Electricity theft detection based on hybrid random forest and weighted support vector data description

Abstract

Improving the detection rate of electricity theft users in smart grids is crucial to the safe operation of the power system and the economic efficiency of the grid. The traditional electricity theft detection relies on manual household inspection, which is costly and inefficient. With the application of smart meters, machine learning-based methods are gradually applied to detect electricity theft users. However, the training of the model is limited by the unbalanced electricity consumption dataset, and the performance of the model is limited in the process of processing high-dimensional data. Therefore, this paper adopts ensemble learning and imbalance data classification model to analyze the electricity theft detection problem in complex grid environment, and proposes an electricity theft detection model based on random forest and weighted support vector description. Firstly, the feature selection algorithm for mixed load data with statistical features based on random forest (FS-MLDRF) is used to extract the features that can best represent the customers’ electricity consumption behavior, so as to reduce the complexity of user data. Finally, an electricity theft detection based on hybrid random forest and weighted support vector data description is constructed (HFR-WSVDD), and the SVDD classifier is trained with the features weighted by feature importance to improve the performance of power theft detection in the case of unbalanced data. The simulation experiment results on the real electricity dataset show that compared with existing feature selection algorithms and anomaly detection algorithms, the algorithms proposed in this paper have excellent feature representation capability and better classification performance.