Self-learning-based secure control of wind power generation systems under cyber threat: Ensuring prescribed performance

Abstract

The increasing prominence of wind energy underscores the need to prioritize cybersecurity measures, with a focus on recognizing vulnerabilities and formulating defensive strategies. Specifically, False Data Injection (FDI) attacks targeted at the communication between rotor speed sensors and wind turbine (WT) controllers can disrupt operations, potentially causing drive-train overload and reduced power generation efficiency. To mitigate these threats, this study introduces an adaptive prescribed performance optimal secure control strategy that employs a reinforcement learning (RL) to compensate the detrimental effects of FDI attack as well as actuator fault. To derive the optimal control policy, the complex Hamilton–Jacobi–Bellman (HJB) equation is solved, facilitated by an actor–critic-based RL approach, where actor and critic neural network (NN) manage control actions and performance assessment. To detect FDI attack, an anomaly detection is developed using a fixed-time disturbance observer. Stability analysis is performed using Lyapunov theory which guarantees semi-global uniformly ultimately bounded (SGUUB) of the error signal. To rigorously validate our approach, we implemented the controller using FAST code for the NREL WindPACT 1.5 MW reference turbine. Simulation results convincingly demonstrate the effectiveness of our proposed control strategy, confirming its potential to enhance both performance and security in real-world WT operations.