patient whether heart disease is present or absent from that particular patient. In addition, the effects of different values for important parameters were investigated in the ANN-based and Neuro-Fuzzy-based systems in order to select the best parameters that obtain the highest performance. Based on the experimental work, it is clear that the Neuro-Fuzzy system outperforms the ANN system using the training data set, where the accuracy for each system was 100% and 90.74%, respectively. However, using the testing data set, it is clear that the ANN system outperforms the NeuroFuzzy system, where the best accuracy for each system was 87.04% and 75.93%, respectively. Keywords:- Heart Disease, ANN, ANFIS, Multilayer Perceptron, Neuro-Fuzzy, Cleveland Data Set
INTRODUCTION Recently, heart disease has become one of the most prevalent diseases which people are being suffered from. According to statistics, it is one of the most important causes of deaths all over the world (CDC’s report). Many factors, such as clinical symptoms and the relation between the functional and the pathologic manifestations of heart diseases and other human organs rather than heart, complicate the diagnosis of it and result in delay in correct diagnosis decision. Therefore, diagnosing of the heart disease is an essential matter in health care industry and many researchers try to develop medical decision support systems (MDSS) to help physicians. These systems are developed to moderate the diagnosis time and enhance the diagnosis accuracy in addition to supporting increasingly complicated diagnosis decision process [1] [2]. Currently, hospital information systems using decision support systems have different tools available to obtain data, but they are still restricted. These tools can just answer some simple queries like “identifying the male patients who are below 20 years old, and single who have been treated for
Automatic Heart Disease Diagnosis System Based on Artificial ...
79
heart attack”. However, they are not able to answer complex queries “given patient records, predicting the probability of patients getting a heart disease” as an example [3]. According to [4], clinical decisions are often made based on doctors’ intuitions and heuristics experience rather than on the knowledge rich data hidden in the database. They lead to unwanted biases, errors and excessive medical costs which affects the quality of treatment provided to patients [5]. Motivated by the necessity of such a system, in this paper, a method is suggested to efficiently diagnose the heart disease, which results in decreasing medical errors and superfluous practice variation, decreasing diagnostic time and enhancing patient safety and satisfaction. This paper presents a decision support system for heart disease classification using neural network. The data set used is the Cleveland Heart Database taken from UCI learning data set repository which was donated by Detrano. The data set is being divided into two classes: 0 corresponding to absence of any disease and 1 corresponding to presence of disease. The rest of the paper is organized as follows. Related works are presented in Section 2. In Section 3, research algorithms and concepts are described. Automated heart disease diagnosis system’s design and implementation details are presented in Section 4. In Section 5, experimental results are presented and discussed in details. The study is finally concluded in Section 6.
RELATED WORKS Until now, various classification algorithms have been employed on heart disease data set and high classification accuracies have been reported in the last decade. Cleveland heart disease database is one of the most accurate existing databases. Robert Detrano created this database in V.A. Medical Center, Long Beach and Cleveland Clinic Foundation in 1988. Since 1988, researchers worked a lot on classification of its data by using various classification algorithms and they obtained different accuracy results. The work presented in [6] used Artificial Immune System (AIS) and resulted in 84.5% classification accuracy. The work in [7] utilized a hybrid Neural Network ANN and fuzzy neural network (FNN) and reached the classification accuracy of 86.8%. On the other hand, [8] developed SAS based software by using neural network ensemble method and obtained 89.01% accuracy in classification. Recently, a research employed an MLP Neural Network by using Back propagation algorithm which classifies the data into 5 categories
80
Soft Computing with NeuroFuzzy Systems
with 97.5% accuracy, whereas the SVM based system achieved 80.41% accuracy. In the presented study, a Multilayer Perceptron Neural Network (MLPNN) with three layers is employed and compared with Support Vector Machine (SVM). Results indicated that a MLPNN with back propagation was more successful than support vector machine for diagnosing heart disease [1]. In addition to artificial neural network, fuzzy expert systems are also used in MDSS. For Instance a fuzzy expert system was proposed to determine heart disease risk of patient in 2007 and the result of this system was 79% [2]. Recently a research designed a fuzzy expert system for heart disease diagnosis, according to the result obtained from designed system, it was correct in 94% [9]. All these previous studies show the applicability of ANN in this selected area.
RESEARCH ALGORITHMS AND CONCEPTS Important concepts, architecture theory, and algorithm for Neural Network and Neuro-Fuzzy are described in this section.
Neural Network Approach Neural Network (NN) also referred to as Artificial Neural Network (ANN) is a computational model where its functions and methods are based on the structure of the brain. Neural network follows graph topology in which neurons are nodes of the graph and weights are edges of the graph. It consists of so many layers that should be finite in order to decrease time of problem solving. In this paper, neural network is used since it has the potential for supporting medical decision support systems. In large data sets, it has costeffective and flexible non-linear modeling since the optimization is easy. In addition, it is accurate in predictive inference. Another important factor is that these models can make knowledge dissemination easier by providing explanation, for instance, using rule extraction or sensitivity analysis [4]. ANN has various models, such as Multilayer Perceptron (MLP), RFB and so forth that are different in terms of architecture and training network which will be discussed in the following sub-sections.
Neural Network Architecture In ANN, neurons can be arranged in various ways and the weights (connection between neurons) can have different patterns which is called neural network architecture. There are different types of architectures,
Automatic Heart Disease Diagnosis System Based on Artificial ...
81
such as feed-forward, feed-back, fully interconnected net, competitive net and so forth. Some of the most important architectures are introduced in [10]. Feed-forward architecture can have single or multi-layer of weights. In single layer feed-forward net, there is only one interconnected weights while in multi-layer feed-forward net, more than one interconnected layers of weights can exist. Figure 1 shows feed-forward multi-layer architecture. Fully recurrent network architecture is the simplest sort of architecture in which every neuron is connected to each other. Simple recurrent network is to somehow like fully recurrent network, except that neurons are not fully connected. Competitive network is the same as single layer feed forward architecture. In addition of all attributes related to single layer feed forward architecture, in competitive network there is connection between outputs. Among aforementioned architecture, feed-forward architecture is the most suitable one in terms of time for a large amount of data.
Network Training The process of training the network aims to achieve the expected output by changing the weights in the connections between network layers. There are three sorts of network training as follows: •
Supervised Training: In this process, a series of sample inputs are available for the network and the resulted output are compared with expected responses. • Unsupervised Training: This process is used for the time that the output of training input vectors are unknown. • Reinforcement Training: This process shows the correctness of output result. In this paper, supervised training is used since it is based on the Cleveland database, whereby all input and expected output data are available.
Figure 1. Multi-layer feed-forward architecture.
82
Soft Computing with NeuroFuzzy Systems
Multilayer Perceptron (MLP) In this research, MLP is used as one neural network model since it follows feed-forward architecture and supervised training. Perceptron network has usually a layer of input, a layer of output and one or more hidden layers in between. Figure 2 represents MLP with two hidden layers. The input layer consists of raw data which is patients’ information in the heart diagnostic system. In addition, the hidden layers have weights and generate output layer. In other words, MLP aims to map the input to the output using historical data. MLP uses back propagation as its training algorithm. This algorithm repeats presentation of the input data to the neural network. In each iteration, the output data is compared with the desired one, error is computed and fed back (back propagated) to the network. This feedback is used to modify the weights of neurons. Finally, the desired output will be generated based on iterations [6].
Neuro-Fuzzy Approach Another model that is used in this work is Neuro-Fuzzy, which is the combination of fuzzy logic and neural networks in order to solve wide variety of real world problems in an effective manner. This combination is for removing the limitation of each model. Since neural networks are good at recognizing patterns and not good at explaining how they achieve their decisions. Fuzzy logic systems that can give inexact reasons, and explain their decisions well but not good at reaching the rules they use to make those decisions [11]. The ability to model a problem domain using a linguistic model instead of complex mathematical is the main advantage of using the Neuro-Fuzzy combination [12]. Therefore, these techniques are complementary to be used together [13]. In this work, a very famous architecture for NeuroFuzzy approach known as Adaptive-Network-based Fuzzy Inference System (ANFIS) is used as introduced in [14]. ANFIS can serve as a basis for constructing a set of fuzzy if-then rules with appropriate membership functions to generate the stipulated input-output pairs. ANFIS tool is embedded now in Matlab, therefore, users have to simply type the command “anfisedit” in Matlab command window in order to use this valuable tool. Figure 3 shows the ANFIS architecture, whereas Figure 4 shows the ANFIS procedure.
Automatic Heart Disease Diagnosis System Based on Artificial ...
83
AUTOMATIC HEART DISEASE DIAGNOSIS SYSTEM’S DESIGN AND IMPLEMENTATION This section highlights all aspects regarding the data set, design, and implementation for the automatic heart disease diagnosis system.
The Cleveland Data Set It is very obvious that data set is an important aspect for developing this kind of systems. The Cleveland data set is very famous and has been widely used as a benchmark for heart disease diagnosis systems. Therefore, the Cleveland data set for heart diseases is used in this project. The Cleveland data set contains a total number of 303 instances with 13 medical attributes (factors) that are acquired from heart disease data set of Cleveland [15]. Table 1 shows some general information of the Cleveland data set, whereas Table 2 shows a detailed attributes description of the Cleveland data set.
Figure 2. MLP with two hidden layers.
Figure 3. ANFIS architecture [14].
84
Soft Computing with NeuroFuzzy Systems
Figure 4. ANFIS procedure. Table 1. General information of Cleveland data set.
Table 2. Attributes’ description of Cleveland data set.
Automatic Heart Disease Diagnosis System Based on Artificial ...
85
It is important to highlight that the Cleveland data set was randomly divided into two main categories namely: Training and Testing data set, which comprise of 80% and 20% of the total Cleveland data set respectively. Experiments with the Cleveland data set have concentrated on simply attempting to distinguish presence (values H1, H2, H3, and H4) from absence (value H0). Therefore, two main outputs are identified, where the value H0 means heart disease is absent from the patient, and the values H1, H2, H3, and H4 mean heart disease is present in the patient. Table 3 depicts the distribution of disease records for Neural Network and Neuro-Fuzzy approaches.
System’s MATLAB Graphical User Interfaces (GUIs) The programming language used for developing the automated heart disease diagnosis system is MATLAB, which is a powerful language for data analysis and visualization. There are many programming languages used in data mining. It is important to know the reasons of choosing MATLAB as data mining tool for this paper. The first advantage of using MATLAB is portability that the users will have the same range of basic functions at their disposal. Second advantage is domain specific representations that points out in MATLAB implementation, all data is the form of matrices [16]. This allows us a variety of algorithms and it is very helpful. In addition, standard neural networks have got multidimensional data which is hard for the human brain to understand. MATLAB makes this problem easy to deal with by 3D graphs and plots [17]. By using MATLAB we can also cluster the data, in which we group the objects that have similar characteristics [18]. Therefore, MATLAB is preferred because of its outstanding data calculation and visual graphic representation function [19]. As stated earlier, there are two main systems, whereby the first one is based on ANN and the second one is based on NeuroFuzzy. Each system has three main modules namely: Training, Testing, and Case-Based Modules.
EXPERIMENTAL RESULTS AND ANALYSIS As stated earlier, there are two main approaches used to develop the automated heart disease diagnosis system, which are the ANN and NeuroFuzzy. Therefore, each system has been experimented and results have been analyzed in order to compare the performance of the two approaches.
86
Soft Computing with NeuroFuzzy Systems
Table 3. Attributes’ description of Cleveland data set.
It is important to highlight that there are two main tests conducted, the first at the training module, where the training data set is tested against the trained Neural Network and Neuro-Fuzzy, and the second at the testing module, where the testing data set is tested against the trained Neural Network and Neuro-Fuzzy. Certain parameters were modified for both systems in order to optimize the systems’ performance and acknowledge their effects on the overall systems’ performance. Finally, the best combination of parameters are selected and used in the systems for future tests. A third test is conducted, where users could input values for a specific case and classify whether the heart disease is present or absent as shown in Table 4.
Experimental Results for ANN System Two main parameters have been explored in training the ANN system, which are the maximum number of epochs and number of hidden neurons. Maximum number of epochs ranges from 1000 to 5000 with an increment of 1000, whereas number of hidden neurons ranges from 5 to 15 with an increment of 5. Table 4 shows the obtained results from different combinations of epochs and hidden neurons. From Table 4, it is clearly found that the ANN system mostly achieved 80% onwards in most combinations of the two parameters. However, it is found that the combination 4000 epochs and 15 hidden neurons achieved the highest accuracy using the training data set, which is 90.74% but did not achieve the highest accuracy using the testing data, which is 85.19%. On the other hand, the combination 5000 epochs and 15 hidden neurons achieved the highest accuracy using the testing data set, which is 87.04% but did not achieve the highest accuracy using the training data, which is 88.43%. In most systems, the testing data is very important and systems are evaluated on how best they can perform when receiving data from of someone who has not been trained earlier. Therefore, if considering this aspect, the combination of 5000 epochs and 15 hidden neurons is selected since it performs the highest using the testing data set.
Automatic Heart Disease Diagnosis System Based on Artificial ...
87
It is also seen that the ANN system could successfully classify the user inputs for a specific case, where through the 15 different experiments, the ANN system could classify that data to “Absent” of heart disease, therefore, the case-based module achieved 100% accuracy as far as Table 4 is concerned.
Experimental Results for Neuro-Fuzzy System Given separate sets of input and output data, “genfis2” parameter generates a Fuzzy Inference System (FIS) structure using fuzzy subtractive clustering. When there is only one output, “genfis2” may be used to generate an initial FIS for ANFIS training by first implementing subtractive clustering on the data. The parameter “genfis2” accomplishes this by extracting a set of rules that models the data behavior. The rule extraction method first uses the MATLAB “subclust” function to determine the number of rules and antecedent membership functions and then uses linear least squares estimation to determine each rule’s consequent equations. This function returns an FIS structure that contains a set of fuzzy rules to cover the feature space. Therefore, the “genfis2” is the only parameter that is investigated in this research and it ranges from 0.1 to 1.0 as shown in Table 5. Table 4. Experimental results for ANN system with different combination of parameters.
From Table 5, it is clearly found that the Neuro-Fuzzy system has excellent achievements using training data set, where all values of “genfis2” from 0.1 to 1.0 achieved 100%. However, the Neuro-Fuzzy system could not achieve higher accuracy using the testing data set. The best performance using the testing data set was at 0.5, which is 75.93%. In fact, “0.5” is the benchmark value for “genfis2” parameter. It is also noticed that as we increase the value for the “genfis2”, the number of generated rules decreases.
88
Soft Computing with NeuroFuzzy Systems
Table 5. Experimental results for neuro-fuzzy system with different values of “genfis2”.
It is also seen that the Neuro-Fuzzy system could not successfully classify all the user inputs for a specific case, where only 6 out of 10 different experiments, the Neuro-Fuzzy system could classify that data to “Absent” of heart disease, and 4 other experiments the Neuro-Fuzzy system failed to successfully classify them. Therefore, the case-based module achieved 60% accuracy as far as Table 5 is concerned. In most systems, the testing data is very important and systems are evaluated on how best they can perform when receiving data from of someone who has not been trained earlier. Therefore, if considering this aspect, the value for the “genfis2” parameter is fixed at “0.5”.
CONCLUSION AND FUTURE WORK This research effort developed two systems based on ANN and Neuro-Fuzzy approaches in order to develop an automatic heart disease diagnosis system. From both Table 4 and Table 5, it is clear that the Neuro-Fuzzy system outperforms the ANN system using the training data set, where the accuracy for each system was 100% and 90.74%, respectively. However, using the testing data set, it is clear that the ANN system outperforms the NeuroFuzzy system, where the best accuracy for each system was 87.04% and 75.93%, respectively. This system can be used at hospital level by doctors and physicians to classify the patient’s heart disease. Future work can be through applying various ANN’s architecture and training algorithms for achieving more accurate results.
Automatic Heart Disease Diagnosis System Based on Artificial ...
89
REFERENCES 1.
Gudadhe, M., Wankhade, K. and Dongre, S. (2010) Decision Support System for Heart Disease based on Support Vector Machine and Artificial Neural Network. IEEE International Conference on Computer and Communication Technology, Allahabad, 17-19 September 2010, 741-745. 2. Yan, H.M., Jiang, Y.T., Zheng, J., Peng, C.L. and Li, Q.H. (2006) A Multilayer Perceptron-Based Medical Decision Support System for Heart Disease Diagnosis. Expert Systems with Applications, 30, 272281. http://dx.doi.org/10.1016/j.eswa.2005.07.022 3. Palaniappan, S. and Awang, R. (2008) Intelligent Heart Disease Prediction System Using Data Mining Techniques. International Journal of Computer Science and Network Security, 8, 343-350. 4. Wu, R., Peters, W. and Morgan, M.W. (2002) The Next Generation Clinical Decision Support: Linking Evidence to Best Practice. Journal Healthcare Information Management, 16, 50-55. 5. Adeli, A. and Neshat, M. (2010) A Fuzzy Expert System for Heart Disease Diagnosis. Proceedings of the International Multiconference of Engineers and Computer Scientists, Hong Kong, 17-19 March 2010. 6. Sivanandam, S.N., Sumathi, S. and Deepa, S.N. (2006) Introduction to Neural Networks Using MATLAB 6.0. McGraw-Hill Education, New York City. 7. Kahramanli, H. and Allahverdi, N. (2008) Design of a Hybrid System for the Diabetes and Heart Diseases. Expert Systems with Applications, 35, 82-89. http://dx.doi.org/10.1016/j.eswa.2007.06.004 8. Das, R., Turkoglu, I. and Sengur, A. (2009) Effective Diagnosis of Heart Disease through Neural Networks Ensembles. Expert Systems with Applications, 36, 7675-7680. http://dx.doi.org/10.1016/j. eswa.2008.09.013 9. Allahverdi, N., Torun, S. and Saritas, I. (2007) Design of a Fuzzy Expert System Determination of Coronary Heart Disease Risk. Proceedings of International Conference on Computer Systems and Technologies, Rousse, June 14-15 2007, 1-8. 10. Lisboa, P.J. (2002) A Review of Evidence of Health Benefit from Artificial Neural Networks in Medical Intervention. Neural Networks, 15, 11-39. http://dx.doi.org/10.1016/S0893-6080(01)00111-3 11. Fuller, R. (1995) Neural Fuzzy Systems. Abo Akademi University, Turku.
90
Soft Computing with NeuroFuzzy Systems
12. Tung, W.L. and Quek, C. (2002) GenSoFNN: A Generic SelfOrganizing Fuzzy Neural Network. Proceedings of IEEE Transactions on Neural Networks Conference, 13, 1075-1086. 13. Alhanafy, T.E., Zaghlool, F. and El Din Moustafa, A.S. (2010) NeuroFuzzy Modeling Scheme for the Prediction of Air Pollution. Journal of American Science, 6, 605-616. 14. Jang, J.S.R. (1993) ANFIS: Adaptive-Network-Based Fuzzy Inference System. IEEE Transactions on Systems, Man, and Cybernetics, 23, 665-685. http://dx.doi.org/10.1109/21.256541 15. KEEL: A Software Tool to Assess Evolutionary Algorithms for Data Mining Problem. http://sci2s.ugr.es/keel/dataset.php?cod=57 16. Trewartha, D. (2006) Investigating Data Mining in MATLAB. Bachelor Dissertation, Department of Science, Rhodes University, Grahamstown. 17. Harrison, R. (2000) Decision Support Technique Developed in MATLAB Improves the Accuracy of Medical Diagnoses. Department of Automatic Control and Systems Engineering, University of Sheffield, Sheffield. 18. Zhao, L., Zheng, X.Q. and Wang, S.Q. (2008) Design and Implementation of Spatial Data Mining System (M-SDM) Based on MATLAB. Journal of Computers, 3, 66-70. 19. Gan, G.J., Ma, C.Q. and Wu, J.H. (2007) Data Clustering: Theory, Algorithms and Applications. ASA-SIAM Series on Statistics and Applied Probability, SIAM, Philadelphia, ASA, Alexandria.
CHAPTER
7
Reliability Estimation of Services Oriented Systems Using Adaptive Neuro Fuzzy Inference System
Ashish Seth1, Himanshu Agarwal2, and Ashim Raj Singla3 1
Department of Computer Science, Punjabi University, Patiala, India
2
University College of Engineering, Punjabi University, Patiala, India
Department of Information Technology, Indian Institute of Foreign Trade, New Delhi, India 3
ABSTRACT In order to make system reliable, it should inhibit guarantee for basic service, data flow, composition of services, and the complete workflow. In service-oriented architecture (SOA), the entire software system consists Citation: Seth, A., Agarwal, H. and Singla, A. (2014), “Reliability Estimation of Services Oriented Systems Using Adaptive Neuro Fuzzy Inference System”. Journal of Software Engineering and Applications, 7, 581-591. doi: 10.4236/jsea.2014.77054. Copyright: © 2014 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY). http://creativecommons.org/licenses/by/4.0
92
Soft Computing with NeuroFuzzy Systems
of an interacting group of autonomous services. Some soft computing approaches have been developed for estimating the reliability of service oriented systems (SOSs). Still much more research is expected to estimate reliability in a better way. In this paper, we proposed SoS reliability based on an adaptive neuro fuzzy inference system (ANFIS) approach. We estimated the reliability based on some defined parameter. Moreover, we compared its performance with a plain FIS (fuzzy inference system) for similar data sets and found the proposed approach gives better reliability estimation. Keywords:- Reliability Estimation, SOA, Fuzzy, Rule-Based, Reliability Model, Soft Computing
INTRODUCTION Reliability is one of the most important non-functional requirements for software. Accurately estimating reliability for service oriented system (SOSs) is not possible. Moreover, soft computing techniques can help to solve problems which are uncertain or unpredictable. Many researchers have proposed different approaches to SOS reliability estimation [1]. IEEE 610.12-1990 [2] defines reliability as “The ability of a system or component to perform its required functions under stated conditions for a specified period of time”. The primary objective of reliability is to guarantee that the resources managed and used by the system are under control. It also guarantees that a user can complete its task with a certain probability when it is invoked. Software reliability management is defined in IEEE 982.1-1988 [3] as “The process of optimizing the reliability of software through a program that emphasizes software error prevention, fault detection and removal, and use of measurements to maximize reliability in light of project constraints such as resources, schedule, and performance”. Thus any reliable system is one that must guarantee and take care of fault prevention, fault tolerance, fault removal, and fault forecasting. The most suitable models for reliability of Service Oriented Architectures (SOAs) are the ones based on architecture. Although the reliability of SOA systems cannot be completely estimated, we can estimate the reliability to a larger extent by analyzing the SOA characteristics and identifying the corresponding requirements. This paper is the result of work done in continuation to our previous study [4] to estimate the reliability of service oriented systems. We started with the identification of important factors for SOA followed by estimating the reliability of such systems through fuzzy inference system (FIS) using Matlab fuzzy tool box,
Reliability Estimation of Services Oriented Systems Using Adaptive ...
93
followed by the present work which we have extended to provide more accurate reliability estimation by using an adaptive neuro fuzzy inference system (ANFIS). The rest of the paper is organized as follows. In Section 1, we discussed the basic definition of SOA, services, fuzzy logic and ANFIS. Section 2 covers the work already done in this area in different research studies. Section 3 discusses the research approach for our work. The experimentation and evaluation results are discussed in Section 4. Finally, the conclusion is drawn in Section 5.
Service Oriented Architecture SOA provides a design framework for realizing rapid and low-cost system development and improving total system quality. SOA uses the Web services standards and technologies and is rapidly becoming a standard approach for enterprise information systems. SOA is a architectural software concept whose core working is based on services, a functionality that can perform any specific task and facilitates to support business requirements. In a SOA environment, resources are made available to other participants within the network as independent services that are accessible across the network in a standardized way. Overall, a business centric, SOA approach delivers a number of benefits, which includes the following: reduced time to market, improved business alignment for growth, reduced costs, reduced business risk. Each Service Oriented Architecture plays one or more of three roles as service brokers, service registers and service providers as follows [5]: •
•
•
A service provider has to make trade-offs between availability & security. It is a web service responsibility for deciding the type of information exposed; Service broker or service register is responsible for making information available to a requestor. A service broker has to decide the amount of information transfer; The service requestor or Web service client requests for a service and binds to the service provider in order to call upon one of its Web services.
Service Services are loosely coupled, autonomous, and reusable. They have welldefined platform-independent interfaces, and provide access to data, business processes, and infrastructure, ideally in an asynchronous manner, so that they can receive requests from any source, making no assumptions as to the
94
Soft Computing with NeuroFuzzy Systems
functional correctness of an incoming request. Service is an implementation of a well-defined business functionality that operates independent of the state of any other service defined within the system. It has a well-defined set of interfaces and operates through a pre-defined contract between the client of the service and the service itself, which must be dynamic and flexible to be able to add, remove, or modify services, according to business requirements [4]. Services can be written today without knowing how it will be used in the future and may stand on its own or be part of a larger set of functions that constitute a larger service. From a dynamic perspective, there are three fundamental concepts that are important to understand: The service must be visible to service providers and consumers; the clear interface for interaction between them is defined; and the real world is affected from interaction between services. These services should be loosely coupled and have minimum interdependency, otherwise they can cause disruptions when any service fails or changes.
Neural Networks and Fuzzy Logic Neural Networks (NNs) and fuzzy logic are the two basic elements of soft computing techniques. Fuzzy means unsure and ambiguous. Fuzzy systems are suitable for approximate reasoning, especially for the system whose mathematical model is hard to derive. Fuzzy logic allows decision making with estimated values under incomplete information. A fuzzy set is a generalization of an ordinary set by allowing a degree (or grade) of membership for each element. The membership-function m(x) of a set maps each element to its degree. A membership degree is a real number on [0, 1]. In extreme cases, if the degree is 0 the element does not belong to the set, and if 1 the element belongs 100% to the set. Neural networks are a form of multiprocessor computer system, with simple processing elements, a high degree of interconnection, adaptive interaction between elements; it is also referred as an “artificial” neural network (ANN). According to Dr. Robert Hecht-Nielsen, a neural network is “...a computing system made up of a number of simple, highly interconnected processing elements, which process information by their dynamic state response to external inputs”. There are many different kinds of learning rules used by neural networks. ANNs can learn from data and feedback and have learning capabilities. On the other hand, fuzzy logic models are rule-based models and do not have learning capabilities, therefore so for learning, fuzzy inference system performs the following operations:
Reliability Estimation of Services Oriented Systems Using Adaptive ...
• • • • •
95
fuzzification of the input variables; determination of membership functions for the parameters; application of the fuzzy operator in the antecedent; implication from the antecedent to the consequent; defuzzification.
Adaptive Neuro Fuzzy Inference System (ANFIS) ANFIS was first defined by J.-S. Roger Jang in 1992. It is a techniques to learn about a data set, in order to compute the membership function parameters that best allow the associated fuzzy inference system to track the given input/output data. The toolbox function “anfis” constructs a fuzzy inference system (FIS) using a given input/output data set, for which membership function parameters are tuned (adjusted) using either a back propagation algorithm alone or in combination with a least squares type of method. ANFIS has the following advantages over an FIS as follows: •
•
•
Through learning algorithms, an ANFIS can optimize the parameters of a given FIS by simulating and analyzing the mapping relation between input and output data; An ANFIS has networks which involve nodes and directional links, along with some learning rules are also associated with these networks whereas an FIS has no network link and its behavior only depends on its membership functions; Learning method in ANFIS is much similar to that of neural networks whereas FIS has no learning capability.
RELATED WORK Most of the research on software reliability engineering focuses on system testing and system-level reliability growth models However, SOA is not taken into account in these approaches. Although there are some soft computing approaches have been developed for estimating the reliability of service oriented systems (SOSs). Goseva-Popstojanova, et al. (2001) and Gokhale (2007) did remarkable work for architecture-based empirical software reliability analysis in relation to architecture-based empirical software reliability analyses [6] [7]. Significant work done in the direction of estimating reliability of SOA is summarized below:
96
Soft Computing with NeuroFuzzy Systems
Danilecki, A., et al. [8] (2011) proposed a model named ReServE, which ensures that business processes are consistently perceived by client and services, transparently recovers the state of a business process. When a service fails, its SPU can initiate the rollback-recovery process. Brosch, F., et al. [9], proposed SAMM (2010) to evaluate the impact of different component topologies on the system reliability, author concludes that not only the hardware, but also different allocation configuration have influence on the reliability prediction. Zibin, Z., et al. [10] (2010) proposed Collaborative Reliability Prediction of Service-Oriented Systems, in his work collaborative framework is proposed for predicting reliability of service-oriented systems which employs past failure data of similar service users for making reliability prediction for the current service user. Wang, L., et al. [11], introduces unified reliability modeling framework (2009) and concludes service pools as backup alternative, reliability of simple services is addressed by considering data reliability, authors used Time Markov Chains (DTMCs) are for analyzing reliability of service composition. Wang, et al. [12] proposed Analyzed-stock market system (SMS) (2006), authors mapped to component failure probabilities and predicts the system reliability. In their work they derived transition probabilities from recorded transitions between components. Tsai, et al. [13] proposed SORM (2004), Service-Oriented Software Reliability Model which tries to determine the reliability of each component and their relationship. It consists of two stages: group testing to evaluate the reliability of atomic services; and evaluation of composite services through the analysis of components and their relationships, author used a group testing technique from the medical field to detect faults.
