“In 15 minutes there is a 90 per cent probability that Turku Motorway westward at Veikkola will be congested.” Congestion forecasts like this may soon be part of everyday life in Finland.
“Together with Digia, we have developed a program that gives quite reliable forecasts of the development of the number and speed of vehicles on public roads. Testing is now underway, and the preliminary results are promising”, says Analytics Expert Pekka Kinnunen at the Finnish Transport Agency.
By promising results, Kinnunen means that the program was able to predict vehicle speeds with a probability of over 99% on the test section of Turku Motorway.
How was this accuracy achieved? Before any prediction could be given, the program had to be taught the causes of congestion. The program has learnt this by itself by studying past records. The program has studied road weather data and data collected by the automatic traffic monitoring systems in 2015 and 2016. This data included, for example, the number and speed of vehicles that had passed the automatic monitoring system.
“In schools, pupils are first taught, and then their knowledge is tested. The machine-learning program was given the test to predict the driving speeds and traffic volumes during the first half of 2017. The program passed the test with flying colours”, says Kinnunen.
Even artificial intelligence has difficulties predicting unforeseen congestions
Artificial intelligence is normally better at predicting than humans, but it has its own limitations. No forecast model is able to predict congestions due to, for example, sudden collisions.
“The current models are not able to predict random situations. However, the program can be developed so that it alerts when the driving speeds on a road become significantly lower than predicted. This would signal that something unforeseen has happened in the traffic”, says Kinnunen.
Utilisation of more accurate congestion predictions
Congestion predictions generated by machine-learning programs are still not used in traffic, but this situation may change in the next few years. In the future, more accurate congestion predictions could be used to at least support traffic control.
“After all, more accurate congestion predictions would result in safer and smoother transport. For example, if we know that Turku Motorway will soon be congested, we can try to facilitate the situation by using signal control”, says Head of Traffic Management Centre Mika Jaatinen at the Finnish Transport Agency.
If the tests show good results, road users, in addition to traffic management centres, may have access to better congestion prediction in the future, for example, through the Finnish Transport Agency’s service, Traffic situation.
Enquiries
Analytics Expert Pekka Kinnunen
Finnish Transport Agency
phone +358 29 534 3250
firstname.lastname(at)liikennevirasto.fi