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

发布时间：2018-10-10 19:12

【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。
[Abstract]:Due to the influence of human factors and field interference, errors will inevitably occur in the measurement and transmission of on-line monitoring data of transformer oil chromatography, which will lead to abnormal data values, resulting in deviation or even error of the results of classical statistical methods. In order to solve the above problems, based on the robust statistical theory and according to the characteristics of abnormal values of oil chromatographic monitoring data, a minimum covariance determinant MCD robust multivariate detection method for abnormal values of H2CO and total hydrocarbons in oil chromatography is proposed. Using the idea of iteration and Mahalanobis distance, a robust covariance estimator is constructed, and the outliers are detected, then the abnormal data and the normal data are classified and processed. The statistical analysis of H _ 2CO and total hydrocarbon shows that the outliers can effectively reduce the interference of the measured values of these three gases to the classical statistical methods and make the statistical law of oil chromatographic data more obvious. By tracking and evaluating outliers, the change of transformer operation state can be more obviously reflected.
【作者单位】：国网河北省电力公司电力科学研究院;华北电力大学高电压与电磁兼容北京市重点实验室;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;
【基金】：国家自然科学基金(51477051) 北京市自然科学基金(3142018)~~
【分类号】：TM406

【参考文献】