Enhancing power grid cybersecurity against FDI attacks via deep Q-network-based moving target defense

Abstract

Cybersecurity threats such as False Data Injection (FDI) attacks pose significant risks to modern power systems, undermining both operational stability and economic efficiency. To address this challenge, we propose an Intelligent Moving Target Defense (iMTD) framework that enhances grid resilience by dynamically modifying the reactances of selected transmission lines using a Deep Q-Network (DQN). This strategy obscures system parameters from potential attackers while ensuring minimal disruption to power flow and cost. Unlike existing methods, such as Pareto-based Multi-Objective MTD (MO-MTD) and the Smallest Principal Angle (SPA) approach, the iMTD model intelligently identifies and perturbs the most influential lines to maximize attack detectability with minimal operational cost impact. A cost-aware reward structure is designed to balance cybersecurity and system efficiency. The proposed framework is evaluated on the IEEE 118-bus test system under both random and adversarial FDI attack scenarios, including stealthy, topology-aware, economic, sparse, adaptive, and coordinated attacks. Simulation results demonstrate that, under random FDI attacks, the iMTD achieves an average attack detection rate of 91.3 % while maintaining an OPF cost increment below 0.0003 %, outperforming SPA and MO-MTD benchmarks by up to 99 % cost reduction. Under worst-case adversarial attacks, detection performance stabilizes at 52.3 % with virtually zero cost increment, highlighting the robustness of the learned defense policy against intelligent attackers. These results highlight the potential of intelligent reinforcement learning techniques in developing adaptive and cost-effective cybersecurity solutions for cyber-physical power systems.