基于多值编码的混合遗传算法的小波神经网络优化  

Optimization of Wavelet Neural Networks with Hybrid Genetic Algorithm Based on Multi-encoding

作　　者：黄敏[1] 方晓柯[1] 王建辉[1] 顾树生[1] 

机构地区：[1]东北大学信息科学与工程学院,辽宁沈阳110004

出　　处：《系统仿真学报》2004年第9期2080-2082,2114,共4页Journal of System Simulation

基　　金：国家自然科学基金资助项目(60274024)

摘　　要：采用多值编码方式构造染色体结构,对小波神经网络的结构和参数进行编码,可以同时确定小波神经网络结构和优化网络参数,简化了问题的求解过程。在遗传算法中嵌入一个梯度下降算子,使得混合算法既有较快的收敛性,又能以较大概率得到全局极值。仿真表明,利用该算法训练小波神经网络,能使网络具有简单的结构形式,较快的收敛速度,较高的逼近精度和较强的泛化能力。A hybrid genetic algorithm based on multi-encoding method for the optimization of wavelet neural networks (WNN) is put forward. This method can be used to optimize the structure and the parameters of WNN in the same training process. Through embedding a gradient descend operator into the generic algorithm, a hybrid algorithm is achieved with fast convergence and great probability for global optimization. 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.

关 键 词：小波神经网络 混合遗传算法 多值编码 梯度下降法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]