基于小波基函数的BP神经网络优化方法

BP Neural Network Optimization Method Based on Wavelet Basis Function

目前,BP神经网络在变形预报中应用广泛,但其容易受到局部极值的影响而致收敛速度缓慢。针对BP神经网络的这一缺点,本文将BP神经网络的激活函数更换为小波基函数,并对BP神经网络的权重及临界值实施改进,形成小波神经网络。小波神经网络拥有优质的时频局域化性质以及自我学习本领,经小波分解实行缩放和平移变换后,可获取与逼近函数性质一致的级数,可以用来做变形预报。同时,经采用缩放和平移两项新的变量后,小波神经网络将比小波分解具备更多的自由度,进而数值模拟精度更佳。实验结果表明,与BP神经网络相比,小波神经网络在变形预报方面收敛效率更高,误差更小,可以达到更好的预测效果。

At present,BP neural network is widely used in deformation prediction,but its convergence speed is slow due to the influence of local extremum.In view of this shortcoming of BP neural network,this paper replaces the activation function of BP neural network with wavelet basis function and improves the weight and critical value of BP neural network to form wavelet neural network.Wavelet neural network has excellent time-frequency localization and self-learning ability.After the scaling and translating transformation of wavelet decomposition,we can obtain the series consistent with the properties of the approximation function,which can be used for deformation prediction.At the same time,after introducing two new variables,scaling and translating,wavelet neural network will have more degrees of freedom than wavelet decomposition,and then the accuracy of numerical simulation will be better.The experimental results show that compared with BP neural network,wavelet neural network has higher convergence efficiency and smaller error in deformation prediction,and can achieve better prediction effect.