Railway EffIcieNcy Fostered by Operations ResearCh EmpoweRing Articificial InteLligence – REINFORCERAIL
REINFORCERAIL
Railway EffIcieNcy Fostered by Operations ResearCh EmpoweRing Articificial InteLligence
Improving railway traffic management
In REINFORCERAIL, we aim to combine Artificial Intelligence (AI) and Operations Research (OR)-based approaches for railway traffic management.<br />In particular, we address two research questions:<br />1. How can AI help find traffic management decisions in real-world circumstances: within short times and for extremely large railway networks?<br />2. How can AI-based approaches for traffic management be best hybridized with OR-based approaches to ensure quality of these decisions and to reduce time and energy spent for the training?<br />To do so, we propose two innovations combining the strengths of the two approaches.<br />First, we will use neural networks to limit the problem size while guaranteeing good network-level railway traffic management.<br />Second, we will hybridize AI approaches with OR techniques. In this hybridization, the AI component will overcome the OR scalability problems, while the OR component will get around the AI struggle in gaining a quality guarantee and<br />the trust required for a fully automated application.
Two methods are proposed.
On the one hand, AI will identify pertinent sub-problems that can be dealt with independently.
On the other hand, AI will drive the configuration of a mixed-integer programming-based algorithm to fit to the features of the traffic situation at hand.
Pertinent KPIs have been defined.
AI methods are to be developed and tested.
1. Sehmisch S., Szymula C., Bešinovic N. Learning to reduce Railway Traffic Management Problems, 2nd Workshop on Data and Transportation Science Cooperation (DATRASCOOP), September 27, 2024, Tenerife (Canary Islands, Spain)
2. Sehmisch S., Szymula C., Bešinovic N. A data-driven Problem Reduction Framework for local Train Rescheduling in Station Areas, 9th AIROYoung Workshop, February 26-28, 2025, Pavia (Italy)
3. C. Bai, P. Pellegrini, J. Rodriguez, M. López-Ibáñez (2025) Learning for RECIFE-MILP: a Parameter Configuration Perspective, RailDresden 2025, April 1-4, 2025, Dresden (Germany).
4. Fouladi A., Bešinovic N. Reinforcement Learning Approach for Train Rescheduling: A Review and Future Directions, RailDresden 2025, April 1-4, 2025, Dresden (Germany)
Challenge
Railways play an important role in the shift towards green mobility of people and goods. Frequent and dense traffic requires better tools for real-time control which are responsible to reduce delays after disturbances or disruptions and will provide punctual and reliable rail services.
Idea
The REINFORCERAIL project proposes a new intelligent Traffic Management System (TMS) component based on Artificial Intelligence (AI). First attempts have shown that AI-based traffic management is principally possible at the cost of designing complex models and performing energy-thirsty training. To alleviate these challenges, two methods enabling real-life automatic railway dispatching are pursued. First, neural networks are designed to identify which trains possibly need rescheduling at which time during perturbed operations, which significantly reduces the problem size. Second, AI agents are hybridized with Operations Research (OR) methods, which allows to give quality bounds and fosters the trust required to implement a fully automated application.
Prospects
The REINFORCERAIL project contributes to the need for a highly automated TMS, to make reliable traffic management decisions after disturbances in very short response times. This need pushes the managers of the largest European national railway infrastructures, in France and Germany, to take a leading role in the shift from computer-assisted manual planning and control to automated operations.
Project coordination
Paola Pellegrini (Université Gustave Eiffel)
The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.
Partnership
Univ eiffel Université Gustave Eiffel
SNCF SOCIETE NATIONALE SNCF
DB DB Netz AG
TU Dresden Dresden University of Technology
Help of the ANR 284,371 euros
Beginning and duration of the scientific project:
May 2023
- 48 Months
