Intensifying the Performance of Nonlinearity Approximation by an Optimal Fuzzy System

Abstract

A technique to optimize the Standard Additive Model (SAM) fuzzy system for nonlinear system approximation is presented. First, fuzzy rules are initialized more much than usual by employing Centroid Neural Network (CNN) and then the genetic algorithm-based optimization process used to exclude unnecessary and redundant rules; thereafter, the fuzzy rule parameters are tuned by the gradient descent method incorporated with momentum technique. Finally, we demonstrate with numerical experiments based on approximating some nonlinear functions and chaotic time series. From the results, we can see that the proposed method is more effective than normal approach in terms of accuracy and training time.