Model predictive control based on Q-learning for magnetic levitation platform system

Article type: Research Article

Authors: Ke, Zhihao^a | Yi, Huiyang^b | Zhang, Penghui^a | Feng, Yuexin^a | Liang, Le^c | Deng, Zigang^{a; c;}

Affiliations: [a] State Key Laboratory of Rail Transit Vehicle System, Southwest Jiaotong University, Chengdu, China | [b] School of Information Science and Technology, Southwest Jiaotong University, Chengdu, China | [c] Research Center for Super-High-Speed Evacuated Tube Maglev Transport, Southwest Jiaotong University, Chengdu, China

Correspondence: [*] Corresponding author: Zigang Deng, State Key Laboratory of Rail Transit Vehicle System, Southwest Jiaotong University, Chengdu 610031, China. E-mail: [email protected]

Abstract: A model predictive control (MPC) method based on Q-learning algorithm, named QMPC, is proposed for weakly damped, nonlinear and open-loop unstable magnetic levitation platform (MLP) systems. In addition, the design of MPC controller for the MLP system, the state space of the MLP system airgap, the action space of the predictive horizon and control horizon, the reward and punishment function are also included in this research. Based on the Simscape and MATLAB/Simulink, the joint simulation of the MLP control system is realized. Compared with PID controller and traditional MPC controller, the simulation results show that QMPC controller has better disturbance rejection ability and tracking performance under six working conditions.

Keywords: Magnetic levitation platform, Q-learning algorithm, model predictive control, disturbance rejection ability, tracking performance

DOI: 10.3233/JAE-240003

Journal: International Journal of Applied Electromagnetics and Mechanics, vol. Pre-press, no. Pre-press, pp. 1-24, 2024