Simulation-Driven Railway Delay Prediction: An Imitation Learning Approach

2025-12-17

AI Summary

To address the challenge of real-time train delay prediction in railway systems, this paper proposes Drift-Corrected Imitation Learning (DCIL). DCIL extends the DAgger framework with a distance-driven covariate shift correction mechanism, enabling robust policy rollout without expert labels or adversarial training. It integrates event-driven modeling with deep data-driven learning and incorporates Monte Carlo simulation for uncertainty-aware forecasting. Evaluated on Infrabel’s million-scale real-world operational dataset, DCIL achieves statistically significant improvements over conventional regression models and behavioral cloning baselines on 30-minute lookahead prediction tasks. It accurately captures both the temporal propagation and stochastic evolution of delays across large-scale rail networks, thereby enhancing prediction robustness and operational efficiency.

Technology Category

• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Planning, Routing, and Scheduling: Model-Based Reasoning
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Reliable prediction of train delays is essential for enhancing the robustness and efficiency of railway transportation systems. In this work, we reframe delay forecasting as a stochastic simulation task, modeling state-transition dynamics through imitation learning. We introduce Drift-Corrected Imitation Learning (DCIL), a novel self-supervised algorithm that extends DAgger by incorporating distance-based drift correction, thereby mitigating covariate shift during rollouts without requiring access to an external oracle or adversarial schemes. Our approach synthesizes the dynamical fidelity of event-driven models with the representational capacity of data-driven methods, enabling uncertainty-aware forecasting via Monte Carlo simulation. We evaluate DCIL using a comprehensive real-world dataset from extsc{Infrabel}, the Belgian railway infrastructure manager, which encompasses over three million train movements. Our results, focused on predictions up to 30 minutes ahead, demonstrate superior predictive performance of DCIL over traditional regression models and behavioral cloning on deep learning architectures, highlighting its effectiveness in capturing the sequential and uncertain nature of delay propagation in large-scale networks.

Problem

• Predict train delays using stochastic simulation modeling
• Mitigate covariate shift in imitation learning for delay forecasting
• Capture sequential uncertainty in delay propagation for railway networks

Innovation

• Imitation learning for stochastic simulation modeling
• Drift-corrected self-supervised algorithm without external oracle
• Monte Carlo simulation for uncertainty-aware delay forecasting