Data Manipulation Attack Mitigation in Power Systems Using Physics-Informed Neural Networks

The growing digitalization of power transmission systems has introduced vulnerabilities to cyberattacks, particularly sensor manipulation, which can compromise control center monitoring and control applications. Conventional monitoring methods, including traditional state estimation and data-driven machine learning approaches, may face challenges in maintaining accuracy and reliability when confronted with such attacks. This paper proposes a physics-informed neural network (PINN)-based state estimation model to enhance robustness. By embedding power system physics, the PINN approach leverages both data-driven insights and physical constraints, improving robustness and accuracy in case of cyberattacks. The model’s performance is evaluated under diverse attack scenarios, comparing it to an equivalently sophisticated purely data-driven machine learning model. The results show that the PINN-based approach significantly improves robustness and accuracy, with or without the presence of attacks, offering a promising solution for robustifying power grid state estimation.