基于马氏距离的双层聚类电力远动异常检测

Electric Power Remote Monitor Anomaly Detection with Two-level Clustering Algorithm Based on Mahalanobis Distance

通过对电力远动监测系统和数据挖掘技术的讨论,提出一种基于马氏距离的双层聚类异常检测算法。针对远动系统数据非球面分布的特点,该算法通过K-means聚类改进算法对数据进行初始分类,然后使用基于马氏距离的Clustering Using Representatives(CURE)聚类改进算法对初始分类结果进行优化,以较少的计算成本去除K值设定的影响,达到预期的检测结果。同时,基于马氏距离的CURE聚类改进算法对球面和非球面分布的数据有非常好的适应能力。

After the discussion of electric power remote monitoring systems and the data mining technology, a two-level clustering anomaly detection algorithm based on Mahalanobis distance is proposed according to the characteristics of the non spherical distribution of data in remote monitoring systems. Firstly, the improved K-means clustering algorithm is used to classify the initial data, and then the algorithm of Clustering Using Representatives based on Mahalanobis distance is used to optimize the classification results to achieve the desired results by removing the effects of parameter K with less computation cost. Meanwhile, the CURE clustering algorithm based on Mahalanobis distance has good adaptability for data both in spherical and aspheric distribution.