by Sara Gestrelius, SICS Swedish ICT
The cost of building a new shunting yard is in the order of €100 million and the expected lifespan is over 50 years. Discussions about new railway infrastructure investments are currently ongoing in Sweden, and SICS Swedish ICT are using mathematical optimization to solve NP-hard shunting yard problems to help the Swedish Infrastructure Manager choose between different yard layouts.
Shunting yards are the hubs in the network that distributes freight cars all over Sweden. Inbound trains arrive at the yard where their cars are sorted into new outbound trains. Shunting yards often consist of different sub-yards: an arrival yard where the trains arrive and are decoupled, a classification bowl where the cars are sorted into new outbound trains, and a departure yard where finished outbound trains can wait for their departure time. The system can be likened to a factory with incoming stock (the arrival yard), a production facility (the classification yard), and outgoing stock (the departure yard). The arrival yard is connected to the classification bowl via a hump, and the cars roll from the arrival yard to the classification bowl by means of gravity. However, not all shunting yards have a departure yard, and the number of tracks in the various sub-yards vary. Figure 1 shows a graphical representation of a typical shunting yard with all three sub-yards.
Figure 1: Graphical representation of the typical lay-out of a shunting yard.
The tracks required in the various sub-yards for successfully building all outbound trains on time are highly dependent on the trains’ arrival and departure times, the car bookings and the operation of the shunting yard. In Sweden, shunting yards are operated in a particular way and the classification bowl tracks are divided into train formation tracks and mixing tracks. Train formation tracks are used for compounding outbound trains while mixing tracks are used as a storage place for early cars whose outbound trains are not yet being built. The mixing tracks are emptied by pulling the cars on the mixing tracks back to the arrival yard and then rolling them over the hump again. It is necessary to do detailed operational planning to know the track requirements, but unfortunately scheduling a shunting yard using Swedish operational practices is an NP-complete problem .
SICS Swedish ICT has previously worked with RWTH Aachen University and ETH Zürich to develop mathematical optimization models and methods for planning shunting yards . A heuristic is used for scheduling the arrival yard, the departure yard, and the times when the mixing tracks are emptied. The output from the heuristic defines an optimization problem for the car sorting in the classification bowl, where the objective is to minimize the number of extra car shunt moves brought about by mixing. In this new project the models are used to generate shunting schedules for different yard layouts. The schedules can be analysed to estimate how much (or little) work that is required to process a certain set of trains given a specific shunting yard layout. Fewer track resources will lead to more shunting work and vice versa. By varying the number of tracks in each sub-yard the trade-off between the number of tracks and the shunting work can be analysed, and the goal is to provide the Infrastructure Manager with data that helps them find a good balance between the cost of building tracks and the cost of shunting. In particular, it is important for the Swedish Infrastructure Manager to identify when the track resources are too scarce to operate the intended traffic in a robust and proper manner.
The first project aiming at investigating shunting yard layouts using optimization was started during spring 2015, and its successor is due to finish at the end of 2016. Within the second project two shunting yards are currently being investigated, Hallsberg Rangerbangård and Sävenäs Rangerbangård. In the Hallsberg case study the departure yard usage is in focus, and in particular questions regarding the trade-off between arrival yard usage and the departure yard usage are being evaluated. Methods for investigating flat shunting are also being developed as a complement to the existing models. It is possible that some freight trains only require car swaps rather than the full resorting offered by shunting yards.
In Sävenäs there are plans for building a completely new yard, and therefore a few different, sometimes completely new and inventive, yard set-ups are assessed. The traffic in Sävenäs also requires some further sorting with respect to the car-order in the outbound trains as there are groups of cars that should be dropped off along the train’s journey. Methods for this extra sorting are being developed.
Another pertinent question regarding freight trains is the routing of cars and models for optimizing the routing are currently being developed. Experimenting with different routings can be useful for improving the overall efficiency of the shunting system, and also for investigating the effect of closing one shunting yard partially or completely. The latter is relevant when discussing re-investment in shunting yards as it may require existing shunting yards to be closed down, which makes the economic cost of the disturbed traffic an important factor when deciding between different investment alternatives.
As stated above, heuristics are part of the current schedule generation. As part of future work the project will therefore also look into the possibility of constructing an optimization model that can handle all three sub-yards.
 M. Bohlin, S. Gestrelius, F. Dahms, M. Mihalák and H. Flier, “Optimization Methods for Multistage Freight Train Formation”, Transportation Science, Articles in Advance, 2015.
Martin Aronsson, Zohreh Ranjbar, Martin Joborn, Sara Gestrelius, Markus Bohlin,
SICS Swedish ICT