chen, yamin (Beijing Jiaotong University), Zhang, Yong (Beijing Jiaotong University), gao, xinjun (Signal and Communication Research Institute of China Academy of )

Research on Prediction Method of TractionBraking Performance of Subway Trains Based on Deep Learning and Model Fusion

Scheduled for presentation during the Regular Session "S33a-Intelligent Control for Next-Generation Railway Systems" (FR-LM-T33), Friday, November 21, 2025, 12:10−12:30, 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

Amid urbanization acceleration and the rapid expansion of urban rail transit systems, the operational efficiency and safety of subway trains have become paramount. As train control systems involve toward greater automation and intelligence, accurate prediction of train traction/braking performance has emerged as a critical requirement, especially in virtual coupling and automatic train control (ATC) systems, where precise performance prediction is essential for safe and coordinated operation. To address this challenge, we propose a hybrid model and data-driven framework that synergizes the robustness of physical models with the adaptability of data-driven approaches. First, a single-point mass dynamics model is established. The Long Short-Term Memory (LSTM) net-work is employed to identify the basic resistance parameters, and the Wasserstein distribution optimization is used to handle uncertainties. Next, a hybrid LSTM-Least Squares (LSTM-LS) framework is proposed for non-linear modeling and parameter fitting of traction/braking characteristic curves. Then, based on the identified traction/braking characteristic curves, key performance indicators are predicted using the automatic train operation(ATO) current and dynamic equations. The simulation results show that the identification error of basic resistance parameters based on Wasserstein optimization is significantly reduced. The LSTM-LS framework achieves >99% prediction accuracy (±1% error tolerance) for traction/braking force, running distance, and braking distance across the test dataset. This method provides theoretical support for the precise control and safe operation of subway trains.