基于改进免疫克隆算法的BP网络权值优化

在简单分析了BP网络的缺陷以及应用遗传算法来优化网络权值存在的不足之后,采用了一种免疫克隆算法对BP网络的权值进行优化,并对算法的变异算子进行了改进,改进后的变异算子把高斯变异与柯西变异有效地结合在一起,兼顾了精确的局部搜索与大范围搜索的优点,在扩大算法搜索范围的同时也保证了算法搜索的精细度.仿真实验结果表明,改进后的免疫克隆算法不仅有效地提高了BP网络学习的速度,还很好地改善了算法收敛性能,克服了遗传算法收敛速度慢,易陷入局部极优的缺点,可以成功地应用于BP网络的权值优化.

Optimization of the Weights of BP Network Based on Modified Immune Clonal Algorithm

After analysing the shortcomings of BP Network and Genetic Algorithm which is used to optimize the weights of network briefly, the paper applies an Immune Clonal Algorithm to optimize the weights of BP Network,and modifies the mutation operator of algorithm. The modified mutation operator combines the Gauss Mutation with Cauthy Mutation effectually,and has the advantages of both the local search and large-scale search. It can also ensure the precision of search when enlarging the area of search. The simulation results show that the modified method impoves the learning velocity of BP Network effectually and the capability of convergence,and it overcomes the shortcomings of Genetic Algorithm, and can be applied to the optimization of BP Network successfully.