Evolving structure and parameters of fuzzy models with interpretable membership functions

Abstract

In this paper, a new evolutionary algorithm for optimization of fuzzy models is proposed. For simultaneous optimization of structure and parameters of a fuzzy model, unique encoding scheme and appropriate evolutionary operators are proposed. There are three important aspects in fuzzy modeling: modeling accuracy, rule compactness, and interpretability of input membership functions. Thus, a new fitness function is proposed to consider the three objectives simultaneously. Through simulations on two well-known modeling problems, it is shown that the proposed algorithm is effective in finding an accurate fuzzy model with compact number of fuzzy rules. In addition, the fuzzy model uses well distributed membership functions that helps to increase interpretability of the fuzzy model.