Ensemble-Learning-Based Decision Support System for Energy-Theft Detection in Smart-Grid Environment

Abstract

Theft of electricity poses a significant risk to the public and is the most costly non-technical loss for an electrical supplier. In addition to affecting the quality of the energy supply and the strain on the power grid, fraudulent electricity use drives up prices for honest customers and creates a ripple effect on the economy. Using data-analysis tools, smart grids may drastically reduce this waste. Smart-grid technology produces much information, including consumers’ unique electricity-use patterns. By analyzing this information, machine-learning and deep-learning methods may successfully pinpoint those who engage in energy theft. This study presents an ensemble-learning-based system for detecting energy theft using a hybrid approach. The proposed approach uses a machine-learning-based ensemble model based on a majority voting strategy. This work aims to develop a smart-grid information-security decision support system. This study employed a theft-detection dataset to facilitate automatic theft recognition in a smart-grid environment (TDD2022). The dataset consists of six separate electricity thefts. The experiments are performed in four different scenarios. The proposed machine-learning-based ensemble model obtained significant results in all scenarios. The proposed ensemble model obtained the highest accuracy of 88%, 87.24%, 94.75%, and 94.70% with seven classes including the consumer type, seven classes excluding the consumer type, six classes including the consumer type, and six classes excluding the consumer type. The suggested ensemble model outperforms the existing techniques in terms of accuracy when the proposed methodology is compared to state-of-the-art approaches.