摘要
BP神经网络已在模拟电路故障诊断领域得到广泛应用,但BP神经网络存在训练速度慢且容易陷入局部最优的问题.由此,本文提出了一种基于混合变异策略的微分进化改进算法,描述了利用微分进化改进算法进行神经网络权值训练的过程和方法,并将微分进化神经网络用于模拟电路故障诊断,文中还对微分进化神经网络与BP神经网络进行了比较.实验结果表明,微分进化神经网络的训练时间和训练精度均优于BP神经网络,其在模拟电路故障诊断中的准确度比BP神经网络提高了7%.
BP neural network has been widely used in analog circuit fault diagnosis. However, it has some inherent flaws： its training efficiency is limited and easy jump into the local optimal points. A modified differential evolution （MDE） algorithm with hybrid mutation strategy is proposed, the procedure and method of neural network training with MDE are described, and then the differential evolution neural network is used in analog circuit fault diagnosis. Meanwhile, the differential evolution neural network and BP neural network are compared. Experiment results show that the training efficiency and precision of the differential evolution neural network are prior to that of the BP neural network greatly, and the accuracy of analog circuit fault diagnosis is improved 7 %.
出处
《测试技术学报》
2012年第1期88-92,共5页
Journal of Test and Measurement Technology
基金
哈尔滨工业大学优秀青年教师培养计划资助(HITQNJS.2009.028)
关键词
微分进化算法
混合变异
神经网络
模拟电路
故障诊断
ifferential evolution algorithm
hybrid mutation
neural network
analog circuit
fault diagnosis
