A Mamdani Fuzzy modeling method via Evolution-Objective Cluster Analysis

In Mamdani Fuzzy modeling, the determination of fuzzy rules is easily influenced by the artificial factor and noise. Therefore the redundant rules are generated, the compatibility of rule base and the distinguishability of the fuzzy partition is decreased. I.e., the interpretability of the Mamdani model is weakened. Considering this, an Envolution-Objective Cluster Analysis-based Mamdani fuzzy modeling method is proposed in this paper. Firstly, the Objective Cluster Analysis algorithm is introduced and enhanced. As a result, the effect from the artificial factor and the noise data on the fuzzy partition is reduced. Furthermore, the compact and initial rule base is obtained by only one pass. Secondly, the criteria of rule covering and Genetic Niching are combined, introduced into the (1+1) Evolutionary Strategy to optimize the semantic values of parameters in the initial rules. Thus both the compatibility among the rules and the distinguishability of the fuzzy partition could be considered in the same time. The compactness, distinguishability and the moderate accuracy of the presented model is demonstrated by the electric application example.