2008 | OriginalPaper | Buchkapitel
Generating Train Plans with Problem Space Search
verfasst von : Peter Pudney, Alex Wardrop
Erschienen in: Computer-aided Systems in Public Transport
Verlag: Springer Berlin Heidelberg
Aktivieren Sie unsere intelligente Suche um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
Planning train movements is difficult and time-consuming, particularly on long-haul rail networks, where many track segments are used by trains moving in opposite directions. A detailed train plan must specify the sequence of track segments to be used by each train, and when each track segment will be occupied. A good train plan will move trains through the network in a way that minimises the total cost associated with late arrivals at key intermediate and final destinations.
Traditionally, train plans are generated manually by drawing trains on a train graph. High priority trains are usually placed first, then the lower priority trains threaded around them. It can take many weeks to develop a train plan; the process usually stops as soon as a feasible train plan has been found, and the resulting plan can be far from optimal.
Researchers at the University of South Australia and WorleyParsons Rail have developed scheduling software that can generate optimised train plans automatically. The system takes a description of the way trains move through the network and a list of trains that are required to run, and quickly generates a train plan that is optimised against key performance indicators such as delays or lateness costs.
To find a good plan, we use a probabilistic search technique called Problem Space Search. A fast dispatch heuristic is used to move the trains through the network and generate a single train plan. By randomly perturbing the data used to make dispatch decisions, the Problem Space Search method quickly generates hundreds of different train plans, then selects the best. The automatic scheduling system can be used to support applications including general train planning, real-time dynamic rescheduling, integrated train, crew and maintenance planning, infrastructure planning and congestion studies.
One of the first applications of the system has been for an Australian mineral railway, to prepare efficient train plans to match mineral haulage requirements. The product is mined at six sites and transported by rail to a port. The numbers and sizes of train loads from each site are determined by grading requirements to meet the product specification for shipping. The train plan is then the orderly translation of these transportation requirements into an efficient timetable which resolves meets and crosses over a long single track railway. These train movements are thus part of an integrated mine-to-ship logistics chain.