摘要
为克服和改进BP算法的不足,文章在分析遗传算法(GA)和粒子群优化(PSO)算法优越性与不足的基础上,提出了一种基于GA和PSO结合的算法——GA-PSO算法,用于训练神经网络权值.算法产生下一代个体时,不仅采用交叉和变异算子,而且在重新定义局部最优粒子的基础上,引入粒子群优化算法,有效地结合了遗传算法的全局收敛性能和粒子群优化算法的局部搜索能力.通过对异或问题和IRIS模式分类问题的学习,仿真结果明显好于单纯地用GA或PSO进行前向神经网络训练,能有效避免早熟收敛的同时,提高搜索精度.
In order to overcome and improve the deficienciesof back-propagation(BP) algorithm,a GA-PSO algorithm was proposed after analying the genetic algorithm(GA) and particle swarm optimization(PSO).Based on redefine local optimization swarm,in GA-PSO,individuals in a new generation are created by not only crossover and mutation operation in GA,also PSO.So it could avoid local minimum and has good global search capacity.Through the learning of XOR and IRIS model sort,the performance of GA-PSO was compared with GA...
出处
《西北民族大学学报（自然科学版）》
2009年第4期27-31,64,共6页
Journal of Northwest Minzu University(Natural Science)
关键词
BP算法
遗传算法
粒子群优化
神经网络
BP algorithm
genetic algorithm
particle swarm optimization
GA-PSO algorithm
