Zhang, Dandan (Tongji University), Sun, Yougang (Tongji University), Qiang, Haiyan (Logistics Engineering College, Shanghai Maritime University), Lin, Guobin (State Key Laboratory of High-speed Maglev Transportation Technol)

Physics-Informed Neural Network Learning-Based Nonlinear MPC for Maglev Vehicles Levitation Systems with Variable Terminal Constraint Set

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

Maglev vehicles are emerging as a promising solution for next-generation transportation, offering advantages such as high speed, low energy consumption, and strong gradient-climbing capabilities. However, their levitation systems are inherently nonlinear, operate within narrow airgap, and are highly sensitive to external disturbances. Additionally, system performance is further degraded by track irregularities, sudden load variations, electromagnetic coupling, and actuator faults can cause significant fluctuations in system dynamics and may even lead to instability. To address the challenge of achieving high-precision and stable levitation control under multi-source uncertain disturbances, this paper presents a physics-informed neural network learning-based nonlinear model predictive control (PINMPC) scheme featuring a variable terminal constraint set. In particular, A deep physics-informed neural network (PINN) is developed to perform identification of system unknown parameters, enabling the controller to adapt to dynamic changes. Based on the updated model, a customized nonlinear model predictive control (NMPC) strategy is formulated to handle system nonlinearities, external disturbances, and input/output constraints. Furthermore, a variable terminal constraint set is designed to mitigate constraint violations caused by input delays and to ensure recursive feasibility. Finally, simulation results validate the effectiveness of the proposed method under input delay and external disturbances.