Resilient-Learning Control of Cyber-Physical Systems Against Mixed-Type Network Attacks
Abstract: This article develops a novel resilient-learning control strategy for cyber-physical systems (CPSs) to mitigate the impact of mixed-type network attacks. These attacks combine false-data-injection (FDI) and replay attacks, which are modeled using Markov jump processes. The attacks are assumed to be uncertain, and a three-layer neural network is employed to learn and approximate them. Based on these approximations, a resilient controller is designed, integrating adaptive laws to estimate the neural network weights in real-time. The proposed control strategy ensures the system's ultimate boundedness and asymptotic stability under attack conditions. To validate the efficacy of the approach, a vertical take-off and landing (VTOL) helicopter model is used for simulation, demonstrating the controller's robustness and effectiveness in maintaining system stability despite the presence of mixed-type network attacks.
Submission Number: 210
Loading