摘要
励磁涌流识别是变压器微机差动保护中的关键问题。在综合考虑多方面因素的前提下,提出了一种基于改进型动态结构神经网络的涌流识别新方法。其主要思想为在采用多层前向反馈传播BP(BackPropagation)网络改进型训练算法(即在调整权值和阈值时,添加Rumelbart惯性冲量技术)的基础上,通过改变隐层的神经元个数使网络结构模型达到最优。仿真试验表明由该方法设计的模型进行涌流识别效果良好。
The identification of magnetizing inrush is a key problem in transformer differential protection.With the integrative consideration of different factors,a new inrush identification method based on dynamic structure NN(Neural Network)is presented,which adopts an improved algorithm of multi-layer forward BP(Back Propagation)neural networks,i.e.the Rumelbart's add-inertia-impulse technology is applied in adjusting the weight value and bias value.The nerve center number of hidden floor is hence changed to optimize the NN struture.The simulative test indicates that the inrush identification based on this method is more effective.
出处
《电力自动化设备》
EI
CSCD
北大核心
2002年第11期67-69,共3页
Electric Power Automation Equipment
