Design of fuzzy logic systems for nonlinear process identification

Abstract

This paper presents a method of constructing accurate nonlinear dynamic MIMO models using a class of fuzzy logic systems (FLSs). Generally, the design of FLS models involves identification of both the FLS structure (linguistic fuzzy IF-THEN rules) and parameter tuning of fuzzy membership functions. Most current techniques treat these parts separately, which may result in a suboptimal solution. The authors propose to optimize the two parts simultaneously using a genetic algorithm (GA), a stochastic global search method which explores the solution space in manner similar to natural evolution. In the authors' work new FLS domain specific operators are introduced that significantly reduce search time, and at the same time, increase the accuracy of results. The FLS model obtained from GA search is further fine-tuned using a conjugate gradient method. To illustrate the proposed method, the authors show that the FLS models compare well to other nonparametric modeling techniques, such as standard feedforward neural networks on an example of a nonlinear system. In addition, a FLS offers qualitative model description in terms of the linguistic fuzzy rules which may contribute to the understanding and control of the process. These merits make the FLS models attractive for use in model predictive control and other process applications.