The optimization of fuzzy rules based on hybrid estimation of distribution algorithms

Abstract

Optimization of fuzzy rules based on numerical data is an important issue in the optimization design of fuzzy system. In this paper, based on an improved estimation of distribution algorithm, an optimization learning method COR_MUMDA for fuzzy rules is proposed. This method can generate fuzzy rules directly from numerical data. The method learn fuzzy rules mainly based on MUMDA (multi-group univariate marginal distribution estimation algorithm). Unlike the general estimation of distribution algorithms, MUMDA can increase the diversity of the population and avoid sticking at local optima. In addition, the elite genetic strategy is used to generate the next population. In this way, it reduces the possibility of losing the optimal solutions. To verify the efficiency of this algorithm, the simulation experiments are performed. The comparative results of three classic examples are given.