SUMP Central

Select Language

Good Practices

Using machine learning to find a solution to recurring congestion

Case study on the effect of the machine learning on congestion handling, Debrecen, Hungary

City: Debrecen
Audience: Medium-sized City
Topic: Traffic and demand management
Step in the SUMP cycle: Step 12: Review and learn lessons

More Information:
Eltis.org

Activity description

Debrecen wanted to develop a self-healing system for the city, able to detect and propose a simple solution for different traffic issues. This means that the machine learning system was developed in order to design reliable Traffic Light Programme (TLP). The system is capable of isolating the out the types of infrastructure improvements that could be implemented, such as; changes to road lanes, traffic lights, new restrictions on traffic turning, etc. The result was a Traffic Light Program (TLP) which can manage the practicalities of peak traffic – to prevent congestion from forming at peak times. There is also s a potential infrastructural change involving a secondary lane extension which was recommended by the ML system.

Lessons learned

The biggest issue for the team was to analyse the whole affected road section. As the road section studied during this project is 2 km long, it proved very difficult to capture all the necessary information at the precise moment congestion started to form – or isolate the exact cause. So the team took several measurements on different sections of the road and then equivalised them to produce realistic simulations. This approach can be easily transferred to any other traffic handling situation analysis.