Anomaly detection for power consumption patterns in electricity early warning system
Huadong Qiu, Ying Tu, Yan Zhang · 2018
In this paper, a monitoring and alarm system is designed, and an anomaly detection algorithm based on Log analysis (ADLA) is proposed. The system architecture includes the modeling system module, the real-time system module, the knowledge base module and the model database module. First, we extract many characteristics that characterize the user's power consumption pattern. We map each user to a two-dimensional plane through the principal component analysis (PCA), which can easily display data and compute local outliers. The grid processing technology selects the abnormal value of the low density region, which greatly improves the efficiency. The results show that by using this abnormal sequence, we can find that most LOF users can find most of the abnormal consumption patterns and output possible fault types. The system has achieved good results in practical operation.