基于L-M法的BP神经网络在电抗器雷电冲击波形调整中的应用  

Application of BP Neural Network in Lighting Impulse Waveform Adjustment of Power Reactor Based on L-M Method

作　　者：李华良[1] 孙敏[2] 李博[1] 钟磊[1] 陆建挺[1] 郝宇亮[1] 

机构地区：[1]西安高压电器研究院有限责任公司,西安710077 [2]西安交通大学电气工程学院,西安710049

出　　处：《高压电器》2012年第3期52-56,62,共6页High Voltage Apparatus

摘　　要：建立了电抗器匝间绝缘雷电冲击调波参数组合寻优模型。为了克服传统BP神经网络训练收敛太慢以及容易陷入局部最小值等问题,采用了正交试验设计方法结合基于Levenberg-Marquardt法的BP神经网络对建立的数学模型进行了仿真训练并对设备雷电冲击调波参数组合进行了寻优,获得了较为精确的标准雷电冲击波形,对实际试验具有一定的指导意义。A reactor＇s turn-to-turn insulation lightning impulse waveform parameter mathematical optimization model has been established.In order to overcome the traditional BP neural network training convergence is too slow and easy to fall into local minimum problems.Orthogonal experimental design method together with BP neural network based on Levenberg-Marquardt method have been adopted to train the experiment data and to find the optimum parameter combinations,and a precise standard lightning impulse waveform has been obtained.Research results have certain significance to the actual test.

关 键 词：神经网络 正交试验设计 匝间绝缘 雷电冲击 

分 类 号：TM866[电气工程—高电压与绝缘技术] TM47