基于PCA和Fast-MCD的油中溶解气体在线监测异常数据识别  

Identification of Abnormal data in online dissolved gas monitoring in oil based on PCA and Fast-MCD

作　　者：王子凌 汪科 柴卫健 李业欣 邹国平 安斯光 WANG Ziling;WANG Ke;CHAI Weijian;LI Yexin;ZOU Guoping;AN Siguang(State Grid Zhejiang Electric Power Co.,Ltd.Research Institute,Hangzhou 310014,China;School of Mechanical and Electrical Engineering,China Jiliang University,Hangzhou 310018,China)

机构地区：[1]国网浙江省电力有限公司电力科学研究院,杭州310014 [2]中国计量大学机电工程学院,杭州310018

出　　处：《浙江电力》2025年第3期100-107,共8页Zhejiang Electric Power

基　　金：国网浙江省电力有限公司科技项目(B311DS230008)。

摘　　要：油中溶解气体在线监测装置对变压器运行状态的准确监测是保证其稳定运行的关键,监测装置若出现故障会导致在线监测数据异常,严重影响在线监测的效果。针对这一问题,提出了一种基于PCA(主成分分析)和Fast-MCD(快速最小协方差行列式)的变压器油中溶解气体异常数据识别方法。首先,利用PCA算法对油中溶解气体时间序列进行降维,有效消除了冗余特征信息。然后,结合稳健统计理论,对降维后的数据进行Fast-MCD稳健分析,实现异常值识别。最后,分析识别出的异常值,判断故障来源。算例研究结果表明,该异常识别方法能够有效识别出油中溶解气体在线监测装置故障,性能优于其他常规方法,准确率高达99.1%。To detect the operational status of transformers by online monitoring of dissolved gases in oil is critical for ensuring its stable operation.However,faults in monitoring devices can result in abnormal data,severely impacting the effect of online monitoring.To address this issue,a method for identifying abnormal data in dissolved gas moni⁃toring based on principal component analysis(PCA)and fast minimum covariance determinant(Fast-MCD)is pro⁃posed.First,the PCA is used to reduce the dimensionality of time-series data for dissolved gas in oil,effectively eliminating redundant features.Next,robust statistical theory is applied to the reduced data using Fast-MCD to iden⁃tify anomalies.Finally,the identified anomalies are analyzed to determine the source of the fault.Case studies dem⁃onstrate that the proposed method effectively identifies faults in online dissolved gas monitoring devices,outperform⁃ing conventional methods with an accuracy of 99.1%.

关 键 词：异常检测 油中溶解气体分析 Fast-MCD 主成分分析 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]