More effective winter road maintenance, fewer ice-related accidents and gentler use of road salt – that’s what scientists at Hof University of Applied Sciences are hoping to achieve with a research project currently underway. With the help of artificial intelligence and current weather data, they are making daily forecasts for the whole of Bavaria about which road sections will freeze first and where the use of gritting services is therefore a particular priority. In the future, the forecasts will be available to Bavarian winter road services free of charge via the weather data management system of the Bavarian State Construction Directorate, which is acting as the university’s project partner.
To be able to generate reliable values about future ice slipperiness with the help of artificial intelligence, the researchers first make use of data from more than 600 weather stations of the German Weather Service for all of Bavaria. “We use this to feed our computers automatically and explicitly include empirical values from the past in our models. Factors such as wind speed, dew point temperature, depth temperature and air temperature are also taken into account. This is then used to create slickness scenarios for the current time, for 3 hours from now and for 18 hours from now – and this is done precisely for road sections 500 meters long,” says project manager Prof. Dr. Heike Markus, explaining the principle. This requires enormously high computer performance. In test runs last winter, however, the forecasts were already tested and confirmed using ground temperature sensors on individual winter service vehicles.
AI forecasts save time and money
On a clearly laid out user interface, those responsible for winter road maintenance can then see where problems due to icy roads are to be expected first and deploy their vehicles and employees accordingly with foresight. According to Prof. Dr. Markus, this has its advantages in rural areas in particular: “In urban areas, main roads, roads with inclines and intersections are generally cleared first, and many winter service drivers have very precise empirical values as to which locations in this narrow environment are particularly critical.” And further:
In rural areas, the places where icy roads occur can vary significantly more due to a wide variety of factors and can surprise even real winter service professionals. Here in particular, our computational models provide an invaluable time advantage – especially given the long distances that have to be covered.”
Prof. Dr. Heike Markus
AI not only makes it possible to react faster in the future and thus avoid accidents. At the same time, the forecasts also provide data on where little danger from icy roads can be expected. “This also allows targeted savings of road salt in these places, which in turn is good for the environment and municipal budgets,” says project member Dr. Ali Fallah Tehrani.
Practical test in the upcoming winter
Of course, the system will now have to prove itself in the coming winter and, if necessary, be further adapted: “Artificial intelligence is machine learning. This means that the model ideally learns with every known error and perfects itself – until it works reliably over the long term,” says Prof. Dr. Heike Markus.
Potential error correction is then again carried out via extensive data collection, which runs in parallel with the delivery of the forecast. “If the quality of the forecast is no longer sufficient for certain weather stations, the model is trained with additional data. In addition, other factors such as the quality of the German Weather Service’s weather forecast strongly influence the road ice forecast because our models use this data,” Prof. Markus said.
Ultimately, however, the new technical possibilities should make the work of winter services much easier.
