Anomaly detection of power Consumption based on waveform feature recognition

In order to identify the fraudulent power consumption, in this paper we propose a waveform feature extraction model for anomaly detection. Based on data mining technology, we proposed a method for detecting abnormal electricity users by combining the line loss and power analysis. By extracting the edge point sequence, we effectively identify behavioral pattern time sequence represented and overall trends. Then applying the vector space cosine similarity to measure the similarity distance between loss changing pattern of regional electricity customers and targeted feeder, sorting based on cosine distance in order to identify high-risk behavior and suspicion electricity users; Finally, by analyzing the examples, we demonstrate the effectiveness and advantages of this method which could be directly used for electricity real-time data monitoring.