An Evolutionary Wavelet Network and its training method
Liu Shou-sheng, Ding Yong, Liu Hai-feng · 2010
Aimed at the problems of the curse of dimensionality and the lack of robustness for wavelet neural network, an optimization method based on principal component analysis for wavelet network structure is presented. However, this method is easily plunge into local minimum points when principal components are solved by Oja iterative algorithm. Genetic Algorithm which has good characteristic in global optimization can remedy the deficiency of Oja algorithm. In this paper, wavelet network is optimized by Hybrid Genetic Algorithm which is composed of genetic algorithm and Oja algorithm. This Evolutionary Wavelet Network(EWN)has effectively solved the problems of wavelet network, such as overmany nods and the lack of robustness. The simulation results show the efficiency of EWN.