基于MCD稳健统计分析的变压器油色谱异常值检测及分布特性  

Outliers Detection and Distribution Characteristics of the Transformer DGA Data Based on MCD Robust Statistics

作　　者：高树国[1,2] 王学磊[2,3] 李庆民[2] 杨芮[2] 

机构地区：[1]国网河北省电力公司电力科学研究院,石家庄050021 [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 [3]山东大学电气工程学院山东省特高压输变电技术与气体放电重点实验室,济南250061

出　　处：《高电压技术》2014年第11期3477-3482,共6页High Voltage Engineering

基　　金：国家自然科学基金(51477051);北京市自然科学基金(3142018)~~

摘　　要：由于人为因素以及现场干扰的影响,变压器油色谱在线监测数据在测量和传输中会不可避免地出现偏误,导致异常数据值,使得经典统计方法的结果出现偏差甚至错误。为了解决上述问题,基于稳健统计理论,根据油色谱监测数据异常值的特点,提出了油色谱H2、CO和总烃3类特征气体异常值的最小协方差行列式MCD稳健多元检测方法。利用迭代和Mahalanobis距离的思想构造一个稳健的协方差估计量,并进行异常值检测,然后再将异常数据和正常数据分类处理。针对H2、CO和总烃这3类特征气体的实例统计分析表明,异常值剔除后可有效减少这3种气体的测量值对经典统计方法的干扰,使得油色谱数据的统计规律更加明显。通过对异常值区间的跟踪评估,还可更加明显地反映变压器运行状态的变化。In the process of on-line monitoring transformers by the dissolved gas analysis（DGA）, biases may be inevitably introduced by human factors and the interference during data measurement and transmission, which eventually leads to abnormal values, i.e. the outliers. Due to existence of the data outliers, the classical statistical methods may give results of large deviation or even mistakes. To deal with this issue, we proposed a DGA data outlier detection method for H2, CO, and the total hydrocarbons based on minimum covariance determinant MCD robust statistics. With integration of an iteration scheme and the Mahalanobis distance, we constructed a robust covariance estimator to detect the data outliers, and then processed the normal data and the outliers, respectively. The statistical analyses of the characteristic gases of H2, CO, and the total hydrocarbons, as well as their application to an on-site test example, prove that the outliers elimination facilitates an approximate normal distribution of the densities of these three gases and the interference caused by measured values of these three gases, and classical statistical methods can be reduced effectively. In addition, a more obvious reflection of the operational condition change of the power transformers can also be achieved by tracking assessment of the outliers section.

关 键 词：变压器 稳健统计 MAHALANOBIS距离 油中溶解气体分析 异常值 统计规律 

分 类 号：TM406[电气工程—电器]