A Q-learning based robust MPC method for DFIG to suppress the rotor overcurrent. (October 2022)

• Record Type: Journal Article
• Authors: Song, Yuyan; Wang, Yuhong; Zeng, Qi; Zheng, Zongsheng; Liao, Jianquan; Liao, Yiben
• Is Part Of: International journal of electrical power & energy systems. Volume 141(2022)
• Publication Date: 2022-10
• Subjects: Model predictive control (MPC) -- Q-learning -- Robust optimization -- Doubly fed induction generator (DFIG) -- Rotor overcurrent; Electrical engineering -- Periodicals; Electric power systems -- Periodicals; Électrotechnique -- Périodiques; Réseaux électriques (Énergie) -- Périodiques; Electric power systems; Electrical engineering; Periodicals; 621.3
• DOI: 10.1016/j.ijepes.2022.108106
• Languages: English
• ISSNs: 0142-0615

Abstract

Highlights: The state space equation of the rotor current in doubly fed induction generator (DFIG)-based wind turbine is established to make the transient DC component of induced electromotive force (IEF) in the rotor observable. The main control based on the robust model predictive control (RMPC) is proposed, which could predict and counteract the transient DC component of IEF with considering the uncertainty of system parameters. The corrective control based on the Q-learning is proposed, which enhances the adaptability and convergence of the main control. A novel Q-RMPC method composed of the main control and the correction control is put forward to suppress the DFIG rotor overcurrent under uncertain operation conditions. Abstract: During power grid faults, the rotor overcurrent in the doubly fed induction generator (DFIG)-based wind turbine may cause the activation of crowbar protection, which enlarges the voltage drop of the point of common coupling (PCC) and influences the power recovery. The existing methods, such as the demagnetization control, have the problems of overshoot and lack of robustness due to the uncertainty of the system. To solve these problems, a Q-learning based robust model predictive control (Q-RMPC) method for DFIG is proposed. A state space equation of the rotor current including the stator current differential term is listed to make the transient DC component of induced electromotive force (IEF) observable. On this basis, a RMPC-based main control to deal with the uncertainty of system is proposed. This control could realize the prediction of rotor current. To enhance the convergence of the main control, a novel correction control based on Q-learning is presented, which can correct the main control by reinforcement learning the decision strategy of perturbation. The Q-RMPC composed of the main control and the correction control is proposed to suppress the rotor overcurrent with considering the precarious of the system. The control effects of the proposed method in the scenarios of wind speed fluctuation, correction control absence, parameter disturbances and parameter offset are discussed. The simulation results verify the correctness and effectiveness of the proposed method.