摘要
遗传算法能有效解决神经网络优化设计存在的问题,但古典的遗传算法有一定的弊端。本文对遗传算法的操作算子进行改进,对结点和连接权采用两种不同的交叉规则,使子代结点个数在两父代之间,而子代个体的权值在较好的父代个体两侧;并增加一个变异概率,增大网络的结构进行突变的几率,这样既加快了搜索进程,在精度上也收到了很好的效果。
Genetic algorithm can resolve some problems during optimal design of neural networks, but there is a little disadvantage when using classic genetic algorithm. This paper does some modification to search operators： using different crossover rules to nodes and weights. This makes that numbers of ficial generation nodes are between the parents, and the weights of these nodes are in the vicinity of the better parental generation. A new mutation probability is added, which increases the mutation chance of the architecture of networks. The modified algorithm shows better effect in searching and precision.
出处
《长春理工大学学报（自然科学版）》
2006年第3期48-50,47,共4页
Journal of Changchun University of Science and Technology(Natural Science Edition)
关键词
遗传算法
遗传算子
神经网络
网络结构
连接权
genetic algorithm
genetic operator
neural networks
networks architecture
weights
