Keeping roads and streets fit for travel using machine learning with YIT

Slippery roads no more

”Weather machine learning” is a system based on machine learning used to manage road maintenance. The system helps YIT predict the probability of slippery conditions and changes in road surface conditions, which is used to optimize maintenance procedures on roads and streets.

Timely tactics using Azure

With this solution, the service vehicle dispatcher has a proactive model of predictive analysis which provides information on the likelihood of road surface slippage in the coming hours. At its best, the system allows the operations to be exactly timely, which means the end user has better conditions when moving around on streets and roads.

For example, the system predicts slipperiness in a whole new way, learning about past actions and their results. It tells the maintenance staff the best time to take action, and how much and which anti-skid materials are needed. "The system is in daily use for all our maintenance contracts, and we are continuously developing the algorithms to better support our operations," says Petri Jansson, project manager responsible for the development of the weather machine learning technology at YIT's maintenance unit.

Among other things, the system enables benchmarking between different contracts. The situation in Oulu can be compared to Helsinki, allowing staff to keep up to date and learn. YIT's 24/7 service center PANU operates as the system administrator, but anyone in charge of the works can see the situation in the whole country.

Weather machine learning is a 24/7 system with a web-based interface which learns about the conditions of worksites and utilizes YIT:s own data as well as external open data. The system works on the Azure platform, using Microsoft's machine learning tools. The solution has been implemented in cooperation with YIT, Louhia and Soikea. The system has been continually developed for several years, as the teachings of the previous winter are processed during the summer and are tested again next winter.

"We're already planning to continue the project by developing new algorithms which should give us even more financial and environmental savings."
- Petri Jansson, Project Manager, YIT