用于信号逼近的自适应时延小波神经网络  

AN ADAPTIVE TIME DELAY WAVELET NEURAL NETWORK FOR SIGNAL APPROXIMATION

作　　者：何振亚[1] 李文化[1] 蔚承建[1] 

机构地区：[1]东南大学无线电工程系,南京210096

出　　处：《电子科学学刊》1998年第5期604-610,共7页

基　　金：国家自然科学基金

摘　　要：小波神经网络是一种强有力的函数逼近工具。本文结合时延神经网络和小波分析概念提出一种新的小波神经网络摸型——自适应时延小波神经网络(ATDWNN:adaptive timedelay wavelet neural network).ATDWNN可以对同一类存在不同时延的多个信号用同一个超小波(superwavelet)进行逼近。为了训练ATDWNN,本文提出一种基于时间机理的竞争学习算法。实验表明,ATDWNN不仅可以成功地对同一类存在不同时延的多个信号采用同一个超小波进行逼近,而且可以用来估计各样本信号的时延。Wavelet neural networks (WNN) is a powerful tool for function approximation. In this paper a new model named adaptive time delay WNN(ATDWNN) is proposed which combines tune delay neural network and wavelet decomposition. ATDWNN is used to approximate signals having different time delays hi the same class. In order to train ATDWNN, time mechanism based competition learning is also proposed. It is shown through experiments that ATDWNN can not only approximate signals having different time delays by th'e same superwavelet, but also detect these time delays successfully.

关 键 词：小波神经网络 信号逼近 时间竞争 ATDWNN 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TN911.7[自动化与计算机技术—控制科学与工程]