Abnormal power consumption detection aims to identify and locate customers in the power system that deviate significantly from regular power consumption behavior. Existing supervised detection methods based on machine learning or deep learning generally require a large amount of manually labeled data, and require transformation for discrete data, thus leading to the loss of important information. FRS theory provides an effective tool for tackling discrete data. Therefore, FRS can be directly applied to the knowledge classification of heterogeneous information that includes continuous and discrete data. This paper proposed an unsupervised anomaly detection method with multi-granularity fuzzy relative differences based on FRS theory, and applied it to detect anomalous power consumption users in smart grid. Specifically, it first used information entropy of fuzzy approximation space to measure the importance of attributes for knowledge classification, then constructed a fuzzy granule sequence based on the attribute set's importance, and defined the fuzzy relative difference of the samples on top of this sequence. Finally, it constructed the anomaly detection method based on multi-granularity fuzzy relative differences and conducted evaluation on public datasets. Experimental results demonstrate the effectiveness and efficiency of the proposed algorithm. The code and data for the experiments are publicly available online(http: //www. github. com/chenbaiyang/FRAD).

[2023-07-12] Accepted Paper /doc/2023.05.0192/v1

[2023-11-05] Printed Article /doc/2023.05.0192/v2