DISCUSSION AND RESEARCH APPROACH This work is an extension of our previous work done to identify the SOA adoption trends & implementation factors [14], followed by estimating the reliability of such systems through fuzzy inference system (FIS) using Matlab fuzzy tool box, followed by the present work which we have extended to provide more accurate reliability estimation by using a adaptive neuro fuzzy inference system (ANFIS). The present work is an extension of our previous research which includes three phases as follows:
Reliability Estimation of Services Oriented Systems Using Adaptive ...
97
1)
In first phase, thorough review of articles and research has been done and identified the factors that are relevant to SOA implementation and the extent to which each factor is crucial to SOA implementation [4]. 2) In second phase using GQM technique, metrics are proposed, and the responses are taken from 125 people in the industry. The data, which is based on the feedback and responses, is defined into the following three parameters [14]: a) AR: adhoc requirements/dynamic binding/agility; b) MG: migration/legacy system integration; c) BI: business and IT collaboration. The rules were defined for the inference engine. Three clusters were formed for the input factors (Low, Medium, and High), and five clusters were formed for the output reliability (Very Low, Low, Medium, High, and Very High). Therefore, with 3 clusters and 3 input factors, a total of 27 rules were formed that yield 33 = 27 sets. These 27 sets or classifications can be used to form 27 rules using fuzzy model. 3) In third phase (i.e. the present work), we followed Sugeno-type inference, defined it for the fuzzy logic toolbox to estimate the reliability of service oriented systems. 3)
In third phase (i.e. the present work), we followed Sugeno-type inference, defined it for the fuzzy logic toolbox to estimate the reliability of service oriented systems.
Reliability Parameters for SOA-Based Systems with Its Constituent Factors 1)
AR: A system capable of fulfilling the ad hoc on-demand changing requirement of the market is assumed to be efficient and reliable. It is based upon the way the rule engine within the model has been trained to perform dynamic binding whenever the demand changes or arises. This also covers agility, which is the important issue when someone moves from present legacy systems to SOA-based systems. It is further concluded that the more the system has capability to handle dynamic binding/ad hoc requirement/agility, the more system is assumed to be reliable. Therefore, SOA reliability α AR. 2) MG: It is observed that, although an SOA system is strong enough in terms of its capacity to handle the ad hoc market, if
Soft Computing with NeuroFuzzy Systems
98
there is no provision of integrating the legacy system or migrating successfully from old system to new one within the system; it is not effective and will not guarantee system reliability. Moreover it is observed that mostly small and medium enterprises (SMEs) using the SOA system be developed from services developed from scratch. Since it is concluded that more migrations affects the system reliability. Therefore, SOA reliability α 1/MG. 3) BI: Within a system, if the collaboration between business process and strategies is aligned with IT capabilities, the system is assumed to be more reliable. Through surveys, it has been observed that, although the powerful IT system is there, it will not be of much valuable to the organization without proper integration within the business strategies. Therefore, SOA reliability α BI. The factors described in the three parameters above assess different properties and characteristics associated with SOA model reliability. The values of these parameters cannot be used independently to measure reliability. Rather, an integrated approach that considers all three parameters and their relative impact is required for estimating a system’s overall reliability [14].
Proposed Approach 1)
Conduct a thorough survey of literature to identify the factors that are relevant to SOA implementation and the extent to which each factor is crucial to SOA implementation. 2) Identify reliability parameters in SOA context among these factors. 3) Cluster reliability parameters into three domain clusters of reliability factors. 4) Assemble a database for the value of these factors. 5) Design an inference engine based on the rule for identifying reliability clusters. 6) Using Sugeno system, perform the following operations • Plot the number of inputs, outputs, input membership functions, and output membership functions. • Load FIS or generate FIS from loaded data using your chosen number of MFs and rules or fuzzy. • Train FIS after setting optimization method, error tolerance, and
Reliability Estimation of Services Oriented Systems Using Adaptive ...
• •
99
number of epochs. Training adjusts the membership function parameters and plots the training (and/or checking data) error plot(s) in the plot region Test data against the FIS model. Anticipate the FIS model output versus the training, checking, or testing data output. We have a training data set that contains desired input/output data pairs of the target system to be modeled. These training and checking data sets are collected based on observations of the target system and are then stored in separate files. It has been observed that only the checking data set is corrupted by noise.
RESULT AND DISCUSSION For present modeling we used the Fuzzy Logic Toolbox neuro-adaptive learning techniques incorporated in the anfis command. The parameters could be chosen so as to tailor the membership functions to the input/output data in order to account for these types of variations in the data values. Our experiments simulated the effect of rules with the MATLAB Fuzzy Logic Toolbox; the reliability for the values obtained is found to be very close to the calculated value, thus result obtained justifies our approach by giving better estimates in comparison to FIS [15]-[17]. For the analysis of result obtained with the experiment we used covariance method to compare the closeness of the value obtained with the experiment with the values collected from original sample data set. Covariance provides a measure of the strength of the correlation between two or more sets of random variants. The covariance for two random variates X and Y, each with sample size N, is defined by the expectation value (1) (2) where and written out explicitly as
are the respective means, which can be
The comparison table for the ANFIS and original data set is shown in the
100
Soft Computing with NeuroFuzzy Systems
table (Table 1). Covariance matrix for (Anfis, ori) =
Since the covariance is positive we can say that we get results closer to the original values. We generated a plot for test data against FIS, the FIS is trained after setting optimization method, error tolerance and number of epochs. Figure 1 shows the plot of testing data against FIS, testing data appear on the plot in blue color while the FIS output is shown in red color. After creating the ANFIS model, we compared the output reliability values for different input sets with the original values. We calculated Average Testing Error for the output obtained by the FIS and the output obtained by the ANFIS with the original output. ANFIS reduces the error to 0.021%. Hence, the ANFIS performs better than the FIS. In ANFIS, we first trained the FIS, on the basis of training data the rules were formed to produce the output of the trained model. We observed during experiments that for large data sets its execution is little complex. Our results show that the ANFIS model gives a more accurate measure of reliability than the FIS model. Table 2 illustrates the comparison chart for FIS, ANFIS and original. Similarly graph shown in Figure 2 indicates ANFIS is closer to original values than FIS. Table 1. Comparison of original data set with ANFIS using Sugeno method.
Reliability Estimation of Services Oriented Systems Using Adaptive ...
101
Figure 1. Plot of testing data: Original vs. ANFIS. Table 2. Plot of FIS, ANFIS and original.
Figure 2. The graph of ANFIS is much close to original values than FIS.
The inference system, inference rules, fuzzy inference system, rule viewer and surface viewer for ANFIS using Sugeno method is shown in appendices.
102
Soft Computing with NeuroFuzzy Systems
CONCLUSION AND FUTURE WORK This paper proposes a neuro fuzzy approach for estimating the reliability of service oriented systems. Proposed approach is based on an ANFIS that requires less computational time than previously proposed FIS and other traditional approaches. Our results show that the ANFIS give more accurate estimation than FIS. Future scope may be to identify other relevant factors that should be used but currently we only have data available for the discussed factors. Our experience documented in this paper will be helpful for practitioners in collecting the data necessary for reliability prediction. Researchers are provided a demonstration on how the fuzzy logic toolbox can be used to find the reliability of such system on the basis of certain SOA features.
Reliability Estimation of Services Oriented Systems Using Adaptive ...
103
REFERENCES 1.
Kirti, T. and Arun, S. (2012) A Rule-Based Approach for Estimating the Reliability of Component Based Systems Advances in Engineering Software. Elsevier, Amsterdam, 24-29. 2. IEEE Standard 610.12-1990 (2014) IEEE Standard Glossary of Software Engineering Terminology. http://ieeexplore.ieee.org/stamp/ stamp.jsp?tp=&arnumber=159342 3. IEEE Standard 982.1-1988 (2014) IEEE Standard Dictionary of Measures to Produce Reliable Software. http://www.baskent.edu. tr/~zaktas/courses/Bil573/IEEE_standards/982_1_2005.pdf http:// standards.ieee.org/findstds/standard/982.1-1988.html 4. Ashish, S., Aggarwal, H. and Singla, A. (2012) Service Oriented Architecture Adoption Trends: A Critical Survey at 5th International Conferences on Contemporary Computing. Proceeding in Communications in Computer and Information Science, Springer, Berlin. 5. Bianco, P., Kotermanski, R. and Merson, P. (2007) Evaluating a ServiceOriented Architecture. Software Engineering Institute. Prepared for The SEI Administrative Agent ESC/XPK, 5, Eglin Street Hanscom AFB. 6. Gokhale, S.S. (2007) Architecture-Based Software Reliability Analysis: Overview and Limitations. IEEE Transactions on Dependable and Secure Computing, 4, 132-140. http://dx.doi.org/10.1109/TDSC.2007.4 7. Goseva-Popstojanova, K. and Trivedi, K.S. (2001) ArchitectureBased Approach to Reliability Assessment of Software Systems. Perform Evaluation, 45, 179-204. http://dx.doi.org/10.1016/S01665316(01)00034-7 8. Danilecki, A., Holenko, M., Kobusinska, A., Szychowiak, M. and Zierhoffer, P. (2011) ReServE Service: An Approach to Increase Reliability in Service Oriented Systems. Parallel Computing Technologies, PaCT 2011, LNCS 6873, 244-256. 9. Brosch, F., Koziolek, H., Buhnova, B. and Reussner, R. (2010) Parameterized Reliability Prediction for ComponentBased Software Architectures. Proceedings of the 6th International Conference on the Quality of Software Architectures (QoSA’10), Springer, New York, 3651. 10. Zheng, Z. and Lyu, M.R. (2010) Collaborative Reliability Prediction
104
11.
12.
13.
14.
15. 16.
17.
Soft Computing with NeuroFuzzy Systems
of Service-Oriented Systems. 2010 ACM/IEEE 32nd International Conference on Software Engineering, Cape Town, 2-8 May 2010, 3544. Wang, W.L., Pan, D. and Chen, M.H. (2006) Architecture-Based Software Reliability Modeling. Journal of Systems and Software, 79, 132-146. http://dx.doi.org/10.1016/j.jss.2005.09.004 Wang, L., Bai, X. and Zhou, L. (2009) A Hierarchical Reliability Model of Service-Based Software System. 33rd Annual IEEE International Computer Software and Applications Conference, Seattle, Washington DC, 20-24 July 2009, 199-208. Tsai, W., Zhang, D., Chen, Y., Huang, H., Paul, R. and Liao, N. (2004) A Software Reliability Model for Web Services. 8th IASTED International Conference on Software Engineering and Applications, Cambridge, 8-11 November 2004. Ashish, S., Aggarwal, H. and Singla, A. (2014) Estimating Reliability of Service-Oriented Systems: A Rule-Based Approach. International Journal of Innovative Computing, Information and Control. ICIC International, 10, 1111-1120. Gershteyn, Y. and Perman, L. (2003) Matlab: ANFIS Toolbox. http:// csrit.gershteyn.net/courses/nn/Presentations/3-MatLab_ANFIS.pdf Arikan, S. (2012) Automatic Reliability Management in SOABased Critical Systems. European Conference on Service-Oriented and Cloud Computing, 1-6. http://dspace.icsy.de:12000/dspace/ bitstream/123456789/367/1/reliability.pdf Becker, S. (2008) Coupled Model Transformations for QoS Enabled Component-Based Software Design. Ph.D. Thesis, University of Oldenburg, Oldenburg.
CHAPTER
8
Prediction of Soil Fractions (Sand, Silt and Clay) in Surface Layer Based on Natural Radionuclides Concentration in the Soil Using Adaptive Neuro Fuzzy Inference System
Saad Al-Hamed1, Mohamed Wahby1, Mohamed Al-Sulaiman2, and Abdulwahed Aboukarima2,3 Department of Agricultural Engineering, College of Food and Agriculture Sciences, King Saud University, Riyadh, KSA 1
2
Community College, Huraimla, Shaqra University, Huraimla, KSA
Agricultural Engineering Research Institute, Agricultural Research Centre, Cairo, Egypt 3
Citation: Al-Hamed, S. , Wahby, M. , Al-Sulaiman, M. and Aboukarima, A. (2014), “Prediction of Soil Fractions (Sand, Silt and Clay) in Surface Layer Based on Natural Radionuclides Concentration in the Soil Using Adaptive Neuro Fuzzy Inference System”. Open Journal of Soil Science, 4, 215-225. doi: 10.4236/ojss.2014.47024. Copyright: © 2014 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY). http://creativecommons.org/licenses/by/4.0
106
Soft Computing with NeuroFuzzy Systems
ABSTRACT In this research, a gamma ray sensor (The Mole) was used to get the natural radionuclides concentration in situ in the surface layer of cultivated soils. For sand, silt and clay predictions, an adaptive neuro fuzzy inference system (ANFIS) was performed to predict such fractions (Sugeno model). The inputs to the system were Potassium (40K), Uranium (238U), Thorium (232Th) and Cesium (137Cs) concentrations. It is concluded that ANFIS structure is acceptable in the prediction of sand, silt and clay considering the studied inputs. Test results and predicted outcomes were compared and acceptable correlations were obtained. Keywords:- Agriculture, ANFIS, Soil Texture, Natural Radionuclides
INTRODUCTION Soil texture refers to the percentage by weight of sand (particles between 0.05 to 2.0 mm), silt (0.002 to 0.05 mm), and clay ( 0.5φVc , a minimum web steel needs to be provided, where b is the width of the beam and s is the spacing of vertical stirrups.
bs 3.5bs A v ≥ Max 0.2 f c' , f y f y (5)
The spacing must not be larger than Min (d / 2, 60) cm; 3) V = s
Region III: If 3φVü ≥ V > φV , shear reinforcement A v f y d Vu = − Vc s φ
(6)
has to be provided to carry the difference and the spacing s must not be
Application of the Adaptive Neuro-Fuzzy Inference System for ...
larger than 4)
Avfy Avfy Min d / 2, 60, , 0.2 f c' b 3.5b
167
cm;
Region IV: If 5φVc ≥ Vu > 3φVc , similarly the shear reinforcement in Equation (6) has to be provided to carry the difference, but the Avfy Avfy Min d / 4,30, , 0.2 f c' b 3.5b
spacing s must not be larger cm. The above statements can be summarized in Figure 3. From the maximum spacing limitations in different regions computed by the selfwritten MATLAB program, the total number of vertical stirrups can be obtained. Because the reaction, in the direction of applied shear, introduces compression into the end regions of a member, the critical section can be assumed at a distance of d from the support, provided that no concentrated loads acts between support face and distance d thereafter. If the factored shear force Vud at a distance d from the face of the support is larger than 5φVc , the beam section must be enlarged. Therefore, the constraint for shear takes the form Vud ≤ 5φVc (7)
Bending Moment For simplicity, this paper assumes that the strain in the tension reinforcement is equal to 0.005; therefore, the section is tension-controlled, that is, the strength reduction factor for moment is fixed at 0.9, not a function of strain in the tension reinforcement any more. Accordingly, the constraint for both positive and negative moment takes the form moment M2 at (3L / 8) from point A orC, as shown in Figure 2(a), and M u ≤ 0 − 9M n,0.005 (8)
1 As f y = As f y d − × M n,0.005 ' 2 0.85f c b (9)
where Mu is the factored moment M1 at the middle support section (negative moment) or the maximum positive
Soft Computing with NeuroFuzzy Systems
168
Figure 3. Maximum spacing limitations for vertical stirrups in different regions.
When the strain in the tension reinforcement is equal to 0.005, the area of the reinforcement is of the form = As
0.85f c' β1 3db × fy 8
(10)
where β1 is the stress block depth factor. To prevent sudden failure with little or no warning when the beam cracks or fails in a brittle manner, the ACI Code also limits the minimum and maximum amount of steel to be A s,min ≤ A s ≤ A s, max (11)
where A s,min =
0.85f c' β1bd 3 (12) fy 7
and 0.8 f ' 14bd c A s,min = max bd, fy fy (13) A s,max in Equation (12) is derived based on the requirement that the tensile strain be equal to 0.004.
Application of the Adaptive Neuro-Fuzzy Inference System for ...
169
Development of Reinforcement According to the ACI Code, at least one-third of the total tension reinforcement provided for negative bending moment at the support should extend beyond the inflections point not less than the effective depth d of the member, 12db, or 1/16 of the clear span. For practical purposes, let span L ≈ clear spam L n . Hence the constraint for the length of the top reinforcement in Figure 2(b) can be expressed as L L + Max d,12d , lü = ≥l ü (14)
where ld is development length of tension reinforcement and db is the nominal diameter of the bar.
Deflections The ACI Code indicates that wherever excessive deflection may adversely affect the service-ability of the structure at service loads, deflections under service load conditions must be computed. Creep and shrinkage will magnify the magnitude of deflection with time. Consequently, design engineers have to evaluate immediate as well as long-term deflection in order to ensure their values satisfy the maximum permissible criteria for the particular structure and its particular use. The additional deflection under sustained loading and long-term shrinkage in accordance with ACI procedure can be calculated by multiplying the immediate deflection by a factor T λ= 1 + 50ρ '
(15)
where ρ ' is the compression reinforcement ratio calculated at midspan for simple and continuous beams and T is a factor that is taken as 1.0 for loading time duration of 3 months, 1.2 for 6 months, 1.4 for 12 months and 2.0 for 5 years or more. Because the beam considered in this paper supports partitions and other construction likely to be damaged by large deflections, the ACI code requires that the long-term deflection ∆ = (∆ i ) L + λ(∆ i ) D ≤
480
(16)
170
Soft Computing with NeuroFuzzy Systems
where (∆ i ) L = immediate live-load deflection and (∆ i ) D = immediate deadload deflection. The formula for the maximum deflection in the beam can be found in Figure 2(a).
NUMERICAL RESULTS The given conditions for the optimal design of two-span continuous singly reinforced rectangular concrete beams with a rectangular cross-section are the span length L, uniformly distributed dead wD and live load wL, compressive strength of concrete fc' and yield strength of steel fy. Design variables are the width b and effective depth d of the beam, the steel ratio ρ1 for the positive moment and the steel ratio ρ2 for the negative moment. The concrete cover for the reinforcement is 4 cm and No. 3 vertical stirrups are used. The objective function is to find the minimum cost in New Taiwan Dollars of concrete and steel used in the two-span continuous beam. In Taiwan, the unit price of concrete is 1800 NT$/m3 and the unit price of steel is 19.5 NT$/kgf. The optimal results found by the genetic algorithm consist of the minimum cost of the two-span continuous beam, the width b and effective depth d of the beam, and the steel ratios for the positive and negative moments. Based on the often-used materials and customs in Taiwan, this paper selects three kinds of yield strength fy of the tension reinforcement: 2800 kgf/cm2 (40 ksi), 3500 kgf/ cm2 (50 ksi) and 4200 kgf/cm2 (60 ksi) as well as three kinds of compressive strength f c1 of the concrete: 210 kgf/cm2 (3000 psi), 280 kgf/cm2 (4000 psi) and 350 kgf/cm2 (5000 psi). Three kinds of span length are chosen: 6 m, 8 m and 10 m; four kinds of uniformly distributed dead load wD are chosen: 2100 kgf/m, 2300 kgf/m, 2500 kgf/m and 2700 kgf/m; uniformly distributed live load wL is fixed at 1800 kgf/m. From the combinatorial analysis, there are totally 108 cases to be designed. This paper adopts the MATLAB toolbox for genetic algorithm [29] to carry out the genetic algorithm. All the constraints are built according to the formulas discussed in Section 4, most of which are highly nonlinear and cause the difficulty using the traditional gradient-based methods to find the optimal solution.
Genetic Algorithms To run the genetic algorithm of the MATLAB software, some parameters need to be selected. Here are the values used in this paper: after a number of trials, the population size is set to be 20, crossover rate 0.8, and elite number 2. Furthermore, all the individuals are encoded as real numbers; “Rank” is
Application of the Adaptive Neuro-Fuzzy Inference System for ...
171
used as the scaling function that scales the fitness values based on the rank of each individual; “Roulette” is the selection function to choose parents for the next generation; “Two-Point Crossover” is used as the crossover method to form a new child for the next generation; the “Adaptive Feasible Function” is chosen as the mutation function to make small random changes in the individuals and ensure that linear constraints and bounds are satisfied. The genetic algorithm is executed 30 times for each case, from which the best is selected. For the use of ANFIS, the total 108 cases of data are divided into 3 groups randomly by a computer algorithm: 64 cases of training data (60%), 22 cases of checking data (20%) and 22 cases of testing data (20%).
Adaptive Neuro-Fuzzy Inference Systems When using ANFIS with MATLAB, there are some restrictions: 1) only first- or zeroth-order Sugeno-type systems are supported; 2) there is only one single output; 3) each rule is of unit weight. The inputs of the adaptive neuro-fuzzy inference system consist of six elements: f y , f c1 , w D , L, b and d . There are three targets: the minimum cost, the steel ratios ρ1 and ρ2 . Because only one output is allowed, ANFIS must be executed for each target individually. To make the Sugeno-type fuzzy inference system more efficient, the “Subtractive Clustering” technique is employed. During the training process, the checking data is also loaded to ANFIS to avoid the overfitting problem. When the model begins to overfit the data, the error on the checking set will typically to rise. When the checking error increases for a specified number of epochs, the training is stopped. The membership function parameters associated with the training epoch that has a minimum checking error are returned. To evaluate the performance of ANFIS, this paper makes use of a linear regression analysis between outputs and targets. While training ANFIS, four algorithm parameters for “Subtractive Clustering” must be provided. This paper uses the default values for the squash factor, accept ratio and reject ratio. As to the range of influence q1, this paper tries a variety of values from 0.1 to 1.5 to obtain the best one because of the complexity of the 7-dimensional data points. Among them, the value of 1.4 is found to have the best results on the whole. The results for the three outputs of the testing data are listed in Tables 1-3, where the symbols m, b and r stand for the slope, the y-intercept and correlation coefficient, respectively. The scatter plots corresponding to q1=1.4 for the steel ratios ρ1 and ρ2 and the minimum cost (103 NT$) are shown in Figures 4-6, respectively.
172
Soft Computing with NeuroFuzzy Systems
Table 1. The linear regression results of the steel ratio ρ1 for the testing data.
Table 2. The linear regression results of the steel ratio ρ2 for the testing data.
The correlation coefficients between the network outputs and targets are 0.9983, 0.9984 and 0.9996 for the steel ratios ρ1 and ρ2 and the minimum cost, respectively. Besides, the slope m is close to 1 and y- intercept b approximately equals 0. Table 3. The linear regression results of the minimum cost for the testing data.
Application of the Adaptive Neuro-Fuzzy Inference System for ...
173
Figure 4. The scatter plot of the steel ratio ρ1 for the testing data.
Based on Figures 4-6 and Tables 1-3, the performance of ANFIS is satisfactory and considered to be excellent. Table 4 lists the number of fuzzy rules for the three outputs with the influence ranges changing, which indicates that the larger the influence range of a cluster center becomes, the fewer fuzzy rules ANFIS results in. Taken as example, the inputs, targets and outputs of ANFIS for some cases of the testing data are shown in Table 5.
Figure 5. The scatter plot of the steel ratio ρ2 for the testing data.
174
Soft Computing with NeuroFuzzy Systems
Figure 6. The scatter plot of the minimum cost (103 NT$) for the testing data. Table 4. The number of fuzzy rules for the three outputs: steel ratios ρ1 and ρ2 and the minimum cost with the influence range changing.
Table 5. Inputs, targets and outputs of ANFIS for some cases of testing data.
Application of the Adaptive Neuro-Fuzzy Inference System for ...
175
CONCLUSION Jiin-Po Yeh, Ren-Pei Yang (ANFIS) is then built based on the data of the given conditions and optimal results of the genetic algorithm. The inputs of this model are the yield strength of steel, compressive strength of concrete, dead load (live load is fixed) and span length, width and effective depth of the beam; targets are the minimum cost, the steel ratios for the positive and negative moments. The inputs of ANFIS are different from the given conditions of the genetic algorithm, which makes ANFI more useful and flexible in the design of beams. This paper proves that ANFIS has excellent performance with correlation coefficients between outputs and targets of the steel ratios for positive and negative moments and the minimum cost of the testing data being 0.9983, 0.9984 and 0.9996, respectively. In addition, the influence ranges of a cluster center from 0.1 to 1.5 for “Subtractive Clustering” to estimate the number of clusters and the cluster centers are explored, among which the value of 1.4 can lead to the best results as a whole, as far as the performance of ANFIS is concerned. In the future, once the input data are provided, ANFIS could quickly yield the minimum cost, steel ratios for the positive and negative moments as well as the spacing of vertical stirrups in each region with high precision, which automatically accomplish the design of the continuous reinforced concrete beams. The ANFIS model for the design of beams is easily implemented and timesaving, because it does not need to build the tedious and complex constraints.
176
Soft Computing with NeuroFuzzy Systems
REFERENCES 1.
Holland, J.H. (1975) Adaptation in Natural and Artificial Systems. The University of Michigan Press, Ann Arbor. 2. Goldberg, D.E. (1989) Genetic Algorithms in Search, Optimization and Machine Learning. Addison Wesley, Reading. 3. Coello, C.C., Hernandez, F.S. and Farrera, F.A. (1997) Optimal Design of Reinforced Concrete Beams Using Genetic Algorithms. Expert Systems with Applications, 12, 101-108. http://dx.doi.org/10.1016/ S0957-4174(96)00084-X 4. Coello, C.A. and Christiansen, A.D. (2000) Multiobjective Optimization of Trusses Using Genetic Algorithms. Computers and Structures, 75, 647-660. http://dx.doi.org/10.1016/S0045-7949(99)00110-8 5. Cheng, J. and Li, Q.S. (2008) Reliability Analysis of Structures Using Artificial Neural Network Based Genetic Algorithms. Computer Methods in Applied Mechanics and Engineering, 197, 3742-3750. http://dx.doi.org/10.1016/j.cma.2008.02.026 6. Belevičius, R. and Šešok, D. (2008) Global Optimization of Grillages Using Genetic Algorithms. Mechanika, 6, 38-44. 7. Šešok, D. and Belevičius, R. (2008) Global Optimization of Trusses with a Modified Genetic Algorithm. Journal of Civil Engineering Management, 14, 147-154. http://dx.doi.org/10.3846/13923730.2008.14.10 8. Chan, C.M., Zhang, L.M. and Jenny, T.N. (2009) Optimization of Pile Groups Using Hybrid Genetic Algorithms. Journal of Geotechnical and Geoenvironmental Engineering, 135, 497-505. http://dx.doi. org/10.1061/(ASCE)1090-0241(2009)135:4(497) 9. McCulloch, W.S. and Pitts, W. (1943) A Logical Calculus of Ideas Immanent in Nervous Activity. Bulletin of Mathematical Biophysics, 5, 115-133. http://dx.doi.org/10.1007/BF02478259 10. Rumelhart, D.E., McClelland, J.L. and The PDP Research Group (1986) Parallel Distributed Processing: Explorations in the Microstructure of Cognition. Volume 1: Foundations. MIT Press, Cambridge. 11. Zadeh, L.A. (1965) Fuzzy Sets. Information and Control, 8, 338-353. http://dx.doi.org/10.1016/S0019-9958(65)90241-X 12. Klaua, D. (1965) Über einen Ansatz zur mehrwertigen Mengenlehre. Monatsberichte der Deutschen Akademie der Wissenschaften zu Berlin, 7, 859-876.
Application of the Adaptive Neuro-Fuzzy Inference System for ...
