Early Conflict Detection

The purpose of this thesis is to design and test an algorithm to predicts conflicts early – up to 2 hours and 45 minutes before they occur. The proposed algorithm tries to bound the delay at the end station using two dispatching strategies. The ‘fixed order strategy’ will not allow trains to pass each other and thus will create the upper bound. The lower bound is constructed through the ‘first opportunity strategy’ where a delayed trains will pass other trains at the first opportunity. When a conflict is predicted by both bounds, the algorithm will assume it has occured. When only the upper bound predicts a conflict, the algorithm will postpone the overtaking of trains in the lower bound untill both bounds will again predict the same conflict. If, in the lower bound, overtaking can be postponed less than 60%, the conflict is assumed to occur in the near future. Using this approach 95% of the conflicts can be predicted prior to them occuring. Those conflicts not predicted can generally be solved without rescheduling. If we force the algorithm to predict conflicts up to 30 minutes in advance this percentage drops to between 52% and 78%, depending on the dispatching strategy used. When increasing this timeframe to 60 minutes the algorithm can only predict about 25% of all conflicts, but the accuracy increases. Note that not all delays occur 30 or 60 minutes prior to the conflict. The increasing accuracy of the algorithm suggests that the earlier a train is delayed the better the outcome will be. Because for dispatchers it is generally the other way around, this algorithm can successfully used to aid dispatchers in their decisions.