A new encoding method of genetic algorithms towards parameter identification of fuzzy expert systems

The membership functions of fuzzy expert systems need a systematic, self-learning method instead of a subjective tuning method in order to increase the performance of the fuzzy model. The genetic-algorithm learning method is consequently employed. The rule-based encoding scheme would bring the redundant information for the genetic algorithm by repeatedly representing the similar membership function in an individual. The new encoding method, which is a parameter-based encoding scheme, would diminish the redundant representation of fuzzy parameters. This method would separate the data structures of fuzzy rules and fuzzy parameters in the genetic-algorithm learning method. This method should efficiently use the memory resources of computers and increase the dimensions of the solved problem. Then, a numerical example and the learning results are demonstrated. Discussions about the effects of population size, reproduction method, crossover rate, mutation rate and fitness scaling are included. Finally, some conclusions are presented.