SNCF uses AI to improve automatic passenger counting on its trains

assistant sncf transports icon app ipa iphone ipadSNCF intends to use artificial intelligence to simplify its method of counting passengers and thus make the travel experience more pleasant for consumers. A formidable challenge at a time when the trains are increasingly crowded.

Artificial intelligence: the great challenge for the SNCF for the years to come

SNCF is currently developing its own artificial intelligence in order to improve the fluidity of its offer. It’s on the website DigitalSNCF that we learn that the pole Innovation & Research implements an algorithm supposed to be able to speed up the automatic counting of passages in trains equipped with this service.

The final objective, to have the necessary data in real time as well as a global vision of the traffic to adjust it accordingly. In other words, be reactive at times when the supply must be high and, conversely, detect quieter periods.

As explained in the article, seen from a distance, it does not seem very complicated. Some will say that it is enough to count the passengers present on board the trains, to register it in tables and to make affluence predictions. Yes, but it’s much more complicated. Firstly, not all trains are equipped with passenger detectors on the doors and, secondly, because this method can sometimes be capricious because of luggage, animals, bicycles, etc.

sncf logo

The data collected with this system is unfortunately too imprecise and errors are frequent, a bit like for the counts in demonstrations. It is therefore vital to reform this part and to have in the future precise and revealing data of the real influx of passengers. With the figures in real time, the SNCF could distribute the passengers in the trains in a much more orderly way and thus improve their travel conditions.

Tom Rousseau, project manager at Artificial Intelligence Platform at the house of Innovation & Research, explains to us why artificial intelligence would be a beneficial daily aid for the company. The acquisition of this data in real time rather than offline seems to be the real asset of this algorithm.

The quality improvement strategy that we adopted breaks down into two steps: first, predict the counts to anticipate possible data loss, then adjust all the counts (actual and predicted).

In 2020, my main challenge was to design a real-time data rectification algorithm by imitating the system currently in force, which only works in deferred time (at D+2/3).

In 2021, I was able to develop a complementary algorithm, aimed at predicting counts and compensating for data deficiencies. By aggregating the counts processed by these two bricks, we obtain a load of on-board passengers that is more reliable than the original.

This AI tool is called FIKA but is not the only one currently in development. Other projects are in progress internally but also externally, in particular for the Affluence project, which is none other than a collaboration between the SNCF, the ENS Paris-Saclay and the Île-de-France region. The ultimate goal will be to democratize artificial intelligence throughout the rail network.

sncf influx algorithm

We would love to thank the author of this write-up for this remarkable web content

SNCF uses AI to improve automatic passenger counting on its trains

Explore our social media profiles and other related pages