A nonlinear system identification approach based on neuro-fuzzy networks

This paper presents a nonlinear system identification approach based on neuro-fuzzy networks. This method consists of two main step: 1) concerns with the structure identification or learning, which includes selection of input variables and determination of the number of fuzzy rules and initial terms for membership functions; and 2) deals with parameter identification or learning. Its task is to adjust the weights of the neuro-fuzzy network, i.e., the antecedent and consequent parameters of rules, so that the error between the desired and real output is minimum. The effectiveness of the proposed technique is confirmed by simulation results.