基于多分类支持向量机的变压器在线监测数据错误模式识别  

Error Pattern Identification of On-line Monitoring Data of Transformer Based on Multi-classification SVM

作　　者：何宁辉 吴旭涛 张佩 沙伟燕 周秀 丁培 杨擎柱 程养春[2] HE Ninghui;WU Xutao;ZHANG Pei;SHA Weiyan;ZHOU Xiu;DING Pei;YANG Qingzhu;CHENG Yangchun(State Grid Ningxia EPRI,Yinchuan 750011,China;Beijing Key Laboratory of High Voltage and EMC,North China Electric Power University,Beijing 102206,China)

机构地区：[1]国网宁夏电力有限公司电力科学研究院,银川750011 [2]华北电力大学高电压与电磁兼容北京市重点实验室,北京102206

出　　处：《高压电器》2024年第7期173-181,共9页High Voltage Apparatus

基　　金：国家电网科学技术项目(5229DK19004Z)。

摘　　要：针对变压器在线油中溶解气体在线监测数据质量问题,统计了200多台监测装置的2020全年数据,总结了3种主要数据错误模式;提出了数据错误模式识别策略和特征参数,构建了多分类支持向量机进行错误数据识别与分类;并利用核主成分分析法和排列组合遍历寻优法对特征向量进行了降维优化。所构建的多分类支持向量机分类器对于H_(2)错误数据识别准确率达到97.5%,对于其他气体达到90%以上。应用所构建的分类器对2020全年数据进行了统计,其中H2的错误数据达到27.14%,C_(2)H_(2)的错误数据达到1.75%。In view of online monitoring data quality of dissolved gas in transformer oil,the annual data of more than 200 monitoring devices in 2020 are counted and three main data error patterns are summarized.The identification strategy and characteristic parameters of data errors pattern are proposed,and a multi-classification support vector machine is constructed to identify and classify the data errors.The Kernel principal component analysis and permutation and combination traversal method are used to reduce dimension of the feature vector.The identification accuracy of the constructed multi-classification support vector machine classifier for H_(2) error data is 97.5%,and for other gasses is more than 90%.The constructed classifier is used to make statistics on the annual data of 2020,in which the error data of H2 and C_(2)H_(2) reach 27.14% and 1.75% respectively.

关 键 词：错误数据 模式识别 支持向量机 在线监测 变压器 油中溶解气体分析 

分 类 号：TM41[电气工程—电器] TP181[自动化与计算机技术—控制理论与控制工程]