Enhancing cybersecurity in wind turbines: A resilient reinforcement learning-based optimal control for mitigating FDI attacks

The growing importance of wind energy underscores the critical importance of cybersecurity protocols, especially in identifying vulnerabilities and developing defenses. In particular, False Data Injection (FDI) attacks targeting the communication link between rotor speed sensors and wind turbine (WT) controllers (WT) pose a significant threat and can lead to operational disruptions such as drive train overload and reduced power generation efficiency. In response to these challenges, this study presents an innovative and robust learning-based control framework for WT systems with state constraints. This framework integrates an actor-critic Reinforcement Learning (RL) mechanism with a backstepping approach that utilizes a Barrier Lyapunov Function (BLF) to limit rotor speeds uniformly within a predetermined range to ensure adaptation to a smoothly feasible set. The learning-based control strategy, augmented by backstepping techniques, is formulated using a constrained Hamilton-Jacobi-Bellman (HJB) function. This architecture incorporates adaptive neural network identifiers that enable iterative updates of both actor and critic components. The key advance of this approach lies in its theoretical foundations, which show that the developed elastic scheme guarantees the boundedness of all system states within the predefined compact set. Finally, empirical results from implementation experiments confirm the effectiveness and robustness of the proposed control methodology.