Reinforcement Learning-Based Optimized Adaptive Secure Control for Constrained Fractional-Order Nonlinear Systems Under FDI Attacks
针对受约束分数阶非线性系统在虚假数据注入攻击下的控制问题，提出一种结合强化学习和模糊逻辑的自适应安全控制方法，实现优化控制并降低通信消耗。
This article considers the adaptive fuzzy optimized secure self-triggered control (STC) problem for constrained fractional-order nonlinear systems (FONSs) subject to unknown false data injection attacks (FDIAs). To fulfill the unilateral full-state constraints (UFSCs), an emerging nonlinear state-dependent function with convexity is utilized by means of its property in removing the feasibility conditions existing in the traditional constraint control. Meanwhile, since the true information of the state variables under attack signals is unavailable, a method of coordinate transformation combining the compromised system and dynamic surface control is used for designing the corresponding controller while alleviating the adverse impacts of the FDIAs. Additionally, by constructing the equivalent auxiliary systems and employing actor-critic neural networks (NNs), a reinforcement learning (RL)-based adaptive fuzzy optimal program is provided to achieve optimal control. Furthermore, considering that event-triggered mechanism needs real-time supervision of the control signals, a STC strategy is introduced to circumvent this drawback and reduce communication consumption. Leveraging the fractional Lyapunov stability theory, it has been confirmed that the devised controller can ensure that the stabilization errors tend toward a small area nearby the origin at the minimum costs and all the signals arising in the closed-loop system (CLS) are bounded. Eventually, two simulation examples are offered to demonstrate the reported control algorithm’s validity.