摘要
为解决神经网络训练中易出现的收敛速度缓慢、陷入局部极小点等问题,提出了一种新的带自适应遗传算子的粒子群神经网络训练算法,通过概率控制,在利用粒子群算法优化神经网络的同时,自适应地对备选粒子进行选择、交叉、变异等遗传操作,最后将算法应用于汽车发动机故障诊断神经网络模型的训练。试验结果显示,本算法继承了遗传算法全局搜索和粒子群算法收敛速度快的优点,能在较少的训练步数内,达到较高的收敛精度,且样本分类正确率比BP算法、遗传算法、粒子群算法显著提高。
A new particle swarm optimization algorithm with adaptive genetic operator（AGPSO） for training ANN was proposed to solve the problems appeared in the train of artificial neural network（ANN） such as the local minimum′s basin of attraction and low speed.Controlled by probability,the particles were operated by genetic operator when ANN is trained by PSO algorithm.This new algorithm was used to train the ANN model of vehicle engine fault diagnosis.The result shows that the neural network trained by AG-PSO algorithm needs least amounts of iterations and achieves the better training accuracy than BP algorithm,GA and PSO algorithm.
出处
《解放军理工大学学报（自然科学版）》
EI
北大核心
2011年第1期70-74,共5页
Journal of PLA University of Science and Technology(Natural Science Edition)
关键词
自适应
遗传算子
粒子群
神经网络
故障
模式识别
adaptive
genetic operator
PSO（particle swarm optimization）
ANN（artificial neural network）
fault
pattern recognition
