Researchers explore ways to make Traffic Models more efficient
Models that predict traffic volume for specific times and places are used to inform everything from traffic-light patterns to apps that tell you how to get from Point A to Point B. Researchers from North Carolina State University have now demonstrated a method that reduces the computational complexity of these models, making them operate more efficiently.
โWe use models to predict how much traffic there will be on any given stretch of road at any specific point in time,โ says Ali Hajbabaie, co-author of a paper on the work and an assistant professor of civil, construction and environmental engineering at NCย State. โThese models work well, but the specific forecasting questions can be so computationally complex that they are either impossible to solve with limited computing resources, or they take so long that the prediction only becomes available when it is no longer useful.โ
The researchersโ starting point for this work was an algorithm designed to help streamline complex computing challenges, but they found it couldnโt be applied directly to traffic problems.
โSo, we modified that algorithm to see if we could find a way to use it in models that predict how much traffic there will be in a specific place and time,โ Hajbabaie says. โAnd the results were gratifying.โ
Specifically, the researchers came up with a modified version of the algorithm that effectively breaks the larger traffic forecasting model into a collection of smaller problems that can then be solved in parallel with one another.
This process significantly reduces run time for the forecasting model. However, the extent of the improved efficiency varies significantly, depending on how complex the forecasting questions are. The more complex the question is, the greater the improved efficiency.
The modified method also improves run time by allowing the model to recognize when it has reached a solution that is good enough โ the solution doesnโt have to be perfect. Traditionally, models will run until they find an optimal solution, or one very close to optimal. But for most purposes, a result that is within 5% โ or even 10% โ of the optimal solution will work fine.
โOur approach here essentially sets error bars around the optimal solution and allows the model to stop running and report a result when it gets close enough,โ Hajbabaie says.
The researchers tested the modified algorithm against a benchmark algorithm used in consumer software to address questions related to traffic forecasting.
โOur modified algorithm outperformed the benchmark in two ways,โ Hajbabaie says. โFirst, our algorithm used far less computer memory. Second, our algorithmโs run time was orders of magnitude faster.
โAt this point, weโre open to working with traffic planners and engineers who are interested in exploring how we can use this modified algorithm to address real-world problems.โ
The paper, โA Distributed Gradient Approach for System Optimal Dynamic Traffic Assignment,โ appears inย IEEE Transactions on Intelligent Transportation Systems. First author of the paper is Mehrzad Mehrabipour, a Ph.D. student at NCย State.
