连续小波过程神经元网络在非线性函数逼近的应用  

Continuous Wavelet Process Neural Networks in the Application of Approximating Nonlinear Functions

作　　者：刘金月[1] 祝宝玲[2] 

机构地区：[1]东北石油大学计算机与信息技术学院,大庆163318 [2]吉林油田公司天然气地面工程建设项目经理部,松原138000

出　　处：《长春理工大学学报（自然科学版）》2010年第3期140-141,158,共3页Journal of Changchun University of Science and Technology(Natural Science Edition)

摘　　要：建立基于BP网络的连续小波过程神经元网络模型,分析网络拓扑结构,给出学习算法。模型根据网络逼近函数特性,选取Morlet函数作为隐层结点激励函数,利用LMS算法训练网络权值、尺度因子和平移因子,输出层采用线性函数,使网络兼具过程神经元网络和小波变换优点。分别用连续小波过程神经元网络和BP网络逼近同一非线性函数,仿真结果表明,小波网络逼近精度明显优于BP网络。The paper established the continuos wavelet process neural networks model based on BP network,analyzed topology struoture of the network and gave learning algorithm. According to approximating function characteristies of the network, the Morlet wavelet function was employed as activation function in the hidden node,network weights,scale factor and displacement factor were trained by using LMS algorithm, linear function was adopted in the output layer,so the continuous wavelet prooess neural networks had better characters oombining with wavelet transform and neural network.The continuous wavelet process neural networks and BP network were used in approximating the same nonlinear function respeetively.the result showed that the wavelet networks had better perrformance than BP network in the approximation accuracy.

关 键 词：小波变换 过程神经元网络 函数逼近 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]