A Hybrid Physics-Deep Learning Load-Altering Attack Detection and Localization Mechanism
Mohammad Ali Sayed, Mohamed Amine Arfaoui, Ribal F. Atallah, Mourad Debbabi, Chadi M. Assi · 2024
Due to their growing popularity, smart high-wattage devices have rapidly proliferated in the modern power grid. Additionally, the grid has witnessed an increasing integration of Information and Communication Technologies (ICTs) to facilitate the shift towards a more efficient smart grid. This integration, however, has made the smart grid and its subsystems prone to cyber-attacks. One especially dangerous family of attacks is the family of Load-Altering (LA) Attacks that manipulate the demand side of the power grid and cause dangerous instabilities and blackouts. To this end, we develop a Hybrid Physics-Deep Learning (DL) LA attack detection and localization mechanism. In this mechanism, the utility collects the generator frequencies and feeds them into the mathematical equations representing the physical behavior of the grid to estimate the load variation patterns on the load buses. These patterns are then fed to a DL algorithm to identify the attacks. The performance of the proposed mechanism is evaluated against several Machine Learning and DL algorithms and detection models in the literature. Our proposed mechanism proved to be more accurate with an accuracy of 99.75%.