Automatic Generation of Fuzzy If-Then Rules using Genetic Algorithms and Hyper-Cone Membership Functions

Abstract

Fuzzy reasoning has developed as a method that can express our knowledge and has been applied to various fields, for example fuzzy control. In fuzzy reasoning, expert knowledge is expressed by fuzzy if-then rules. But, there is the problem that it is very difficult to tune its membership functions and reasoning rules. In addition, it is difficult to obtain expert knowledge in large-scale systems with many inputs and outputs. Therefore, many techniques have been suggested to automatically tune or generate membership functions and fuzzy if-then rules.In this paper, we present an automatic generation technique for fuzzy if-then rules by genetics-based machine learning. We consider flexible shape and location of fuzzy rules to obtain fuzzy rules of nearly human sense. In this method, the antecedent part and the consequent part of each rule are expressed by a hyper-cone membership function. The genetic algorithm decides the location and the radius of each hyper-cone membership function in the input and output spaces. We add genetic information whether each rule is fired or not, and delete needless rules using this information and a method of forgetting. We applied this method to the car pursuit control problem and the trailer-track back-up control problem.