摘要
提出一种用于非线性函数逼近的小波神经网络算法 ,分析了网络的拓扑结构 ,给出了网络的参数估计方法 .采用遗忘因子法训练网络的权值 ,利用具有优良渐近性质的递推预报误差算法训练尺度因子和平移因子 ,分析并给出两种小波元的个数选择方法 .该算法用于非线性函数逼近时优于同等规模的 BP神经网络 .仿真研究表明 ,该方法具有收敛速度快 ,逼近精度高等优点 ,在为非线性系统建模提供一种新方法的同时 。
A kind of wavelet neural network used in approaching non linear functions is given. Geometrical structure of the network is analyzed and the method of parameter estimation of the network is given. Weights of the network are trained by the method of forgetting factor, and scale factor and displacement factor are studied by the predicting error method with an exellent recursive character excellent. Two kinds of methods in selecting the quantity of wavelet element were analyzed and given in detail. It is better in approaching the non linear functions than the traditional BP neural network. The results of simulation indicate that the method is fast in its convergence speed and has a good approaching precision. It provides a new method in modelling non linear systems besides offering a beneficial reference to the identification of complex non linear systems.
出处
《北京理工大学学报》
EI
CAS
CSCD
北大核心
2002年第3期274-278,共5页
Transactions of Beijing Institute of Technology
基金
兵科院预研项目
关键词
小波神经网络
BP神经网络
函数逼近
wavelet neural networks
BP neural networks
function approach