177
13. Bezdek, J.C. and Harris, J.D. (1978) Fuzzy Partitions and Relations: An Axiomatic Basis for Clustering. Fuzzy Sets and Systems, 1, 111127. http://dx.doi.org/10.1016/0165-0114(78)90012-X 14. Kuzmin, V.B. (1982) Building Group Decisions in Spaces of Strict and Fuzzy Binary Relations. Nauka, Moscow. 15. De Cock, M., Bodenhofer, U. and Kerre, E.E. (2000) Modelling Linguistic Expressions Using Fuzzy Relations. Proceedings 6th International Conference on Soft Computing, Iizuka, 1-4 October 2000, 353-360. 16. Jang, J.S.R. (1993) ANFIS: Adaptive-Network-Based Fuzzy Inference System. IEEE Transactions on Systems, Man and Cybernetics, 23, 665-685. http://dx.doi.org/10.1109/21.256541 17. Klir, G.J. and Yuan, B. (1995) Fuzzy Sets and Fuzzy Logic: Theory and Applications. Prentice-Hall International, Inc., Englewood Cliffs. 18. Chang, F.J. and Chang, Y.T. (2006) Adaptive Neuro-Fuzzy Inference System for Prediction of Water Level in Reservoir. Advances in Water Resources, 29, 1-10. http://dx.doi.org/10.1016/j.advwatres.2005.04.015 19. Valizadeh, N., El-Shafie, A., Mukhlisin, M. and El-Shafie, A.H. (2011) Daily Water Level Forecasting Using Adaptive Neuro-Fuzzy Interface System with Different Scenarios: Klang Gate, Malaysia. International Journal of the Physical Sciences, 6, 7379-7389. http:// dx.doi.org/10.5897/IJPS11.1314 20. Folorunsho, J.O., Iguisi, E.O., Mu’azu, M.B. and Garba, S. (2012) Application of Adaptive Neuro Fuzzy Inference System (Anfis) in River Kaduna Discharge Forecasting. Research Journal of Applied Sciences, Engineering and Technology, 4, 4275-4283. 21. Lin, L.C. and Chang, H.K. (2008) An Adaptive Neuro-Fuzzy Inference System for Sea Level Prediction Considering Tide-Generating Forces and Oceanic Thermal Expansion. Terrestrial Atmospheric and Oceanic Sciences, 19, 163-172. 22. Heydari, M. and Talaee, P.H. (2011) Prediction of Flow through Rockfill Dams Using a Neuro-Fuzzy Computing Technique. The Journal of Mathematics and Computer Science, 2, 515-528. 23. Wang, A.P., Liao, H.Y. and Chang, T.H. (2008) Adaptive Neuro-Fuzzy Inference System on Downstream Water Level Forecasting. 2008 Fifth International Conference on Fuzzy Systems and Knowledge Discovery, 3, 503-507. http://dx.doi.org/10.1109/FSKD.2008.671
178
Soft Computing with NeuroFuzzy Systems
24. Mukerji, A., Chatterjee, C. and Raghuwanshi, N.S. (2009) Flood Forecasting Using ANN, Neuro-Fuzzy and Neuro-GA Models. Journal of Hydrologic Engineering, 14, 647-652. http://dx.doi.org/10.1061/ (ASCE)HE.1943-5584.0000040 25. Kwong, C.K., Wong, T.C. and Chan, K.Y. (2009) A Methodology of Generating Customer Satisfaction Models for New Product Development Using a Neuro-Fuzzy Approach. Expert Systems with Applications, 36, 11262-11270. http://dx.doi.org/10.1016/j. eswa.2009.02.094 26. Leung, K.F., Leung, F.H.F., Lam, H.K. and Ling, S.H. (2007) Application of a Modified Neural Fuzzy Network and an Improved Genetic Algorithm to Speech Recognition. Neural Computing and Applications, 16, 419-431. http://dx.doi.org/10.1007/s00521-0060068-4 27. Yeh, J.P. and Chang, Y.C. (2012) Comparison between Neural Network and Adaptive Neuro-Fuzzy Inference System for Forecasting Chaotic Traffic Volumes. Journal of Intelligent Learning Systems and Applications, 4, 247-254. http://dx.doi.org/10.4236/jilsa.2012.44025 28. ACI (2008) Building Code Requirements for Structural Concrete (ACI 318-08) and Commentary (ACI 318R-08). American Concrete Institute, Farminton Hills. 29. The MathWorks Inc. (2012) Global Optimization Toolbox: User’s Guide. The MathWorks, Inc., Natick. 30. Conn, A.R., Gould, N.I.M. and Toint, Ph.L. (1991) A Globally Convergent Augmented Lagrangian Algorithm for Optimization with General Constraints and Simple Bounds. SIAM Journal on Numerical Analysis, 28, 545-572. http://dx.doi.org/10.1137/0728030 31. Conn, A.R., Gould, N.I.M. and Toint, Ph.L. (1997) A Globally Convergent Augmented Lagrangian Barrier Algorithm for Optimization with General Inequality Constraints and Simple Bounds. Mathematics of Computation, 66, 261-288. http://dx.doi.org/10.1090/S0025-571897-00777-1 32. Jang, J.S.R., Sun, C.T. and Mizutani, E. (1997) Neuro-Fuzzy and Soft Computing: A Computational Approach to Learning and Machine Intelligence. Prentice-Hall, Upper Saddle River. 33. Sugeno, M. (1985) Industrial Applications of Fuzzy Control. Elsevier Science Pub. Co., Amsterdam.
Application of the Adaptive Neuro-Fuzzy Inference System for ...
179
34. Chopra, S., Mitra, R. and Kumar, V. (2006) Analysis of Fuzzy PI and PD Type Controllers Using Subtractive Clustering. International Journal of Computational Cognition, 4, 30-34. 35. Chiu, S. (1994) Fuzzy Model Identification Based on Cluster Estimation. Journal of Intelligent and Fuzzy Systems, 2, 267-278.
CHAPTER
12
Comparison between Neural Network and Adaptive Neuro-Fuzzy Inference System for Forecasting Chaotic Traffic Volumes
Jiin-Po Yeh1 and Yu-Chen Chang2 1 Department of Civil and Ecological Engineering, I-Shou University, Kaohsiung City, Taiwan; 2 Institute of Civil Engineering Technology, National Kaohsiung University of Applied Sciences, Kaohsiung City, Taiwan.
ABSTRACT This paper applies both the neural network and adaptive neuro-fuzzy inference system for forecasting short-term chaotic traffic volumes and compares the results. The architecture of the neural network consists of the
Citation: J. Yeh and Y. Chang, “Comparison between Neural Network and Adaptive Neuro-Fuzzy Inference System for Forecasting Chaotic Traffic Volumes,” Journal of Intelligent Learning Systems and Applications, Vol. 4 No. 4, 2012, pp. 247-254. doi: 10.4236/jilsa.2012.44025 Copyright: © 2014 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY). http://creativecommons.org/licenses/by/4.0
182
Soft Computing with NeuroFuzzy systems
input vector, one hidden layer and output layer. Bayesian regularization is employed to obtain the effective number of neurons in the hidden layer. The input variables and target of the adaptive neuro-fuzzy inference system are the same as those of the neural network. The data clustering technique is used to group data points so that the membership functions will be more tailored to the input data, which in turn greatly reduces the number of fuzzy rules. Numerical results indicate that these two models have almost the same accuracy, while the adaptive neuro-fuzzy inference system takes more time to train. It is also shown that although the effective number of neurons in the hidden layer is less than half the number of the input elements, the neural network can have satisfactory performance. Keywords:- Neural Network; Adaptive Neuro-Fuzzy Inference System; Chaotic Traffic Volumes; State Space Reconstruction
INTRODUCTION It has been known for decades that chaotic behaviors exist in traffic flow systems Gazis et al. [1] developed a generalized car-following model, known as the GHR (Gazis-Herman-Rothery) model, whose discontinuous behavior and nonlinearity suggested chaotic solutions for a certain range of input parameters. Due to the capacity dimension [2] of the attractor being fractal and first Lyapunov exponent [3] being positive, Disbro and Frame [4] showed the presence of chaos in this General Motors’ model without signals, bottlenecks, intersections, etc. or with a coordinated signal network. Chaos was observed in a platoon of vehicles described by the traditional GHR model modified by adding a nonlinear inter-car separation dependent term [5,6], Poincaré maps of which appear as a cloud of points without any repeat. Traffic volume collected at 2-min interval on the Beijing Xizhimen highway, China, was also found to posses chaotic behaviors [7]. Because of nonperiodic behaviors, chaotic time series seem to be unpredictable, but a variety of short-term forecast models have been attempted and proven to be successful, such as models employing Kalman filtering theory [8], the local linear model using information based on past values [9], the polynomial model [10], neural network-based black-box models [11-15], a model consisting of a fuzzy C-means clustering and a radial-basis-function neural network [16], etc. This paper also tries to forecast the short-term chaotic traffic volume at the intersection. Two kinds of models are presented for comparison. One is the neural network, where the delay coordinates [2,17,18] of the reconstructed state space of the traffic flow system are
Comparison between Neural Network and Adaptive Neuro-Fuzzy ...
183
used as the input vector of the neural network and the first delay coordinate of next state as the target of the neural network. The other model is the adaptive neuro-fuzzy inference system [19,20], where inputs and targets are identical to the first one, but membership functions and fuzzy rules [21,22] replace neurons in the neural network. The number and the shapes of the membership functions are decided and tuned by a data clustering technique and backpropagation neural network, respectively, which is different from the Park’s model [16] in the ways of data clustering and learning process.
DIAGNOSIS OF CHAOS The Poincaré map, time series, autocorrelation function, etc., can often provide graphic evidence for chaotic behavior, while the fractal dimension and largest Lyapunov exponent are two principal quantitative measures of chaos. This paper selects the fractal dimension, largest Lyapunov exponent and autocorrelation function to show the existence of chaos of the traffic flow. Brief introduction to them is as follows.
Fractal Dimension If there is only one measurement available for a system, delay coordinates are usually used to reconstruct its state space [17]. Given a time series x(t) and time delay τ, an n-dimensional state space can be reconstructed with the delay coordinates: . To get the appropriate dimension for reconstructing the state space of a chaotic dynamical system, the first step is to obtain the fractal dimension of the chaotic attractor in the state space. There are a number of ways to measure the chaotic attractor dimension [2]. Among them, this paper chose the method of correlation dimension, because it is much easier to implement and not time-consuming. Consider an orbit discretized to a set of N points in the state space. A sphere of radius r is poisoned at each point of the orbit and the number of points within each sphere with Euclidean distance less than r is counted. A correlation function is then defined as [2,23] (1) where is the Euclidean distance between points Xi and Xj and H is the Heaviside function (or unit step function ). For many attractors, this function C(r) exhibits a power law dependence on r, as r → 0 ; that is
Soft Computing with NeuroFuzzy systems
184
lim C(r) = ar d (2) r →0
Hence, a correlation dimension is defined by the expression (3) The chaotic attractor dimension will often approach an asymptote d with the dimension of the reconstructed state space gradually increasing. To embed a d-dimensional chaotic attractor, the state space may be reconstructed with dimension greater than or equal to 2d + 1, which according to Takens [24] will be sufficient to have generic delay plots.
The Largest Lyapunov Exponent The largest Lyapunov exponent of a chaotic orbit is defined by the expression [3] (4) The calculation is initiated by locating the nearest neighbor to the first point of a reference trajectory in the reconstructed state space and the distance between them is denoted by d0i. This pair of points is then propagated through the attractor for a fixed short time ∆ and its final separation di is computed. After that, a replacement for the propagated pair is attempted by the following procedure: 1) The distance of each delay coordinate point in the attractor to the propagated point of the reference trajectory is determined; 2) Points closer than a given length and away from another much smaller length (to avoid noise) are examined to see if the angle between the original pair and attempted pairs is less than a given small angle (e.g. 0.3 radians); and 3) The attempted pair with the smallest angle is used as replacement for the next propagation. The repeating of propagating and replacing are carried out for m cycles.
Autocorrelation Function To find out the resemblance of the signal x (t) with itself as time passes, the autocorrelation function (5) is an often-seen tool to achieve this purpose. If R(η ) approaches the square of the mean of the function x (t) as η → ∞ , it means that the signal
Comparison between Neural Network and Adaptive Neuro-Fuzzy ...
185
is only correlated with its recent past [25], i.e., sensitive to the initial conditions. Furthermore, the time lagη at which R(η ) first crosses the square of the mean of the function x (t) is usually considered as the time delay τ for reconstructing the state space.
FORECASTING MODELS There are two models applied in this paper to forecast short-term chaotic traffic volumes: the feedforward backpropgation neural network and the adaptive neuro-fuzzy inference system, which are described as follows.
Feedforward Backpropagation Neural Network Model The first forecasting model used in this paper is a feedforward neural network with the backpropagation training algorithm, as shown in Figure 1. The transfer function in the single hidden layer is the tan-sigmoid function (6) where, of the input vector, s is the number of
are the elements
Figure 1. The structure of the feed forward back propagation neural network.
neurons, are the weights connecting the input vector and the ith neuron, and bi is the bias of the ith neuron. The output layer with a single neuron is given by the linear function (7) where , are the weights connecting the neurons of the hidden layer and the neuron of the output layer, and B is the bias of the output neuron.
186
Soft Computing with NeuroFuzzy systems
There are many variations of the backpropagation algorithm, aiming to minimize the network performance function, i.e., the mean square error between the network outputs and the targets, which is (8) where tj and aj are the jth target and network output, respectively. This paper chooses the Levenberg-Marquardt algorithm [26-28] as the training function to minimize the network performance function. This algorithm interpolates between the Newton’s algorithm and the gradient descent method. If a tentative step increases the performance function, this algorithm will act like the gradient descent method, while it shifts toward Newton’s method if the reduction of the performance function is successful. In this way, the performance function will always be reduced at each iteration of the algorithm. To avoid the problem of overfitting, there are two methods to improve the network generalization: Bayesian regularization [29] and early stopping. The Bayesian regularization can provide a measure of how many network parameters (weights and biases) are being effectively used by the network. From this effective number of parameters, the number of neurons required in the single hidden layer of the neural network can be derived by the following equation (9) where R is the number of elements in the input vector, s is the number of neurons in the hidden layer, and P is the effective number of parameters found by the Bayesian regularization. In the strategy of early stopping, the available data is divided into three sets: the training set, validation set and testing set. The training set is used for computing the gradient and updating the network weights and biases, while the error of the validation set is monitored during the training process. When the network begins to overfit the training data, the error on the validation set typically begins to rise. Once the validation error keeps increasing for a specified number of iterations, the training is stopped and the weights and biases at the minimum of validation error are returned. The testing set is not used during the training, but is used to check the performance of the trained network. To evaluate the performance of the trained network, this paper performs linear regression analysis between the network outputs and the corresponding targets, and computes the correlation coefficient [30].
Comparison between Neural Network and Adaptive Neuro-Fuzzy ...
187
Adaptive Neuro-Fuzzy Inference System Model (10) The output level zi of each rule is weighted by the firing strength wi of the rule, which is (11) where F1 ( x) and F2 ( y) are the membership functions of inputs 1 and 2, respectively. Finally the output of the inference system yields
Figure 2. The structure of the adaptive neuro-fuzzy inference system.
(12) where N is the number of the rules. Because the number of the input variables and data sets are large in this paper, the “subtractive clustering” technique [32] is adopted to cluster the data and assign every data point a membership grade for each cluster. According to the number of membership functions and input variables, the number of rules is then decided. Due to the fact that membership functions are more tailored to the input data, the fuzzy inference system will end up having much fewer rules than that without clustering.
NUMERICAL RESULTS The eastbound traffic volumes at the intersection of Jiouru Road and Ningsia Street, Kaohsiung City, Taiwan, are taken as examples. The traffic volume data were collected by the vehicle traffic counter in November, 2008, totaling 14 days excluding weekends. Due to data being recorded every five minutes, three time intervals are chosen: 5-min, 10-min and 15-min. The data are divided into three sets: training data (8 days), validation data
188
Soft Computing with NeuroFuzzy systems
(4 days) and testing data (2 days). As mentioned previously, training and validation data are used to train forecasting models, while testing data are used to examine how good the trained models are. All forecasts are only one time interval ahead of occurrence, i.e., 5-min, 10-min or 15- min ahead of time. The MATLAB software [33] is applied to build the neural network and adaptive neurofuzzy inference system. To get a reasonable time delay for reconstruction of the traffic flow system, the autocorrelation function R(η) is plotted. Figures 3-5 show the autocorrelation function for 5-min, 10-min and 15-min traffic volumes, respectively, where the dotted horizontal line represents the square of the mean of time series of the traffic volume. All the three curves tend to approach the dotted horizontal line and the time lag η for the autocorrelation to first cross the dotted horizontal line is found approximately at 300 min for all these three time intervals. Hence the time delay τ to reconstruct the flow system is 60 for 5-min interval, 30 for 10-min interval and 20 for 15-min interval. By using the corresponding time delay and gradually increasing the dimension n of the state space, the correlation dimension of the chaotic attractor will reach an asymptote as n increases. These processes are shown in Figures 6 to 8 for 5-min, 10-min and 15- min, respectively. These figures indicate that the correlation dimension d for 5-min interval is 6.687, for 10-min interval is 6.766 and for 15-min interval is 6.637. Therefore, the embedding dimension (≥ 2d + 1) is 15 for these three time intervals. Aside from the fractal dimension, the largest Lyapunov exponent of the attractor is also calculated to show the presence of chaos. The largest Lyapunov exponents are all positive for different time intervals and almost identical for each time interval with different evolution steps, as shown in Table 1. Only after obtaining the required embedding dimension and time delay can the forecasting commence. The training input/ output data is a structure whose first component is a 15- dimensional input: Table 1. The largest Lyapunov exponent found for different time intervals and evolution steps.
Comparison between Neural Network and Adaptive Neuro-Fuzzy ...
189
Figure 3. Autocorrelation function of 5-min traffic volume.
Figure 4. Autocorrelation function of 10-min traffic volume.
Figure 5. Autocorrelation function of 15-min traffic volume.
, where x (i) is the observation of the time series of traffic volume and τ is the time delay, and whose second
190
Soft Computing with NeuroFuzzy systems
component is the output: x (i+1) . As mentioned previously, the time delay is chosen to be 60, 30 and 20 for 5-min, 10-min, and 15-min traffic volumes, respectively. Numerical results for the neural networks and adaptive neurofuzzy inference system are discussed as follows.
Neural Networks By using the Bayesian regularization, the effective network parameters (weighs and biases) can be found and the number of effective neurons in the hidden layer is then calculated from Equation (9). The results for three time interval are listed in Table 2, which shows the number of neurons actually required in the hidden layer is indeed less than half the number of input elements. The performance of a trained network can be measured to some extent by the errors on the training, validation and test sets. One option is to perform a regression analysis between the network response and the corresponding targets. Through linear regression analysis, the correlation coefficients between outputs and targets for different time intervals and data sets are obtained and shown in Table 2, ranging from 0.951 to 0.985.
Adaptive Neuro-Fuzzy Inference System By using the “subtractive clustering” technique, the minimum inference rules are found for 5-min, 10-min and 15- min intervals, respectively.
Comparison between Neural Network and Adaptive Neuro-Fuzzy ...
191
Figure 6. (a) The curves of lnC r( ) vs with state space dimension increasing from 3 to 23 (up to bottom); (b) The curve of the correlation dimension vs for 5- min traffic volume.
Figure 7. (a) The curves of lnC ( r ) vs ln( r ) with state space dimension increasing from 3 to 23 (up to bottom); (b) the curve of the correlation dimension dc vs n for 10-min traffic volume.
192
Soft Computing with NeuroFuzzy systems
Table 2. The number of effective neurons in the hidden layer of the neural network and the correlation coefficient.
The results are shown in Table 3. The number of rules found by clustering technique is indeed much fewer than that without clustering. Through the learning process, the parameters of the membership functions in the antecedent and the constants in the equation of the consequent of each rule are decided. After simulating the fuzzy inference, the correlation coefficients between outputs and targets for different time intervals and data sets are found, as shown in Table 3, ranging from 0.951 to 0.990.
CONCLUSION The phenomena of the fractal dimension, the positive largest Lyapunov exponent and the autocorrelation approaching the square of the mean of the time series confirm the existence of chaos in the traffic flow system. Two forecasting models of the chaotic traffic flow presented in this paper prove to be very successful with satisfactory accuracy. The Bayesian regularization applied to the neural network to get effective number of neurons in the hidden layer and the subtractive clustering technique applied to the adaptive neuro-fuzzy inference system to get the minimum number of fuzzy rules are both quite useful and effective. The numerical results show that the prediction accuracies of these two modes are almost the same, as far as the correlation coefficient is concerned, but the adaptive neuro-fuzzy inference system requires more time to train, because more parameters need to be determined and that the number of effective neurons in the hidden layer is usually less than the number of elements in the input vector.
Comparison between Neural Network and Adaptive Neuro-Fuzzy ...
193
Figure 8. (a) The curves of lnC (r)vs ln(r) with state space dimension increasing from 3 to 23 (up to bottom); (b) The curve of the correlation dimension dc vs nfor 15-min traffic volume. Table 3. The number of inference rules of the adaptive neurofuzzy inference system and the correlation coefficient.
194
Soft Computing with NeuroFuzzy systems
REFERENCES 1.
D. C. Gazis, R. Herman and R. W. Rothery, “Nonlinear Follow-theLeader Models of Traffic Flow,” Operational Research, Vol. 9, No. 4, 1961, pp. 545-567. doi:10.1287/opre.9.4.545 2. F. C. Moon, “Chaotic and Fractal Dynamics: An Introduction for Applied Scientists and Engineer,” John-Wiley and Sons Inc., New York, 1992. 3. A. Wolf, J. B. Swift, H. L. Swinney and J. A. Vastans, “Determining Lyapunov Exponents from a Time Series,” Physica D, Vol. 16, No. 3,1985, pp. 285-317. 4. J. E. Disbro and M. Frame, “Traffic Flow Theory and Chaotic Behavior,” Transportation Research Record, Vol. 1225, 1989, pp.109-115. 5. P. S. Addison and D. J. Low, “Order and Chaos in the Dynamics of Vehicle Platoons,” Traffic Engineering Control, Vol. 37, No. 7-8, 1996, pp. 456-459. 6. P. S. Addison and D. J. Low, “A Novel Nonlinear CarFollowing Model,” Chaos, Vol. 8, No. 4, 1998, pp. 791- 799. doi:10.1063/1.166364 7. P. Shang, X. Li and S. Kamae, “Chaotic Analysis of Traffic Time Series,” Chaos, Solitons & Fractals, Vol. 25, No. 1, 2005, pp. 121-128. doi:10.1016/j.chaos.2004.09.104 8. I. Okutani and Y. J. Stephanedes, “Dynamic Prediction of Traffic Volume through Kalman Filtering Theory,” Trans portation Research Part B: Methodological, Vol. 18, No. 1, 1984, pp. 1-11. doi:10.1016/01912615(84)90002-X 9. J. D. Farmer and J. J. Sidorowich, “Predicting Chaotic Time Series,” Physical Review Letters, Vol. 59, No. 8, 1987, pp. 845-848. doi:10.1103/ PhysRevLett.59.845 10. L. A. Aquirre and S. A. Billings, “Validating Identified Nonlinear Models with Chaotic Dynamics,” International Journal of Bifurcation and Chaos in Applied Sciences and Engineering, Vol. 4, No. 1, 1994, pp. 109-125. doi:10.1142/S0218127494000095 11. J. C. Principe, A. Rathie and J. M. Kuo, “Prediction of Chaotic Time Series with Neural Networks and the Issue of Dynamic Modeling,” International Journal of Bifurcation and Chaos in Applied Sciences and Engineering, Vol. 2, No. 4, 1992, pp. 989-996. doi:10.1142/ S0218127492000598 12. A. M. Albano, A. Passamante, T. Hediger and M. E. Farrell, “Using
Comparison between Neural Network and Adaptive Neuro-Fuzzy ...
13.
14.
15.
16.
17. 18.
19.
20.
21.
22. 23.
195
Neural Nets to Look for Chaos,” Physica D, Vol. 58, No. 1-4, 1992, pp. 1-9. doi:10.1016/0167-2789(92)90098-8 G. Deco and B. Schurmann, “Neural Learning of Chaotic System Behavior,” IEICE Transactions, Fundamentals, Vol. E77-A, No. 11, 1994, pp.1840-1845. R. Bakker, J. C. Schouten, F. Takens and C. M. van den Bleek, “Neural Network Model to Control an Experimental Chaotic Pendulum,” Physical Review E, Vol. 54A, No. 4, 1996, pp. 3545-3552. doi:10.1103/ PhysRevE.54.3545 E. I. Vlahogianni, M. G. Karlaftis and J. C. Golias, “ShortTemporal Short-Term Urban Traffic Volume Forecasting Using Genetically Optimized Modular Networks,” Computer-Aided Civil and Infrastructure Engineering, Vol. 22, No. 5, 2007, pp. 317-325. doi:10.1111/j.1467-8667.2007.00488.x B. Park, “Hybrid Neuro-Fuzzy Application in Short-Term Freeway Traffic Volume Forecasting,” Transportation Research Record, Vol. 1802, 2002, pp. 190-196. K. T. Alligood, T. D. Sauer and J. A. Yorke, “Chaos: An Introduction to Dynamical Systems,” Springer-Verlag, New York, 1997. L. W. Lan, J.-B. Sheu and Y.-S. Huang, “Investigation of Temporal Freeway Traffic Patterns in Reconstructed State Spaces,” Transportation Research Part C, Vol. 16, No. 1, 2008, pp. 116-136. doi:10.1016/j. trc.2007.06.006 J.-S. R. Jang, “ANFIS: Adaptive-Network-Based Fuzzy Inference System,” IEEE Transactions on Systems, Man and Cybernetics, Vol. 23, No. 3, 1993, pp. 665-685. doi:10.1109/21.256541 J.-S. R. Jang, C.-T. Sun and E. Mizutani, “Neuro-Fuzzy and Soft Computing: A Computational Approach to Learning and Machine Intelligence,” Prentice-Hall, Upper Saddle River, 1997. G. J. Klir and B. Yuan, “Fuzzy Sets and Fuzzy Logic: Theory and Applications,” Prentice-Hall International, Inc., Englewood Cliffs, 1995. L. A. Zadeh, “Fuzzy Sets,” Information and Control, Vol. 8, No. 3, 1965, pp. 338-353. doi:10.1016/S0019-9958(65)90241-X P. Grassberger and I. Proccacia, “Characterization of Strange Attractors,” Physical Review Letters, Vol. 50, No. 5, 1983, pp. 346349. doi:10.1103/PhysRevLett.50.346
196
Soft Computing with NeuroFuzzy systems
24. F. Takens, “Detecting Strange Attractors in Turbulence,” Lecture Notes in Mathematics, Vol. 898, 1981, pp. 366- 381. doi:10.1007/ BFb0091924 25. C. Y. Yang, “Random Vibration of Structures,” John Wiley & Sons, New York, 1986, pp. 44-59. 26. M. T. Hagan and M. Menhaj, “Training Feedforward Networks with the Marquardt Algorithm,’’ IEEE Transactions on Neural Networks, Vol. 5, No. 6, 1994, pp. 989- 993. doi:10.1109/72.329697 27. K. Levenberg, “A Method for the Solution of Certain Problems in Least Squares,” Quarterly of Applied Mathematics, Vol. 2, 1994, pp. 164-168. 28. D. Marquardt, “An Algorithm for Least Squares Estima tion of Nonlinear Parameters,” SIAM Journal on Applied Mathematics, Vol. 11, No. 2, 1963, pp. 431-441. doi:10.1137/0111030 29. D. J. C. MacKay, “Bayesian Interpolation,” Neural Computation, Vol. 4, No. 3, 1992, pp. 415-447. doi:10.1162/neco.1992.4.3.415 30. W. Mendenhall, R. L. Scheaffer and D. D. Wackerly, “Mathematical Statistics with Applications,” 3rd Edition, Duxbury Press, Boston, 1986. 31. M. Sugeno, “Industrial Applications of Fuzzy Control,” Elsevier Science, Amsterdam, 1985. 32. S. Chiu, “Fuzzy Model Identification Based on Cluster Estimation,” Journal of Intelligent and Fuzzy Systems, Vol. 2, No. 3, 1994, pp. 267278 33. H. Demuth, M. Beale and M. Hagan, “Neural Network Toolbox User’s Guide,” The MathWorks Inc., Natick, 2010.
CHAPTER
13
The Development of an Alternative Method for the Sovereign Credit Rating System Based on Adaptive Neuro-Fuzzy Inference System
Hakan Pabuçcu1and Tuba Yakıcı Ayan2 1 Department of Business Administration, Bayburt University, Bayburt, Turkey 2 Department of Econometrics, Karadeniz Technical University, Trabzon, Turkey
ABSTRACT The main purpose of this article is to determine the factors affecting credit rating and to develop the credit rating system based on statistical methods, fuzzy logic and artificial neural network. Variables used in this study were
Citation: Pabuçcu, H. and Ayan, T. (2017), “The Development of an Alternative Method for the Sovereign Credit Rating System Based on Adaptive Neuro-Fuzzy Inference System”. American Journal of Operations Research, 7, 41-55. doi: 10.4236/ ajor.2017.71003. Copyright: © 2017 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY). http://creativecommons.org/licenses/by/4.0
198
Soft Computing with NeuroFuzzy Systems
determined by the literature review and then the number of them was reduced by using stepwise regression analysis. Resulting variables were used as independent variables in the logistic model and as input variables for ANN and ANFIS model. After evaluating the models and comparing with each other, the ANFIS model was chosen as the best model to forecast credit rating. Rating determination was made for the countries that haven’t had a credit rating. Consequently, the ANFIS model made consistent, reliable and successful rating forecasts for the countries. Keywords:- Credit Rating, Logistic Regression (LR), Neural Networks (ANN), Adaptive Neuro-Fuzzy Inference System (ANFIS), Comparative Studies
INTRODUCTION Problems which occur during a global recession or decline periods affect all countries first and foremost developing countries. Countries lose their debt discharging competence and economic woes have seen almost each point of economic life. Independent credit rating agencies evaluated for government bonds in the point of predicting these kinds of problems determine credit ratings for countries and evaluate possibilities debtors failed to pay. For credit rating studies some statistical methods such as regression analysis [1] [2], logit and probit regression analysis [3] and discriminant analysis [4] are often used. But nowadays, determining non-linear relationships [5] instead of determining the existence of only linear relationships between variables gained importance. For that reason, advanced techniques such as neural networks, support vector machine [6], have been applied. When several studies in literature were examined, credit rating was handled on the basis of enterprises and classification problem not on the basis of countries. Besides, any credit rating estimation was made for the countries which haven’t got credit rating except [2]. In this study, rating application was carried out on the basis of countries in order to fill the gap in the literature. While performing this, assessments of three large credit rating agencies (Moody’s, S&P and Fitch) were considered. Also statistic, math and econometric models were used together with both comparative and supporting each other for credit rating application. Finally, credit prediction of 21 countries of which credit rating was not determined until today was carried out. Consequently, it is thought that this study was original and important either in respect to the methods used or in respect to determining the supremacy and deficiency of a method.
