High-speed maglev train levitation systems face significant stability and safety challenges due to their inherent nonlinear open-loop instability and the complex operating environment caused by aerodynamic lift force and track irregularities. Existing model-based control methods rely on precise mathematical models and manual parameter tuning, struggling to adapt to complex dynamic environments. Learning-based approaches, meanwhile, suffer from difficulties in convergence under strong disturbances and insufficient safety guarantees. To address these limitations, this paper proposes a safe reinforcement learning control method based on higher-order control barrier functions (HOCBF) and disturbance observer (DOB). The proposed method employs a hierarchical design: reinforcement learning (RL) adaptively learns the optimal policy from data; HOCBF construct a safety layer to modify the RL agent's actions, ensuring system safety; and the DOB compensates for external disturbances like aerodynamic lift, enhancing convergence stability under strong disturbances. Simulation results validate the effectiveness of the proposed method under three conditions, demonstrating significant improvement in the levitation system's disturbance rejection capability and control accuracy.

Maglev train,Levitation system,Safe reinforcement learning,Control barrier function,Disturbance observer