An Intelligent Machine Learning Approach for Smart Grid Theft Detection

Abstract

Smart grids are an improvement of the traditional electric grids. They allow a much higher degree of automation and more efficient power distribution. Nonetheless, due to automation, these grids become more vulnerable to cyber attacks. Hence, cyber security becomes a major milestone to overcome before we can permanently shift to smart grids. Electric theft is one of the most dangerous cyber attacks in a smart grid. It allows users to lie about their load profiles and decrease their electricity bills. Several research studies have been conducted regarding the detection of such cyber attacks in a smart grid, but none of them consider weather information as a feature. This paper proposes a novel machine learning-based approach to smart grid electricity theft detection using both the load profile of a household and the weather features. The results show that our current approach using both load and weather information perform much better than previous approaches that only use load information.