In May 2021, we were preparing to renovate the Miramas marshalling yard in south-eastern France—but we couldn’t commit until we had quantified the infrastructure investments needed to handle the expected increase in traffic. In the past, those operations studies would have taken a good 3 weeks—enough time for a dozen people to run a complex series of calculations. But that was before...

... before SNCF researchers had grappled with the problem of optimizing marshalling yard management. In 2020, they developed a software program that took just minutes to calculate timetables for the marshalling yard movements that keep wagon transfers on time. This tool was code-named ROC TRI, combining the French words for “operations design research” and “marshalling”.

“The ROC TRI calculation engine creates timetables that let locomotives carry out marshalling operations—disassembling and assembling trains and returning them to service—in a way that maximizes the number of transfers for wagons in transit,” explains Juliette Pouzet, Modelling and Decision Support project leader. “With this tool, we’re sure we’ve got the optimum solution,” she adds.

ROC TRI is an excellent example of how algorithms can be used to reduce employee workload and save substantial amounts of money and energy. “In short, we identify the best solution using mathematics,” says Christelle Lérin, head of the Modelling and Decision Optimization unit in SNCF’s Technology, Innovation and Group Projects Division.

In addition to operations research, SNCF experts use another important tool: simulation. In simulation, we use mathematics to reproduce the workings of a system and test our hypotheses. One example is a software program that “predicts” arrival timetables by simulating future rail traffic, detecting potential conflicts, and trying to resolve them. We’re also exploring the potential of AI-based machine learning techniques to improve decision support models.

A train is running late. What should we do? Stagger the trains behind it? Skip stops at select stations to make up the delay? Lay on an extra train? The answer depends on many factors: the relative number of passengers affected, overall network traffic, train and employee availability, and more. Algorithms take all of these parameters into account and present solutions—and their consequences—to our employees. The end result is invaluable support for operators and better service for passengers.

In a nutshell, decision science lets us calculate the best choices—the ones that deliver the best system performance. This might mean producing a given service at the best possible cost, using as few resources as possible, or delivering the best possible service quality with fixed resources.

We can calculate the optimum size for a fleet of trainsets to provide service to our customers, or choose the optimum routes for transporting freight. If there’s a disruption, we can find the best solution for regulating and prioritizing trains in real time. In each case, we work closely with our information systems divisions, which supply us with data and scale up prototypes. We also partner with Synapses, SNCF’s internal network of 80 experts. They help us with optimization and operations issues.

The future is “quantum optimization”. Today, when a problem is highly combinatorial, calculating power is limited. So we’re already working on quantum algorithms, on the assumption that they’ll be used in tomorrow’s rail industry. Looking ahead is another reason why research is important for SNCF. 

• Partners at SNCF: TGV Intercités, Transilien, Fret, SNCF Réseau and the Group’s Information Systems Divisions
• Top partners in academia: in France, we work with Université Gustave Eiffel, École des Mines de Saint-Etienne, École des Ponts ParisTech, CentraleSupélec. We also partner with IVADO in Montreal to analyse industrial problems and identify the best solution strategies for algorithms under development