TLDR: DISPLIB is a new library introducing a common problem definition and file format for train re-routing and re-scheduling. It aims to standardize research in train dispatching by providing openly available real-world problem instances and a framework for comparing optimization algorithms. The initiative includes the DISPLIB 2025 benchmark set, derived from various railway operators, and a computational competition that validated the framework’s effectiveness and the challenging nature of the instances. This effort seeks to spur innovation and improve efficiency in railway operations by fostering reproducible research and advanced algorithmic solutions for managing train delays.
Train travel is a cornerstone of modern transportation, but even the most meticulously planned timetables can be disrupted by unexpected delays. When these disruptions occur, train dispatchers face the daunting task of manually re-routing and re-scheduling trains to prevent a cascade of further delays. This manual process is not only difficult, especially during major disturbances, but often leads to less-than-optimal decisions made under pressure.
A new initiative, DISPLIB, aims to transform this critical aspect of railway operations. DISPLIB, which stands for a library of train dispatching problems, introduces a standardized problem definition and file format designed to capture all the essential features of train re-routing and re-scheduling. This standardization is a significant step forward, as it addresses a long-standing challenge in the research community: the lack of shared code, data, and common problem definitions, which has hindered reproducibility and the ability to compare different algorithmic solutions effectively.
Inspired by successful communities built around other complex optimization problems like Mixed-Integer Linear Programming (MILP) and the Traveling Salesman Problem (TSP), DISPLIB provides a unified framework. This allows researchers and developers to work on train dispatching problems without needing direct industrial connections, fostering a collaborative environment for empirical comparisons between different solvers and algorithms.
Understanding the DISPLIB Framework
At its core, DISPLIB defines train dispatching as an optimization problem. It considers a set of trains, each performing a sequence of operations. These operations have minimum durations, and specific time windows (lower and upper bounds) within which they must start. A crucial aspect is the concept of resources, such as sections of railway track (block sections), which trains require exclusive access to for a certain period. The problem then involves finding a feasible solution – a route and schedule for all trains – that respects all operational constraints and resource limitations.
The framework also incorporates a cost function, typically designed to minimize delays. This can include penalties for trains starting or arriving late, with flexible options for defining how these costs accumulate over time. The goal is to find an optimal solution that minimizes this total cost while maintaining feasibility.
DISPLIB is designed to model real-time train dispatching scenarios. For instance, it can represent how railway infrastructure is divided into block sections, which are then treated as resources. It also accounts for various real-world complexities, such as trains already in motion, intricate cost functions with piecewise linear dependencies, trains longer than a single block section, and even more advanced concepts like rolling stock correspondence and cancellation decisions.
The DISPLIB 2025 Benchmark Set
To facilitate robust testing and comparison, DISPLIB has gathered a diverse set of problem instances, forming the DISPLIB 2025 benchmark set. These instances come from multiple real-world use cases and railway infrastructure managers across different countries. They are categorized based on their origin and characteristics, including data from Siemens Mobility Italy, SINTEF (representing parts of the Norwegian national railway network), Swiss Federal Railways (SBB), and Wabtec.
These categories offer a wide range of scenarios, from small sets of trains with complex interactions to large networks with hundreds of trains and tens of thousands of operations. The instances vary in difficulty, ensuring that they challenge even the most advanced state-of-the-art algorithms for train scheduling. The ambition is for an algorithm to produce near-optimal solutions for all these instances in under a minute, which would mark a significant achievement in the field.
Also Read:
- AI’s Adaptive Approach to Complex Optimization Challenges
- New Algorithm G-CSEA Enhances Infeasibility Diagnosis in Workforce Scheduling Models
A Platform for Progress: The Computational Competition
To generate interest and validate the DISPLIB framework, a computational competition was held from October 2024 to April 2025. This competition encouraged researchers to develop new train dispatching algorithms and submit their solutions. Participants were given a maximum of 10 minutes of computation time per instance, a limit set to encourage innovative algorithmic ideas, even if not fully optimized for real-time deployment.
The competition successfully validated that the DISPLIB problem definition and file format are general enough to represent diverse data sources and are easily understandable by the community. Crucially, none of the participating teams managed to solve all instances to optimality, confirming that the DISPLIB 2025 instances are indeed challenging and highlight the ongoing need for improved algorithms in this domain.
The DISPLIB initiative, with its standardized problem definition, openly available real-world instances, and a reference solver implementation, provides a crucial infrastructure for advancing the state of the art in train dispatching algorithms. All materials, including problem instances and competition details, are available online at displib.github.io.
