A Deep Reinforcement Learning Based Control Approach for Suspension Systems of Maglev Trains

Abstract

The magnetic suspension control system is one of the core components of the maglev trains. However, because of the system's unstable open loop, strong nonlinearity, and model uncertainty, the design of maglev suspension control method is challenging. In this paper, a third-order maglev train suspension system dynamics model is established firstly. Then the affine nonlinear model of the magnetic suspension system through the nonlinear coordinate transformation theorem is obtained. Subsequently, without making any linear approximation, the nonlinear integral sliding mode controller (NISMC) is directly developed and the stability analysis is performed. To eliminate the influence of system disturbance on control performance, RBF neural network and Actor-Critic algorithm combined to construct a modified deep reinforcement learning method, which is used to optimize controller parameters in real time and enhance system robustness. Numerical simulation results are provided to demonstrate the effeteness of the proposed deep reinforcement learning method.