Combined Construction of Wavelet Neural Networks for Nonlinear System Modeling

ABSTRACT A combined selection-elimination constructive approach of wavelet networks is presented. To obtain a compact non-parametric model of nonlinear systems, the forward growth first selects some most useful hidden nodes to achieve required modeling accuracy. Then in the backward pruning process, some hidden units are removed and the network's modeling ability is preserved at the same time. Simulation examples are given to illustrate the proposed method