Data-driven surrogate models can replace expensive traffic simulations in maintenance planning, making it practical to optimize when and where to schedule road work based on traffic impact.
This paper develops fast machine learning models to predict traffic flow patterns when roads are under maintenance. Instead of repeatedly solving complex traffic equations (which is slow), the authors train surrogate models on real traffic data to quickly estimate how road closures affect traffic. They test this on Newark, NJ traffic data as a foundation for automated maintenance scheduling.