The Development of an Alternative Method for the Sovereign ...
199
SOVEREIGN CREDIT RATING AND RELATED WORKS There is a growing interest to sovereign credit ratings in recent years. The risk assessment performed by the rating agencies that represent the obligation of governments. A rating is a prospective forecast of the default risk. Sovereign ratings are not “country ratings”. There is an important differentiation between them. Sovereign rating is the credit risk of the national governments not the specific default risk of other issuers [7]. Governments usually look for credit ratings to ease their own reach to international capital markets. Sovereign credit ratings are determined by using some macroeconomicqualitative factors. Although sovereign credit ratings are assigned by the credit rating agencies, lots of questions are in governments minds related with the ratings’ rationale and consistency. The real question is “how clear the factors affect sovereign credit ratings are” [1]. In this study, it is tried to determine the factors affecting the sovereign credit ratings and an alternative model to assign ratings for the countries is developed. Related works with the sovereign-credit ratings are presented as follows. Basic variables such as income per capita, gross domestic product (GDP), growth rate, inflation, fiscal balance, external balance, external debt, default background and development sorting which are used for credit valuableness in the study carried out by [1] which was shown as a reference point for almost all studies regarding the credit rating. Direction and volume of Country ratings and relationship between the factors determining the ratings were tried to determine by regression analysis. [8] used the data of S&P and Moody’s in order to analyze the numeric data of credit ratings by dividing into two parts as countries have higher ratings and lower ratings. As the most important result of the study, it has been expressed that ratings can’t be explained only with economic and financial indicators. In [3] classification methodology with neural networks and aligned probit analysis comparing was carried out for credit rating. Credit ratings were used as dependent variable, foreign debt/export, financial balance, external balance, inflation rate, gross national product (GNP) per capita, growth rate, development situation of the country was used as an independent variable. [9] used variables of political stability, government efficiency, superiority of law and illegality in order to determine how official bodies affect the political organization’s credit rating. [10] included corruption perception index, the default history, commercial gap, position of a country in the world, democracy index, source of commercial laws, population and petrol production situation along with macro-economic variables into their models and determined the existence of a negative
200
Soft Computing with NeuroFuzzy Systems
relationship between corruption perception. In the study, where political violence, diplomatic pressure, illegality, military domination, religious trends, effectiveness of laws, structure of democracy and foreign direct investment data were used [11] not only the macro-economic variables but also the qualitative variables were provided. In this study, methodologies of credit rating agencies were handled and even GDP per capita variable could affect credit rating 80%. [12] discussed credit rating agencies and their effects on developing countries. Handled mentioned agencies within the frame of Basel 1 - 2 criteria, examined their methodologies and tried to explain qualitative and quantitative methods in details. [2] determined ratings by using regression analysis for the countries which were not included in to Credit rating by Moody’s Fitch and S&P. Credit ratings and GNP per capita, GDP per capita, reserve ratios, GDP per capita volatility, inflation and superiority of law were used as explanatory variables. Most of the predictions made for the countries which have no rating were predicted as “B” and over. [13] examined the relationships between economic freedom, credit ratings and situation of default of country. Probit analysis and Tobit analysis were used for this study. [14] used the method to compare economic growth which was considered to affect credibility, human development and political stability. As a result of the multiple regression analysis, GDP per capita, internal debt, current account balance and human development which was adapted to inequity, affected debt discharging liability negatively and unemployment and political stability affected debt discharging liability positively. [15] searched the reasons of change of credit ratings in developing markets and how changes of credit ratings in countries affected the bank credibility. Credit rating changes were used as dependent variable and economic freedom index, corruption perception index, property rights, income per capita, inflation, current account balance, financial balance and external debt were used as independent variable.
DESCRIPTION OF SELECTED MODELS Variable Selection Methodology According to [16], the most significant point in credit rating studies existing variables which symbolize the problems as possible as fewer and has strong representational ability. Due to the variable selection, the dimensions of variable space are decreased to provide effective working of algorithm [17]. When the studies carried out for variable selection were assessed, it is
The Development of an Alternative Method for the Sovereign ...
201
observed that many methods were used. E.g. [18] [19] used Kernel principal components analysis, [20], [4] used linear discriminant analysis, [17] used genetic algorithm, [21] used one way variance analysis (ANOVA), [22] used one way ANOVA, factor analysis and stepwise regression methods. In this study four different methods were implemented separately for variable selection and best considered stepwise regression analysis were preferred.
Logistic Regression Analysis Simple and multiple linear regression models could be predicted by ordinary least squares (OLS). But due to the dependent variable is qualitative (discontinuous, categorical) OLS predicts are not reliable. For that reason, alternative models were developed such as logit and probit [23]. Logistic regression is a special condition of linear regression. But while a dependent variable could take any numeric value in logistic regression, this value should be dual or categorical [24]. In [16] and [25], logistic regression analysis is mentioned which successfully implemented statistical technique on credit rating studies and many fields.
Artificial Neural Network Artificial neural network (ANN) is an advanced mathematical technique which uses intelligent learning paradigms and having several implementation fields such as social, science and engineering fields. The architectural structure of model consists of three layers such as input, hidden and output [26]. In [27] each node in input layer transfers the value belonging to independent variable to the intermediate node and data coming to intermediate layer are combined in determined rules and transformed then mapped to target value in output layer. There is only one node in artificial neural network output layer which has been founded for credit rating. Artificial neural networks not to require to provide independent variables distributive characteristics or assumptions and they could model all non-linear relationships between input-output variables [28].
Adaptive Neuro Fuzzy Inference System Ordinary mathematical method and tools are insufficient to model the systems which haven’t been defined well or uncertain systems. In contrary, fuzzy inference systems (FIS) can obtain better results by including human knowledge and reasoning process by fuzzy “if-then” rules. Fuzzy modeling was developed firstly by [29] and implemented several supervision, predict
202
Soft Computing with NeuroFuzzy Systems
mechanisms. Adaptive Neuro Fuzzy Inference System (ANFIS) structure is the condition where FIS adapted to adaptive networks. Due to ANFIS hybrid learning algorithm fuzzy “if, then” rules and human knowledge reveal input output structure. ANFIS model which was firstly developed by [30] was used then for modeling the non-linear problems, control systems and solution of chaotic time series problems. Double input (x and y) and double fuzzy “if, then” rule could be mentioned by Takagi Sugeno type ANFIS Equations (1) & (2). (1)
(2)
Functions of node in each layer of ANFIS architecture and so the functions of layers are following respectively [31] [32] [33] [34] [35]. 1st layer: Each node in this layer transfers the input signals to another layer without implementing any collecting or activation process. 2nd layer: Node shown by square in this layer represent Ai and Bi fuzzy clusters. The output values of these nodes are the membership levels bounded to input values and used membership functions (Equation (3)). (3) There are total four nodes for both inputs in the second layer. In this layer generally continuous or partial triangle, trapezoidal or bell shaped curve membership functions is used as membership function. Equation (4) which was formed by using a bell shape curve (Gaussian) membership function density function could be used in each node for Ai and Bi expressions in order to calculate 0-1 membership levels.
(4) Here mi and σi show the mean and standard deviation of the bell shaped curve membership function respectively. Parameters which are used in the meaning of “premise parameter” are adjusted while the network was being trained in this layer. 3rd layer: Each node was labeled by π in this layer and represents the multiply of all input signals. Here each node outputs which represents the firing strength of each rule is calculated by Equation (5).
The Development of an Alternative Method for the Sovereign ...
203
(5) 4th layer: Outputs of node shown by N labeled circle in this layer means the normalized threshold of the rules. Mentioned threshold could be calculated by Equation (6). (6) 5th layer: Each output of i node which was shown by square in this layer could be calculated Equation (7). (7) In equity, strength. parameter.
is the output of fourth layer and shows normalized firing Parameter sets are used in the meaning of consequent
6th layer: Node in this layer was expressed by circle and labeled by Σ. Total output consisted in this layer (f) is calculated by Equation (8) as the sum of all coming signals. (8)
EMPIRICAL STUDY Data Data of credit ratings of countries were compiled from the reports of three largest credit rating agencies as dependent variables. For that reason, the letter points of Moody’s, Standard & Poor’s and Fitch were transformed to numeric ratings by using a scale transformation and their averages were taken. A variable pool was organized for the factors affecting credit rating, so country risk after literature research result and data belonging to those variables and 2011-2013 years were collected for 180 countries over the world. Data belonging to three years is used as cross section data as separate units by not considering the time factor. The purpose of this increasing safe estimate ratio by providing ANFIS models better learning with more samples and neural networks. Due to deficiency of some data belonging to some countries, some countries were removed from the analysis. Data belonging to totally 230 units could be collected completely.
204
Soft Computing with NeuroFuzzy Systems
Variable Selection Methodology Factor analysis, linear discriminant analysis, square discriminant analysis, stepwise regression analysis implementations were carried out in order to decrease the numbers of variable by selecting from the variable pool which was formed after a literature search. As a result of factor analysis, determined four factors. But we realized that if the number of samples increases, factor loads will change. So we decided not to use factor analysis in variable selection. When common variables which were used in similar studies were considered study was continued by the variables obtained from the regression analysis. The data set should provide some assumption in parametric analysis. For that reason, variables were examined whether they are suitable for normal distribution and normality assumption was tried to provide by logarithmic transformations. Transformed variables were renamed by adding “LG” code. After the normality test, multi-collinearity problem examination was carried out and some variables which have higher Variance Inflation Factor (VIF) values were removed from the analysis.
Credit Rating Prediction Models Credit rating prediction models are generally used as classification or clustering except in this study. We investigate this problem as directly credit rating prediction as points, by using some determinants of ratings. There have been lots of determinants for the credit rating problem so we decided to decrease the number of determinants. After completing the variable selection process by using stepwise regression analysis; logistic regression analysis, artificial neural network and ANFIS model were implemented in order to determine the relationship of selected variables to credit rating. Finally, model performances were evaluated and predicts were made. In logistic regression analysis, which is implemented for credit rating estimate, the dependent variable was coded as (0 - 1) and possibility values which were calculated as analysis result were evaluated as default risks. 76% of 230 pcs data were shared for training, 13% were shared for test and 11% were shared for validation for ANN and ANFIS models. For assessing model performances, total correct classification percentage and mean absolute error (MAE) scales were used. After the analysis, credit rating estimates were made for the countries which haven’t got credit ratings and models were compared.
The Development of an Alternative Method for the Sovereign ...
205
EMPIRICAL RESULTS Variable Selection Model Results Variables determined by stepwise regression analysis in order to use for estimation models were submitted in Table 1. Here; LGGCI is global competitiveness index, LGCAB% is current account balance % of export, LGRL is the rule of law, LGDebtS is debt service and LGDEF is GDP Deflator. All variables are logarithmic forms. The “grade” is the dependent variable in all models that represents credit rating of countries.
Logistic Regression Results At this stage, variables obtained from stepwise regression analysis were used as explanatory variables. The categorical dependent variable has been formed with the classification which was made by credit rating agencies such as “investible” and “non-investible”. For that reason, “1” has been designated for the units which has credit note over 60 and “0” has been designated for the other units. As a result of logistic regression analysis, group memberships of countries were estimated and non-default possibility (being investible country) has been calculated. In Table 2, several explanatory coefficients used for model selection are seen and in Table 3, it is seen that the model estimated in fourth step according to the results of the Hosmer Lomeshow test which was implemented in order to assess model data compatibleness. Table 1. Stepwise regression coefficient.
206
Soft Computing with NeuroFuzzy Systems
Table 2. Model summary.
Table 3. Hosmer-Lemeshow test.
In Table 4, below logistic regression model obtained by forward selection method is as Equations (9)-(10). A correct classification percentage which was realized by using that model as 0.88. This value is significantly higher. (The correct classification percentage was realized 72% in [36] ANN model, 75% in [2] and 92.4% in [37].) But the possibilities obtained from logistic regression model and credit notes comply with each other. In another saying, it is not proper to transform possibilities to credit notes. That reduces the advantage of the model.
(9)
(10)
ANN Results Here the relationship between dependent and independent variables was determined by the multi-layer perception (MLP). Graphic examinations revealed that credit note was in the same direction with RL, CAB and GCI variables and opposite direction with DebtS and DEF variables. Parameters
The Development of an Alternative Method for the Sovereign ...
207
about ANN architectural structure and obtained error values are provided in Table 5.
ANFIS Results The variables selected by stepwise regression analysis were used in ANFIS model. But ANFIS model doesn’t generate good results if there are more than 4 inputs due to the rule base enlarging as [30]. For that reason, in order to determine the variable which causes the highest error, tests were carried out with quart variable groups (five group) and it has been decided to use the GCI, CAB, RL and Debt variables. Also model with five variables were trained by two membership function, but good results were not taken. Table 4. Forward selection method results.
Table 5. ANN model parameters and error results.
When mean absolute percentage errors (%MAE) related to ANFIS model in Table 6 were examined and compared with in Table 5, it is understood that it was trained well compared to ANN and could make better estimations with lower error percentage. Figure 1 and Figure 2 shows the training errors trend of ANFIS model addition to MAE statistics in Table 6 which is an indicator of whether the network was trained or not. As seen both errors in percentage (E%) and absolute percentage errors (AE%) trends approach to zero by decrease rapidly. The horizontal axis represents the sample number.
208
Soft Computing with NeuroFuzzy Systems
Also, it is seen in the graphics that errors belong to 60 and 62 numbered units are extremely large. When these units were excluded to obtain a more realistic value, MAE% value for training data was reduced 0.0771 to 0.0684. While prediction success was 87.13% in ANN model, the higher accuracy percentage was obtained as 90.66% in ANFIS model. Here it is seen that ANFIS model can comprehend nonlinear relationships between variables successfully.
CONCLUSIONS The sovereign credit rating is very important for country and other issuers, macro and micro level. Investors want to make their investment in the country have a good rating. Good rating means, lower stock cost, lower interest rate, lower investment cost and higher profit for any country or firm. In this research LR, ANN and ANFIS models were compared and credit rating predictions were made for the countries which have not. When several studies were examined about credit rating, these are seen as micro rating applications such as company or bond rating. Most of these studies handled the problem as a classification problem. Designed models here handled the credit rating problem as country risk graded and configured directly for predicting credit rating estimation. This study is considered to contribute to literature in respect to trying different statistical methods during variable selection, determining ANFIS model as the best among several prediction methods, success of credit rating prediction and risks of 21 countries firstly measured. Table 6. ANFIS model parameters and error results.
The Development of an Alternative Method for the Sovereign ...
209
Figure 1. ANFIS percentage training error.
Figure 2. ANFIS absolute percentage training error.
Besides the main targets of this study, another examined issue is whether group membership possibilities could have transformed into credit ratings. Even group membership shows significantly good predictions for countries which have higher credit ratings; it was not same for the countries which have lower credit ratings. It is clear that LR credit rating prediction which has 88% correct accuracy classification success is unsuccessful in respect to credit rating estimation. ANN and ANFIS are the models which learn the relationships over case studies and generate predictions. For that reason, it should be noted that as the number of samples increases, the correct estimation rates increase. Also, it is important that obtained data should be reliable. Relationships learned from incorrect data cause incorrect predictions. These models are considered useful to determine more complicated relationships instead of simple linear relationships. But due to these complicated relationships, it cannot be expressed by a simple equation; prediction equation could not be revealed after analysis. ANFIS method which was the best among three prediction models with 91% accuracy percentage was used for first credit rating predictions of 21 countries (Appendix B). Thereby the developed ANFIS model proved its prediction success on the verification data set. Consequently, it was revealed that ANFIS model is the most proper model in order to measure country risks and assign credit ratings and it could be used trustfully. ANFIS model is selected as the best prediction model because of its mathematical hybrid structure and learning algorithm. All possible structure
210
Soft Computing with NeuroFuzzy Systems
and parameters are tried for LR and ANN. For the future research, it’s possible to apply another artificial intelligence technic or heuristic-metaheuristic search technic for variable selection and credit rating prediction.
The Development of an Alternative Method for the Sovereign ...
211
REFERENCES 1.
Cantor, R. and Packer, F. (1996) Determinants and Impact of Sovereign Credit Ratings. Journal of Fixed Income, 6, 76-91. https:// doi.org/10.3905/jfi.1996.408185 2. Ratha, D., De, P.K. and Mohapatra, S. (2011) Shadow Sovereign Ratings for Unrated Developing Countries. World Development, 39, 295-307. https://doi.org/10.1016/j.worlddev.2010.08.006 3. Bennell, J.A., Crabbe, D., Thomas, S. and Gwilym, O.A. (2005) Modelling Sovereign Credit Ratings: Neural Networks versus Ordered Probit. Expert System with Applications, 30, 415-425. https://doi. org/10.1016/j.eswa.2005.10.002 4. Subasi, A. and Gursoy, I.M. (2010) EEG Signal Classification Using PCA, ICA, LDA and Support Vector Machines. Expert System with Applications, 37, 8659-8666. https://doi.org/10.1016/j. eswa.2010.06.065 5. Harris, T. (2015) Credit Scoring Using the Clustered Support Vector Machine. Expert System with Applications, 42, 741-750. https://doi. org/10.1016/j.eswa.2014.08.029 6. Jones, S., Johnstone, D. and Wilson, R. (2015) An Empirical Evaluation of the Performance of Binary Classifiers in the Prediction of Credit Ratings Changes. Journal of Banking & Finance, 56, 72-85. https://doi. org/10.1016/j.jbankfin.2015.02.006 7. Beers, T.D. and Cavanaugh, M. (2004) Sovereign Credit Ratings: A Primer. Standard & Poor’s, New York. 8. Bissoondoyal-Bheenick, E. (2005) An Analysis of the Determinants of Sovereign Ratings. Global Finance Journal, 15, 251-280. https://doi. org/10.1016/j.gfj.2004.03.004 9. Butler, A.W. and Fauver, L. (2006) Institutional Environment and Sovereign Credit Ratings. Financial Management, 35, 53-79. https:// doi.org/10.1111/j.1755-053X.2006.tb00147.x 10. Depken, C., LaFountain, C. and Butters, R. (2007) Corruption and Creditworthiness: Evidence from Sovereign Credit Ratings. Working Papers 0601, Department of Economics, University of Texas at Arlington. 11. Busse, M. and Hefeker, C. (2007) Political Risk, Institutions and Foreign Direct Investment. European Journal of Political Economy, 23, 397-415. https://doi.org/10.1016/j.ejpoleco.2006.02.003
212
Soft Computing with NeuroFuzzy Systems
12. Elkhoury, M. (2009) Credit Rating Agencies and Their Potential Impact on Developing Countries. Compendium on Debt Sustainability, 165-180. 13. Roychoudhury, S. and Lawson, R.A. (2010) Economic Freedom, the Credit Ratings and the Sovereign default spread. Journal of Financial Economic Policy, 2, 149-162. https://doi. org/10.1108/17576381011070201 14. Bundala, N.N. (2013) Do Economic Growth, Human Development and Political Stability Favour Sovereign Creditworthiness of a Country? A Cross Country Survey on Developed and Developing Countries. International Journal of Advances in Management and Economics, 1, 32-46. 15. Williams, G., Alsakka, R. and Gwilym, O. (2013) The Impact of Sovereign Rating Actions on Bank Ratings in Emerging Markets. Journal of Banking & Finance, 37, 563-577. https://doi.org/10.1016/j. jbankfin.2012.09.021 16. Leonard, K.J. (1995) The Development of Credit Scoring Quality Measures for Consumer Credit Applications. International Journal of Quality & Reliability Management, 12, 79-85. https://doi. org/10.1108/02656719510087346 17. Hajek, P. and Michalak, K. (2013) Feature Selection in Corporate Credit Rating Prediction. Knowledge-Based Systems, 51, 72-84. https://doi. org/10.1016/j.knosys.2013.07.008 18. Li, W., He, Q., Fan, X. and Fei, Z. (2012) Evaluation of Driver Fatigue on Two Channels of EEG Data. Neuroscience Letters, 506, 235-239. https://doi.org/10.1016/j.neulet.2011.11.014 19. Zhao, C., Zheng, C., Zhao, M., Tu, Y. and Liu, J. (2011) Multivariate Autoregressive Models and Kernel Learning Algorithms for Classifying Driving Mental Fatigue Based on Electroencephalographic. Expert System with Applications, 38, 1859-1865. https://doi.org/10.1016/j. eswa.2010.07.115 20. Boostani, R., Sadatnezhad, K. and Sabeti, M. (2009) An Efficient Classifier to Diagnose of Schizophrenia Based on the EEG Signals. Expert System with Applications, 36, 6492-6499. https://doi. org/10.1016/j.eswa.2008.07.037 21. Huang, Z., Chen, H., Hsu, C.J., Chen, W.H. and Wu, S. (2004) Credit Rating Analysis with Support Vector Machines and Neural Networks:
The Development of an Alternative Method for the Sovereign ...
22.
23. 24. 25.
26.
27. 28.
29.
30.
31.
32.
213
A Market Comparative Study. Decision Support Systems, 37, 543-558. https://doi.org/10.1016/S0167-9236(03)00086-1 Kim, K.S. and Han, I. (2001) The Cluster-Indexing Method for CaseBased Reasoning Using Self-Organizing Maps and Learning Vector Quantization for Bond Rating Cases. Expert System with Applications, 21, 147-156. https://doi.org/10.1016/S0957-4174(01)00036-7 Gujarati, D.N. (2003) Student Solutions Manual for Use with Basic Econometrics. 4th Edition, McGraw Hill, New York. Hosmer, D.W. and Lemeshow, S. (2004) Applied Logistic Regression. 2th Edition, Wiley, New York. Hand, D.J. and Henley, W.E. (1997) Statistical Classification Methods in Consumer Credit Scoring: A Review. Journal of Royal Statistical Society, 160, 523-541. https://doi.org/10.1111/j.1467985X.1997.00078.x Yu, L., Wang, S. and Lai, K.K. (2008) Forecasting Crude Oil Price with an EMDBased Neural Network Ensemble Learning Paradigm. Energy Economics, 30, 2623- 2635. https://doi.org/10.1016/j. eneco.2008.05.003 Smith, K.A. and Gupta, J.N.D. (2002) Neural Networks in Business: Techniques and Applications. IGI Global, New York. Jardin, P. and Séverin, E. (2012) Forecasting Financial Failure Using a Kohonen Map: A Comparative Study to Improve Model Stability over Time. European Journal of Operational Research, 221, 378-396. https://doi.org/10.1016/j.ejor.2012.04.006 Takagi, T. and Sugeno, M. (1985) Fuzzy Identification of Systems and Its Applications to Modeling and Control. IEEE Transactions on Systems Man & Cybernetics, 15, 116-132. https://doi.org/10.1109/ TSMC.1985.6313399 Jang, J.S.R. (1993) ANFIS: Adaptive-Network-Based Fuzzy Inference system. IEEE Transactions on Systems Man & Cybernetics, 23, 665685. https://doi.org/10.1109/21.256541 Jang, J. (1991) Fuzzy Modeling Using Generalized Neural Networks and Kalman Filter Algorithm. Proceeding of 9th National Conference Artificial Intelligence, 91, 762-767. Jang, J.S.R. (1993) ANFIS: Adaptive-Network-Based Fuzzy Inference System. IEEE Transactions on Systems Man & Cybernetics, 23, 665685. https://doi.org/10.1109/21.256541
214
Soft Computing with NeuroFuzzy Systems
33. Jang, J.S.R. (1996) Input Selection for ANFIS learning. Proceeding IEEE 5th International Fuzzy Systems, 2, 1493-1499. https://doi. org/10.1109/FUZZY.1996.552396 34. Fullér, R. (1995) Neural Fuzzy System. 35. Lin, C.T. and Lee, C.S.G. (1991) Neural-Network-Based Fuzzy Logic Control and Decision System. IEEE Transaction on Computers, 40, 1320-1336. https://doi.org/10.1109/12.106218 36. Leshno, M. and Spector, Y. (1996) Neural Network Prediction Analysis: The Bankruptcy Case. Neurocomputing, 10, 125-147. https://doi. org/10.1016/0925-2312(94)00060-3
SECTION 4
NEURO-FUZZY CONTROLLERS
CHAPTER
14
Implementation of Adaptive Neuro Fuzzy Inference System in Speed Control of Induction Motor Drives
K. Naga Sujatha, K. Vaisakh Department of Electrical Engineering, AU College of Engineering, Andhra University, Visakhapatnam, India.
ABSTRACT A new speed control approach based on the Adaptive Neuro-Fuzzy Inference System (ANFIS) to a closed-loop, variable speed induction motor (IM) drive is proposed in this paper. ANFIS provides a nonlinear modeling of motor drive system and the motor speed can accurately track the reference signal. ANFIS has the advantages of employing expert knowledge from the
Citation: K. Sujatha and K. Vaisakh, “Implementation of Adaptive Neuro Fuzzy Inference System in Speed Control of Induction Motor Drives,” Journal of Intelligent Learning Systems and Applications, Vol. 2 No. 2, 2010, pp. 110-118. doi: 10.4236/ jilsa.2010.22014. Copyright: © 2010 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY). http://creativecommons.org/licenses/by/4.0
218
Soft Computing with NeuroFuzzy Systems
fuzzy inference system and the learning capability of neural networks. The various functional blocks of the system which govern the system behavior for small variations about the operating point are derived, and the transient responses are presented. The proposed (ANFIS) controller is compared with PI controller by computer simulation through the MATLAB/SIMULINK software. The obtained results demonstrate the effectiveness of the proposed control scheme. Keywords:- ANFIS Controller, PI Controller, Fuzzy Logic Controller, Artificial Neural Network Controller, Induction Motor Drive
INTRODUCTION Over the last three decades, variable speed drives are the most complex of all power electronic systems. Drive technology has been a confluence of many professionals from other fields, such as electrical machines, control systems and traditional power engineering. To a traditional power electronics engineer with expertise in the design of, such as thyristor phasecontrolled converters, switching mode power supplies, or uninterruptible power supply systems, the technology is incomprehensible because of its complexity and multidisciplinary characteristics. Modern variable speed drive applications require steeples control and suitable dynamic response and accuracy. These considerations have been met to a large extent in the past decade by thyristor-controlled dc machines. However, the dc machine remains expensive in relation to the types of rotating machines. For the higher power drives in industries, the lighter, less expensive, reliable simple, more robust and commutator less induction motors are desirable and these motors are being applied today to a wider range of applications requiring variable speed. Unfortunately, accurate speed control of such machines by a simple and economical means remains a difficult task. With the development of the siliconcontrolled rectifier, triac and related members of the thyristor family, it has become most feasible to design variable-speed induction motor drives for a wide variety of applications. Different techniques have been used, using SCR controllers. A back-to back connected SCR’ are used in series with the rotor phases to control their effective impedance [1-4]. A chopper-controlled external resistance is used to control the speed by varying the duty cycle of the chopper. A controlled rectifier is used in the rotor circuit to feed the external resistance, and by varying the firing angle, the effective rotor impedance is controlled.
Implementation of Adaptive Neuro Fuzzy Inference System in ...
219
Generally, variable speed drives for Induction Motor (IM) require both wide operating range of speed and fast torque response, regardless of load variations. This leads to more advanced control methods to meet the real demand. Very recently, the artificial intelligence tools, such as expert system, fuzzy logic and neural network are showing impact on variable frequency drives. They are applied to important fields such as variable speed drives, control systems, signal processing, and system modeling. Artificial Intelligent systems, means those systems that are capable of imitating the human reasoning process as well as handling quantitative and qualitative knowledge. It is well known that the intelligent systems, which can provide human like expertise such as domain knowledge, uncertain reasoning, and adaptation to a noisy and time-varying environment, are important in tackling practical computing problems. ANFIS has gain a lot of interest over the last few years as a powerful technique to solve many real world problems. Compared to conventional techniques, they own the capability of solving problems that do not have algorithmic solution. Neural networks and fuzzy logic technique are quite different, and yet with unique capabilities useful in information processing by specifying mathematical relationships among numerous variables in a complex system, performing mappings with degree of imprecision, control of nonlinear system to a degree not possible with conventional linear systems [5-11]. To overcome the drawbacks of Neural networks and fuzzy logic, Adaptive Neuro-Fuzzy Inference System (ANFIS) was proposed in this paper. The ANFIS is, from the topology point of view, an implementation of a representative fuzzy inference system using a Back Propagation neural network structure. The purpose of this paper is to present a general method for estimating both the nature of the dynamic response and the values of the significant parameters and operating constraints of typical induction machines controlled by SCR controllers [12,13]. The dynamic behavior of a closed-loop speedcontrol system with deltaconnected SCR’s in the rotor is discussed. The various functional blocks of the feedback system which governs the system behavior for small variations about the operating point are derived, and responses for speed perturbations are obtained analytically and simulated.
