基于BP神经网络的变压器故障诊断模型  

Fault Diagnosis Model of Transformer Based on BP Neural Network

作　　者：禹建丽[1] 卞帅[1] 

机构地区：[1]郑州航空工业管理学院,郑州450015

出　　处：《系统仿真学报》2014年第6期1343-1349,共7页Journal of System Simulation

基　　金：河南省自然科学基金项目(132102210091;142102210077;142102210105)

摘　　要：研究一种基于BP神经网络的变压器故障诊断方法。以绝缘油中6种气体作为神经网络的输入,建立可对变压器低能放电、高能放电、局部放电、低温过热、中温过热、高温过热以及混合型故障高温过热兼高能放电等7种故障进行诊断的BP神经网络模型。通过全面调节隐层神经元的个数对网络进行训练,优化网络结构和参数,用最大隶属原则处理网络输出,对多达60个的检验样本进行实际故障诊断,准确率达到了95%。建立的故障诊断模型是最为简单的BP神经网络模型,因此该模型便于在实际中应用,通过与其它改进的神经网络方法进行比较研究,结果表明,BP神经网络故障诊断模型的各项性能指标均处于较好水平。According to the characteristics of fault types of the transformer, BP neural network was used to diagnose transformer fault. Six gases were regarded as inputs of the neural network and established BP neural network model which could diagnose seven transformer faults： low discharge, high energy discharge, partial discharge, low temperature heat, medium temperature overheat, high temperature overheat, high temperature overheat and mixed fault of high temperature overhead and high energy discharge. By adjusting the number of hidden layer neurons, the network was trained to optimize the network structure and parameters. The output of the neural network was processed according to the maximum subordination principle. Simulation studies show that the rate of the transformer fault diagnosis accuracy reaches95%. Additionally, the BP neural network model is easy to establish and applicable to use.

关 键 词：BP神经网络 变压器 故障诊断 仿真 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]