Graph Neural Networks Approach for Cyberattack Detection in Smart Grids

Due to the development and applications of cyber-physical systems (CPS) and industrial Internet of Things (IoT), smart grids have significantly evolved the traditional power grids. Their increased utilization also exposes the vulnerabilities in smart grids, such as cyberattacks. Because of the impressive performance of machine learning models across various application domains, they are increasingly being applied for detecting cyberattacks in smart grids. However, most existing machine learning-based methods do not utilize the grids’ topology information leading to inefficiency in detecting network-level attacks in the smart grids. In this paper, we propose a new approach to cyberattack detection by utilizing the graph topology of smart grids. We define the network topology of smart grids as graphs to exploit spatial features of smart grids’ topological information, enabling detection of cyberattacks using a graph neural network (GNN)-based approach. Experimental results indicate that GNNs outperform other traditional machine learning models, highlighting the potential of exploiting GNNs on cybersecurity-related works for smart grids.