STATE SPACE APPROACH A Set of nonlinear differential equations can describe the behavior of the induction motor [14-16]. If a complete solution of the dynamic behavior of
220
Soft Computing with NeuroFuzzy Systems
the induction machine is desired, these equations must be solved in detail. By linerarizing these questions about a steady state operating condition, the resulting equations in state form can describe the dynamics, and provide the future state and output of the system. Perturbations in reference voltage or firing angle and load torque leads to changes in rotor speed. The analytical results used to investigate these speed changes are obtained considering the various previous functional blocks, where the different input and output variables are denoted by X1, X2, X3 and X4. These variables are defined as follows: (1) The differential equations, which govern the small variations about the operating point, are written in terms of the above variables and representing in matrix form in Equation (2), where
SYSTEM DESCRIPTION The system consists of a slip-ring induction motor with three equal external resistances, each connected to the rotor phase and three delta-connected phase-controlled SCR’s placed at the open star point of the rotor as shown in Figure 1. In variable speed ac induction motor drives, a continuous monitoring or control of slip speed or slip frequency is required. A permanent magnet tachogenerator is mounted on the rotor shaft to provide a dc signal proportional to the rotor speed to the feedback control circuit. The block diagram of the feedback control scheme of the induction motor is shown in Figure 2. The induction motor stator is supplied with constant voltage, constant frequency supply. The rotor speed is controlled and adjusted by advancing or retarding the firing angle α of the SCRs. The tachogenerator output voltage proportional to the rotor speed and is compared with a fixed dc level VR which represents the set speed.
(2)
Implementation of Adaptive Neuro Fuzzy Inference System in ...
221
Figure 1. Schematic diagram of phase controlled SCR’s in delta (Δ) configuration.
Figure 2. Block diagram of feedback system.
The error voltage is forwarded to the controller. The set peed is changed by varying VR automatically or manually. The controller may be a proportional, or proportional integral or proportional integral derivation type. The function of the controller is to give the required control voltage which will adjust the firing angle to the suitable value and can be used also as a stabilizing signal if more than one controller is used. The simulink block diagram of feedback control scheme of the induction motor is shown in Figure 3.
Transfer functions for the functional blocks: The transfer functions for the various functions blocks of the feedback system are shown in Figure 4, and given in details as follows: 1)
Tachogenerator and filter: The transfer function of this block is represented by: (3)
Soft Computing with NeuroFuzzy Systems
222
where K1 is the combined gain of the tachogenerator and the associated filter, and is the effective time constant of the filter. 2)
Controller: The change in the output voltage of the tachogenerator is compared with the reference voltage VR and the resultant error voltage is fed to the controller. The controller output voltage is corrected in accordance with the input change in voltage. The change in the controller output voltage is denoted as Vc. The transfer function of the proportional integral controller is: (4)
3)
Firing Circuit: The firing circuit decides the change in firing angle in accordance with the change in control voltage Vc. It consists of a ramp generator and a comparator. The ramp is synchronized with the signal available across the slip-rings of the machine. For a given change in the control voltage Vc, the change in firing angle is given by: (5)
where m is the slope of the ramp. For the present study, the firing circuit transfer function can be written as (6) where K3 is equal to l/m, and the time constant is equal to one half of the maximum expected delay. If the slip of the rotor at the operating point is s, then the time constant T3 is given by (7)
Figure 3. The simulink block diagram of feedback control scheme of the induction motor drive.
Implementation of Adaptive Neuro Fuzzy Inference System in ...
223
Figure 4. Functional blocks of closed-loop system.
4)
Induction Motor: The torque developed by the machine at a given operating point is a function of speed of the machine and the firing angle of the thyristors. The difference between the developed torque and the load torque is applied to the rotating elements. The torque developed by the machine is presented by (8)
where ω is the rotor speed in rad/sec, and α is the firing angle. For the dynamic behavior of the induction machine about any operating point for a given perturbation, the small change in the developed torque can be represented in terms of the small changes in rotor speed and firing angle as: (9) or (10) The constants K4 and K5 depend upon the operating point and are to be obtained from the steady-state characteristics of the system. The resultant change in the developed torque is represented as the summation of the outputs of the two blocks (4) and (5). The change in the developed torque is compared with the change in load torque and the resultant value is forwarded to the mechanical system, whose transfer function can be expressed as: (11) where F is the frictional constant in N.m/rad/s, and J is the moment of inertia of the rotating system in KG ⋅m2.
224
Soft Computing with NeuroFuzzy Systems
ANFIS BASED SPEED CONTROLLER Artificial Intelligent tools such as Fuzzy Logic and Artificial Neural Networks have shown great potential on variable frequency drives. Artificial Neural Networks are concerned with adaptive learning, nonlinear function approximation, and universal generalization; fuzzy logic with imprecision and approximate reasoning [17,18]. But they share some common shortcomings that hinder them from being used more widely. For example, neural networks, often suffer from a slow learning rate. This drawback renders neural networks less than suitable for time critical applications. Therefore, new and enhanced methods can be put forward. The fuzzy neural network is constructed to merge fuzzy inference mechanism and neural networks into an integrated system so that their individual weaknesses are overcome. The ANFIS system determines a control action by using a neural network which implements a fuzzy inference. In this way, the prior expert’s knowledge can be incorporated easily. The controller has two states, a learning state and a controlling state. In the learning state, the performance evaluation is carried out according to the feedback which represents the process state. If input-output training data is available, the performance can be assessed easily, and supervised learning can be employed.
ADAPTIVE NEURO-FUZZY PRINCIPLE The fuzzy inference commonly used in ANFIS is first order Sugeno fuzzy model because of its simplicity, high interpretability, and computational efficiency, builtin optimal and adaptive techniques. A typical architecture of an ANFIS is as shown in Figure 5. Among many FIS models, the Sugeno fuzzy model is the most widely applied one for its high interpretability and computational efficiency, and built-in optimal and adaptive techniques. For a first order Sugeno fuzzy model, a common rule set with two fuzzy if-then rules can be expressed as:
where Ai and Bi are the fuzzy sets in the antecedent, and pi, qi and ri are the design parameters that are determined during the training process. Layer 1: Every node in this layer contains membership functions. (12)
Implementation of Adaptive Neuro Fuzzy Inference System in ...
225
(13) where
can adopt any fuzzy membership function (MF).
Figure 5. Adaptive neuro fuzzy structure.
Layer 2: This layer chooses the minimum value of two input weights. (14) Layer 3: Every node of these layers calculates the weight, which is normalized. (15) where
is referred to as the normalized firing strengths
Layer 4: This layer includes linear functions, which are functions of the input signals. (16) where is the output of layer 3, and {pi, qi, ri} is the parameter set. The parameters in this layer are referred to as the consequent parameters. Layer 5: This layer sums all the incoming signals. (17) The output z in Figure 5 can be rewritten as:
(18)
226
Soft Computing with NeuroFuzzy Systems
In this paper the normalized membership functions of input variables and output variable are shown in Figures 6 and 7. The Three-dimensional plot of Fuzzy Control surface is shown in Figure 8.
SIMULATION RESULTS In this paper, performance of the proposed ANFIS speed controller is evaluated and is compared with PI controller and without any controller. The controller parameters are chosen to optimize the performance criterion of the dynamic operation, and then the tuning was empirically improved. The simulation is carried out to observe the performance of the system at different load perturbations.
Figure 6. Triangular membership functions for input variables e and ∆e.
Figure 7. Triangular membership functions for output variable.
Figure 8. Three-dimensional plot of control surface.
Implementation of Adaptive Neuro Fuzzy Inference System in ...
227
The software environment used for this simulation is Matlab ver. 7.1, with simulink package.The change in rotor speed is due to the perturbations in reference voltage or firing angle and load torque. The analytical results used to investigate these speed changes are obtained considering the various previous functional blocks, where the different input and output variables are denoted by X1, X2, X3 & X4. The differential equations which govern the small variations about the operating point in terms of above variables are given in Equation (2).The perturbation studies were carried out at different operating points with different system parameters (gains and time constants). Studies are carried out at operating points with various system parameters (gains and time constants). The simulation results give the present perturbation study for step change in the load torque and reference voltage. From the Figures 9 to 11 the starting transients are realized for ANFIS controller at different operating conditions. It can be observed from the figures that the performance of the ANFIS gives better response compared with PI controller and without any controller.
CONCLUSIONS A framework for tuning and self-organizing ANFIS controller has been presented. This approach has been contrasted without any controller and with PI controller. The dynamic behavior of a closed-loop, variable speed induction motor drive which uses three silicon controlled rectifiers has been studied in this paper. Transfer function blocks of the system for small variations about an operating point are derived, and the transient responses with the analytical studies have been carried out. Comparison of ANFIS controller, without any controller and with PI controller under normal operation for a given load torque and reference speed perturbations has been presented.
Figure 9. Variation of speed deviation at 5% load change.
228
Soft Computing with NeuroFuzzy Systems
Figure 10. Variation of speed deviation at 10% load change.
Figure 11. Variation of speed deviation at 15% load change.
It has been demonstrated that the proposed method gives a good response, regardless of parameter variations or external force. Simulation results have shown the capabilities of the proposed controller in tracking predetermined desired speed trajectory.
Implementation of Adaptive Neuro Fuzzy Inference System in ...
229
REFERENCES 1.
R. P. Basu, “A Variable Speed Induction Motor Using Thyristors in the Secondary Circuit,” IEEE Transactions on Parer Apparatus and Systems, Vol. 90, 1971, pp. 509-514. 2. M. Ramamoorthy and M. Arunachalam, “A Solid-State Controller for Slip Ring Induction Motors,” The IEEE Industry Applications Society Annual Meeting, Los Angeles, California, October 2-6, 1977. 3. M. Ramamoorthy and M. Arunachalam, “Dynamic Performance of a Closed Loop Induction Motor Speed Control System with PhaseControlled SCR’s in the Rotor,” IEEE Transactions on Industry Applications, Vol. 15, No. 5, 1979, pp. 489-493. 4. Y. Hsu and W. Chan, “Optimal Variable-Structure Controller for DC Motor Speed Control,” IEEE Proceedings D on Control Theory and Applications, Vol. 131, No. 6, 1984, pp. 233-237. 5. B. S. Zhang and J. M. Edmunds, “On Fuzzy Logic Controllers,” IEEE International Conference on Control, Edinburg, UK, 1991, pp. 961965. 6. H. Ying, W. Siler and J. J. Buckley, “Fuzzy Control Theory: A nonlinear Case,” Automatica, Vol. 26, No. 3, 1990, pp. 513-520. 7. D. Dirankov, H. Hellendorn and M. Reinfrank, “An Introduction to Fuzzy Control,” Springer-Verlag, New York, 1993. 8. M. Maeda and S. Murakami, “A Self-Tuning Fuzzy Controller,” Fuzzy sets and Systems, Vol.51, No. 1, 1992, pp. 29-40. 9. T. J. Procyk and E. H. Mamdani, “A Linguistic SelfOrganizing Process Controller,” Automatica, Vol. 15, No. 1, 1979, pp. 53-65. 10. R. Storn and K. Price, “Differential Evolution-A Simple and Efficient Adaptive Scheme for Global Optimization over Continuous Spaces,” ICSI Technical Report, March 1995. 11. D. Karaboga and S. Okdem, “A Simple and Global Optimization Algorithm for Engineering Problems: Differential Evolution Algorithm,” Turk Journal of Electrical Engineering, Vol. 12, No. 1, 2004, pp. 53-60. 12. D. Borojevic, L. Garces and F. Lee, “Performance Comparison of Variable Structure Controls with PI Control for DC Motor Speed Regulator,” IEEE Industry Applications Conference, 1984, pp. 395405.
230
Soft Computing with NeuroFuzzy Systems
13. J. Zhao and B. K. Bose, “Evaluation of Membership Functions for Fuzzy Logic Controlled Induction Motor Drive,” IEEE 2002 28th annual Conference of the Industrial Electronics Society, Vol. 1, 2002, pp. 229-234. 14. A. S. A. Farag, “State-Space Approach to the Analysis of DC Machines Controlled by SCRs,” IEEE Proceeding Publication-on the Control of Power Systems Conference, Oklahoma, March 10-12, 1976, pp. 157163. 15. N. Mohan, “Electric Drives: An Integrative Approach,” Minnesota Power Electronics Research & Education, Minnesota, 2003. 16. N. Mohan, “Advanced Electric Drives: Analysis, Control and Modeling using Simulink®,” Minnesota Power Electronics Research & Education, Minnesota, 2001. 17. B. K. Bose, “Fuzzy Logic and Neural Network Applications in Power Electronics,” Proceedings of the IEEE, Vol. 82, No. 8, 1994, pp. 13031323. 18. M. G. Simoes and B. K. Bose, “Neural Network Based Estimation of Feedback Signals for Vector Controlled Induction Motor Drive,” IEEE Transactions on Industry Applications, Vol. 31, No. 3, 1995, pp. 620629.
CHAPTER
15
Neuro-Fuzzy Based Interline Power Flow Controller for Real Time Power Flow Control in Multiline Power System
A. Saraswathi1 and S. Sutha2 Department of Electrical and Electronics Engineering, University College of Engineering Villupuram, Anna University, Chennai, India 1
Department of Electrical and Electronics Engineering, University College of Engineering Dindigul, Anna University, Chennai, India 2
ABSTRACT This article investigates the power quality enhancement in power system using one of the most famous series converter based FACTS controller like IPFC (Interline Power Flow Controller) in Power Injection Model (PIM).
Citation: Saraswathi, A. and Sutha, S. (2016), “Neuro-Fuzzy Based Interline Power Flow Controller for Real Time Power Flow Control in Multiline Power System”. Circuits and Systems, 7, 2807-2820. doi: 10.4236/cs.2016.79239. Copyright: © 2016 by authors and Scientific Research Publishing Inc. This
work is licensed under the Creative Commons Attribution International License (CC BY). http://creativecommons.org/licenses/by/4.0
232
Soft Computing with NeuroFuzzy Systems
The parameters of PIM are derived with help of the Newton-Raphson power flow algorithm. In general, a sample test power system without FACTs devices has generated more reactive power, decreased real power, more harmonics, small power factor and poor dynamic performance under line and load variations. In order to improve the real power, compensating the reactive power, proficient power factor and excellent load voltage regulation in the sample test power system, an IPFC is designed. The D-Q technique is utilized here to derive the reference current of the converter and its D.C link capacitor voltage is regulated. Also, the reference voltage of the inverter is arrived by park transformation technique and its load voltage is controlled. Here, a sample 230 KV test power system is taken for study. Further as the conventional PI controllers are designed at one nominal operating point they are not competent to respond satisfactorily in dynamic operating conditions. This can be circumvented by a Fuzzy and Neural network based IPFC and its detailed Simulink model is developed using MATLAB and the overall performance analysis is carried out under different operating state of affairs. Keywords:- FACTS Controller, Fuzzy Logic Controller, Interline Power Flow Controller, Reactive Power Compensation (RPC), Voltage Stability
INTRODUCTION In current scenario, the transmission and distribution networks are densely populated due to the surplus demand of global energy consumption. Constructing a new transmission network becomes a challenging one, because of the environmental impacts related with law and legislative cries. Also, the cost effective technical challenges are presented in deregulation of the power network. In general, the distance between generating units and load centers are far away. For that reasons losses are huge. So as to minimize the power losses and also to ensure highquality power supply up to the end user, concern must be taken. To rectify such troubles, embedding FACTS controllers in power system is play a vital role and may offer proficient control with safe working conditions. Even though FACTS technology (based on thyristor operation) is invented in 1986 by N.G. Hingorani in Electric Power Research Institute (EPRI) USA, and tremendous improvement in high power electronics like IGBT and GTO is achieved, the converter based FACTS technology come to picture after one decade. Inter line power flow controller (IPFC) was proposed by Gyugi in 1998 [1] and its specialty is not only its multi functionality but also its controllability of multi transmission lines in a
Neuro-Fuzzy Based Interline Power Flow Controller for Real Time ...
233
substation. Controlling the converter operation and hence real power flow in specific corridor and reactive power compensation without huge AC capacitor bank and bulk reactor is possible. The controllable converter output voltage is injected into the transmission lines using properly designed series transformer. This is most fortune to emulate capacitive/inductive effect in the transmission lines. Hence only controlling the converter parameters, power flow control in the system is obtained. In the case of deregulation energy market, where transmission lines are utilized up to its thermal limits, this innovative idea is highly appreciable. IPFC may consist of m number of inverters connected in different transmission lines using separate m number of series transformers and share a common DC source. In other words it consists of more than one SSSC with common DC link [2]. In a typical transmission system with the exchange of reactive power between the series converter and the network, control the flow of active power through the transmission line is achieved [3]. In general the stability limit is defined as the maximum power which can flow through a point in the system without causing loss of stability. After a small slow disturbance a system may regain and maintain its operating condition is known as steady state stability and the same for large and sudden disturbance it is known as transient state stability. The latter one is all ways greater than the previous one [4]. Another important stability is dynamic stability and during its study a large disturbance like short circuit/ loss of generation/loss of load is analyses for long period of time like 4 - 10 seconds. Especially, when large rating generators are connected to a long transmission lines, due to its large steady state synchronous reactance, poor load-voltage characteristics may occur and leads to significant drop in synchronizing torque which promote to transient stability related problems. Moreover there is a high degree of uncertainty present in the power system behavior, after and before a fault occurrence [5]. Hongesombut et al. [6] applied phase plane fuzzy logic to control the frequency robustly and to improve the power system stability due to the uncertainty of power produced from the installed wind turbine in the system. In [7] a hybrid algorithm is formed by the contribution of genetic algorithm (GA) and gravitational search algorithm (GSA) for optimal setting of IPFC. Amir Ghorbani et al. first time investigate the effect of the GIPFC on the measured impedance by loss of excitation (LOE) relay of synchronous generation [8]. A new concept of SIPFC is in adopted version of IPFC which eliminates the common dc link is presented in [9]. In [10] security constrained unit commitment
234
Soft Computing with NeuroFuzzy Systems
(SCUC) problem is solved using ABC algorithm incorporating IPFC. The injection model for congestion management and total active power failure minimization in electric power system has been developed [11]. Mustafa et al using intelligent techniques for the selection of parameters of IPFC as an optimization problem and the effectiveness of the IPFC in diminishing transmission power losses is demonstrated [12]. In this paper ATC enhancement of test power system with IPFC is studied with MATLAB software. The reviewed power injection model (PIM) of IPFC is considered for steady state analysis. The paper is organized, in Section 1; the introduction is given in terms of the necessity of FACTS controllers and need of IPFC. Section 2 summarizes the basic concepts, operating principle followed by problem formulation in Section 3. The state space modeling of IPFC is given in Section 4. A standard five bus system is considered as test system for performance analysis. The need of the Fuzzy based controller, fuzzy inference system and its rules are given in Section 5. The neural network based controller for slave converter of IPFC is described in Section 6. Eventually in Section 7 the simulation results with discussion and important conclusions drawn from the simulation studies are given.
WORKING PRINCIPLE OF IPFC An elementary IPFC scheme consists of two back-to-back DC-to-AC converters; each compensating a transmission line by series voltage injection is shown in Figure 1. The reactive power control can be totally independent in each converter whereas the real power flowing into or out of each converter has to be coordinated in such a way that the DC link voltage is kept constant also the overall surplus power from the under-utilized lines can be used by other over loaded lines for real power compensation. The DC to AC converter is basically voltage source converter and can inject a controllable voltage into the transmission line irrespective of the transmission line current. Hence the effective impedance of the transmission line is changed as either inductive or capacitive in nature that is reactive series compensation is obtained [13]. In general the transmission lines are inductive in nature as its resistance is very small compared to its inductive reactance. Hence The real power transfer from the sending end is given by (1)
θ = 90ο.
Neuro-Fuzzy Based Interline Power Flow Controller for Real Time ...
235
where, Vs—is the magnitude of sending end voltage,
Vr—is the magnitude of receiving end voltage,
δ—is the phase difference between sending and receiving end voltages. When series connected converter inject voltage in quadrature with the transmission line current it can emulate either inductive or capacitive reactance in the line. Consider Equation (1) and assume the series converter inject a controllable voltage in the transmission line in such a way to emulate capacitance effect. Hence the net effective reactance of the line is reduced and power transmission capacity is increased. (2) It is clear that for the same values of V and δ the transmittable real power Ps3 is higher than the Ps2. The increase in transfer power is given by (3)
Figure 1. Schematic block diagram of IPFC.
The factor K is known as degree of compensation percentage (%).
PROBLEM FORMATION The basic building block of IPFC is series connected voltage source converter (VSC). In the case of the series connected VSC, the current flowing through the transmission line has an impact on the dc circuit by means of average switching functions. Since under the steady sate condition, the converter at the AC terminals can exchange only reactive power with the system, when neglecting losses the series injected voltage always has to be perpendicular
236
Soft Computing with NeuroFuzzy Systems
to the current flowing through the transmission line. But at wide transmission angles or large line current even a small displacement from the balance position can cause major displacement from the balance position and can cause major changes in the voltage across the dc condenser. This fact has to be taken into consideration when deriving a control algorithm and selecting the size of condenser. The equivalent circuit model of IPFC in simplified form is shown in Figure 2. In case of shunt connected converter in STATCOM there is almost linear dependence between the output voltage amplitude of the shunt VSC and the reactive current component. Also the linearity is maintained between the phase angle and active current component flowing between the shunt compensator and the network [14]. But in the case of series connected voltage source converter this relationship is not clearly evident. For the given transmission angle it is evident from the presented steady state characteristics that the active component of the series current primarily and almost depends on the average switching function in the direction. Another important constrain is the active power invariance of IPFC. That is the sum of real power exchange between IPFC and System is zero with loss less converter valves and no independent energy storage. In general by controlling the injection voltage magnitude and phase angle the power flow control in the transmission line is achieved. The injection of a controlled voltage from converter1 results in exchange of active and reactive power between Convertyer1 and line1. The active power component Psc1 imposes a demand to the DC terminals. But converter2 is forced by the active power invariance constraint. Hence unlike converter1 the operation of converter2 has its freedom degrees reduced; its series voltage V2pq can compensate only partially to its own line and this imposes a restriction mainly on its power flow. Also the converter2 has another task of regenerating the DC link voltage. So the Psc2 component of converter2 is predefined. Under this condition the primary system will have priority over the secondary system in achieving its set point requirements. Hence a decoupling algorithm is used to control the IPFC operation [15]. In this work convert1 is meant for both real and reactive power flow control. For this a fuzzy based controller is designed because it can handle nonlinearity more robust than conventional nonlinear controllers. Converter2 is meant for reactive power compensation and to maintain DC link voltage constant. A Neural network based controller is designed to meet out the converter2 controlling operation. For this a multilayer feed-forward neural network based controller with back propagation learning process is used [16].
Neuro-Fuzzy Based Interline Power Flow Controller for Real Time ...
237
Figure 2. Equivalent circuit model of IPFC.
STATE SPACE MODELING OF IPFC The outputs of two converters are coupled to the transmission lines by using two different series transformer. The injected voltage vector is controllable irrespective of the transmission line current. Hence it can stimulate effective impedance of the transmission lines and power flow control is obtained even power from underutilized lines to heavily loaded lines. The test system considers here is a single machine infinite bus system. In practical the power system is highly interconnected with large number of synchronous generators and huge number of loads. For easy analyzing purpose the power system may be grouped as power exporting area and power importing area. The exporting area simply represented by a single equivalent generator. Similarly the power importing area is represented by an equivalent synchronous motor. Hence a multi machine power system model reduced into two machine problem. It can further reduce to single machine connected to an infinite bus-a constant voltage constant frequency system [17]. The dynamic modeling of IPFC is considered for analysis [18].
(4)
(5)
238
Soft Computing with NeuroFuzzy Systems
(6)
The dynamic model of the power system may be expressed by these nonlinear equations. (7) (8) (9) (10) Then the state model of the power system with IPFC is (11) where X is the state vectors, U is the control vectors, A is the system matrix and B is the control matrix respectively [7]. (12) (13)
(14)
(15)
Neuro-Fuzzy Based Interline Power Flow Controller for Real Time ...
239
Assume that initially the power system is operating under synchronous equilibrium that is the rotor angles of the different machines are fixed one with respect to the rotating synchronous reference frame. Also the system may be in frequency equilibrium state which means that the operating frequency of the power system is constant. An infinite bus is one whose voltage and frequency are unaffected by a disturbance in the machine under consideration. In general synchronous reference frame SRF-PLL uses Park’s transformation as the phase differentiator and hence not able to track instantaneous evolution of the voltage vector when the PLL bandwidth is low. In case of unbalanced condition as components in d and q axis components make it difficult to track phase angle. In this paper a DSRF control which is works well even for unbalanced condition is presented. It is a combination of two SRF-PLLs which can detect the positive and negative sequence components of voltage vector separately [19]. The D-Q control for Power Flow control mode of IPFC is shown in Figure 3. At the receiving end the real and reactive power is sensed and the d and q component of the current vector is calculated. The actual and reference current vectors are compared and the corresponding error signals are processed by the PI controllers. The feed forward cross coupling reduced errors due to unbalanced components in the current vectors. Next the required direct and quadrature component of the series voltage is determined and injected in to the transmission line. The overall simulation circuit of IPFC with test power system is developed. The specifications of the test system are given in Table 1. The controller is support the system under steady state condition by tracking the set reference values as shown in Figure 8 and Figure 9. But during transient condition like sudden load variation the results proven that the needs of efficient controller, as PI parameters are tuned to a particular operating condition.
Figure 3. D-Q control diagram of IPFC.
240
Soft Computing with NeuroFuzzy Systems
Table 1. Specification of the test system.
FUZZY CONTROL SCHEME FOR MASTER CONVERTER OF IPFC Fuzzy logic is introduced by LotfiZadeh in 1965 to process imprecise data and the basic concept is attempted to mimic human control logic. The mathematical modeling of FACTS controller is complex but in case of fuzzy logic controller there is no need of accurate mathematical model of the system going to be controlled and can work with imprecise inputs. Practically it doesn’t need fast processors and it needs less data storage in the form of membership functions and rules than conventional look up table for nonlinear controllers [20]. Power converters are inherently nonlinear. The causes of nonlinearity in the power converters include a variable structure within a single switching period, saturating inductances, voltage clamping etc. Even the FACTS controllers are highly accurate and fast dynamic in nature, they interact with other static controllers always accompanied with non-linearity and uncertainties. For the sake of power quality it is forced to defeat the stability. The Fuzzy Logic Controller (FLC) has many advantages than the conventional PI controllers. It can handle nonlinearity and more robust than conventional nonlinear controllers. It is a controller based on the concepts of fuzzy sets, linguistic variables and approximate reasoning to evaluate rules. The set of rules are represents the control decision mechanisms required to adjust the effect of certain cases. The internal structure of FLC is shown in Figure 4. FLC are range to range and range to point controllers and the input of FLC is membership function and the output also depends on membership function. In conventional controller we have control gain or control loss, which are combination of numerical values. In FLC instead of fixed gain or loss, the updated variables are used in terms of rules and they are linguistic in nature [11]. Such a typical rule can be written as follows: (16)
Neuro-Fuzzy Based Interline Power Flow Controller for Real Time ...
241
where, Ai, Bi and Ci are the labels of linguistic variables of Error (E), Change of Error (CE) and output respectively. E, CE and output represent degree of membership. The internal structure FIS Fuzzy editor with four inputs and two outputs are shown in Figure 5. A basic architecture of FLC contains fuzzification, inference mechanism, fuzzy rule base, and defuzzification. Fuzzification converts binary data into fuzzy data it has two processes that is derive the membership function for input and output variables. Defuzzification process is to derive the desired crisp output value by combining the membership functions with fuzzy rules. It converts fuzzy o/p values to control signals. It derived by two category, one is off line defuzzification, here all input and output membership functions are based on actual experience, i.e., specified application. Online method is real time controllability has higher control accuracy. Fuzzy set allows objects or members to represent a smooth boundary whereas classical represent sharp boundary and the membership function take the value in the interval 0 and 1. It is derived by different methods like triangular waveform, trapezoidal, sigmoidal, guassian, bell etc. In this study the min inference rule and centroid defuzzzification technique have been used [13].
