NS uses smart software for planning at shunting yards

NS is deploying a new AI-based method to plan trains at shunting yards. Thanks to this approach, NS can respond more quickly and effectively to unexpected changes or disruptions in the timetable. The planning method contributes to a better train journey, with cleaner trains and a higher chance of getting a seat. The algorithm is based on research by Roel van den Broek, who completed his PhD on the subject at Utrecht University.

Each day, NS operates hundreds of train units. Outside the timetable, these units must be parked, cleaned, inspected and coupled before being put back into service. This takes place at shunting or stabling yards. Partly because space at these yards is limited, planning is complex for NS.

Planning at these sites is still largely done manually by a team of around 120 planners. They prepare the plans more than a month before execution, based on the information available at the time about the eventual real-life situation, which is still limited. If the timetable changes in the final month before implementation, for example due to damage to the track, a month’s worth of plans has to be rescheduled. This creates considerable pressure on the planning and operational control departments.

Hundreds of plans within minutes

PhD researcher Roel van den Broek developed an algorithm that significantly accelerates this process. For smaller stations, the system can now generate plans within minutes, taking into account all the aspects a human planner must consider: available tracks and staff, required service tasks, and rules for safe shunting.

Joris Snijders, responsible at NS for implementing the new software, explains: “The algorithm uses, among other things, a local search method. It generates hundreds of possible plans, from which the best is selected. The algorithm then creates hundreds of variations on that plan, with small adjustments to the original. Again the best option is chosen, which is then refined further. Planners subsequently polish the algorithm’s best plan into a final schedule.”

The so-called Hybrid Integrated Planning Method (HIP) makes NS “logistically more agile”, according to Snijders. “Because of the time savings we achieve, we move from planning weeks in advance to planning days in advance. That allows us to incorporate changes at a later stage. It also creates more room for planning passenger services. Ultimately, the planning method therefore contributes to the quality of the train journey for customers: cleaner trains and a higher chance of getting a seat.”

AI is a support tool

For the time being, the algorithm can be used at smaller hubs, such as the stations of Enkhuizen, Leeuwarden and Vlissingen. The ambition is to eventually deploy it at larger hubs as well, including Amsterdam and Utrecht. Snijders says: “Ultimately we want to plan train movements at all 32 stabling yards in this way. Within a European subsidy project we are working to further develop the method.”

Artificial intelligence will not fully take over the planner’s work, Snijders emphasises. The system has been designed so that users can still adjust plans as they see fit. “Humans remain in control. AI is a support tool.”

There is already interest from abroad in our planning method

The new planning method has not gone unnoticed. “In the Netherlands we have a unique situation: large numbers of passengers on a very busy rail network, with shunting yards located on the edges of cities. That combination of factors is currently typical of the Dutch situation. But that could change in the future. There is already interest from abroad in our planning method.”

Ivoren toren

According to Snijders, the project clearly demonstrates what collaboration between academia and practice can achieve. “This algorithm proves that fundamental research can find its way into practical application. The impact of using this smart software on our planning should not be underestimated. The algorithm will become increasingly important in the planning process.”

Han Hoogeveen, Associate Professor at Utrecht University and closely involved in the project, is also enthusiastic about the collaboration. He proposed the new approach to NS. “Roel van den Broek subsequently demonstrated brilliantly that this approach really works, and that convinced the NS board to invest in it. To me, it is a wonderful example of what such collaboration can achieve. It also shows that the university is not merely an ivory tower.”