Abnormal electricity detection method based on multi-dimensional unsupervised learning

In order to reduce the operation cost of power companies and assist the marketing inspection management in more efficient power consumption inspection, evidence collection, analysis and treatment, aiming at non-technical loss (NTL), this paper proposes an abnormal power consumption detection method combining "algorithm anomaly analysis" and "empirical method principle analysis". "Algorithm anomaly analysis" uses local outlier factor algorithm to detect from four perspectives: community evolution anomaly, group behavior anomaly, individual power anomaly and association feature anomaly. "Rule of thumb analysis" further screens the results of "algorithm anomaly analysis" from the perspective of three-phase voltage and current imbalance correction. The experimental results show that compared with the existing mainstream abnormal power consumption detection methods, the proposed method can more accurately diagnose the abnormal power consumption behavior of users from power big data.