改进朴素贝叶斯模型在电力变压器故障定位中的应用  

Application of Improved Naive Bayes Model in Fault Location of Power Transformer

作　　者：范慧芳 咸日常[1] 王涛 高鸿鹏 陈蕾 张冰倩 FAN Huifang;XIAN Richang;WANG Tao;GAO Hongpeng;CHEN Lei;ZHANG Bingqian(College of Electrical and Electronic Engineering,Shandong University of Technology,Shandong Zibo 255049,China;State Grid Shandong Electric Power Company Zaozhuang Power Supply Company,Shandong Zaozhuang 277000,China)

机构地区：[1]山东理工大学电气与电子工程学院,山东淄博255049 [2]国网山东省电力公司枣庄供电公司,山东枣庄277000

出　　处：《高压电器》2023年第2期190-197,共8页High Voltage Apparatus

摘　　要：电力变压器故障能否精准定位一直是制约其状态检修有效开展的技术瓶颈。文中针对目前已有故障定位模型存在的不足,借助变压器故障类型与特征状态量之间的内在关系,将朴素贝叶斯网络模型进行特征属性加权改进,并将其扩展为改进的双层朴素贝叶斯网络模型应用至电力变压器故障定位中。在这一过程中,考虑到特征属性与类别之间和各特征属性之间的依赖关系,采用ReliefF算法和相关系数法分别对特征属性进行加权处理,构造出改进的朴素贝叶斯网络模型,并在MATLAB软件中进行了诊断对比预测,得到了较好的预测结果,文中最后利用实际案例进一步验证了所提模型与分析方法的有效性,可为电力变压器故障诊断提供技术指导。Accurate location on fault of power transformer has always been a technical bottleneck constraining the effective development of its state maintenance. In view of the shortcoming of the presently existed fault location model,the inherent relationship between the type of transformer fault and the characteristic state is used to improve the characteristics attribute weight of the Naive Bayes network model,is expanded into an improved two-layer Naive Bayes network model and applied to the fault location of power transformers. During this process,considering the dependency between feature attributes and categories and between each feature attributes,the ReliefF algorithm and the correlation coefficient method are used to weight the feature attributes respectively and to construct an improved Naive Bayesian network model. The model is diagnosed,compared and predicted in MATLAB software and good prediction results are obtained. Finally,the actual cases are used to further verify the effectiveness of the proposed model and analysis method,which can provide technical guidance for fault diagnosis of power transformer.

关 键 词：电力变压器 朴素贝叶斯 属性加权 故障定位 RELIEFF算法 

分 类 号：TM41[电气工程—电器] O212.8[理学—概率论与数理统计]