自适应多层小波神经网络建模方法

Nonlinear system modeling of self-adaptive multilayer wavelet neural network

基于小波多分辨率分析,提出了一种自适应多层小波神经网络的建模方法.该网络由平滑子网和多层细节子网组成.为改善模型精度,可递推并入新的细节子网,并且新网的训练不影响以前网络训练结果.应用遗传算法辨识多层小波网络的结构,用带遗忘因子的递推最小二乘法辨识网络的权值,较好解决了小波网络的结构优化问题.仿真表明:随着分阶层数的增加,网络的逼近误差逐渐下降,三层自适应小波网络即能满足建模精度要求.

This paper proposes a self-adaptive multilayer wavelet neural network based on multiresolution analysis theory of wavelet.This network consists of the smooth sub-network and multilayer details sub-network and can recursively incorporate new details sub-network to improve the accuracy.However,training the new sub-network does not affect the structure of the initial network. Multilayer wavelet network structure is determined by genetic algorithm and the weights are identified by Recursive Least Squares with forgetting factor.The method can effectively solve the problem of the structure of wavelet neural network.The application shows the approach is effective.