基于深度学习融合证据理论的电缆接头故障检测方法研究  

Research on cable joint fault detecting method based on deep learning fusion evidence theory

作　　者：王晓康 丁雷 何嘉斌 马学灵 张童瑞 梁睿 WANG Xiaokang;DING Lei;HE Jiabin;MA Xueling;ZHANG Tongrui;LIANG Rui(Wuzhong Power Supply Company of State Grid Ningxia Electric Power Co.,Ltd.,Wuzhong 751199,China)

机构地区：[1]国网宁夏电力有限公司吴忠供电公司,宁夏吴忠751199

出　　处：《绝缘材料》2025年第3期125-130,共6页Insulating Materials

基　　金：国家电网有限公司科技项目(5229WZ230003)。

摘　　要：为了解决目前电力电缆监测手段相对单一、缺少多参量综合监测判断的问题,本文集成电缆接头的各类局放信号,提出基于深度学习融合证据理论的电缆接头故障监测方法,对电缆接头无缺陷、内部绝缘缺陷、接头受潮三类情况进行了故障识别研究。根据卷积神经网络算法模型,分别对各类局放信号图谱进行训练和测试,再通过D-S证据理论进行数据融合,实现故障类型识别。结果表明:对于内部绝缘缺陷、接头受潮情况,单一信息源识别效果最好的为高频局部放电,其平均识别率能达到85.6%,超声法识别率最低,其平均识别率为78.7%,而本文提出的识别方法平均识别率达到95.7%;当多维信息源中某一信息源的识别结果出现误判时,通过D-S证据理论融合能够排除错误信息源的干扰,准确识别放电类型。To solve the current problem that the monitoring methods of power cable are relatively simple and lack of multiparameter comprehensive monitoring and judgment,this paper integrated various partial discharge signals of cable joints,and proposed a cable joint fault monitoring method based on deep learning fusion evidence theory.Then fault identification research was conducted for three situations,including no defects,internal insulation defects,and joint moisture in cable joints.According to convolutional neural network algorithm model,the partial discharge signal graphs were trained and tested separately,and data fusion was carried out by D-S evidence theory to realize fault type identification.The results show that for the situation of internal insulation defects and joint moisture,the best recognition effect of single information source is high frequency partial discharge,and its average recognition rate can reach 85.6%,and the average recognition rate of ultrasonic method is the lowest,which is 78.7%,while the average recognition rate of the identify method proposed by this paper can reach 95.7%.When there is a misjudgment in the recognition result of one of the multi-dimensional information sources,D-S evidence theory fusion can eliminate the interference of the wrong information source,accurately identify the discharge type.

关 键 词：电缆接头 深度学习 信息融合 故障诊断 

分 类 号：TM215[一般工业技术—材料科学与工程] TM247[电气工程—电工理论与新技术]