Optimal Design of Type_1 TSK Fuzzy Controller Using GRLA F. Naderi, A. A. Gharaveisi and M. Rashidinejad Electrical Engineering Dept. of Shahid Bahonar University of Kerman Kerman, Iran a_gharaveisi@yahoo.com ABSTRACT A new methodology for designing optimal systematic GA-based fuzzy controller is presented in this paper. Our design is based on Genetic Reinforcement Learning Algorithm (GRLA), unlike conventional GA that is based on the competition between chromosomes only to survive, this method is based on competition and cooperation between chromosomes, GA tries to find good chromosomes and good combination for them to form an optimal fuzzy controller. The proposed GRLA design method has been applied to the cart-pole balancing system. The controller was capable of balancing the pole for initial conditions up to 80 ° . As a comparison we applied a Mamdani controller which is designed through normal GA and uses five membership functions for inputs and output variables to the same problem. the results show the efficiency of the proposed method. Keywords : Genetic Reinforcement Learning Algorithm (GRLA), TSK Fuzzy Controller I. INTRODUCTION Fuzzy theory has been developed by L.A.Zadeh in 1965 [1]. Since conventional control schemes are limited in their range of practical applications, fuzzy logic controllers are receiving increased attention for intelligent control applications [2]. Fuzzy control systems employ a mode of approximate reasoning that resembles the decision-making process of humans. The behavior of a fuzzy controller is easily understood by a human expert as knowledge is expressed by means of intuitive, linguistic rules. In the design of a fuzzy controller the definition of membership functions and the establishment of control rules (if-then rules) are very important. In [5], Procyk and Mamdani show that a change in the membership function (mf) may alter the fuzzy control system (FCS) performance significantly. Unfortunately the human experts are not sometimes able to express their knowledge in the form of fuzzy if-then rules. So this fact has forced researchers to find a method that automatically determines the parameters. Some papers propose automatic methods using neural networks (NN) [7,8] ,Fuzzy clustering [6] ,genetic algorithms(GA) [3,9,10,11,12] , gradient methods[13,14], or Evolutionary Algorithm (EA) [4,22,23]Karr used a GA to generate membership function for a fuzzy controller [15] in Karr’s work , the user has to define a method to set the rules at first or hand-design this exhaustive task, then use the GA to design the mf only. Since the mf and rule set are highly dependent, hand-design a one, and GA- design of the other, does not use the GA to its full advantage. So in most automatic designs For an FCS the optimization of both the Membership function (mf) and the control rules of the FCS are required. Berenji [16, 17] introduced a method that learns to adjust the fuzzy mf of the linguistic labels used in different control rules through First Joint Congress on Fuzzy and Intelligent Systems Ferdowsi University of Mashhad, Iran 29-31 Aug 2007 Intelligent Systems Scientific Society of Iran Intelligent Systems Scientific Society of Iran ﻫ ﻔ ﺳﻴﺴﺘﻤﻬﺎي ﻛﻨﻔﺮاﻧﺲ ﺘﻤﻴﻦ ﺳﻴﺴﺘﻤﻬﺎي ﻛﻨﻔﺮاﻧﺲ ﺘﻤﻴﻦ ﻓﺎزي ﻓﺎزي، ، 9 - 7 ﺷﻬﺮﻳﻮر ﺷﻬﺮﻳﻮر1386 1386 ﻣﺸﻬﺪ ﻓﺮدوﺳﻲ داﻧﺸﮕﺎه، ﻣﺸﻬﺪ ﻓﺮدوﺳﻲ داﻧﺸﮕﺎه،