基于最具影响粒子群优化的BP神经网络训练  

Back-propagation neural network training based on particle swarm optimization with best influential partical

作　　者：王慧[1] 刘希玉[1] 

机构地区：[1]山东师范大学信息科学与工程学院,济南250014

出　　处：《计算机工程与应用》2007年第18期69-71,86,共4页Computer Engineering and Applications

基　　金：山东省自然科学基金(the Natural Science Foundation of Shandong Province of China under Grant No.Z2004G02) ;山东省教育厅资助科研课题(the research Project of Department of Education of Shandong Province; China under Grant No.J05G01) ;"泰山学者"建设工程专项经费资助。

摘　　要：系统地介绍了粒子群优化算法,将粒子群优化算法用于BP神经网络的学习训练,提出了一种改进的粒子群算法——最具影响粒子PSO算法BIPSO,并利用复合适应度即均方误差和误差均匀度之和作为BIPSO训练神经网络的指标,并对它与其他的神经网络训练算法诸如BP算法、GA算法、PSO算法进行了比较。实验结果表明:BIPSO性能优于其他算法,更容易找到全局最优解,具有更好的收敛性。This paper systematically introduces particle swarm optimization algorithm and applies it to the training of neural networks.h advances an improved particle swarm optimization algorithm-Best Influential PSO （BIPSO）,which regards multiple fitness degree（SE and EU） as the guildline of BIPSO and compares it with BP,GA and PSO algorithms.The result of experiment indicates that BIPSO has better performance than other algorithms,finds global best solution more easily and has better astringency.

关 键 词：粒子群优化算法 神经网络 学习算法 最具影响粒子 复合适应度 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]