Electricity Theft Detection and Classification Method based on D-S Feature Fusion and IALO-SVM

Electricity theft and detection and classification are playing an essential role in smart grid operation. Aiming at solving the problem that massive data, including voltage, current, and power factor, have not been widely applied in electricity theft detection, a multi-layer Support Vector Machine (SVM) based on Dempster-Shafer (D-S) feature fusion and improved Antlion algorithm (ALO) is proposed. Fuzzy-C-Mean (FCM) based Synthetic Minority Oversampling Technique (SMOTE) is applied to balance the dataset. Then, an improved D-S evidence framework is applied to fuse the abnormal characteristics of multi-dimensional electricity consumption data, based on which the abnormal customers can be detected. Further, a multi-layer SVM classification framework is proposed, and an improved ALO is applied for tuning. Numerical analysis on real datasets verifies that the proposed method has superior classification performance. Besides, the training time is reduced by about 50% compared with traditional SVM.