Conflict Prevention Strategies for Real-time Railway Traffic Management

Train timetables are built such that trains can drive without any delay. However, in real-time, unexpected events such as overcrowded platforms or small mechanical defects can cause conflicts, i.e., two trains requiring the same part of the infrastructure at the same time. Currently, such conflicts are typically resolved by experienced dispatchers. However, it is impossible for them to fully anticipate the impact of their actions on the entire network. Conflict detection and prevention tools embedded in a Traffic Management System (TMS) can help them in making informed decisions.

Though some advanced train movement prediction and conflict detection software has been introduced in the last years, there still exists a need for conflict prevention strategies capable of delivering conflict resolutions based on retiming, reordering or rerouting some of the trains in real-time.

This dissertation presents three Conflict Prevention Strategies (CPS, in short), aimed at assisting dispatchers in making informed decisions, that can easily be integrated in a TMS. All strategies proposed in this dissertation start by looking for possible rerouting options by using an optimization model. If no solution is found, a solution based on delaying one of the trains is suggested.

The first strategy presented here is called CPS-0, and examines the possibility of resolving multiple conflicts simultaneously. Because conflicts are often not detected at the same time, the CPS has to wait until sufficient conflicts are detected before it can start. This waiting time leads to less conflict resolutions available, because the simulation continues in the meantime. The resulting total train delay therefore increases significantly. Together with an increasing computation time, it is decided that resolving multiple conflicts simultaneously in this manner is not interesting.

The second conflict prevention strategy is called `original CPS' (or `oCPS'). oCPS significantly benefits from an offline calculation for determining related, frequently occurring conflicts. Based on this data, a so-called `dynamic impact zone' is created online for each conflict. Therefore, the heuristic can focus on the most relevant consequences of a given conflict. The performance of this oCPS is compared to a common dispatching strategy, other heuristics and an exact method. Extensive experiments on a large part of the Belgian rail network show that by considering this dynamic impact zone the total train delay can be decreased by 67 % on average compared to the basic First Come, First Served decision rule. The computation time for returning a resolution to a conflict with oCPS is, in 95 % of the conflicts, less than 2 seconds, and at most 26 seconds. When only considering the retiming heuristic in oCPS, the maximum computation time is limited to 4 seconds but the solution quality is slightly worse.

The third conflict prevention strategy is called `iCPS-BEST', both improving and extending the previous oCPS in order to make the strategy even more applicable in practice by including rolling stock connections. iCPS-BEST can take passenger numbers into account and allow trains to be canceled. Extensive experiments on the largest network considered in this dissertation show that by applying iCPS-BEST the total train delay can be reduced by 45 % while delivering conflict resolutions in 2.4 seconds on average. It is also shown that iCPS-BEST improves the previous strategy oCPS by 11% in total train delay, while reducing the average computation time by 23 %.