基于遗传算法的前向神经网络结构优化  

Structural Optimization of Feedforward Neural Network Based on Genetic Algorithms

作　　者：王宏刚[1] 钱锋[1] 

机构地区：[1]华东理工大学自动化研究所,上海200237

出　　处：《控制工程》2007年第4期387-390,共4页Control Engineering of China

基　　金：国家973计划资助项目(2002CB3122000);上海市自然科学基金资助项目(05ZR14038);上海科委科技攻关基金资助项目(04DZ11010;05DZ11C02);上海市科委重大基础研究资助项目(05DJ14002)

摘　　要：对近几年应用遗传算法(Genetic Algorithm,GA)优化设计前向神经网络结构的研究进行了评述。指出了神经网络结构优化设计的重要性和目前各种方法存在的不足。介绍了神经网络结构设计原理和应用GA优化设计神经网络应着重考虑的两个问题:即结构表达策略和适应度函数设计。分别对近来应用GA优化设计多层感知器、径向基函数神经网络和径向基概率神经网络结构的研究进行了细致介绍和分析。指出了目前研究工作的不足和未来研究工作的发展方向。A review of the current research state on applying genetic algorithm to stuctural optimization for feedforward neural networks in resent years is presented. The significance of structural design optimization for ANN and the drawback of the current methods are analyzed. The principle of ANN structural design is introduced and two critical issues on architecture encoding strategy and fitness function design are proposed. The research on applying genetic algorithm to design optimization for multilayer perception, radial basis fimction and radial basis probabilistic neural networks are introduced and analyzed respectively in detail. According to the problems in current research, the future research directions are pointed out.

关 键 词：遗传算法 前向神经网络 结构优化 多层感知器 径向基函数 径向基概率神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]