Improving Train Routing Selection for Railway Traffic Management

Railway transportation plays a central role in modern mobility systems, serving a growing demand for both passenger and freight services. Train movements are scheduled in advance through timetables to ensure safe and efficient operations. However, unexpected events may disrupt the planned schedule, requiring real-time interventions to restore feasibility, such as retiming trains, modifying their order on shared tracks or assigning alternative routes. This problem is known as the real-time Railway Traffic Management Problem (rtRTMP). Solving the rtRTMP within the strict time limits of real-time operations is particularly challenging in congested areas, where numerous routing alternatives lead to large-scale optimization problems. To improve tractability, this thesis focuses on a preprocessing step known as the Train Routing Selection Problem (TRSP), which selects an optimized subset of routes for each train, thereby reducing the search space of the rtRTMP. The TRSP evaluates routing options through a cost estimation model that approximates delays associated with different route configurations. Because it operates as a rapid preprocessing step, the model relies on simplified representations of train interactions, which may introduce inaccuracies affecting the ranking of candidate configurations and the quality of the reduced search space. This thesis investigates the sources of such discrepancies in the pACO-TRSP framework, which combines a graph-based formulation with a parallel Ant Colony Optimization algorithm. Targeted refinements are introduced to improve the predictive accuracy of the cost estimation model and its alignment with rtRTMP outcomes. Their impact is assessed through an extensive computational study on real-world railway infrastructures. The results demonstrate that the proposed refinements improve the reliability of route preselection, particularly in dense traffic scenarios, while preserving performance in less congested settings.