一种免疫克隆选择算法在BP网络学习中的应用  

Application of Immune Clonal Selection Algorithm in BP Network Learning

作　　者：姜新农[1] 王文香[1] 

机构地区：[1]中国科技大学计算机科学技术系,合肥230027

出　　处：《计算机工程》2007年第5期164-165,167,共3页Computer Engineering

摘　　要：在分析BP网络学习存在的问题后,采用了一种免疫克隆选择算法对BP网络的权值进行优化学习,并提出了一种新的变异方法,该变异方法可以根据亲和力的大小自适应调整抗体变异的幅度,与传统的高斯变异相比,不但简化了抗体的编码,还很好地体现了克隆选择算法抗体变异的特点,提高了算法的搜索能力和收敛性能。仿真实验表明,基于这种变异方法的免疫克隆选择算法可以很好地提高BP网络的学习速度,有效地避免算法过早收敛的问题。After analyzing the shortcomings of learning BP network simply, the paper applies an immune clonal selection algorithm（ICSA）to optimizing the weights of BP network, and puts forward a new method of mutation. The method of mutation can adjust the extent of mutation adaptively by the affinity of antibody, in comparison with Gauss mutation, it can not only simplify the coding of antibody, but also incarnate the characteristic of mutation of ICSA and improve the capability of search and convergence of algorithm. The simulation results show that ICSA based on this method of mutation can improve the rapidity of learning BP network well and avoid prematurity effectively.

关 键 词：BP网络 免疫克隆选择算法 变异 运动学逆问题 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]