遗传算法优化神经网络的拓扑结构与权值  

Optimization of Topological Structure and Weight Value of Artificial Neural Network Using Genetic Algorithm

作　　者：谷小青[1] 易当祥[1] 刘春和 

机构地区：[1]西北工业大学经济研究中心,陕西西安710072 [2]北京清河大楼子八,北京100085

出　　处：《广东工业大学学报》2006年第4期64-69,共6页Journal of Guangdong University of Technology

摘　　要：遗传算法(GA)和人工神经网络(ANN)的相互结合有辅助式和合作式两种方式.本文在此基础上提出了融合、BP_GA和GA_BP三种算法,并采用GA_BP算法同时优化BP神经网络的结构、权值和阈值,研究和实现了一套先进的编码技术和进化策略,克服了传统BP神经网络经验尝试方法的盲目性.实例优化与检验结果表明:遗传算法优化获得的神经网络比由经验尝试法得到的BP网络性能更优异,方法更合理.In crder to make use of the respective advantages of Genetic Algorithm （GA） and BP neural network ,integrated algorithm , BP-GA algorithm and GA-BP algorithm are brought for ward. Genetic algorithm is used to optimize the topological structure ,the weight values and the biases of BP neural network. Furthermore, a set of advanced coding technique and evolution tactic are studied and realized to overcome the blindness resulted from experiences and tries of traditional BP network. The simulation results of instances show that optimised metwork is of better fitting capability and of more reasonable method than traditional BP network.

关 键 词：遗传算法 神经网络 拓扑结构 权值 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]