Efficient Electricity Theft Detection Using Hybrid CNN-XGBoost Model

Abstract

Non technical losses especially in distributed networks play key roles in electricity theft that pose serious challenges to power grids. As central electricity is distributed through the power grid to connect all consumers, any fraudulent usage is capable of interfering with the operations of the grid, produce low-quality supply, and even destroy the overall system. This means, as the data volume increases it becomes arduous to identify such fraudulent activities. Smart grids provide a solution in this aspect since electricity flow is bidirectional providing a channel for detection, correction and application of the corrective measures to the flow of the electrical data. Today’s electricity theft detection techniques incorporate one-dimensional (1-D) electric data leading to maximum possible imprecision. This work proposes a model that integrates CNN and XGB known as CNN-XGB. To supplement 1-D theft detection framework, the proposed model includes both 1-D and 2-D power usage data. A comparison with existing benchmark methods, using experimental sample results, shows that the proposed model delivers accurate results for the task, which was the main objective of designing the model.