基于邻域保持嵌入-主成分分析的高压电缆状态数据异常检测及分析

为发现高压电缆异常状态并及时地发出异常告警,提出了一种基于邻域保持嵌入(neighborhood preserving embedding,NPE)和主成分分析(principal component analysis,PCA)的高压电缆异常状态检测方法。针对PCA只能保留数据全局结构信息的缺陷,提出将流形学习算法NPE与PCA结合,从而实现数据全局和局部特征信息的全方面提取;然后利用T2和SPE统计量作为电缆状态特征量,其控制限作为状态异常阈值判据,并推导出不同异常状态特征指标的贡献度,确定高压电缆主要异常指标;接着通过计算高压电缆各分段统计量的值,确定电缆异常区域;最后利用广东珠海供电局辖区内220 k V高压电缆统计资料验证所提策略的正确性。

Anomaly Detection and Analysis of State Information of High Voltage Cable Based on Neighborhood Preserving Embedded and Principal Component Analysis

In order to discover the abnormal state of the high voltage cable and issue an abnormal warning in time, this paper proposed an abnormal state detection method of high voltage cable based on neighborhood preserving embedded (NPE) and principal component analysis (PCA). Since PCA can only detect data global structure information, the manifold learning algorithm NPE and PCA are combined to achieve the global and local feature information extraction from all aspects. Then, T2 and SPE statistics are used as the characteristics of cable anomaly detection and their control limits are taken as the criterion of abnormal state, and the main abnormal index of high voltage cable is determined based on the contribution degree of different abnormal state indexes. the abnormal area of the cable is determined by calculating the statistics of each segment. Finally, the correctness of the proposed strategy is verified by the statistical data of 220kV high voltage cable in Zhuhai, Guangdong.