Data-Driven Optimal Controller Design for Maglev Train: Q-Learning Method

The maglev train is an open-loop and unstable complex nonlinear system. Generally, design of offline controllers based on a single operating state. However, the system of maglev train will influence by various complex factors in actual operation. When the system model changes, the controller is designed and tuned offline will suffer severe performance degradation that will affect the system's stability. This paper proposes a Data-Ddriven Optimal Controller (DDOC) based on the Q-learning theory in reinforcement learning in response to this problem. The controller does not need to know the model information of the controlled object, only calculates iteratively based on the system's real-time input, output data, which has the advantages of fewer tuning parameters and fast convergence speed. For the problem that system model change during operation, the method proposed in this paper makes the system accurately track the given reference signal by dynamically and rapidly changing the parameters of feedback gain matrix through the real-time data of the system, thus ensuring the stability and reliability of the control system.