|
Abstract:Cyber-physical systems (CPS) are complex systems comprised of physical and computational components, which are susceptible to various disturbances and attacks, leading to system failures and security breaches. In recent years, CPS resilience has garnered increasing attention, with some studies proposing CPS resilience methods. However, existing methods overlook the interdependence between different components of the information and physical layers in the CPS network, and exhibit limitations in scalability, adaptability, and efficiency. To address these issues, this paper introduces a multilayer cascade resilience recovery framework based on deep reinforcement learning. Firstly, the high degree of interaction between the information and physical layers in CPS resilience recovery is comprehensively synthesized, and this correlation relationship is modeled using an association matrix. Secondly, a hybrid resilience recovery strategy is proposed to segment the association matrix into horizontal and vertical slices, treating its resilience strategy solution as an optimization problem. Finally, a deep reinforcement learning algorithm centered on resilience prioritization is presented to solve the optimal policy for hybrid resilient recovery. |