基于遗传算法的函数型小波网络  

The Functional Wavelet Neural Network Based on Genetic Algorithm

作　　者：陈健[1] 谭阳红[1] 尹新[1] 周野[1] 

机构地区：[1]湖南大学电气与信息工程学院,湖南长沙410082

出　　处：《计算机仿真》2010年第2期180-183,268,共5页Computer Simulation

基　　金：国家自然科学基金(50277010);教育部博士基金(20020532016)

摘　　要：针对传统小波神经网络易陷入局部极小等缺陷,采用遗传算法对神经网络进行优化。提出了一种结合实数编码与二进制编码的多值编码遗传算法,上述算法在同一条染色体上同时使用实数编码与二进制编码,有机结合了两者的优点,并把遗传算法用于优化函数型小波网络的结构中,可获得具有更好泛化能力的小波网络。仿真实验结果表明,利用该遗传算法训练小波神经网络,能使网络具有简单的结构形式,较高的逼近精度和较强的泛化能力,并证实了网络的有效性和优越性能。There are some disadvantages in traditional wavelet neural networks, such as falling into local minimum point easily. To avoid that disadvan -rage, networks are optimized by genetic algorithm （GA）. A type of Genetic Algorithm based on multiple varibles coded with real and binary number is proposed in this paper. In this algorithm,real and binary numbers are presented simultaneously in a chromosome, which integrate the advantages of both. And then this genetic algorithm is put in the optimization of the functional wavelet neural network. So the more strengthened wavelet network is obtained. Simulation results show that WNN with hybrid genetic algorithm has a comparatively simple structure, and that it can both meet the precision request and enhance the generalization ability. The availability and superiority are testified through the result of the simulational experiment.

关 键 词：多值编码 遗传算法 函数型小波网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]