Reinforcement learning-based fixed-time resilient control of nonlinear cyber physical systems under false data injection attacks and mismatch disturbances

Abstract

This paper presents a fixed-time adaptive resilient control framework based on reinforcement learning (RL), mismatch disturbance observer and nonsingular fast terminal sliding mode scheme for nonlinear cyber-physical systems (CPS) to enhance cyber-security under False Data Injection (FDI) attacks, as well as match and mismatch disturbances. The RL algorithm contains an actor-critic neural network (NN) in which the actor NN is employed to estimate the uncertainty while the critic NN is applied to evaluate the performance cost function. Next, to compensate the detrimental effects of attacks and mismatch disturbances, a new sliding mode-based adaptive disturbance observer is designed. To achieve high precision trajectory tracking within a fixed-time interval, a nonsingular fast terminal sliding mode scheme is designed. This scheme ensures fixed-time convergence of the tracking error and facilitates disturbance attenuation and attack mitigation which are the key features of the proposed fixed-time secure control strategy. Finally, the fixed-time stability analysis of is performed through Lyapunov theory. Results reveal the effectiveness of the proposed resilient control for wind turbine system as the nonlinear CPS.