Topology Attack Detection in Smart Grids Through Learning Graph Signal Smoothness

This paper focuses on the investigation of topology information attacks within power systems. These attacks are treated as a specialized form of false data injection attacks, aiming to manipulate topological information to create the false impression of a system topology change. These manipulations have the potential to disrupt the performance of models that support power system operations, such as state estimation models. In this work, the impact of topology information attacks on a state estimation model is presented to demonstrate the need to advance our understanding of these attacks and their effects and to develop methods capable of promptly and accurately detecting them. Next, a graph signal processing approach is under-taken that analyzes the smoothness properties of power system graph signals under various topology information attack scenarios. The graph smoothness features of graph signal measurements are then utilized to train machine learning models to detect such attacks. The performance of these models are evaluated under various scenarios and locations of topology information attacks.