Safe Reinforcement Learning Control for Maglev Train Levitation System with Stability Guarantee and Safety Constraint

High-speed maglev trains represent a new intelligent transportation mode that combines environmental friendliness with high speed. Under high-speed operations of 600 km/h, the maglev train levitation system faces numerous challenges, such as track irregularity and response delay. Most existing levitation control methods are problem-specific and lack sufficient adaptability. While reinforcement learning-based methods offer improved adaptability, they do not guarantee theoretical stability. This paper presents a safe reinforcement learning control method for the levitation system based on Lyapunov stability theory and control barrier function (CBF). First, the system model is developed, considering various disturbances and time delays. Subsequently, stability constraints were constructed based on Lyapunov theory, and a control law was designed by integrating deep reinforcement learning algorithm. Considering physical constraints, a cost function is constructed and a CBF is introduced to guide the system away from constrained boundary. Through rigorous mathematical analysis, it is proven that the proposed method guarantees the stability and safety of the levitation system. Experimental results show that under track irregularity and actuator time-delay disturbances, the root mean square error of the air gap controlled by the proposed method is reduced by 35.7% and 24.3%, respectively, compared with the adaptive sliding mode method, and it exhibits stronger robustness and better time-delay resistance.