小波神经网络用于非线性函数逼近的研究

Wavelet neural networks in the research of approximating nonlinear functions

神经网络具有良好的学习特性。而小波变换具有良好的时频局部化特性.将二者结合在一起构成小波神经网络,网络隐层采用morlet小波函数,输出层采用线性函数,使得该网络兼具神经网络和小波变换的优点.分别用小波网络和BP网络逼近一非线性函数,其结果表明,在相同的误差条件下,小波网络的收敛速度要远远快于一般的BP网络.

Neural network has a good learning character, and wavelet transform has a good localization character both in time-domain and frequency-domain .We can get wavelet neural network by combining them. The morlet wavelet function and linear function are employed as an activation function in the hidden and output layer respectively. So the wavelet neural networks have better characters comparing with wavelet transform and neural network. In this paper, we contrast the wavelet neural network and BP network in approximating nonlinear function, and the results indicate that the rate of convergence of the wavelet neural network is faster than the speed of BP network in the same errors.