Research On Two Hidden-layer Wavelet Networks

Wavelet neural network is the one of the effective tools to deal with the signal at present. A lot of scholars have done a large amount of research to it. Because it inherit successfully the advantages of RBF neural network's structure and the wavelet approximation, the research of the theory of wavelet neural networks is focus on the improvement of Learning algorithm and obtains few development in the structure that adopts the classical RBF structure (single hidden layer) at present. On the other hand, although wavelet neural network make and focus one's attention upon success, not only can't overcome the shortcoming of slow convergence; but also the network precision can not get raising further. This has influenced further development of wavelet neural network. This text has put forward and studied a kind of new two-hidden layers wavelet neural network frame. Based on the frame, this text proposes several kinds of novel wavelet neural network models and the corresponding learning algorithms. The results that apply the models to the Function Learning show that: Compared with the other neural networks, wavelet networks, the models reach a hundredfold improvement in speed and its generalization performance has been greatly improved.