基于点对称变换的乙丙橡胶电缆终端缺陷诊断  

Defect Diagnosis of EPR Cable Terminal Based on Symmetrized Dot Pattern

作　　者：周利军[1] 刘聪[1] 权圣威 曹伟东 项恩新 Zhou Lijun;Liu Cong;Quan Shengwei;Cao Weidong;Xiang Enxin(College of Electrical Engineering Southwest Jiaotong University,Chengdu 611756 China;Electric Power Research Institute Yunnan Power Grid Co.Ltd,Kunming 650217 China)

机构地区：[1]西南交通大学电气工程学院,成都611756 [2]云南电网有限责任公司电力科学研究院,昆明650217

出　　处：《电工技术学报》2022年第9期2388-2398,共11页Transactions of China Electrotechnical Society

基　　金：四川省科技计划(2020JDTD0009);广东省基础与应用基础研究基金(2020B1515130001)资助项目。

摘　　要：为解决高速动车组车载电缆健康状况的智能化诊断受限于天窗期短的问题,该文提出一种基于点对称(SDP)的乙丙橡胶(EPR)电缆终端缺陷局部放电诊断方法。首先,搭建局部放电试验平台获取局部放电信号;然后,提出一种车载电缆局部放电信号的SDP参数确定方法,并基于SDP变换将不同类型缺陷局部放电信号映射到极坐标系中形成SDP图像;最后,对比三种常见的深度学习网络——卷积神经网络(CNN)、栈式自编码器(SAE)及深度置信网络(DBN)提取不同类型缺陷的SDP图像深层特征,并基于网络尾端Softmax分类器进行识别。结果表明:针对四种典型的电缆缺陷,DBN网络与SDP图像的结合效果最佳,缺陷识别率达到了96.1%,相比于传统诊断方法,识别准确率提高了10%左右,由此验证了通过深度学习算法自适应提取SDP图像特征的方法,可有效应用于电缆缺陷诊断领域,具有较好的工程应用前景。In order to solve the problem that the intelligent diagnosis of the health condition of onboard cables of high-speed electric multiple units(EMU)in China is limited by the short skylight period,a partial discharge(PD)diagnosis method based on symmetrized dot pattern(SDP)for the terminal defects of ehglene propylene rubber(EPR)cables is proposed.Firstly,the partial discharge test platform was built to obtain the partial discharge signal.Then,a method to determine the SDP parameters of the on-board cable partial discharge signals was proposed,and based on SDP transformation,the partial discharge signals of different types of defects were mapped to the polar coordinate system to form SDP images.Finally,three common deep learning networks—convolutional neural network(CNN),stack autoencoder(SAE)and deep belief network(DBN)were compared to extract deep features of SDP images of different types of defects,and Softmax classifier at the end of the network was used to identify them.The results shows that:Aiming at four typical cable defect,DBN network combined with the SDP image effect is best,defect recognition rate reached 96.1%,compared with the traditional diagnosis methods,identification accuracy increase by about 10%,thus verified through deep learning algorithm of adaptive SDP image feature extracting method,can be effectively used in cable defect diagnosis,and it has a good prospect of engineering application.

关 键 词：乙丙橡胶 深度学习 对称点模式 状态识别 

分 类 号：TM247[一般工业技术—材料科学与工程]