2504. A Machine Learning-based Selection Strategy for Railway Crew Scheduling with Trip Shifting
Invited abstract in session TA-57: Railways applications, stream Transportation.
Tuesday, 8:30-10:00Room: Liberty 1.12
Authors (first author is the speaker)
| 1. | Anne Schönhofen
|
| Department of Operations Management, University of Cologne | |
| 2. | Ulrich Thonemann
|
| Supply Chain Management, Universtiy of Cologne |
Abstract
Traditionally, timetabling and crew scheduling are subsequent tasks in the workforce planning process in airline, railway, and public bus transportation. Trip shifting offers an opportunity to enhance flexibility in crew schedule planning by allowing the scheduled time of trips to be adjusted within predefined discrete time windows. This approach is widely acknowledged in the literature for generating highly cost-efficient schedules. In graph-based solution approaches, time shifts are typically modeled by duplicating arcs corresponding to scheduled trips. From a computational perspective, this leads to a significant increase in graph size and problem complexity. In practice, planners seek to maximize cost savings while shifting only a small number of trains. Motivated by this practical insight, we address the computational complexity by training a machine learning model to predict the most promising time shift options before optimization. By including only promising shifts, we keep the graph’s size manageable and ensure the resulting crew schedule remains applicable in practice. We integrate our approach into a column generation heuristic and test it on real-world datasets provided by a major European railway freight carrier. The results demonstrate that our approach realizes a large share of the optimization potential and achieves greater cost reductions than a known method from the literature while having a comparable network size.
Keywords
- Railway Applications
- Scheduling
- Machine Learning
Status: accepted
Back to the list of papers