Ajuste de regras e funções de pertinência utilizando algoritmos genéticos

Abstract

The purpose of this paper is to present a strategy for an automatic rules fitting and an automatic membership function fitting using genetic algorithms. For that, an algorithm has been developed and implemented to transform the rules and the fuzzy membership function in chromosomes, which are submitted to an evolution, crossover and mutation. The main idea is to fit a new family of rules and membership functions, which can establish a better system control, optimizing the final result. The proposed algorithm has been incorporated to the existent working Computational Package for Teaching Fuzzy Control, with the objective of teaching Fuzzy Logic to the control students. The package contains all required instructions for the users to gain an understanding of fuzzy control principles. The main objective of the package is to park a car, approaching from any direction, in a parking lot. To accomplish this task the students must first develop sets of fuzzy-control rules and and membership functions which define the trajectory of the car. Processes, such as fuzzification and defuzzification of the variables, are performed by the program without the interference of the user. The choice of the optimization method genetic algorithms is because of the better they bring to the performance of a fuzzy controller, proved for the obtained results of this work. These algorithms have been used with success in many varied problems of optimization and machine learning.