Reinforcement-Learning-Based Fuzzy Adaptive Finite-Time Optimal Resilient Control for Large-Scale Nonlinear Systems Under False Data Injection Attacks
Jipeng Zhao, Guang‐Hong Yang
Abstract
In this article, the reinforcement learning (RL)-based finite-time adaptive optimal resilient control issue is studied for uncertain large-scale nonlinear systems under unknown sensor false data injection (FDI) attack. Due to the state information of the nonlinear system being corrupted by an additional attack signal, the true state information is unavailable for controller design. To circumvent these obstacles, with the help of RL-based actor–critic architecture, a novel finite-time adaptive optimal control algorithm for each subsystem is developed to alleviate the negative impacts of cyberattacks that deliberately tamper with sensor signals. Furthermore, the proposed resilient adaptive optimal control approach for a compromised nonlinear system ensures that all the signals of the overall system remain bounded in finite time. In contrast to the current results, the presented control method not only addresses the finite-time optimal resilient control issue of large-scale nonlinear interconnected systems with backlash-like hysteresis but also eliminates the continuous excitation conditions commonly required in existing RL-based optimal control schemes. Finally, simulation results confirm the efficiency of the developed methodology.