Evolutive identification of fuzzy systems for time-series prediction

Abstract

This paper presents a new algorithm for designing fuzzy systems. It automatically identifies the optimum number of rules in the fuzzy knowledge base and adjusts the parameters defining them. This algorithm hybridizes the robustness and capability of evolutive algorithms with multiobjective optimization techniques which are able to minimize both the prediction error of the fuzzy system and its complexity, i.e. the number of parameters. In order to guide the search and accelerate the algorithm’s convergence, new specific genetic operators have been designed, which combine several heuristic and analytical methods. The results obtained show the validity of the proposed algorithm for the identification of fuzzy systems when applied to time-series prediction.