Multi-Layered Optimization for Adaptive Decoy Placement in Cyber-Resilient Power Systems Under Uncertain Attack Scenarios

The increasing reliance on digital infrastructures in power systems, combined with the rising penetration of renewable energy sources (RES), has heightened their vulnerability to sophisticated cyber-physical attacks, particularly false data injection a ttacks (FDIAs). These attacks exploit state estimation processes to disrupt grid operations while remaining undetected. This paper presents a novel multi-layered optimization framework to enhance the resilience of cyber-physical power systems against FDIAs under uncertain attack scenarios. The framework employs a tri-level Stackelberg optimization approach to model the interactions between defenders, attackers, and system operations. The defender's strategy focuses on optimal resource allocation and adaptive decoy placement to misdirect attacker efforts while minimizing operational costs. The middle level simulates attacker strategies using generative adversarial networks (GANs) to generate stealthy and adaptive attack vectors. The lower level incorporates physical and operational constraints of the grid, ensuring realistic scenario modeling. Advanced methodologies, including multi-agent deep reinforcement learning (MADRL), Bayesian inference, and distributionally robust optimization, are integrated to address dynamic uncertainties and evolving attack patterns. The proposed framework is validated on a modified IEEE 123-bus system with synthesized attack scenarios, demonstrating significant improvements in grid resilience. Results indicate an average reduction in attack success rates by 40% and an enhancement in resilience metrics by 35%, achieved through optimized defense budget allocation and adaptive decoy strategies. This research contributes to the field by bridging game theory, robust optimization, and machine learning, offering a comprehensive solution to ensure the security and reliability of modern power systems under extreme cyber-physical threats.