Optimal Scheduling of Road Maintenance Jobs Considering Impact on Traffic Flows

Imagine the city as a giant, living puzzle where every road is a piece and every driver is a player trying to solve it. When a road gets closed for repairs, it's like pulling a piece out of the puzzle; suddenly, everyone has to scramble to find a new spot, and the whole picture shifts. This is the world of traffic engineering, a field dedicated to understanding how people move through these complex networks. At the heart of this puzzle is a concept called "equilibrium," which is just a fancy way of saying that everyone has found a route where they can't get to their destination any faster by switching paths. If you've ever been stuck in traffic while your GPS reroutes you, only to find the new path is just as jammed, you've experienced this balance in action. The big question for city planners is: how do we schedule road repairs so that we don't turn a small pothole fix into a city-wide traffic nightmare? Traditionally, figuring this out has been like trying to solve a massive math equation every single time you consider a new repair plan, a process so slow and heavy that it often stops planners from making the best decisions.

This paper, presented at the 2026 IISE Annual Conference, tackles that heavy math problem by teaching computers to guess the answer instead of calculating it from scratch every time. The authors, Charitha Nandepu and their team, looked at a real-world traffic network in Newark, New Jersey, specifically focusing on a busy corridor known as the Garden State Parkway. They started by running a super-accurate, but very slow, computer simulation to figure out exactly how traffic would flow if roads were closed. Think of this as the "gold standard" or the "ground truth"—the perfect answer key. They used this data to train four different types of artificial intelligence (AI) models, essentially teaching them to look at where people want to go (Origin-Destination demand) and instantly predict how the traffic would behave without needing to do the heavy math.

The team tested four different "brains" for their AI: a standard multi-layered network (MLP), a pattern-spotter (CNN), a network-structure expert (GNN), and a model that pays attention to relationships between all parts of the system (Attention-based NN). The results were clear: the model that paid attention to the big picture won. The "Attention-based" neural network was the star of the show, correctly predicting traffic patterns about 98% of the time (an R² score of 0.9835) and making the fewest mistakes on the busiest roads. In contrast, the other models, especially the standard ones, stumbled significantly when traffic got heavy, with errors jumping to nearly 1,000 vehicles per hour on critical links. The graph-based model (GNN) did a decent job, but the attention-based model was the most accurate, with an error rate of just 35.87 vehicles per hour overall.

The paper doesn't claim to have solved the entire problem of scheduling road repairs forever. Instead, it suggests that using these fast, data-driven AI models is a promising way to speed up the planning process. By replacing the slow, heavy calculations with a quick AI prediction, city managers could potentially test many more repair schedules in the time it used to take to test just one. The study shows that for the Newark area, this approach works well enough to be a useful tool, offering a way to keep traffic flowing smoothly even when the roads are under construction. It's a step toward making our cities less gridlocked, one smart prediction at a time.