Fuzzy modeling with genetic algorithms

Recent applications of fuzzy control have created an urgent demand for fuzzy modeling techniques. Several methods for identification of fuzzy models from numerical input-output samples have been proposed. Among them, Sugeno and Yasukawa's method [6], which employs fuzzy c-means clustering, holds significant promises. This paper improves the method of Sugeno and Yasukawa. Identified fuzzy models are tuned at various stages by means of genetic algorithms, i.e., the numbers of input variables and rules are reduced and membership function parameters are adjusted. The technique, when applied to a nonlinear system, demonstrates its efficiency in a comparison with the original method of Sugeno and Yasukama.