Transfer Learning-Based Deep Reinforcement Learning for Adaptive Control of Maglev Trains

As a key component of a maglev train, the magnetic suspension control system ensures that the airgap between the train and its guideway is stable. Although current suspension controllers meet basic engineering requirements, problems related to their performance often arise during long-term operations. These problems can seriously affect the stability and reliability of a maglev system and may lead to partial suspension-point failure. Hence, this paper uses transfer learning-based deep reinforcement learning to develop an adaptive nonlinear suspension system controller that enables automatic adjustment of control strategies. First, suspension control based on deep reinforcement learning is mathematically modeled using Markov decision processes, and the nonlinear state space of a single electromagnet suspension control system is established as an agent–environment interaction with the developed deep reinforcement learning controller. Then a twin delayed deep deterministic policy gradient algorithm in an actor–critic framework is adopted to solve the Markov decision processes. To address the dispersion caused by nonlinear suspension control, a transfer learning-based two-stage training process is devised that first trains the twin delayed deep deterministic policy gradient networks on a linearized model and then transfers the networks to a nonlinear model. The effectiveness of the new controller is verified by comparing it with a conventional proportional–integral–derivative controller and a neural network based adaptive super-twisting sliding mode controller. The robustness of the transfer learning–deep reinforcement learning controller is examined in the presence of uncertainty, such as train load changes and disturbance forces in the suspension system. Additionally, experiments have also been conducted to confirm the effectiveness of the TL–DRL controller on a magnetic levitation system.