小波神经网络的参数初始化研究

A Research on the Initialization of Parameters of Wavelet Neural Networks

随机产生的初始参数往往使小波神经网络的学习次数大幅度地增加,甚至不收敛.为了加快网络的学习速度,本研究提出了一种将小波网络的初始参数设置和小波类型、小波时频参数和学习样本等联系起来的小波神经网络的初始参数设置方法.学习实例结果表明,按照这一方法不但可以获得高几率的优秀初始参数,而且能大大加快小波网络的后续学习速度.

Parameters obtained at random tend to increase the training times of wavelet neural networks and even make the training course unconvergent. In order to accelerate the training speed of the wavelet neural networks,a method of how to set the initial parameters of the wavelet neural networks is proposed in this paper. This method integrates the setting of initial parameters with the wavelet type,time-frequency parameters of the wavelet and the training samples. The training example shows that super initial parameters can be obtained with high probability by this method and, as a result, the network' s training speed can be accelerated very quickly.