ANN CONTROL SCHEME FOR SLAVE CONVERTER OF IPFC Neural Network (NN) has been widely used to solve complex problems of pattern recognition and they can recognize patterns very fast after they have already been trained. The training process requires a training algorithm, which uses the training data set to adjust the networks weights and bias so as to minimize an error function, such as the mean squared error function (MSE), and try to classify all patterns in the training data set [21]. The internal structure of ANN control is depicted in Figure 6. It is very important to choose a proper algorithm for training a neural network. In this paper Back Propagation (BP) training algorithm is used.
Figure 4. Internal structure of fuzzy logic controller.
242
Soft Computing with NeuroFuzzy Systems
Figure 5. Fuzzy editor with 4 inputs and 2 outputs.
Figure 6. The structure of ANN control.
SIMULATION RESULTS AND DISCUSSION In this section, the circuit model of IPFC with four bus test system is presented. The open loop IPFC circuit model has been developed. Two three phase converters designed and connected back to back each other and their controllers were built in MATLAB/Simulink. The AC side controlled voltage is fed to the transmission lines by using two different series transformer. It is clear that by injecting proper series voltage the impedance of the transmission line is changed either inductive or capacitive effect as shown in Figure 11. The DC link voltage, AC side voltage and current waveforms are shown in Figure 7. The real and reactive power at receiving end is measured. Also at steady sate, the IPFC enhance the real power by reactive power compensation as shown in Figure 8. The IPFC with PI controllers are working well in the limited variation but at higher degree of variation it is failed to hold the stability of the system. To study the performance of
Neuro-Fuzzy Based Interline Power Flow Controller for Real Time ...
243
the test system with IPFC a voltage fluctuation both voltage swell and sag are externally created and the corresponding power flow at four bus are measured. Up to 5% of voltage variation the series voltage compensation is in effective and maintained constant voltage. But at the same time the THD value at load 2 is worse than the uncompensated line at load1, which can be improved by placing a LC filter circuit. The total harmonic distortion is 12.5%, which is greater than IEEE standard value hence a filter is designed to reduce the harmonics and therefore the heating in the injecting transformer is reduced. Further improvement in THD is obtained by introducing a LC filter at the load side as in Figure 9. The impedance response of IPFC both in capacitive and inductive mode is shown in Figure 10. Also, the transmission line where master converter is connected has upper hand than the other one in holding settling point. Figure 11 and Figure 12 show this effect clearly with load variation. This is due to the presence of common interlinking capacitor. The fault clearance time is improved due to the presence of IPFC as depicted in Figure 13. The peak overshoot of the real and reactive power is comparatively reduced as shown in Figure 14. The current through the capacitor and voltage across it is shown in Figure 15. The overall power flow during fluation in the supply side with the presence of IPFC is depicted in bar chart as Figure 16. At 20% change of supply voltage in the positive direction is the worst case compared to the reverse direction that is −20% change of supply voltage.
Figure 7. AC, DC voltage and current of VSC.
244
Soft Computing with NeuroFuzzy Systems
Figure 8. Real and reactive powers.
Figure 9. THD of VSC.
Neuro-Fuzzy Based Interline Power Flow Controller for Real Time ...
Figure 10. Frequency response of IPFC.
Figure 11. P & Q under load variation.
Figure 12. P & Q under load variation.
245
246
Soft Computing with NeuroFuzzy Systems
Figure 13. Line voltage during Fault condition.
Figure 14. Q-response during Fault condition.
CONCLUSION In this paper, a detailed circuit model of two converter IPFC for real power flow control and reactive power compensation with designed controller (like: FLC, ANN, and reference calculation D-Q method) has been successfully demonstrated using MATAB/Simulink software platform. A FLC based converter operation is meant for independent real and reactive power control.
Figure 15. DC link voltage and current.
Neuro-Fuzzy Based Interline Power Flow Controller for Real Time ...
247
Figure 16. Power Flow during fault.
The ANN based converter limited operation is meant for reactive power compensation and to maintain constant DC link voltage. The performance of entire system with IPFC is studied and the simulation results show that using FLC and ANN enhanced the system performance in steady state region and limited in transient state region. In future, the training algorithm is to be modified for ANN and number of rules increased in FLC to meet out the set-back in transient state performance.
248
Soft Computing with NeuroFuzzy Systems
REFERENCES 1.
Gyugyi, L., Sen, K.K. and Schauder, C.D. (1999) The Inter Line Power Flow Controller Concept: A New Approach to Power Flow Management in Transmission Systems. IEEE Transactions on Power Delivery, 14, 1115-1123. http://dx.doi.org/10.1109/61.772382 2. Zhang, J. and Yokoyama, A. (2007) Application of Inter Line Power Flow Controller to ATC Enhancement by Optimal Power Flow Control. IEEE Lausanne Power Tech, Lausanne, 1-5 July 2007, 1226-1231. 3. Papic, I. (2000) Mathematical Analysis of FACTS Devices Based on a Voltage Source Converter Part I: Steady State Operational Characteristics. EPSR, 56, 139-148. 4. Kundur, P., Paserba, J., et al. (2004) Definition and Classification of Power System Stability. IEEE Transaction in Power Systems, 19, 1387-1401. 5. Kumar, A. and Yadav, R.S. (2011) Fault Polerance in Real Time Distributed System. International Journal on Computer Science and Engineering, 3, 933-939. 6. Hongesombut, K., Kerdphol, T. and Weerakamaeng, Y. (2013) Robust Inter Line Power Flow Controller Using Phase Plane Fuzzy Logic Control. 10th International Conference on Electrical Engineering/ Electronics, Computer, Tele communications and Information Technology (ECTI-CON), Krabi, 15-17 May 2013, 1-4. 7. Sai Ram, I. and Amarnath, J. (2013) Optimal Setting of IPFC for Voltage Stability Improvement Using (GA-GSA) Hybrid Algorithm. Nirma University International Conference on Engineering (NUiCONE), Ahmedabad, 28-30 November 2013, 1-6. 8. Ghorbani, A., Soleymani, S. and Mozafari, B. (2015) A PMU-Based LOE Protection of Synchronous Generator in the Presence of GIPFC. IEEE Transactions on Power Delivery, 31, 551-558. 9. Yuan, Z.H., DeHaan, S.W.H. and Ferreira, B. (2008) A New Concept of Exchanging Active Power without Common DC Link for Inter Line Power Flow Controller (S-IPFC). Power and Energy Society General Meeting—Conversion and Delivery of Electrical Energy in the 21st Century, IEEE, 1-7. 10. Sreejith, S., Simon, S.P. and Padhy, N.P. (2013) Estimation of Recovery Cost with the Incorporation of an IPFC in a SCUC Problem. IEEE Power & Energy Society General Meeting, Vancouver, 21-25 July
Neuro-Fuzzy Based Interline Power Flow Controller for Real Time ...
11.
12.
13.
14.
15.
16. 17. 18. 19.
20.
21.
249
2013, 1-5. Naresh, B. and Kummar, J.A. (2013) Performance Evaluation of FACTS Controller for Power Flow Management in Transmission System Using IPFC Applied to Induction Derive. International Journal of Electrical and Electronics Engineering, 4, 2231-5184. Al Khabbaz, M.M. (2014) Transmission Power Loss Reduction Using Intelligent Techniques-Regulated IPFC. IEEE Conference on Energy Conversion (CENCON), Johor Bahru, 13-14 October 2014, 423-428. Saraswathi, A. and Sutha, S. (2015) Fuzzy Logic Controller Based Interline Power Flow Controller and Its Performance Analysis. Applied Mathematical Sciences, 9, 3651-3658. Papic, I. (2000) Analysis of FACTS Devices Based on a Voltage Source Converter Part II: Steady State Operational Characteristics. EPSR, 56, 149-157. Saraswathi, A. and Sutha, S. (2014) Detailed Simulink Modeling and Analysis of Interline Power Flow Controller for Multiline System. ICIRET. Rutkowski, L. (2008) Computational Intelligence Methods and Techniques. Springer-Verlaq. Padiyar, K.R. (2007) FACTS Controllers in Power Transmission and Distribution. New Age International Pvt. Limited, New Delhi. Spandana, M. and Rao, T. (2014) Dual Damping Controller Using Interline Power Flow Controller. IJATCSE, 3, 363- 368. Salem, S. and Sood, V.K. (2007) Simulation and Controller Design of an Interline Power Flow Controller in EMTP RV. International Conference on Power Systems Transients (IPST’07), Lyon, 4-7 June 2007, 1-8. Ross, T.J. (2010) Fuzzy Logic with Engineering Applications. 2nd Edition, Wiley India Pvt. Limited, New Delhi. PMid:20386946 PMCid:PMC2914240 Shahgholilan, G., Mahadavian, M., et al. (2014) Design of a New IPFC-Based Damping Neurocontrol for Enhancing Stability of a Power System Using Particle Swarm Optimization. IJSEE, 3, 73-78.
CHAPTER
16
Controlling Speed of DC Motor with Fuzzy Controller in Comparison with ANFIS Controller
Aisha Jilani1, Sadia Murawwat1 and Syed Omar Jilani2 Electrical Engineering Department, Lahore College for Women University, Lahore, Pakistan 1
2
Electrical Engineering Department, University of Lahore, Lahore, Pakistan
ABSTRACT Machines have served the humanity starting from a simple ceiling fan to higher industrial applications such as lathe drives and conveyor belts. This research work aims at providing an appropriate software based control system because it provides computer featured applications, prevents rapid
Citation: Jilani, A. , Murawwat, S. and Jilani, S. (2015), “Controlling Speed of DC Motor with Fuzzy Controller in Comparison with ANFIS Controller”. Intelligent Control and Automation, 6, 64-74. doi: 10.4236/ica.2015.61008. Copyright: © 2015 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY). http://creativecommons.org/licenses/by/4.0
252
Soft Computing with NeuroFuzzy Systems
signal loss, reduces noise while also significantly improves the steady state and dynamic response of the motor. In this research paper, we have worked on DC motors due to its significant advantages over other types of machine drives. We have first individually studied Fuzzy and ANFIS (Adaptive Neuro-Fuzzy Interference System) controller in controlling speed for a separately excited DC motor. Afterwards, we have analyzed both results to conclude that which technique is better to be adopted for precisely controlling the speed of DC motor. Outcomes from MATLAB fuzzy logic toolbox for simulation of our schematic has been provided in this research work. Our study parameters include input voltage of DC motor, its speed, percentage overshoot and rising time of the output signal. Our proposed research has interpreted the outcomes that ANFIS controller is better than Fuzzy controller because it produces less percentage overshoot and causes less distortion of the output signal as the overshoot percentage of ANFIS controller is 8.2% while that of Fuzzy controller is 14.4%. Keywords:- ANFIS, DC Motor, Fuzzy Logic, Percentage Overshoot, Rising Time
INTRODUCTION Motor drives have been in use for long time and are an efficient way of transferring mechanical energy into de sirable output in industries. Although there are two types of motor drives currently being used in every industry but DC motors can be considered much better than AC motors especially when considering transportation equipment because of their maximum torque producing quality at stalls which is very poor in AC motors. Also energy recovery mechanism observed in DC motors is much better than in AC motors [1]. Moreover dc motors provide a low horsepower rating at a much cheaper rate than AC drives [2]. To achieve maximum productivity, every single thing of a machine should be taken into account and analyzed accordingly. In motor control systems, hundreds of problems are faced such as change in load dynamics. The most important affecting factor will be noise parameter which is too much various and unpredictable affecting the functioning of the machine [3]. Similarly, another main factor is speed which should be monitored constantly according to the requirement for a desirable and reliable output.
A DC motor as the name indicates is a motor initiated usually by direct current and is converted into mechanical energy according to the requirement. DC motors are ruling the world due to their extensive use in modern technologies and in almost every industry such as to operate steel rolling mills, electric screw drivers, sewing machines, hard disk drives, air compressors, reciprocating machine etc. [4].
Controlling Speed of DC Motor with Fuzzy Controller in ...
253
DC motors are generally classified into two types: • Self-excited DC motor; • Separately excited DC motor. The basic difference between the two types is that self-excited DC motor is initiated by its own field circuit while separately excited DC motor is initiated by an external input supply. Our research work aims at speed controlling of separately excited DC motor rather than self-excited DC motor. We need variable speed drives in our everyday industries such as automotive, petrochemical, food and beverage etc. However, position control of machine drive is also important but once a position is adjusted by some mechanism then its need not to be changed accordingly again and again. However, speed of an object needs to be changed as required such as of motor used in blender. Sometimes it is required to blend the mixture at high speed and sometimes at medium or low speed. Therefore, a technique should be devised for variable speed control rather than variable position control. The major reason of working on separately excited DC motor is that initiation of the motor is independent of internal circuitry of the machine. This gives us an advantage of generating output as desired by varying input supplied voltage with accurate and better speed control as compared to self-excited DC motors. Separately excited DC motors are now extensively ruling the industries due to their marvelous inventions such as OEM battery-powered applications separately excited electric golf car dc motor etc. When considering hardware of a separately excited DC motor, speed can be controlled by following methods: • By controlling Armature voltage of the machine [5]; • By adding variable resistance to armature circuit resistance [6]. Research studies have been done on using different controllers to control speed of separately excited DC motor. Several mathematical models have been used to control speed of drive as discussed in [7] [8]. Different types of controllers used are Proportional Integral Controller (PI), Proportional Integral Derivative Controller (PID) etc. The speed of motor is usually dependent on the type of motor used. Usually when a motor starts, it draws a higher current than an expected value due to a static friction associated with the motor. The higher current will always remain proportional to the input voltage. The input voltage verses speed of a specified machine drive has been analyzed carefully and
254
Soft Computing with NeuroFuzzy Systems
corresponding outputs have been studied. Results are then analyzed to conclude that which controller is better. Fuzzy and ANFIS controller are monitored so as to give less percentage overshoot and less rising time to make an efficient system and to minimize distortion. Our research work has aimed on achieving precise and accurate speed control of separately excited DC motor by using Fuzzy Controller and ANFIS Controller. The purpose is to provide a better speed control method by comparative study of two controllers.
BACKGROUND Machines have successfully replaced uncountable human efforts into efficient and reliable output. Both DC and AC machines are equally important suiting to the required application. Several researches have been done on improving reliability and efficiency in machines. They have not discussed that despite of so many advantages of AC machines for why only DC motors speed should be precisely and accurately controlled and how are they better than AC machines [9]-[11]. Similarly, no comparison of separately excited DC motor and self-excited DC motor is shown [12]-[15]. All parameters of DC motors are correlated such as load dynamics, angular machines, speed of drive etc. Angular position can be affected by changes in load and speed until and unless ideal case is assumed. Publication such as [16] has not focused on how angular position will be affected with variable speed of the drive. DC motors speed can be controlled by various methods of which most commonly used is fuzzy controller based on Mamdani and Sugeno systems. [17] [18] have not discussed the reason of using Mamdani system rather than preferring Sugeno system. A feasible, proficient, workable and ultra-efficient system should always be designed for negligible percentage overshoot and minimum rising time. However, lowering of percentage overshoot and minimizing rise time often contradict and is not possible at one time. Publications like [19]-[21] have not shown any conclusion for controlling speed of DC drives by considering percentage overshoot and rising time. To incorporate any method in a practical and systematic way for efficacious, operative and dynamic performance, results should be analyzed on the basis of comparison to conclude which method is better to be implemented. [22] does not contain any comparative technique.
Controlling Speed of DC Motor with Fuzzy Controller in ...
255
METHODOLOGY In this research work, MATLAB Simulink has been used to implement working of controllers. Speed is monitored by using following two types of controllers and then their comparison on various factors is taken into account. • •
Fuzzy Controller; ANFIS Controller.
Implementation of Fuzzy Controller A Fuzzy Logic Controller (FLC) is formed by interpreting the analog or continuous values of 0 and 1 despite analyzing the digital values. FLC basically controls a process by assimilation of expert human knowledge into a pattern containing a relationship between inputs and outputs. Fuzzy control rules (mostly conditional rules) are then applied on the pattern of input and output.
MATLAB Fuzzy Controller Implementation Process First of all type “fuzzy” in command window or have to click on “start” menu and open the “fuzzy editor” from “toolboxes”. Now click “Edit” and add an input from “Add variable”. Now you can select the “sugeno” FIS from “File”. There are two types of Fuzzy logic systems which can be used in control systems: • Mamdani; • Sugeno. Here, we have used Sugeno system in our research work because output membership functions of sugeno shows us either constant or linear result, which is not possible when using mamdani systems. As shown in Figure 1, Fuzzy controller has four main processes to operate: 1. Fuzzification; 2. Fuzzy base rules; 3. Interference engine; 4. Defuzzification. Fuzzification converts our measured data (e.g. speed of car is 15 mph) into rhetorical data (car is moving too slow). Fuzzy base rules define some rules on the information provided by fuzzification. The interference engine provides us appropriate coherence and analysis for an output simulation.
Soft Computing with NeuroFuzzy Systems
256
Defuzzification gives us an output on the basis of defined set of membership function and rules. The complete process is shown in Figure 1. Following graphical tools are used to create, analyze and view output of Fuzzy logic Controller: •
•
•
Fuzzy Inference System (FIS) Editor: It has a command on handling basic issues of the control system such as defining of input and output variables. Fuzzy logic toolbox can hold unlimited amount of inputs but corresponding there will be a huge number of membership functions which will become difficult for us to handle. Membership Function Editor: It defines the appearance and shape of membership function as per input. This is shown in Figure 2 and Figure 3. Rule Editor: It builds and edits the set of rules which is associated with the behavior of the system.
Figure 1. Flow chart of fuzzy controller simulation process. https://www.scirp.org/pdf/ICA_2015012211042821.pdf
Controlling Speed of DC Motor with Fuzzy Controller in ...
257
Figure 2. Overlapping of Input: 1 of DC motor.
Figure 3. Overlapping of Input: 2 of DC motor.
• •
Rule Viewer: It is used to examine and view the controller output on the basis of defined rules. This is shown in Figure 4. Surface Viewer: It generates a 3-D linkage of output associated with the particular number of inputs. Afterwards, a rule viewer for DC motor can be seen in Fuzzy editor, which defines set of rules for Input: 1 ANDed with Input: 2.
Implementation of ANFIS Controller ANFIS is a hybrid network which consists of a combination of two controllers; Fuzzy logic and neural network. These both controllers result in a single entity which enhances the features of controlling machine than using a single controller alone. MATLAB ANFIS Controller Simulation Process ANFIS editor window opens by typing “anfisedit” in MATLAB command window. The complete process is shown in Figure 5. ANFIS GUI involves the following steps: •
Load Data: This will load our previously saved data from .dat extension file. After loading the data, ANFIS editor will be displayed as shown in Figure 6.
258
Soft Computing with NeuroFuzzy Systems
Figure 4. Rule viewer for DC motor.
Figure 5. Flow chart of ANFIS controller simulation process.
Controlling Speed of DC Motor with Fuzzy Controller in ...
259
Figure 6. ANFIS Editor for DC motor.
•
Generate FIS: FIS model can be generated by using any one of the following techniques. 1) Grid partition: It generates data via grid portioning. 2) Sub clustering: It generates data by analyzing the number of clusters in the given set of data. ANFIS structure can be observed clicking option of “Structure” as shown in Figure 7. • Training and validation of FIS: This process trains the FIS model generated, repeats itself until and unless required number of epoch is reached and goal of training error is attained. Put epochs = 25 as given in load data, then “Train now”. “Train now” shows the value of “Epochs error”. This is shown in Figure 8. Testing of the FIS is carried out by clicking “Test now”. “Test now” shows the value of “Average testing error”. This is shown in Figure 9.
MATLAB Simulink Separate responses of DC motor for Fuzzy and ANFIS are then analyzed on MATLAB Simulink as shown in Figure 10 and Figure 11 respectively. Then their comparative interpretations are done. In the circuit shown in Figure 11 the ANFIS file is loaded in ANFIS block which was saved earlier. In ANFIS circuit, system is initialized by taking step input. The circuit has been made steady and stabilized through Gain6 and Gain7. Also specific oscillations during simulation have been adjusted successfully through these gains. The specified blocks of Tapped delay and Gain are employed as the differential circuit. Gain1 and Gain3 will amplify the signal. In Transfer Function block we can insert the transfer function of DC machine. Each Gain is calculated through specified transfer functions which are selected on
260
Soft Computing with NeuroFuzzy Systems
the basis of technique used. That’s why the value of certain gains is different in both the circuits. The comparison of both of these circuits is shown in Figure 12. Tabular analysis of rising time and percentage overshoot is calculated from graphical response of Fuzzy and ANFIS controller. This is shown in Table 1. Fuzzy Controller output response has given us the percentage overshoot of 14.4% which is 6.2% greater than that of ANFIS controller. However, it gives a rising time of 0.072 sec less than that of ANFIS controller which can be counted as a drawback of ANFIS controller.
Figure 7. Structure of DC motor.
Figure 8. Train output for DC motor.
Figure 9. Test output for DC motor.
Controlling Speed of DC Motor with Fuzzy Controller in ...
Figure 10. Fuzzy model of DC motor.
Figure 11. ANFIS model of DC motor.
Figure 12. Output response of fuzzy and ANFIS performance. Table 1. Comparison between Fuzzy and ANFIS performance.
261
262
Soft Computing with NeuroFuzzy Systems
CONCLUSIONS Machine drives work when their speed is controlled precisely. For analysis a lot of issues come across such as changes in load dynamics, variable inputs, noise propagation and certain unknown parameters which result in unpredictable output of machines. Moreover, a reliable load regulating response with low and almost negligible noise propagation is necessary for a proficient system. Our research work has first analyzed Fuzzy and ANFIS controller separately and then we have interpreted both outcomes to show a comparison that which technique should be used in controlling speed of DC motor. A system should always be designed for less percentage overshoot and less rising time. Often there is a contradiction when adjusting percentage overshoot for minimum rising time. Different research work has been done for lessening percentage overshoot and minimizing rising time. [23] has controlled the speed of DC motor by giving a less percentage overshoot of 10.1% and 2.1% at two different reference speeds. The comparative study in this research paper has shown that ANFIS controller is much better than Fuzzy controller as it gives a percentage overshoot of 8.2% than that of fuzzy controller which is 14.4%. Percentage overshoot indicates an outcome when a signal surpasses its steady-state value. ANFIS technique gives a lower percentage overshoot because of phases such as epoch and training involved in its simulation. Training phase repeats itself until and unless minimum error is reached. This minimum error limit reached is synchronized with given value of epoch which gives a low percentage overshoot then Fuzzy technique. However, a minimum adjustment of 0.072 sec rising time has to be made with ANFIS controller but due to less percentage overshoot, it should be considered a more adoptable technique.
Controlling Speed of DC Motor with Fuzzy Controller in ...
263
REFERENCES 1.
2. 3.
4.
5.
6.
7.
8.
9. 10. 11.
12.
13.
Bansal, U.K. and Narvey, R. (2003) Speed Control of DC Motor Using Fuzzy PID Controller. Advance in Electronic and Electric Engineering, 3, 1209-1220. Afrasiabi, N. and Yazdi, M.H. (2013) DC Motor Control Using Chopper. Global Journal of Science, Engineering and Technology, 8, 67-73. Atri, A. and Ilyas, Md. (2012) Speed Control of DC Motor Using Neural Network Configuration. International Journal of Advanced Research in Computer Science and Software Engineering, 2, 209-212. Saleem, F.A. (2013) Dynamic Modeling, Simulation and Control of Electric Machines for Mechatronics Applications. International Journal of Control, Automation and Systems, 1, 30-42. Kushwah, M. and Patra, A. (2014) Tuning PID Controller for Speed Control of DC Motor Using Soft Computing Techniques—A Review. Advance in Electronic and Electric Engineering, 4, 141-148. Waghmare, R.J., Patil, S.B., Shid, U.S. and Siddha, U.Y. (2013) Need of Electronic Starter for DC Motor. International Journal of Science, Engineering and Technology Research, 2, 1203-1206. Addasi, E.S. (2013) Modelling and Simulation of DC-Motor Electric Drive Control System with Variable Moment of Inertia. ACEEE International Journal on Electrical and Power Engineering, 4, 52-57. Gupta, R. and Ruchika (2013) Thyristor Based DC Motor Control with Improved PF & THD. International Journal on Electrical Engineering and Informatics, 5, 519-536. Suradkar, R.P. and Thosar, A.G. (2012) Enhancing the Performance of DC Motor Speed Control Using Fuzzy Logic. IJERT, 1, 103-110. Munde, V. and Jape, V.S. (2013) Fuzzy Logic for Controlling Speed of DC Motor. IJES, 2, 33-39. Salim, Ohri, J. and Naveen (2013) Speed Control of DC Motor using Fuzzy Logic based on LabView. International Journal of Scientific & Research Publications, 3, 1-5. Dewangan, A.K., Shukla, S. and Yadu, V. (2012) Speed Control of a Separately Excited DC Motor Using Fuzzy Logic Control Based on Matlab Simulation Program. International Journal of Scientific & Technology Research, 1, 52-54. Omar, B.A., Haikal, A.Y. and Areed, F.F. (2012) An Adaptive
264
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
Soft Computing with NeuroFuzzy Systems
Neuro-Fuzzy Speed Controller for a Separately Excited DC Motor. International Journal of Computer Applications, 39, 29-37. Saini, M. and Sharma, N. (2012) Speed Control of Separately Excited DC Motor Using Computational Method. International Journal of Engineering Research & Technology (IJERT), 1, 86-93. Kushwah, R. and Wadhwani, S. (2013) Speed Control of Separately Excited DC Motor Using Fuzzy Logic Controller. International Journal of Engineering Trends and Technology (IJETT), 4, 2518-2523. Sailan, K. and Kuhnert, K-D. (2013) DC Motor Angular Position Control Using PID Controller for the Purpose of Controlling the Hydraulic Pump. International Conference on Control, Engineering & Information Technology (CEIT’13), Proceedings Engineering & Technology, 1, 22-26. Thorat, A.A., Yadav, S. and Patil, S.S. (2013) Implementation of Fuzzy Logic System for DC Motor Speed Control Using Microcontroller. International Journal of Engineering Research and Applications (IJERA), 3, 950-956. Ahmed, H., Singh, G., Bhardwaj, V., Saurav, S. and Agarwal, S. (2013) Controlling of DC Motor Using Fuzzy Logic Controller. Conference on Advances in Communication and Control Systems 2013 (CAC2S 2013), DIT University, 6-8 April 2013, 666-670. Dechrit, M., Benchalak, M. and Petrus, S. (2011) Wheelchair Stabilizing by Controlling the Speed Control of Its DC Motor. World Academy of Science, Engineering and Technology, 310-314. Gupta, R., Lamba, R. and Padhee, S. (2012) Thyristor BasedSpeed Control Techniques of DC Motor: A Comparative Analysis. International Journal of Scientific and Research Publications, 2, 14-18. Meha, S.A., Haziri, B., Gashi, L.N. and Fejzullahu, B. (2011) Controlling DC Motor Speed Using PWM from C# Windows Application. 15th International Research/Expert Conference Trends in the Development of Machinery and Associated Technology TMT 2011, Prague, 12-18 September 2011. Singhal, R., Padhee, S. and Kaur, G. (2012) Design of Fractional Order PID Controller for Speed Control of DC Motor. International Journal of Scientific and Research Publications, 2, 1-8. Sadiq, A., Mamman, H.B. and Ahmed, M. (2013) Field Current Speed Control of Direct Current Motor Using Fuzzy Logic Technique. International Journal of Information and Computation Technology, 3, 751-756.
CHAPTER
17
A Neuro-Fuzzy Controller for Collaborative Applications in Robotics Using LabVIEW
Hiram E. Ponce, Dejanira Araiza, and Pedro Ponce Escuela de Graduados en Ingeniería y Arquitectura, División de Ingeniería y Arquitectura , Instituto Tecnológico y de Estudios Superiores de Monterrey, Campus Ciudad de México, Mexico City 14380, Mexico
ABSTRACT A neuro-fuzzy controller was designed and implemented using LabVIEW over a mobile robotic platform. The controller is based on fuzzy clusters, neural networks, and search techniques. Also, wireless communication with
Citation: Morabito F., Ponce H. E., Araiza D., Ponce P., “A Neuro-Fuzzy Controller for Collaborative Applications in Robotics Using LabVIEW”, Journal on Applied Computational Intelligence and Soft Computing, Volume 2009 , Article ID 657095, https://doi. org/10.1155/2009/657095. Copyright: © 2009 by authors and Hindawi. This work is licensed
under the Creative Commons Attribution International License. http:// creativecommons.org/licenses/by/4.0
266
Soft Computing with NeuroFuzzy Systems
Bluetooth protocol was used to communicate the robot with the controller running in LabVIEW, allowing a simple collaborative task that consisted in pick and place objects, through knowing the position of the robot and measuring the distance to the objects. The neuro-fuzzy controller was split in two parts: the position controller and the evasion controller against collisions.
