基于差分进化小波神经网络的多维非线性系统辨识  

Identification of multi-dimensional nonlinear systems based on differential evolution wavelet neural network

作　　者：李目[1,2] 何怡刚[2] 谭文[1] 

机构地区：[1]湖南科技大学信息与电气工程学院,湘潭411201 [2]湖南大学电气与信息工程学院,长沙410082

出　　处：《电子测量与仪器学报》2010年第7期599-604,共6页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(编号:50677014;60876022)资助项目;高校博士点基金项目(编号:20060532016)资助项目;教育部新世纪优秀人才支持计划(编号:NCET-04-0767)资助项目;湖南省自然科学基金项目(编号:06JJ2024)资助项目

摘　　要：提出一种基于差分进化小波神经网络(DE-WNN)的多维非线性系统辨识方法。利用差分进化算法同时优化小波神经网络的结构和参数,简化网络结构,提高小波神经网络的学习精度和收敛速度。将该方法应用于多维非线性系统的辨识,并与RBF神经网络和遗传小波神经网络的辨识结果进行了比较,实验结果表明,差分进化算法优化的小波神经网络隐层节点为6,迭代次数为30,网络训练时间为0.58s,辨识均方误差达到1.02×10?4,所提出的方法具有更高的辨识精度和收敛速度,能够更好的辨识出多维非线性系统。A new method based on differential evolution wavelet neural network（DE-WNN） was presented for identification of multi-dimensional nonlinear systems.The structures and parameters of wavelet neural network were optimized by DE-WNN algorithm at the same time.The algorithm could get a best network structure and improve learning accuracy and convergence rate.The proposed method was applied to identify the multi-dimensional nonlinear systems,and the results were compared with those of RBF neural network and genetic wavelet neural network.The simulation results have shown that the hidden layer nodes is 6 and the iteration number is 30 in the DE-WNN model,the training time of the network is 0.58s and mean square error is 1.02×10^-4,the proposed method have more identifying accuracy and convergence rate,it can better identify the multi-dimensional nonlinear systems.

关 键 词：差分进化算法 小波神经网络 多维非线性系统 辨识 

分 类 号：TP27[自动化与计算机技术—检测技术与自动化装置]