基于不平衡数据扩充的接地变压器故障诊断方法研究  

Research on Diagnosis Method of Grounding Transformers Based on Expansion of Imbalanced Data

作　　者：张伟 安勇 李海涛 杜银景 张红敏 盛雨 ZHANG Wei;AN Yong;LI Haitao;DU Yinjing;ZHANG Hongmin;SHENG Yu(Heze Power Supply Company,State Grid Shandong Electric Power Company,Heze 274000,China)

机构地区：[1]国网山东省电力公司菏泽供电公司,山东菏泽274000

出　　处：《电工技术》2023年第20期146-148,154,共4页Electric Engineering

基　　金：国网山东省电力公司科技项目“接地变压器运行状态在线检测关键技术研究与应用”(编号520614220004)。

摘　　要：接地变压器为中性点不接地系统提供中性点,可有效减小配电网发生接地短路故障时的对地电容电流。接地变压器运行状态直接关系到配电系统供电可靠性。现有智能故障诊断方法通常忽略了实际诊断时训练数据不平衡问题,增加了其落地应用难度。为此,引入过采样算法——MWMOTE算法对不平衡数据进行扩充。实验结果表明,MWMOTE算法缓解了不平衡数据带来的分类偏差,能够有效提高机器学习算法在接地变压器不平衡数据故障诊断中的诊断性能。A grounding transformer provides neutral point for a neutral ungrounded system,which can effectively reduce line-to-earth capacitive current when the distribution network undergoes a grounding fault.The operation status of grounding transformers directly affects power supply reliability of distribution system.Existing intelligent fault diagnosis methods usually ignore the imbalance of training data in the case of practical diagnosis,which increases the difficulty of its application.Therefore,this paper introduces an oversampling algorithm,i.e.,MWMOTE algorithm,to expand imbalanced training data.The experimental results show that MWMOTE can effectively alleviate the classification deviation caused by imbalanced data,and can effectively improve the diagnostic performance of machine learning algorithm when coping with faults with imbalanced data.

关 键 词：接地变压器 故障诊断 不平衡数据 过采样 机器学习 

分 类 号：TM933[电气工程—电力电子与电力传动]