Zhao, Xiaoning (Tongji University), Sun, Yougang (Tongji University), Xu, Junqi (Tongji University), Qiang, Haiyan (Logistics Engineering College, Shanghai Maritime University)

Safe Reinforcement Learning Control for Maglev Train Levitation System with Robust Safety Assurance

Scheduled for presentation during the Regular Session "S33b-Intelligent Control for Next-Generation Railway Systems" (FR-EA-T33), Friday, November 21, 2025, 14:50−15:30, Southport 3

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

High-speed maglev trains represent a novel intelligent transportation, where the performance of the levitation system directly determines the safety and stability of train operation. To address the issues of poor adaptability in traditional methods and insufficient safety in reinforcement learning approaches for high-speed maglev train levitation control systems, this paper designs a safe reinforcement learning (RL) method for the levitation system based on high-order control barrier functions (CBF). A hierarchical control architecture is constructed to integrate a CBF safety correction layer, and differentiable optimization methods are employed to embed CBF into the RL algorithm, enabling safety constraints and stable convergence of the levitation control system. Additionally, Gaussian processes are introduced to predict external disturbances, enhancing robustness and theoretically ensuring robust safety during both training and operation. Simulation results demonstrate that the proposed method maintains stable levitation under static levitation initiation and irregularity disturbance conditions, with the maximum error reduced by 47.3% compared to traditional methods in certain scenarios, verifying its superiority in safety and adaptability.