Electric Power Remote Monitor Anomaly Detection with a Density-Based Data Stream Clustering Algorithm

Abstract

Nowadays data streams are more and more involved in the real industry. In this paper, the authors apply the data stream clustering to the electric power remote anomaly detection and propose a new data stream clustering algorithm based on density and grid (density-based data stream clustering algorithm, DBClustream). The double-frame analysis model is used in the proposed algorithm. In the online component, the authors optimize the initialization of the parameters for the K-means algorithm with a method based on density and grid and use the kernels to represent the micro clusters as the result of the online component. In the offline component, the time fading weight and the dynamic threshold to optimize the performance of the DENCLUE algorithm are proposed. To evaluate the performance of the proposed algorithm, both the evaluation of the anomaly detection and the evaluation of the data stream clustering are adopted. As the experiment result demonstrates, compared with the others algorithms, DBClustream can resolve the multi-density data stream and keep the high detection rate as well as the low false positive rate.