前馈网络的遗传混沌搜索耦合学习算法研究

Research on a Coupled Learning Algorithm of Genetic Algorithm and Chaotic Search of Forward Neural Network

针对神经网络的BP算法易陷入局部极小的问题,提出了遗传混沌搜索耦合的学习算法。其原理是在遗传操作中加入混沌替换因子以防止算法早熟,而后对由遗传算法进行"粗搜索"所得的结果进行混沌"细搜索",有效地利用了遗传算法和混沌寻优的全局性的优点。普通的遗传编码是以一条长字符串为染色体,该方式存在搜索时间长、破坏了神经网络权值和阈值的整体性的缺点,提出的基于矩阵的细胞体编码方式克服了这一缺点。

This paper puts forward a coupled learning algorithm of genetic algorithm and chaotic search of forward neural network in view of the easily sinking into the local minimum of BP＇ s network. Its principle is that joining the chaos substitution factor into the Genetic Algorithm in order to prevent the emergence of precocity, and then carries on the chaotic ＂thin search＂ bases on the result of the chaotic ＂ thick search ＂. This algorithm has used the global optimization ability of Genetic Algorithm and chaotic optimization efficiently. Long character string forms the chromosome in the ordinary codes, so the search time is long and the wholeness of neural network＇ s weight and domain are destroyed. The method introduces in this paper bases on the matrix has overcome the shortcoming abovementioned.