Reinforcement Learning-based Suspension Control for Electromagnetic Suspension Maglev Trains

Abstract:

In order to ensure the safe and reliable operation of maglev trains， the suspension control of suspension systems under the condition of parameter perturbation was studied. Firstly， the basic suspension unit of a electromagnetic suspension （EMS） maglev train was modeled and the current control model was given. Then， the reinforcement learning environment and the soft actor-critic （SAC）agent were established for the suspension system， and a reward function and an anti-suspension-contact strategy were designed for accelerating the training process. Finally， a corresponding reinforcement learning method was proposed. Compared with the conventional proportional-integral-derivative （PID） control method， the proposed method has faster dynamic response and better tracking accuracy under the condition of 50% loss of coil turns and the pole area change of the magnet.