Design of Information Granulation-Based Fuzzy Radial Basis Function Neural Networks Using NSGA-II

This paper is concerned with information granulation-based fuzzy radial basis function neural networks (IG-FIZBFNN) and its multi-objective optimization by means of the nondominated sorting genetic algorithms II (NSGA-II). By making use of the clustering results, the ordinary least square (OLS) learning is exploited to estimate the coefficients of polynomial. In fuzzy modeling, complexity and interpretability (or simplicity) as well as accuracy of model are essential issues. Since the performance of the 1G-RBFNN model is affected by some parameters such as the fuzzification coefficient used in the FCM. the number of rules and the orders of polynomials of the consequent part of fuzzy rules, we require to cam, out both structural as well as parametric optimization of the network. In this study, the NSGA-II is exploited to find the fuzzification coefficient, the number of fuzzy rules and the type of polynomial being used in each conclusion part of the fuzzy rules in order to minimize complexity and simplicity as well as accuracy of a model simultaneously.