Effective rail traffic management is necessary to mitigate the impact of unforeseen train service disturbances. Traditional decomposition methods, while effective in managing complexity, often struggle to maintain global optimality and real-time responsiveness. In this paper, we propose a novel approach that decomposes the rescheduling problem by means of a selforganising paradigm where trains are intelligent autonomous agents deciding on their decisions after reaching a consensus. The proposed Self-Organized Train Rescheduling (SOTR) algorithm is inspired by the Distributed Constraint Optimization Problem (DCOP) framework. This algorithm treats trains as intelligent agents capable of constructing their own traffic plans, communicating with neighbouring agents, and making decisions that lead to an optimal timetable. Each train, acting as an agent, assesses its situation, predicts conflicts, and negotiates with other trains to find the most efficient solution in regard to total delay. This distributed decision-making process allows for rapid adaptation to dynamic disturbances and ensures scalability to large networks. We validate the effectiveness of our approach by using a micro-simulation tool, demonstrating its ability to minimize secondary delays and maintain network continuity in perturbation scenarios.

The railway scheduling problem concerns the determination of trains' scheduled departure and arrival times at stops, and the allocation of capacity in the network. The timetable must be both conflict-free given infrastructure constraints, and stable enough for trains to recover from delays that could occur in normal operations. Existing methods for tactical scheduling contain a tradeoff between having an accurate (microscopic) representation of signalling constraints, and having a simple-enough (macroscopic) infrastructure representation to scale to real-world problem instances. This creates issues for infrastructure managers looking to run more trains on their infrastructure by migrating to Distance-To-Go (DTG) signalling systems (e.g. ETCS Level 2), and to exploit the capabilities of Connected Driver Advisory Systems (C-DAS) and Automatic Train Operation (ATO) to control trains more precisely. In this paper, we present a methodology for incorporating the capabilities of DTG signalling in conjunction with C-DAS and ATO systems into a disjunctive scheduling model for both periodic and nonperiodic instances. We show that the resulting model has both a microscopic infrastructure representation, and a macroscopic computational complexity, allowing railways to quickly compute conflict-free and stable timetables for large problem instances. The resulting model also accurately represents the computation of the brake indication point for both conventional and DTG signalling as a function of the trains' current speed. Tests on a large-scale periodic scheduling instance in the UK show that the model produces timetables with reasonable computation time.

This paper reviews the concept of self-organisation as defi ned in diff erent fi elds and attempts to provide a defi nition of goal-oriented self-organization that can be applied in the context of railway traffi c. Based on the provided defi nition a modelling approach for self-organising rail traffi c is then proposed to set the basis for future research and exploration of such a concept which could revolutionise the current rail transport to meet long-term capacity and competitiveness goals envisaged by the railway industry.