改进粒子群算法对BP神经网络的优化  

Improved Particle Swarm Optimization Algorithm

作　　者：沈学利[1] 张红岩[1] 张纪锁[1] 

机构地区：[1]辽宁工程技术大学电子与信息工程学院,辽宁葫芦岛125105

出　　处：《计算机系统应用》2010年第2期57-61,共5页Computer Systems & Applications

摘　　要：介绍一种基于改进粒子群算法优化BP网络的权值调整综合方法。该算法在传统BP算法的误差反传调整权值的基础上,引入粒子群算法的权值修正,并且在训练神经网络权值的同时优化其连接结构,删除冗余连接,从而建立了基于粒子群算法优化的BP网络新模型。结果表明,改进算法不仅可以克服传统BP算法收敛速度慢和易陷入局部权值的局限,而且很大程度地提高了结果精度和BP网络学习能力。A new method to adjust weights of BP network is proposed. The new model is based on the weight adjustments of traditional BP algorithm by tuning the structure and connection weights of BP network and improved particle swarm optimization simultaneously. The result shows that the improved algorithm can not only overcome the limitations in both the slow convergence and the local extreme values of traditional BP algorithm, but also improve the precision of the result and the learning ability greatly.

关 键 词：粒子群算法 惯性权值 神经网络 BP算法 优化 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TM774[自动化与计算机技术—控制科学与工程]