采用机器学习的变压器分层故障诊断  

Multi-level Fault Diagnosis of Power Transformer Based on Machine Learning

作　　者：王子鉴 秦瑜瑞 李景丽[1] WANG Zijian;QIN Yurui;LI Jingli(School of Electrical Engineering,Zhengzhou University,Zhengzhou 450001,China;Zhengzhou Power Supply Company,State Grid Henan Electric Power Company,Zhengzhou 450000,China)

机构地区：[1]郑州大学电气工程学院,郑州450001 [2]国网河南省电力公司郑州供电公司,郑州450000

出　　处：《电力系统及其自动化学报》2022年第7期20-25,共6页Proceedings of the CSU-EPSA

摘　　要：油中溶解气体分析DGA(dissolved gas analysis)是进行变压器故障诊断的重要依据。针对传统变压器故障诊断未能有效利用特征气体与不同故障类型之间的相关性大小,对特征气体进行筛选,以致出现故障诊断结果不准确的问题,文章基于变压器分层故障诊断的方法,提出了利用卡方检验在每个故障诊断层中选取最优气体,剔除冗余气体,并采用不同机器学习分类器对所选择的最优气体进行分类的新方案。进行交叉验证后的结果表明,卡方检验能够有效提取特征气体,采用不同分类器进行分层故障诊断的效果优于用单个分类器对小类故障直接诊断。Dissolved gas analysis(DGA)is an important criterion for the fault diagnosis of power transformers.The tra⁃ditional fault diagnosis methods fail to screen the feature gases by effectively utilizing the correlation between feature gases and different fault types,leading to inaccurate fault diagnosis result.In this paper,a novel scheme is proposed based on the multi-level fault diagnosis method for transformers,in which the optimal gases at each fault diagnosis level are selected by means of the chi-square test,the redundant gases are eliminated,and the selected optimal gases are classified by using different machine learning classifiers.The results of cross-validation show that the chi-square test can effectively extract feature gases,and the effect of multi-level fault diagnosis using different classifiers is better than that using one single classifier to directly diagnose small fault classes.

关 键 词：油浸变压器 溶解气体分析 卡方检验 分层诊断 机器学习 

分 类 号：TM411[电气工程—电器]