Search Results

Delays and disruptions are commonplace in public transportation. An important tool to limit the impact of severely delayed vehicles is the use of short turns, where a planned trip is shortened in order to be able to resume the following trip in the opposite direction as close to the schedule as possible. Short turns have different effects on passengers: some suffer additional delays and have to reschedule their route, while others can benefit from them. Dispatchers therefore need decision support in order to use short turns only if the overall delay of all affected passengers is positively influenced. In this paper, we study the planning of short turns based on passenger flows. We propose a simulation framework which can be used to decide upon single short turns in real time. An experimental study with a scientific model (LinTim) of an entire public transport system for the German city of Stuttgart including busses, trams, and local trains shows that we can solve these problems on average within few milliseconds. Based on features of the current delay scenario and the passenger flow we use machine learning to classify cases where short turns are beneficial. Depending on how many features are used, we reach a correct classification rate of more than 93% (full feature set) and 90% (partial feature set) using random forests. Since precise passenger flows are often not available in urban public transportation, our machine learning approach has the great advantage of working with significantly less detailed passenger information.