Adaptive electricity theft detection method based on load shape dictionary of customers

With the application of the advanced measurement infrastructure in power grids, data driven electricity theft detection methods become the primary stream for pinpointing electricity thieves. However, owing to anomaly submergence, which shows that the usage patterns of electricity thieves may not always deviate from those of normal users, the performance of the existing usage-pattern-based method could be affected. In addition, the detection results of some unsupervised learning algorithm models are abnormal degrees rather than “0-1” to ascertain whether electricity theft has occurred. The detection with fixed threshold value may lead to deviation and would not be sufficiently flexible to handle the detection for different scenes and users. To address these issues, this study proposes a new electricity theft detection method based on load shape dictionary of users. A corresponding strategy for tunable threshold is proposed to optimize the detection effect of electricity theft, and the efficacy and applicability of the proposed adaptive electricity theft detection method were verified from numerical experiments.