GA-Optimized Wavelet Neural Networks for System Identification
Jinhua Xu · 2006
In this paper, a genetic algorithm is proposed to design WNNs for nonlinear system identification. The model structure of a high dimensional system is decomposed into some submodels of low dimensions. By introducing a connection switch to each link between a wavelet and an input node, the decomposition is done automatically during the evolutionary process. GA is used to train the wavelet parameters and the connection switches. In this way, both the structure and wavelet parameters of WNNs can be optimized simultaneously. The proposed WNNs can handle nonlinear identification problems in high dimensions.