基于均值漂移聚类的开关柜局部放电异常检测

Anomaly Identification of Switchgear Insulation Condition Based on Mean-shift Clustering

针对开关柜现场带电检测数据,提出一种基于多维特征量的均值漂移聚类算法,对开关柜的局部放电进行异常识别。采用局部放电检测数据的离散度、平均距离百分比、集中度和最大波动率指标全面量化开关柜局部放电的程度,并构建多维特征数据库;通过自动搜索偏移量的均值漂移聚类算法对开关柜的状态进行划分,并通过所给定的开关柜簇标签隶属度函数判定是否为异常点,由此实现开关柜的绝缘状态异常检测。对现场带电检测实际数据进行实例分析,验证该方法的可行性,为开关柜的绝缘状态异常识别提供一定的理论依据。

Aiming at the on-site live detection data of switchgear,a mean-shift clustering algorithm based on multi-dimensional feature quantity was proposed to identify the abnormal partial discharge of switchgear.The switchgear partial discharge detection data which include dispersion,average distance percentage,concentration and maximum volatility were used to quantify the insulation of switchgear to construct multi-dimensional feature database.The insulation condition of the switchgear was divided by mean-shift clustering algorithm which can automatically search for the offset,and the abnormal point was determined by the membership function of the cluster labels to realize the abnormal detection of the switchgear insulation condition.The feasibility of the algorithm was verified by the live detection data,which can provide a certain theoretical basis for the switchgear evaluation of the switchgear.