1176. Integrating Reinforcement Learning methodologies into a generic CDCR Heuristic to effectively resolve occupancy conflicts at railway nodes in real-time.
Invited abstract in session WB-51: Railway Traffic Management, stream Public Transport Optimization.
Wednesday, 10:30-12:00Room: M5 (building: 101)
Authors (first author is the speaker)
| 1. | Arturo Crespo Materna
|
| Department of Civil and Environmental Engineering, Technical University of Darmstadt, Institute of Railway Engineering | |
| 2. | Cedric Steinbach
|
| Department of Civil and Environmental Engineering, Technical University of Darmstadt, Institute of Railway Engineering | |
| 3. | Shanqing Chai
|
| Institut für Bahnsysteme und Bahntechnik, TU Darmstadt | |
| 4. | Keren Zhou
|
| TU Darmstadt | |
| 5. | Andreas Oetting
|
| TU Darmstadt | |
| 6. | Hendrik Speh
|
| TU Darmstadt |
Abstract
This contribution presents a novel approach to enhance the real-time resolution of occupancy conflicts at railway nodes by integrating Reinforcement Learning (RL) methodologies into a generic Conflict Detection and Conflict Resolution (CDCR) heuristic. Given the intricate nature of railway systems, the occurrence of stochastic events cannot be avoided. In this context, conflicting operations between train journeys must be resolved. These conflicts are harder to resolve within nodes than in lines due to the larger number of possible resolution combinations. The proposed approach foresees the integration of RL into an existing CDCR heuristic approach for resolving occupancy conflicts in nodes. The generic CDCR Heuristic provides a baseline for identifying and resolving conflicts, while RL enriches the system's intelligence by providing real-time feedback learned from historical data. The resulting hybrid approach facilitates adaptive decision-making, optimizing train movements, and minimizing the overall delay in the system. The integration of RL within the existing CDCR heuristic aims to enhance the system's efficiency and effectiveness, ultimately improving the overall efficiency of railway operations. The proposed approach is evaluated in a real-world scenario, demonstrating its potential to enhance real-time schedule adjustments by providing robust, optimized solutions to occupancy conflicts.
Keywords
- Decision Support Systems
- Machine Learning
- Expert Systems and Neural Networks
Status: accepted
Back to the list of papers