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.
出处
《高压电器》
CAS
CSCD
北大核心
2012年第3期52-56,62,共6页
High Voltage Apparatus
关键词
神经网络
正交试验设计
匝间绝缘
雷电冲击
neural network
orthogonal experimental design
turn-to-turn insulation
lightning impulse
