A comparative study on initial parameterization methods of fuzzy wavelet neural networks for time delay systems identification

Due to the great effect of initial parameterization on the accuracy of approximation result, when a Fuzzy Wavelet Neural Network (FWNN) is used for system identification, the number of sufficient fuzzy rules, the number of needed wavelet functions and the initial values of structure parameters have to be correctly determined. The aim of this paper is to present a new efficient method for determination of all these network parameters. Based on the powerful properties of wavelets: the time-scale distribution of wavelet energy and the admissibility of wavelet kernels, we present a method for initial parameterization of FWNN. Taking nonlinear dynamical systems with longer input delays as simulated example, a comparative study is done in order to prove the effectiveness of the proposed method. It is seen that the proposed initialization method achieves higher accuracy and has good performance comparing to other techniques.