Evolving fuzzy systems

Abstract

Fuzzy rule-based systems have proved to be a convenient tool for modeling complex systems. This is due to their capacity to capture their typical imprecision, which makes classical methods inefficient. At present, Fuzzy Logic Controllers (FLC) are considered one of the most important applications of fuzzy rule-based systems. However, the learning process of proper rules for a given problem is still an important research issue. In this direction, different solutions for this problem have been developed, many of them based on Evolutionary Algorithms. Nevertheless, the preservation of fuzzy system rules semantics during the evolving process - more specifically during the recombination - is not always assured. This paper proposes a codification for fuzzy systems together with the proper genetic operators in order to achieve a balance between the searching process carried out by the evolving algorithm and the preservation of the recombined fuzzy system. Such codification and the proposed genetic operators have been used in an evolving algorithm, and its behavior in real function approximation has been tested, with successful results.