论文部分内容阅读
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 network is a powerful function approximation tool. In this paper, a novel wavelet neural network model called ATDWNN (adaptive time delay wavelet neural network) is proposed based on the concept of time-delay neural network and wavelet analysis. The signals are approximated by the same superwavelet. In order to train ATDWNN, this paper presents a competitive learning algorithm based on time mechanism. Experiments show that ATDWNN can not only successfully apply the same superwavelet to multiple signals of the same type with different delays, but also can estimate the delay of each sample signal.