INTRODUCTION Many problems that affect society can be solved using software and hardware that includes powerful tools from artificial intelligence. For example, determining the amount of money necessary in a specific bank according to the demand, having a database to know the most common sicknesses in Mexican children to have enough medicine and institutions to provide health care assistance, making decisions in small businesses, all this means the possibility to use technological resources to reach development and improvement of life’s quality among people in different countries [1]. For all the mentioned above, the generation of generic and global tools to develop algorithms and programs is a challenging area that deserves the attention of the academic community, as more trained professionals are needed in order to develop and use the tools to improve situations and to solve problems. This project is focused in using modern tools like fuzzy control, neural networks, computational search, wireless communication, and others, to generate material and ideas for implementing solutions. The results of it can be extrapolated beyond the task that was selected to develop the collaborative work, but can be done in any other field as examples described in [2, 3]. The main objective of this project includes implementing a collaborative task in a group of microrobots, implicating autonomous navigation in a determined area, through the use and development of algorithms. In order to achieve this, the work was divided in three goals: designing and implementing a neuro-fuzzy controller based on previous works [4, 5] to get the autonomous navigation inside a determined area and implementing the collaborative work. Additionally, some specific objectives were stated: using LabVIEW [6] to design and implement the algorithms to evaluate their use, designing a
A Neuro-Fuzzy Controller for Collaborative Applications in ...
267
collaborative task for the micro robots used in the project, and generating a new structural design, cheap and flexible, for future projects. The first steps in the project implicated research about the theory that sustains this work, mainly about fuzzy controllers, neural networks, fuzzy clusters, and search techniques. Subsequently, the efforts were focused in designing the neuro-fuzzy controller [7], including a model of the system to validate the controller, and making the proper adjustments to the physical structure of the robots. Finally, implementation of intercommunication was achieved using Bluetooth technology, and the collaborative task was designed, followed by several tests and measurements to validate the work.
PROTOTYPE DESCRIPTION Robots used in this project contemplate a 15 × 15 cm design, with a plastic base, two differential modified servomotors, a simple wheel for support, and different boards including a microprocessor Basic Stamp 2, three Ping ultrasonic range sensors for navigation place them one in front and the other two at left and right sides, and an Embedded Bluetooth Transreceiver for wireless communication, on each robot. This system also includes a magnet for the collaborative task. In order to control the avoidance of obstacles and the movement of the robot, a neuro-fuzzy controller was implemented to run over the LabVIEW platform [6], in communication with the robots through Bluetooth protocol, using a Belkin USB-Bluetooth adapter.
NEURO-FUZZY CONTROLLER A neuro-fuzzy controller is used in robots in order to obtain the desired movements on them, that is, reaching a final position getting from an initial position. Figure 1 is a block diagram of the neuro-fuzzy controller proposed. It is based on trigonometric series [4] and an interface to communicate the robot with the processing system. In fact, a collaborative work has to be in a real-time-based processing. Then, trigonometric artificial neural networks [4] are implemented because their training phase only requires one epoch. In addition, this controller allows obstacles avoidance and adaptation of parameters according to interactions between the robot and its environment.
268
Soft Computing with NeuroFuzzy Systems
Figure 1. Proposed neuro-fuzzy controller [4].
The robot is a MIMO system in which we have to control the rotation of the two wheels via two digital pulses coming from the microcontroller. So, it can be divided in two neuro-fuzzy controllers to simplify the designing of the fuzzy controller part: one that avoids collision against obstacles, and one to control the position through the motion in the servos, considering a position as the set point.
Neural Networks Based on the definition of a fuzzy controller and its parts, and referring to Sugeno approach in the inference mechanism [8], neural networks can be applied to the clusters obtained through the Fuzzy Cluster Means (FCM) method. Fuzzy inferences of the form if_then, and a neural network in the output, can be expressed like (1) where, yi = gi can be a group of trigonometric functions [4], considering that Fourier series can be used to model the input in a neural network, and this method is useful to obtain the coefficients of the weights [9]. Considering a Fourier series defined with (2) and similarities which describes neural networks as (3) it makes evident the possibility of its description, where, coefficients of (2) are the Fourier Coefficients: (2) (3)
A Neuro-Fuzzy Controller for Collaborative Applications in ...
269
The topology includes two layers of neurons, the first composed by neurons with a trigonometric activation function and a weight that depends directly on the frequency. The second layer adds the trigonometric functions, multiplied by their respective weights plus a constant, as Figure 2 shows [9]. The coefficients or weights in the neural network can be calculated considering if the signal is even or odd. A minimum squares method is used to find the coefficients analytically, through (4) converted in a matrix [9]:
(4)
Figure 2. Topology of a trigonometric artificial neural network [9].
Fuzzy Cluster Means In the inputs, it has been proved that better results are obtained using the method called Fuzzy Cluster Means to adapt and tune membership functions according to the environment surrounding a robot. This method allows the minimization of the distance between the elements of each cluster and the maximization of the distances that separate the centroids in the clusters. This can be defined by the objective function in (5). Neural networks based on trigonometric series can be applied after obtaining the clusters, to soften the shapes of the membership functions, and to approach them to trigonometric series [4]: (5)
270
Soft Computing with NeuroFuzzy Systems
where μx,i is the membership value of the element x = {1, 2, ... , N} in the fuzzy cluster i = {1, 2, ... ,c}; vi is the centroid in each cluster; zx, with x = {1, 2, ... , N}, is the group of data; m is the fuzzyfication value; d2(zx, vi) is the Euclidian distance between zx and vi; N is the number of samples in data; c is the number of clusters [10].
Tabu Search Method The Tabu search method is a heuristic way to find a best solution combining several strategies: descendent method, long and short terms memory, and diversification strategies. As the search is happening, some of all the possible solutions are named as “taboo” and included in a list through the memories, during a defined number of iterations. These solutions will not be visited in some points of the search; meanwhile better solutions are found to replace an initial proposed solution [11]. An algorithm to find a better set of fuzzy rules than the proposed ones was developed and programmed, and it is included in Figure 3.
Design of Each Controller The procedure of designing is described in Figures 4 and 5, for both the position controller and the evasion one.
Figure 3: Tabu search algorithm.
A Neuro-Fuzzy Controller for Collaborative Applications in ...
271
The final neuro-fuzzy controller can be explained as following: (i) Two neuro-fuzzy controllers run parallel: one for avoiding obstacles and one for correcting the position of robots. (ii) Each neuro-fuzzy controller has the structure of Figure 1, in which inputs are fuzzified by memberships obtained from a fuzzy clustering determined by the environment and trigonometric neural networks. Then, a set of optimal rules obtained by the Tabu search method are evaluated in order to obtain a desired fuzzy output. Finally, outputs are computed by trigonometric neural networks. As seen, membership functions and rules are changing depending on the environment, given to robots the characteristic of adaptation. (iii) Avoidance obstacles neuro-fuzzy controller uses the left, right, and center measured distances between the robot and the nearest obstacle. Outputs are the pulses for correcting the position of the robot generated by some obstacle. (iv) Position neuro-fuzzy controller uses the final position as inputs and outputs are the pulses for correcting the position of the robot. (v) Avoidance obstacle controller has the highest priority
Figure 4: Design of the position controller.
Figure 5: Design of the evasion controller.
272
Soft Computing with NeuroFuzzy Systems
Figure 6: Fixed parameters for the collaborative task
COLLABORATIVE TASK The general idea of the collaborative task selected implies the detection and collection of small objects through the use of basic sensors–ultrasonic and infrared sensors. After localizing and collecting the objects, the robots move to deposit them in a certain place, using intercommunication between the computer and the robots to control the movements. Certain physical parameters have been stated for the collaborative task: a plane surface with 2 × 2 m per side, 15 cm tall walls around the surface, five spherical objects with metallic incrustations, and half illuminated space. Other specific parameters are 10 cm minimum of separation between each object, with a predetermined distribution for the objects and also for the robots. This is illustrated in Figure 6. The proposed algorithm for the scenario is shown in a schematic way in Figure 7. The initial position of each robot is called home, and it is the place where the objects will be deposited after the recollection. In particular, the algorithm was proved with two robots. Actually, scaling the number of robots is also possible with the implementation of this neurofuzzy controller.
RESULTS Neuro-Fuzzy Controller The main result is the generic neuro-fuzzy controller. It was designed based on a Sugeno approach for generic fuzzy controllers, and it was programmed on LabVIEW [6] as the frame to create a generic method that allows to
A Neuro-Fuzzy Controller for Collaborative Applications in ...
273
create any kind of neuro-fuzzy controller only changing basic parameters like the number of membership functions. This generic controller allows the scalability on the number of robots in the collaborative task. The total controller implemented on LabVIEW is shown in Figure 8.
Position Controller With all programs ready in LabVIEW [6], several tests were made to verify the position controller error in different trajectories and also to prove the right behavior of the evasion controller using the most common obstacle cases in the proposed scenario. For the position controller, a linear desired position results in the trajectory shown in Figure 9, with an error of 5.46%. On the other hand, a diagonal desired position has the trajectory of Figure 10 and an error of 6.2%. The set of fuzzy rules proposed was verified using Tabu search method, with this set of rules and other selected randomly, with the results shown in Tables 1 and 2 for both cases, respectively.
Mathematical Model of the Plant In order to finish the design and adjustment of the position control, a mathematical model from the navigation behavior was needed. For this, the servomotors were characterized first, and then a model that predicts the position according to the real motion of the servos was developed. This model describes the relation between the pulses applied to the servos and the displacement considering the direction of the turn. Characterization of servomotors was done.
Figure 7: Proposed algorithm for the simple collaborative task.
274
Soft Computing with NeuroFuzzy Systems
The result of the measures is shown in Figure 11, and this can be expressed as(6). (6) Define the input variable u = [u1 u2]T as the vector of machine times that are needed to move the wheels, and the output variable y = [y1 y2 y3]T as the position vector in the 2D navigation plane, with direction angle extension respecting from a coordinated plane M in R2 defined in the initial states. In general, the action of the plan can be divided in two main blocks: the angular displacement model and the placement in the M plane, as Figure 12 shows. The angular displacement model considers an input variable u = [u1 u2] , which contains the information about the activation time in the rotation of each wheel, u1 is the activation time in the left wheel, and u2 represents the activation time in the right wheel. The output variable yd = θR in this block is the relative angular displacement of plane R in R2, with its center in the origin GM, mapped referring to the robot itself. This block determines the relative angular displacement according to the previous location and the new position after the movement (Figure 11). T
A series of criteria was determined to delimitate the angular displacement model. (a) The robot can move forward or backward and the vector u has its elements with the same value. (b) The robot can turn to the left, and . (c) The robot can turn to the right, and . (d) The robot can remain static and the vector u has 0 values. (e) A movement forward takes place if the vector is positive. (f) A movement backward takes place if the vector is negative.
Figure 8: Controller developed in LabVIEW (Block diagram).
A Neuro-Fuzzy Controller for Collaborative Applications in ...
Figure 9: Trajectory for a linear desired position.
Figure 10: Trajectory for a diagonal desired position.
Figure 11: Measurement of angles in the servomotor.
Figure 12: Block diagram of the model.
275
276
Soft Computing with NeuroFuzzy Systems
Table 1: Tabu method results for the proposed FAM.
Table 2: Tabu method results for the random FAM.
A Neuro-Fuzzy Controller for Collaborative Applications in ...
277
Figure 13: Angular displacement model alter applying a group u of pulses to the wheels. Point GM represents the point of interest in the robot, and plane R is defined from it. (a) Plane R before the pulse input, (b) plane R after the movement produced by the pulses.
The absolute displacement depends on the actual position in plane M; so a point GM is defined as the origin of Euclidian plane R (Figure 14). Point GM has the direction value measured in plane M. Mathematically, GM is a position vector (7): (7)
4The transformation between plane R and plane M is done using homogeneous matrices [12, 13], where the mapping ϕ : R → M starts with (8): (8)
(9)
The coordinates in ϕ correspond to the new position in plane M. The new angle of direction is (10):
(10)
278
Soft Computing with NeuroFuzzy Systems
Figure 14: Relation between plane M and plane R, both in R2.
Figure 15: Different cases for the evasion controller evaluation.
The output vector is defined as y = [y1 y2 y3]T, where the values of its element imply (11):
(11) Twenty observations were taken to validate the proposed model.
Evasion Controller Four different cases were tested to verify the evasion against obstacles, which are shown in Figure 15 together with the trajectory followed by the robot on each case. The evaluation of this controlleris acceptable.
Wireless Communication In order to send and receive the numerical data from the microcontroller to the computer and back, a basic protocol was designed. This protocol sent characters such as “go” or “hi” as the control signal, and then the respective digits of the correction signal or the feedback.
A Neuro-Fuzzy Controller for Collaborative Applications in ...
279
To establish and implement the communication through Bluetooth, a routine was designed based on predetermined blocks including exploration to find the device, sending, and receiving.
Designing a New Robot Searching for a more flexible and less expensive platform, a new structure was designed and implemented as shown in Figure 16, using plastics and polymers to substitute aluminum and other heavier materials. The saving was of 16% in costs and 10% in weight.
Figure 16: Implementation of the new structural design.
Figure 17: Trajectories followed by the two robots during the collaborative task.
Figure 18: A robot taking an object with the magneto.
280
Soft Computing with NeuroFuzzy Systems
Collaborative Task Several experiments were made to observe the performance of the proposed system during the collaborative task, and the results of the trajectories followed by the robots are shown in Figure 17, according to the algorithm proposed previously. The total time was 17 minutes and 20 seconds. According to the results, time might be enhanced by proving other communication protocols and removing the master station. In that way, robots have to have a better microcontroller that allows programming the entire neuro-fuzzy controller inany one. On the other hand, Figure 18 shows how a robot is catching an object with the magneto at the front part in this collaborative task. However, the neuro-fuzzy controller has its main characteristic that can be used in other tasks different from that one, because of its inherent adaptation to the environment.
CONCLUSIONS A neural fuzzy-controller topology was proposed to control the autonomous navigation of a robot, and it was implemented in LabVIEW [6] to be tested in a collaborative task. The controller acts as expected, collecting the objects and evading obstacles. The proposed blocks of algorithms allowed adjusting parameters as many times as it was necessary, and the whole methodology gave a possibility to extrapolate the results to other situations and tasks. The final prototype of the generic neuro-fuzzy controller was programming on LabVIEW, and it can be extended to other situations [2, 3].
A Neuro-Fuzzy Controller for Collaborative Applications in ...
281
REFERENCES 1. 2.
3.
4.
5.
6. 7.
8. 9. 10. 11. 12. 13.
L. A. García Fernández, “Usos y aplicaciones de la inteligencia artificial,” La Ciencia y el Hombre, vol. 17, no. 3, p. 1, 2004.View at: R. Ranjan, A. Awasthi, N. Aggarawal, and J. Gulati, “Applications of fuzzy and neuro-fuzzy in biomedical health sciences,” in IEEE International Conference on Electro Information Technology, pp. 60– 65, East Lansing, Mich, USA, 2006. P. P. Bhogle, B. M. Patre, L. M. Waghmare, and V. M. Panchade, “Neuro fuzzy temperature controller,” in Proceedings of the IEEE International Conference on Mechatronics and Automation (ICMA ‘07), pp. 3344–3348, 2007. P. P. Cruz et al., “A Novel Neuro-Fuzzy Controller Based on Both Trigonometric Series and Fuzzy Clusters,” Tecnológico de Monterrey, México, 2006. D. Méndez and F. D. Ramírez, “Sistema de Navegación Neuro-Difuso para Robots Móviles,” Instituto Tecnológico y de Estudios Superiores de Monterrey, Campus Ciudad de México, 2007. National Instruments. LabVIEW 8.5, Programming software. United States of America, 2007. K. S. Rattan and G. S. Sandhu, “Design of a proportional plus derivative neuro fuzzy controller,” in Proceedings of 18th International Annual Conference of the North American Fuzzy Information Processing Society (NAFIPS ‘99), pp. 874–878, New York, NY, USA, 1999. H. Nguyen et al., A First Course in Fuzzy and Neural Control, Chapman & Hall, Boca Raton, Fla, USA, 1st edition, 2002. P. P. Cruz and R. S. M. Suárez, “Neural Networks Based on Fourier Series,” Tecnológico de Monterrey, México, 2005. J. C. Bezdek and S. K. Pal, Fuzzy Models for Pattern Recognition, Institute of Electrical & Electronics Enginee, New York, NY, USA, 1992. F. Glover and M. Laguna, Tabu Search, Kluwer Academic Publishers, Boston, Mass, USA, 1997. N. Efimov, Formas cuadráticas y matrices, Mir Editions, Moscou, Russia, 1970. Departamento de Arquitectura y Tecnología de Computadoras, “Matrices homogéneas,” Power Point presentation. Universidad de Sevilla, Spain.
CHAPTER
18
An Adaptive Fuzzy Sliding Mode Control Scheme for Robotic Systems
Abdel Badie Sharkawy Shaaban Ali Salman Mechanical Engineering Department, Faculty of Engineering, Assiut University, Assiut, Egypt
ABSTRACT In this article, an adaptive fuzzy sliding mode control (AFSMC) scheme is derived for robotic systems. In the AFSMC design, the sliding mode control (SMC) concept is combined with fuzzy control strategy to obtain a model-free fuzzy sliding mode control. The equivalent controller has been substituted for by a fuzzy system and the uncertainties are estimated on-
Citation: A. Sharkawy and S. Salman, “An Adaptive Fuzzy Sliding Mode Control Scheme for Robotic Systems,” Intelligent Control and Automation, Vol. 2 No. 4, 2011, pp. 299-309. doi: 10.4236/ica.2011.24035. Copyright: © 2011 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY). http://creativecommons.org/licenses/by/4.0
284
Soft Computing with NeuroFuzzy Systems
line. The approach of the AFSMC has the learning ability to generate the fuzzy control actions and adaptively compensates for the uncertainties. Despite the high nonlinearity and coupling effects, the control input of the proposed control algorithm has been decoupled leading to a simplified control mechanism for robotic systems. Simulations have been carried out on a two link planar robot. Results show the effectiveness of the proposed control system. Keywords:- Sliding Mode Control (SMC), Adaptive Fuzzy Sliding Mode Control (AFSMC), Fuzzy Logic Control (FLC), Adaptive Laws, Robotic Control
INTRODUCTION Performance of many tracking control systems is limited by variation of parameters and disturbances. This specially applies for direct drive robots with highly nonlinear dynamics and model uncertainties. Payload changes and/or its exact position in the end effector are examples of uncertainties. The control methodologies that can be used are ranging from classical adaptive control and robust control to the new methods that usually combine good properties of the classical control schemes to fuzzy [1,2], genetic algorithms [3], neuro-fuzzy [4,5] and neural network [6] based approaches. Classical adaptive control of manipulators requires a precise mathematical model of the system’s dynamics and the property of linear parameterization of the system’s uncertain physical parameters [7]. The study of output tracking problems has a longstanding history. Sliding mode control (SMC) is often favored basic control approach, because of the insensitivity to parametric uncertainties and external disturbances [7-10]. The theory is based on the concept of changing the structure of the controller to achieve a desired response of the system. By using a variable high speed switching feedback gain, the trajectory of the system can be forced on a chosen manifold, which is called sliding surfaces or switching surfaces, and remains thereafter. The design of proper switching surfaces to obtain the desired performance of the system is very important and has been the topic of many previous works [11,12]. With the desired switching surface, we need to design a SMC such that any state outside the switching surface can be driven to the switching surface in finite time. Generally, in the SMC design, the uncertainties are assumed to be bounded. This assumption may be reasonable for external disturbance, but it is rather restrictive as far as unmodelled dynamics are concerned.
An Adaptive Fuzzy Sliding Mode Control Scheme for Robotic ...
285
Nowadays, fuzzy logic control (FLC) systems have been proved to be able to solve complex nonlinear control problems. They provide an effective means to capture the approximate nature of real world. Examples are numerous; see [13] for instance. While non-adaptive fuzzy control has proven its value in some applications [1,2,14], it is sometimes difficult to specify the rule base for some plants, or the need could arise to tune the rule-base parameters if the plant changes. This provides the motivation for adaptive fuzzy control, where the focus is on the automatic on-line synthesis and tuning of fuzzy controller parameters. It means the use of on-line data to continually “learn” the fuzzy controller, which will ensure that the performance objectives are met. This concept has proved to be a promising approach for solving complex nonlinear control problems [15,16]. Recently, adaptive fuzzy sliding mode control design has drawn much attention of many researchers. Because, control chattering, an inherent problem associated with SMC, can evoke un-modeled and undesired high frequency dynamics, Ho et al. [17] have proposed an adaptive fuzzy sliding mode control with chattering elimination for nonlinear SISO systems. The adaptive laws, however, rely on the projection algorithms, which can hardly be satisfied in practical problems. In [18], the authors have established an adaptive sliding controller design based on T-S fuzzy system models. The fuzzy system used is rather complicated and the upper bound of the uncertainty is needed to synthesize the controller. A robust fuzzy tracking controller for robotic manipulator which uses sliding surfaces in the control context can be found in [19]. The control scheme, however, depends heavily on the properties of the dynamic model of robotic manipulators and similar to [17], the authors use the projection algorithms which have practical limitations. More recently, Li and Huang [20] have designed a MIMO adaptive fuzzy terminal sliding mode controller for robotic manipulators. In the first phase of their work, the fuzzy control part relied on some expert knowledge and a trial-and-error procedure is needed to determine the output singletons. In the second phase, they designed an adaptive control scheme that determines these parameters on-line. The rule base, however is restricted to five rules per each joint and the fuzzy singletons should have values within specified ranges to enforce stability. In this work, an adaptive fuzzy sliding mode control (AFSMC) scheme is proposed for robotic systems. The scheme is based on the universal approximation property of fuzzy systems and the powerfulness
286
Soft Computing with NeuroFuzzy Systems
of SMC theory. A one dimensional adaptive FLC is designed to generate the appropriate control actions so that the system’s trajectories stick to the sliding surfaces. Adaptive control laws are developed to determine the fuzzy rule base and the uncertainties. With respect to SMC, the proposed algorithm eliminates the usual assumptions needed to synthesize the SMC and better performance can be achieved. The paper is organized as follows. In Section 2, the equivalent control method is used to derive a SMC for rigid robots. Section 3 introduces the proposed AFSMC which is a model free approach. Simulation results which include comparison between AFSMC and SMC are presented in Section 4. Section 5 offers our concluding remarks.
SLIDING MODE CONTROL (SMC) Design In this Section, the well-developed literature is used to demonstrate the main features and assumptions needed to synthesis a SMC for robotic systems. SMC employs a discontinuous control effort to derive the system trajectories toward a sliding surface, and then switching on that surface. Then, it will gradually approach the control objective, the origin of the phase plane. To this end, consider a general n-link robot arm, which takes into account the friction forces, unmodeled dynamics, and disturbances, with the equation of motion given by
(1)
where x ∈Rn joint angular position vector of the robot; τ ∈ Rn applied joint torques (or forces);
M(x) Rn*n inertia matrix, positive definite;
C(x, x) x ∈ R n effect of Coriolis and centrifugal forces;
G(x)∈Rn gravitational torques; Fd∈Rn*n diagonal matrix of viscous and/or dynamic friction coefficient;
Fs(x) ∈Rn vector of unstructured friction effects and static friction terms;
Td∈Rn vector of generalized input due to disturbances or unmodeled dynamics.
The controller design problem is as follows. Given the desired trajectories with some (or all) system parameters being unknown, derive a
An Adaptive Fuzzy Sliding Mode Control Scheme for Robotic ...
287
control law for the torque (or force) input τ (t) such that the position vector
x and the velocity vector x can track the desired trajectories, if not exactly then closely. For simplicity, let (1) rewritten as: (2) where the vector The following assumptions are needed to synthesis a SMC: Assumption 1: The matrix M(x) is bounded by a known positive definite ˆ matrix M(x) .
Assumption 2: There exists a known estimate fˆ (x, x) for the vector
function f (x, x) in (2). The tracking control problem is to force the state vector to follow desired state trajectories . Let xd(t) be the tracking error vector. Further, let us define the linear time-varying surface s(t) [21],
(3)
where is a time varying linear function. Thus from (2) and (3), we can get the equivalent control (also called ideal controller): (4) where is equivalently the average value of τ(t) which maintains the system’s trajectories (i.e. tracking errors) on the sliding surface s(t)=0. To ensure that they attain the sliding surface in a finite time and thereafter maintains the error e(t) on the sliding manifold, generally the control torque e(t) consists of a low frequency (average) component τeq(t) and a hitting (high frequency) component τht as follows (5) The role of τht (t) acts to overcome the effects of the uncertainties and bend the entire system trajectories toward the sliding surface until sliding mode occurs. The hitting controller τht (t) is taken as [8,21] (6)
288
Soft Computing with NeuroFuzzy Systems
Where,
, and
To verify the control stability, let us first get an expression for (3)-(5), the first derivative of (3) is:
. Using
(7) Choosing a Lyapunov function (8) and differentiating using (6) and (7), we obtain:
(9) which provides an exponentially stable system. Since the parameters of (2) depend on the manipulator structure and payload it carries, it is difficult to obtain completely accurate values for these parameters. In SMC theory, estimated values are usually used in the control context instead of the exact parameters. So that (4) can be written as: (10) where are bounded estimates for M(x), and respectively. As mentioned earlier in Assumption 1 and 2, they are assumed to be known in advance. In sliding mode, the system trajectories are governed by [9]: (11) So that, the error dynamics are determined by the function β (t). If coefficients of β(t) were chosen to correspond to the coefficients of a Hurwitz polynomial, it is thus implying that . This suggests β (t) taking the following form: (12) So that, in a sliding manifold, the error dynamics is:
An Adaptive Fuzzy Sliding Mode Control Scheme for Robotic ...
289
(13) and the desired performance is governed by the coefficients c1 and c2
In summary, the sliding mode control in (5), (6) and (10) can guarantee the stability in the Lyapunov sense even under parameter variations. As a result, the system trajectories are confining to the time varying surfaces (3). With this in hand, the error dynamics is decoupled i.e. each degree of freedom is dependent on its perspective error function, (13). The control law (10) however, shows that the coupling effects have not eliminated since the control signal for each degree of freedom is dependent on the dynamics of the other degrees of freedom. Independency is usually preferred in practice. Furthermore, to satisfy the existence condition, a large uncertainty bound should be chosen in advance. In this case, the controller results in large implementation cost and leads to chattering efforts.
DECOUPLED ROBOT TRACKING CONTROL DESIGN In this Section, we propose a fuzzy system that would approximate the equivalent control (4). The main challenge facing the application of fuzzy logic is the development of fuzzy rules. To overcome this problem, an adaptive control law is developed for the on-line generation of the fuzzy rules. The input of the fuzzy system is the sliding surfaces (3), and the output is a fuzzy controller, which substitutes for the equivalent (4). With this choice, no bounds are needed about the system functions. Furthermore, the uncertainties are estimated and continuously compensated for, which means that the hitting controller mht (6) is adaptively determined on-line. The coming Subsection gives a brief introduction to fuzzy logic systems and characterizes them with the type, which is utilized in this contribution.
Fuzzy Logic Systems A fuzzy logic system consists of a collection of fuzzy IF-THEN rules. A one-input one-output fuzzy system has the following form: (14) where l = 1,2, ,L is the rule number, s and τf are respectively, the input and output variables. Al is the antecedent linguistic term in rule l; and θl, l = 1…,L is the label of the rule conclusion, a real number called fuzzy singleton. The conclusion of each rule (control action), a numerical value
290
Soft Computing with NeuroFuzzy Systems
not a fuzzy set, can be considered as pre-defuzzified output. Defuzzification maps output fuzzy sets defined over an output universe of discourse to a crisp output, τf. In this work, we have adopted singleton fuzzifier, product inference, the center-average defuzzifier which reduces the fuzzy rules (14) into the following fuzzy logic system: (15) where µAl is the membership grade of the input s into the fuzzy set AI. In (15), if θ1’s are free (adjustable) parameters, then it can be rewritten as: (16) where is the parameter vector and a regression vector given by
is
(17) Generally, there are two main reasons for using the fuzzy systems in (16) as building blocks for adaptive fuzzy controllers. Firstly, it has been proved that they are universal approximators [22]. Secondly, all the parameters in ξ(s) can be fixed at the beginning of adaptive fuzzy systems expansion design procedure so that the only free design parameter vector is ϑ. In this case, τ(θ, s) is linear in parameters. This approach is adopted in synthesizing the adaptive control law in this paper. Without loss of generality, Gaussian membership functions have been selected for the input variables. A Gaussian membership function is specified by two parameters {c,σ}:
where c represents the membership function’s center and σ determines its width. The fuzzy system used in this contribution is one input one output system, (14). The input of the fuzzy system is normalized using L number of equally spaced Gaussian membership functions inside the universe of discourse. Slopes are identical, see Figure 1. The described fuzzy system is used to approximate the nonlinear dynamics of robotic systems. In a decoupled manner, the control action is
An Adaptive Fuzzy Sliding Mode Control Scheme for Robotic ...
291
computed for each degree of freedom, based on the corresponding sliding surface. The control actions θ1 (output singletons) which are contained in the parameter vector e θ should be known. In the coming Subsection, adaptive laws are derived to do this task. The antecedent part is fixed with Gaussian membership functions.
