Ma, Shaoyang (Beijing Jiaotong University), Zhu, Li (Beijing Jiaotong University), Cao, Fang (Beijing Jiaotong University), gao, xinjun (Signal and Communication Research Institute of China Academy of )

Real-Time Anomaly Detection for Train Movement Authority Using TACformer

Scheduled for presentation during the Invited Session "S16b-Control, Communication and Emerging Technologies in Smart Rail Systems" (TH-EA-T16), Thursday, November 20, 2025, 13:30−13:50, Southport 1

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

With the acceleration of urbanization, the safety and reliability of urban rail transit systems have become critical challenges in operational management. The Train Movement Authority (MA), as a key component in ensuring train operation safety, can cause serious accidents if anomalies occur in its data. Previous studies have focused on how to generate MA data, while lacking attention to the detection of MA anomalies caused by various factors. With the increasing computational resources of trains, anomaly detection in MA has become both feasible and necessary. However, traditional anomaly detection methods, relying on fixed thresholds, struggle to adapt to the complex and dynamic railway environment. Although deep learning approaches have improved accuracy, they face limitations in real-time performance and computational efficiency. To address these issues, we propose TACformer, a fast anomaly detection model for movement authority. TACformer addresses key challenges by: filtering obvious anomalies through a pre-detection head, capturing periodic dependencies in MA data via an autocorrelation module, and achieving a balance between anomaly detection accuracy and computational efficiency. Experiments conducted with simulation data from Beijing Subway Line 19 demonstrate that the proposed model outperforms existing methods with a macro F1 score of 0.9.