Deep Reinforcement Learning Based Moving Target Defense for Mitigating False Data Injection Attacks in Power Grids

The integration of physical and cyber elements in Cyber-Physical Power Systems (CPPS) is revolutionizing traditional infrastructure operations, enabling better monitoring, control, and automation. However, this integration also exposes systems to complex cybersecurity vulnerabilities, particularly False Data Injection (FDI) attacks, which can compromise the integrity and functionality of the power grid. The integration of several energy sources, modern communication technologies, and the unpredictability of variations in energy output and demand all contribute to the increasing complexity of CPPS, highlight the short-comings of traditional cybersecurity techniques, and emphasize the pressing necessity for adaptable protection measures. Moving Target Defense (MTD) is a strategy involving system configuration changes to complicate the attackers’ reconnaissance efforts. Recently, MTD has been applied to power system State Estimation (SE), incorporating Distributed Flexible AC Transmission System (D-FACTS) devices, thereby enhancing the security and resilience of CPPS against threats like FDI attacks. This thesis expands on the employment of MTD by proposing a defensive mechanism using Deep Reinforcement Learning (DRL) to create an intelligent and adaptable system. The approach uses the dynamic nature of DRL to develop an adaptive and intelligent system capable of identifying and mitigating FDI attacks with a high detection rate. The effectiveness of this approach is examined on the IEEE 14-bus system using a real-world dataset. The simulation results show promising outcomes, highlighting the potential of DRL-based MTD solutions for enhancing the resilience of power grids against stealthy FDI assaults.