基于CNN+D-S证据理论的多维信息源局部放电故障识别  

Partial discharge fault identification of multi-dimensional information sources based on CNN+D-S evidence theory

作　　者：王磊 张磊 牛荣泽 孙芊 李丰君 张周胜 WANG Lei;ZHANG Lei;NIU Rongze;SUN Qian;LI Fengjun;ZHANG Zhousheng(State Grid He'nan Electric Power Research Institute,Zhengzhou 450052,China;School of Electrical Engineering,Shanghai University of Electric Power,Shanghai 200090,China)

机构地区：[1]国网河南省电力公司电力科学研究院,河南郑州450052 [2]上海电力大学电气工程学院,上海200090

出　　处：《电力工程技术》2022年第5期172-179,共8页Electric Power Engineering Technology

基　　金：国家电网有限公司科技项目“面向配电物联网的多维监测与运行辅助决策关键技术研究”(5400-202024116A-0-0-00)。

摘　　要：基于多维信息源融合的局部放电故障识别方法对提高故障识别的准确性和容错性具有重要意义。文中以开关柜中的典型局部放电类型为识别对象,设置4种典型的局部放电模型(电晕放电、沿面放电、悬浮放电和气隙放电),利用超声波(Ultra)法、甚-特高频(V-UHF)法以及脉冲电流法(PCM)采集不同放电类型产生的局放信号。首先利用深度卷积神经网络(CNN)算法对不同传感器测量数据进行训练,之后利用Dempster-Shafer(D-S)证据理论对多维信息源识别结果进行融合,并作出最终决策。结果表明,相比于基于单一信息源的故障识别模式,基于多维信息源的故障识别模式准确率更高,且当多维信息源中某一信息源出现误判时仍能正确识别放电类型,对信息源的容错性更好,识别效果良好。The partial discharge fault identification method based on the fusion of multi-dimensional information sources can greatly improve the accuracy and fault tolerance in the fault identification of power equipment.In this paper,four typical partial discharge models,namely corona discharge,suspended discharge,floating discharge and air-gap discharge are prepared.The partial discharge signals generated by different discharge models are collected by ultrasonic(Ultra),very-ultra high frequency(V-UHF)and pulse current method(PCM)sensors.Firstly,the deep convolutional neural network(CNN)algorithm is used to train the measurement data of different sensors,and then the Dempster-Shafer(D-S)evidence theory is used to perform fusion calculation on the recognition results of multi-dimensional information sources.Finally,according to the fusion calculation results,the identification conclusion is made.The results show that the fault identification model based on multi-dimensional information sources constructed in this paper has higher accuracy than that based on single information source.When a misjudgment occurs in one of the multi-dimensional information sources,the model can still correctly identify the type of discharge,which indicates that the model has better fault tolerance for the information sources and the recognition effect is good.

关 键 词：局部放电 故障识别 深度卷积神经网络(CNN) Dempster-Shafer(D-S)证据理论 多维信息源 信息融合 

分 类 号：TM591[电气工程—电器]