Figure 1. Input fuzzy sets.
The Adaptation Mechanism Fuzzy systems are universal function approximators. They can approximate any nonlinear function within a predefined accuracy if enough rules are used. This implies the necessity of using expert knowledge in the form of large number of rules and suitable membership functions. Usually trial and error procedure is needed to achieve the requested accuracy. Assigning parameters of the fuzzy systems (or some of them) adaptively greatly facilitates the design (e.g. reduce the number of rules) and enhances the performance (saves the computation resources). In this subsection, we derive an adaptive control law to determine the consequent part (control actions contained in parameter vector θ) of the fuzzy system which is used to approximate the unknown nonlinear dynamics of robotic systems. The proposed scheme saves the need to expert knowledge and tedious work needed to assign parameters of the fuzzy system. Furthermore, disturbances, approximation errors and uncertainties are determined compensate for on-line leading to a stable closed loop system. Lyapunov stability analysis is the most popular approach to prove and evaluate the convergence property of nonlinear controllers, e.g., sliding mode control, fuzzy control system. Here, Lyapunov analysis is employed to investigate the stability property of the proposed control system. By the universal approximation theorem [22], there exists a fuzzy controller τf (s, θ) in the form of (16) such that
292
Soft Computing with NeuroFuzzy Systems
(18)
where εi is the approximation error and is bounded by a fuzzy controller
. Employing
to approximate
as: (19) where is the estimated value of the parameter vector θi. Now, the SMC in (5) can be rewritten as: (20) where the fuzzy controller is designed to approximate the equivalent controller . Define use (17), then it is obtained that (21) An expression for
can be expressed as follows:
(22) Substituting from (19-21): (23) where Now, assume that can be approximated by finite diagonal matrix known constant positive de . Unlike constant con- schemes (see [23,24] for example), this assumption has been taken into account as follows. Equation (23) can be rewritten as (24) Where Ei is the sum of approximation errors and uncertainties. A control goal would be the on-line determination of its estimate, error is defined by Define a Lyapunov functions as
(25)
.The estimation
An Adaptive Fuzzy Sliding Mode Control Scheme for Robotic ...
293
(26)
where and are positive constants. that
Differentiating (25) with respect to time and using (23), it is obtained
To satisfy
, the adaptive laws can be selected as (27) (28)
Using (20) (29) then (22) can be rewritten as (30) Therefore, V2 is reduced gradually and the control system is stable which means that the system trajectories coverage to the sliding surfaces s(t) while θˆ and Eˆ remain bounded. Now, if we let
(31)
and integrate Γ(t) with respect to time, then it is shown that
(32)
Because is bounded and bounded, it implies that
is non-increasing and
(33) Furthermore, Γ is bounded, so that by Barbalat’s lemma [7], it can be t
shown that lim t ←∞ ∫0 Γ(τ)dτ . That is s(t) → 0 , as t → 0 . As result, the proposed AFSMC is asymptotically stable. Hence, the control law (18) can be rewritten as follows
294
Soft Computing with NeuroFuzzy Systems
(34) In summary, the adaptive fuzzy sliding mode controller (34) has two terms; given in (19) with parameter the adjusted by (27) an AFSMC d the uncertain ap nals to the robotic system may result in chattering caused ties and proximation bound adjusted by (29). By applying these adaptive laws, the AFSMC is model free and can be guaranteed to be stable for any nonlinear system has the form of (2). It should be noted that implementing the algorithm implies that the both error dynamics and control signals has been decoupled, since each of them is dependent only on the perspective sliding surface. Unlike SMC, the proposed AFSMC does not require any knowledge about the system functions nor their bounds. It adaptively determines and compensates for the unknown dynamics and external disturbances leading to a stable closed loop system. Figure 2 shows the main elements of the control system. Remark 1. Since the control laws (6) and (34) contain the sign function, direct application of such control signals to the robotic system may result in chattering caused by the signal discontinuity.
Figure 2. The closed loop control system utilizing AFSMC.
To overcome this problem, the control law is smoothed out within a thin boundary layer φ [7,21] by replacing the sign function by a saturation function defined as:
An Adaptive Fuzzy Sliding Mode Control Scheme for Robotic ...
295
SIMULATION RESULTS In this section, we simulate the AFSMC and SMC on a two link robot; Figure 3. Simulation tests are carried out using MATLAB R2009a, version 7.8 under Windows 7 environment. A two link robot arm with varying loads is used to generate data in the simulation tests. The arm is depicted as 2-input, 2-ouput nonlinear system. The control architecture shown in Figure 2 represents the closed loop system, in which the robot is the plant to be controlled. The detailed descriptions of the matrices M(x), in (1) for this robot are given in Appendix A. We consider the state variable vector as the joint positions; i.e. . They are usually available feedback signals through encoders mounted on the motor shafts. Link parameters are , where the mass of link one m1 and link two m2 are randomly varied; rand(1) is a pseudorandom number ranges from . Figure 4(a) shows their time history. A random disturbance torque has been added to the gravity torque of link two, such that , Figure 4(b). Dynamic and static friction torques were selected as follws:
Figure 3. A two link rigid robot.
The friction and disturbance torques were unknown to the algorithm. Random signals were generated by the rand function in MATLAB.
296
Soft Computing with NeuroFuzzy Systems
The desired trajectories for x1 and x2 were set as: with A1 = 1.2 rad , A2 = 1.6 rad , ω1= π/2 rad s-1, ω2=π rads-1 Initially, the arm is assumed at rest, i.e
rad s−1 and position of links as rad s-1 and position of links as rad, which is initial position error degree and velocity error rad s-1 The AFSMC has been simulated under the following settings. Two rules were implemented to determine each of the two equivalent control components, i.e. L=2 in (14). Each rule base has one input, Si and one output, teqi, where the subscript i=1,2 donates the joint number. This means that a total of 4 rules were used to determine the two equivalent torques. This is relatively a quite small number of rules. In a similar study, i.e. adaptive fuzzy sliding mode control for nonlinear systems [25], the rule base consists of 36 rules for a one degree of freedom system (the inverted pendulum). Coefficient of the sliding surfaces in (12) were picked as c1=[40, 40]T and c2=[3, 3]T. After few simulation tests, the learning rates were adjusted as and . The estimated errors in (28) have been initiated as . As mentioned earlier, the sign function in (6) and (28) has been replaced by the saturation function with fi=1, i=1,2.
Evolution of the parameter vectors is given in Figure 5(a). Zero were used to initiate their elements. The superscripts denotes the rule number, 1 and 2. The rates of adaptation for the parameter vectors are depicted in Figure 5(b). As it can be noticed, the rate of adaption of rule 1 is very close to rule 2 for the same joint. This remark was noticed by the authors from an enlarged version of Figure 5(b).Time history of the estimated errors is shown in figure 6. With respect to SMC in (5), (6) and (10), we have simulated it under the following settings. The control system has been initiated with the same initial conditions (i.e. e and e ) followed by the AFSMC. Similar to what we did with respect to the AFSMC, the sign function in (6) has been replaced by the saturation function. The gain K of the hitting controller gain in (6) was set as K = 70I where I is × 22 identity matrix. This value of K has been selected as the maximum, maximum possible rate of possible one which means convergence.
An Adaptive Fuzzy Sliding Mode Control Scheme for Robotic ...
297
Figure 5. Time history of (a) parameter vectors (i.e. control actions) and (b) adaptation rate.
Larger value results in chattering. To synthesize the SMC, and in (10) were selected as follows: ˆM = 5I which means that it is a time-independent matrix and
where Fd,Fs and Td are defined above
Similar to AFSMC, the friction and disturbance torques were unknown to the control algorithm. Results are shown in Figures 7-12. A close look to these Figures shows that the AFSMC was little-bit faster the SMC. Figure 12 depicts the control signals. In the transient phase, the maximum input torques of the SMC exhibits larger values than those of the AFSMC. In order to quantify the performance of the two controllers, we have used the following three criteria. 1)
Integral of the absolute value of error (IAE):
2)
Integral of time multiplied by the absolute value of the error (ITAE)
Soft Computing with NeuroFuzzy Systems
298
Figure 6. Time history of the estimated errors.
3)
Integral of the square value (ISV) of the control input
Figure 7. The desired joint angles, xd and actual angles x.
Figure 8. Time history of the sliding surfaces.
Figure 9. Phase plots.
An Adaptive Fuzzy Sliding Mode Control Scheme for Robotic ...
299
Figure 10. Velocity tracking errors.
Figure 11. Trajectory tracking errors.
Figure 12. The input torques. Table 1. The performance indices.
Finally, it can be concluded that all signals of the proposed control system are bounded, the states have converged to the equilibrium points and the control have been met.
300
Soft Computing with NeuroFuzzy Systems
CONCLUSIONS In this article, we utilized the universal approximation property of fuzzy systems and powerfulness of SMC theory to compose an AFSMC scheme for robotic systems. Optimal parameters of the fuzzy system and uncertainty bound are generated on line .The proposed control scheme has the following advantages: 1) does not require the system model; 2) guarantees the stability of the closed loop system, 3) uses a simple rule base (one-input one-output fuzzy system). The adaptive control law generates On-line the fuzzy rules. Furthermore, the uncertainties are learned on-line and adaptively compensated for. In comparison with SMC, the proposed control scheme is decoupled and has eliminated the assumptions, which are usually needed to synthesize a SMC. The control scheme has been simulated on a two link planar robot. The fuzzy system needs only two rules per joint to determine the control signal. The approach significantly eliminates the fuzzy data base burden and reduces the computing time, thereby increasing the sampling frequency for possible implementation. It should be emphasized that, the developed adaptive laws learn the fuzzy rules and uncertainties. Zeros have been used to initiate them. Results show the effectiveness of the overall closed-loop system performance.
An Adaptive Fuzzy Sliding Mode Control Scheme for Robotic ...
301
REFERENCES 1.
A. B. Sharkawy, H. El-Awady and K. A. F. Moustafa, “Stable Fuzzy Control for a Class of Nonlinear Systems,” Tranactions of the Institute of Measurement and Control, Vol. 25, No. 3, 2003, pp. 265-278. doi:10.1191/0142331203tm086oa 2. J. Lin, R.-J. Lian, C.-N. Huang and W.-T. Sie, “Enhanced Fuzzy Sliding Mode Controller for Active Suspension Systems,” Mechatronics, Vol. 19, No. 7, 2009, pp. 1178-1190. doi:10.1016/j.mechatronics.2009.03.009 3. P. C. Chen, C. W. Chen and W. L. Chiang, “GA-Based Modified Adaptive Fuzzy Sliding Mode Controller for Nonlinear Systems,” Expert Systems with Applications, Vol. 36, No. 3, 2009, pp. 58725879. doi:10.1016/j.eswa.2008.07.003 4. S. Kaitwanidvilai and M. Parnichkun, “Force Control in a Pneumatic System Using Hybrid Adaptive Neuro-Fuzzy Model Reference Control,” Mechatronics, Vol. 15, No. 1, 2005, pp. 23-41. doi:10.1016/j. mechatronics.2004.07.003 5. L.-C. Hung and H.-Y. Chung, “Decoupled Sliding-Mode with FuzzyNeural Network Controller for Nonlinear Systems,” International Journal of Approximate Reasoning, Vol. 46, No. 1, 2007, pp. 74-97. doi:10.1016/j.ijar.2006.08.002 6. Y. Zhiyong, W. Jiang and M. Jiangping, “Motor-Mechanism Dynamic Model Based Neural Network Optimized Computed Torque Control of a High Speed Parallel Manipulator,” Mechatronics, Vol. 17, 2007, pp. 381-390. doi:10.1016/j.mechatronics.2007.04.009 7. J.-J. Slotine and W. Li, “Applied Nonlinear Control,” Printice-Hall International, Inc., Upper Saddle River, 1991. 8. V. I. Utkin, “Sliding Modes and Their Applications in Variable Structure Systems,” Mir Publishers, Moscow, 1978. 9. V. I. Utkin, “Sliding Modes in Control and Optimization,” SpringerVerlag, Berlin, 1992. 10. A. Harifi, A. Aghagolzadeh, G. Alizadel and M. Sadeghi, “Designing a Sliding Mode Controller for Slip Control of Antilock Brake Systems,” Transportation Research Part C, Vol. 16, No. 6, 2008, pp. 731-741. doi:10.1016/j.trc.2008.02.003 11. B. Yao, S. P. Chan and D. Wang, “Variable Structure Adaptive Motion and Force Control of Robot Manipulator,” Automatica, Vol. 30, No. 9, 1994, pp. 1473-1477. doi:10.1016/0005-1098(94)90014-0
302
Soft Computing with NeuroFuzzy Systems
12. B. Yao, S.P. Chan and D. Wang, “Unified Formulation of Variable Structure Control Schemes for Robot Manipulators,” IEEE Transactions on Automatic Control, Vol. 39, No. 2, 1994, pp. 371-376. doi:10.1109/9.272337 13. K. Passino and S. Yurkovich, “Fuzzy Control,” Addison- Wesley Longman, Inc., Boston, 1998. 14. R. Ordonez, J. Zumberge, J. T. Spooner and K. M. Passino, “Adaptive Fuzzy Control: Experiments and Comparative Analysis,” IEEE Transactions on Fuzzy Systems, Vol. 5, No. 2, 1997, pp. 167-188. doi:10.1109/91.580793 15. F. H. Hsiao, C. W. Chen, Y. W. Liang, S. D. Xu and W. L. Chiang, “T-S Fuzzy Controllers for Nonlinear Interconnected TMD Systems with Multiple Time Delays,” IEEE Transactions on Circuits and Systems, Vol. 52, No. 9, 2005, pp. 1883-1893. doi:10.1109/TCSI.2005.852492 16. F. H. Hsiao, J. D. Hwang, C. W. Chen and Z. R. Tsai, “Robust Stabilization of Nonlinear Multiple Time-Delay Large-Scale Systems via Decentralized Fuzzy Control,” IEEE Transactions on Fuzzy Systems, Vol. 13, No. 1, 2005, pp. 152-163. doi:10.1109/TFUZZ.2004.836067 17. H. F. Ho, Y. K. Wong and A. B. Rad, “Adaptive Fuzzy Sliding Mode Control with Chattering Elimination for SISO Systems,” Simulation Modeling Practice and Theory, Vol. 17, No. 7, 2009, pp. 1199-1210. doi:10.1016/j.simpat.2009.04.004 18. C.-C. Cheng and S.-H. Chien, “Adaptive Sliding Mode Controller Design Based on T-S Fuzzy System Models,” Automatica, Vol. 42, No. 1, 2006, pp. 1005-1010. doi:10.1016/j.automatica.2006.02.016 19. H. F. Ho, Y. K. Wong and A. B. Rad, “Robust Fuzzy Tracking Control for Robotic Manipulators,” Simulation Modelling Practice and Theory, Vol. 15, No. 7, 2007, pp. 801-816. doi:10.1016/j.simpat.2007.04.008 20. T.-H. S. Li and Y.-C. Huang, “MIMO Adaptive Fuzzy Terminal SlidingMode Controller for Robotic Manipulators,” Information Sciences, Vol. 180, No. 23, 2010, pp. 4641-4660. doi:10.1016/j.ins.2010.08.009 21. Y. Stephaneko and C.-Y. Su, “Variable Structure Control of Robot Manipulators with Nonlinear Sliding Manifolds,” International Journal of Control, Vol. 58, No. 2, 1993, pp. 285-300. doi:10.1080/00207179308923003 22. L. X. Wang, “Adaptive Fuzzy Systems and Control,” PTR Prentice Hall, Upper Saddle River, 1994.
An Adaptive Fuzzy Sliding Mode Control Scheme for Robotic ...
303
23. S.-J. Huang and W.-C. Lin, “Adaptive Fuzzy Controller with Sliding Surface for Vehicle Suspension Control,” IEEE Transactions on Fuzzy Systems, Vol. 11, No. 4, 2003, pp. 550-559. doi:10.1109/ TFUZZ.2003.814845 24. A. Poursamad and A. H. Davaie-Markazi, “Robust Adaptive Fuzzy Control of Unknown Chaotic Systems,” Applied Soft Computing, Vol. 9, No. 3, 2009, pp. 970-976. doi:10.1016/j.asoc.2008.11.014 25. M. Roopaei, M. Zolghadri and S. Meshksar, “Enhanced Adaptive Fuzzy Sliding mode Control for Uncertain Nonlinear Systems,” Communications in Nonlinear Science and Numerical Simulation, Vol. 14, No. 9-10, 2009, pp. 3670-3681. doi:10.1016/j.cnsns.2009.01.029
INDEX
A ACO algorithm 29, 31, 33 Adaptive control laws 284 Adaptive fuzzy sliding mode control (AFSMC) scheme 281 Adaptive laws 282 Adaptive learning 40 Adaptive network 129 Adaptive neural network 109 Adaptive neuro-fuzzy inference systems 54 Advanced control methods 217 Advanced mathematical technique 199 Agile supply chain (ASC) 143, 144 Agriculture applications 107 Airborne pollution 21 Air compressors 250 Air quality 22
Analyzed-stock market system (SMS) 96 ANFIS architecture 56, 146, 148 ANFIS controller 250, 252, 256, 258, 260 ANFIS Controller 216 ANFIS model 100, 152 ANFIS system 28 ANFIS technique 260 Angular displacement model 272 Angular machines 252 Ant colony optimization 21, 22, 28, 35 Ant colony optimization (ACO) 22 Aquatic systems 63 Architectural software 93 Architecture-based empirical software reliability analyses 95 Artificial immune system (AIS) 79 Artificial intelligence 54
306
Soft Computing with NeuroFuzzy systems
Artificial intelligence tools 217 Artificial intelligent tools 222 Artificial neural network controller 216 Asset management efficiency 145 Authentication mechanisms 52 Autocorrelation function 186 Average testing error 100
B Back propagation algorithm 79 Backpropagation gradient descent methods 131 Back propagation learning process 234 Backpropagation method 8 Back propagation neural network structure 217 Bayesian regularization 180, 184, 188, 190 Bending moment 167 Biosorbents materials 65 Biotechnology 64 Biotic surface 66 Bluetooth technology 265 Builtin optimal 222 Business applications 51 Business strategies 98
C Chaotic attractor dimension 181 Chaotic behaviors 180 Chaotic traffic volumes 180 Chopper-controlled external resistance 216 Clay surfaces 114 Cleveland database 81 Cleveland data set 77, 83, 84, 85, 86 Closed-loop speed-control system
217 Closed loop system 293 Closed-loop system performance 298 Clustering 24, 25, 34 Communication protocols 278 Competitive network 81 Complex nonlinear control problems 283 Complex systems 54 Computational tools 66 Computer based modeling techniques 124 Computer simulation 216 Construction industry 144 Controller parameters 224 Conventional linear systems 217 Conventional nonlinear controllers 234 Coordinated signal network 180 Cost function 25 Credit analyst 5 Credit rating 195, 196, 202, 209 Credit rating agencies 201 Credit rating prediction models 202 Credit rating problem 206 Credit risk management 3, 4 Credit scoring problem 5 Customer satisfaction models 159
D Data analysis 85 Data behavior 72 Data clustering 162, 181 Data clustering technique 180 Data reliability 96 Data sets 23 Debt service 203 Decision support systems 80
Index
Defuzzification method 23 Diagnosis decision process 78 Distribution networks 230 Dynamic system 152
E Effective thermal conductivity (ETC) 123 Electric energy 124 Electric screw drivers 250 Equality constraints 161 Equivalent circuit model 234 Euclidian plane 275 Evasion controller 269 External debt 197
F Factor analysis 202 Failure mode and effects analysis (FMECA) 44 Fault diagnosis 38, 40 Fault diagnosis systems 45 FCM algorithm 25 FCM function 26 Feed-forward architecture 82 Fiber-reinforced polymeric composite 126 Financial balance 198 Financial systems 39 Flexibility assessment model 152 Flexibility assessment tool 152 Flexural reinforcement 158 Flood forecasting 159 Forecasting model 183 Foreign direct investment data 198 Forward selection method 204 Fungal and yeast metabolism 65 Fuzzification 7 Fuzzy associative memories (FAM)
307
128 Fuzzy cluster means (FCM) method 266 Fuzzy C-Means (FCM) 24 Fuzzy control rules 253 Fuzzy expert systems 80 Fuzzy inference 147 Fuzzy inference mechanism 222 Fuzzy inference system (FIS) 23, 146 Fuzzy inference system model 67 Fuzzy knowledge 69 Fuzzy language information 147 Fuzzy logic 4, 5, 11, 18 Fuzzy logic based web service selection 53 Fuzzy Logic Controller 216 Fuzzy logic controller (FLC) 238 Fuzzy logic models 94 Fuzzy logic principle 159 Fuzzy logic sets 57 Fuzzy logic systems 82 Fuzzy logic toolbox 254 Fuzzy membership function (MF) 223 Fuzzy modeling 26 Fuzzy modeling procedure 67 Fuzzy neural network (FNN) approach 38 Fuzzy-Neuro model 3, 4 Fuzzy numbers 53 Fuzzy production rule 67 Fuzzy region 23 Fuzzy rules 4 Fuzzy sets 9 Fuzzy sets tuning 6 Fuzzy subsets 44 Fuzzy subtractive clustering 87 Fuzzy technique 260
308
Soft Computing with NeuroFuzzy systems
G Gamman software 109 Gamma ray sensor 106 Gamma-ray spectrometer 108 Gaussian functions 147 Gaussian membership functions 288, 289 Generalized Bell-shaped function 55 Generalized car-following model 180 Genetic algorithm 157 Global optimization toolbox 160, 161, 176 GQM technique 97 Gradient descent technique 159 Gradual assessment 159 Gravitational search algorithm (GSA) 231 Gross domestic product (GDP) 197 Gross monthly income (GMI) 13 Gross national product (GNP) 197
H Hazy system 72 Heart disease 77, 78, 89 Heart disease diagnosis systems 83 Heavy metal 65 Heavy metal removal efficiency 72 Heavy metals 63, 64 Heuristic-metaheuristic search 207 Heuristics experience 79 HFNN model 4, 6, 9, 10, 11, 12, 13, 14, 15, 16, 17 Hierarchical NF networks 39 High-density polyethylene (HDPE) 126
High-efficiency lattice Boltzmann solver 126 Hospital information systems 78 Human reasoning 217 Hybrid algorithm 148, 231 Hybrid fuzzy logic 3, 4, 6 Hybrid genetic algorithms 158 Hybrid intelligent system 127 Hybrid learning algorithm 112, 123
I Induction motor (IM) 215 Industrial process 45 Industrial wastes 65 Inflation rate 197 Information processing 51 Initial values 43 Input vector matrix 150 Interline power flow controller 229, 230, 247 International capital markets 197 Inter-organizational processes 145
K Kalman filtering theory 180
L Lagrangian multiplier 161 Learning ability 38 Learning algorithms 57 Learning process 181, 190 Learning rules 94 Least-squares method 161 Legacy system integration 97 Lewis-Nielson model 125 Linear discriminant analysis 4, 199, 202 Linear low-density polyethylene (LLDPE) composites 125
Index
Linear programming 53 Linear regression analysis 184, 188 Linguistic expressions 54 Linguistic variables 42 Logistic model 196 Logistic regression 4 Logistic regression analysis 203 Low-cost wastewater treatment 66 Low-density polyethylene (LDPE) 127 Lyapunov analysis 289 Lyapunov exponent 180, 181, 182, 186, 190 Lyapunov stability analysis 289
M Mamdani system 252 Manufacturing systems (MS 38 Marginal accounts 17 Mathematical programming models 142 Matlab programme 149 Mean absolute deviation (MAD) 112 Mean absolute error (MAE) 202 Mechanical energy 250 Medical decision support systems (MDSS) 78 Mercury bioavailability 66, 72 Metal ion bioavailability 66 Metal pollution control 64 Microcontroller 276 MIMO system 266 Minimum squares method 267 Mobile robotic platform 263 Modern variable speed drive applications 216 Motor control systems 250 Mthematical modeling techniques
309
107 Multi-collinearity problem 202 Multi-layered feed forward network 54 Multi-layer feed-forward neural network 234 Multilayer perceptron 78 Multilayer perceptron (MLP) neural network 16 Multilayer perceptron (MLP) structure 78 Multilayer perceptron neural network (MLPNN) 80 Multi machine power system model 235 Mutation function 169
N Natural ecosystems 66 Natural genetics 158 Natural radionuclides 105, 106, 113, 120 Natural systems 67 Natural waters components 71 Network generalization 184 Network performance function 184 Network structure 26 Neural fuzzy-controller topology 278 Neural network architecture 38, 80 Neural network (HFNN) model 4, 6 Neural networks 4, 18, 19 Neuro-adaptive learning techniques 99 Neurofuzzy approach 57 Neuro-Fuzzy combination 82 Neurofuzzy controller 270 Neuro-fuzzy inference system (ANFIS) 22
310
Soft Computing with NeuroFuzzy systems
Neuro-fuzzy logic 50 Neuro-Fuzzy models 22 Neuro-fuzzy structures 39 Neuro-fuzzy system 31 Non-adaptive fuzzy control 283 Non-exhaustive techniques 28 Non-linear dynamical system 38 Nonlinear multiple regression analyses 107 Nonlinear systems 40 Non-linear viscoelastic behavior 126 Non-metal filled polymer composites 135
Power network 230 Power system 229 Power transmission capacity 233 Probit analysis 198 Process integration as 145 Prognosis system 38 Programming language 85 Projection algorithms 283 Property rights 198 Proportional integral controller (PI) 251 Proportional integral derivative controller (PID) 251 Pruned system 10
O
Q
Obstacle controller 269 OEM battery-powered applications 251 Optimization algorithm 28 Ordinary least squares (OLS) 199 Outperforms regression 151
P Parallel processing 39 Pattern recognition 5, 239 Percentage overshoot 250 Perturbations 218 PI Controller 216 Political violence 198 Politic-economic environment 141 Polyamide composites 125 Polymer composites 124 Polyvinylidene fluoride (PVDF) 125 Position control system 45 Power flow control 231, 234, 235, 244 Power injection model (PIM) 232
QoS generic description 51
R Radial-basis-function neural network 180 Radionuclide activity 114 Radionuclides 113 Rating determination 196 Reactive power 230 Reactive power compensation (RPC) 230 Reciprocating machine 250 Reflectance measurements 106 Regression analysis 188 Relative angular displacement 272 Reliability analysis 158 Reliability estimation 92 Reliability model 92 Risk management 142 Robotic control 282 Root mean square error (RMSE) 112 Rotating machines 216
Index
Rule-based modeling 127
S Security constrained unit commitment (SCUC) 232 Self-excited DC motors 251 Service-oriented architecture (SOA) 91 Service-oriented software reliability model 96 Service oriented systems 102 Service oriented systems (SOSs) 92 Service selection mechanisms 53 Short-term chaotic traffic 179 Signal discontinuity 292 Sign function 292 Siliconcontrolled rectifier 216 Simulink model 71 Single machine infinite bus system 235 Singleton fuzzifier 288 Sliding mode control (SMC) 282 Small and medium enterprises (SMEs) 98 SMC theory 284, 286, 298 SOA-based systems 97 Soft computing 92 Soft computing techniques 92, 107 Soft-computing tool-based approach 127 Software based control system 249 Software reliability engineering 95 Software reliability management 92 Soil bulk density 108 Soil natural radionuclides 113 Soil particles analysis 108 Soil permeability 107 Soil texture 106 SOS reliability estimation 92
311
Sovereign credit rating 206 Sovereign rating 197 State space reconstruction 180 Statistical-based credit scoring systems 5 Statistical mechanics 125 Step-threshold activation function 158 Stepwise regression methods 199 Subtractive clustering 163 Subtractive clustering technique 185 Sugeno approach 266, 270 Sugeno system 98 Sugeno-type fuzzy inference system 162, 169 Supply chain flexibility 145 Supply chain flexibility evaluation 146 Supply chain management evaluation 152 Supply chain production 142, 154 Supply chain system 143 Surface layer 106 Symbolic knowledge 38
T Tabu search method 268, 269, 271 Takagi-Sugeno type fuzzy rules 40 Tangent sigmoid function 130 Task system 33 Technological challenges 57 Thermal conductivity 126 Thyristor family 216 Tobit analysis 198 Tracking control systems 282 Traditional enterprise 50 Traditional gradient search methods 160
312
Soft Computing with NeuroFuzzy systems
Traditional power engineering 216 Traffic flow system 180, 186 Training samples (TR) 15 Transmission system 231 Trigonometric activation function 267 Trigonometric artificial neural networks 265 Trigonometric functions 267 T-S fuzzy system 283 TSK fuzzy model 56 TSK models 39 TSK Neuro-Fuzzy systems 38 Tuning process 8
U UCI learning data set 79 Ultra-efficient system 252 Universal description, discovery, and integration (UDDI) 50 Universal generalization 222
V Variance inflation factor (VIF) 202 Voltage source converter 232 Voltage source converter (VSC) 233
W Web service 49, 50, 58, 59 Wireless communication 264