In studies carried out to identify traffic hotspots in the U.S., Los Angeles typically ranks. And, the city is notorious for its traffic jams is no secret. Estimates suggest people in Los Angeles are stuck in traffic for an additional 120 hours each year. In such a scenario, the L.A. transportation system has its own advantages, if a new system is devised to quickly predict and redirect that traffic.
Under the umbrella of a larger project, researchers at the Argonne National Laboratory, U.S. Department of Energy set out to do this. The project led by collaborators at Lawrence Berkeley National Laboratory at the Department of Energy involved design and planning of mobility systems.
Using machine learning, the research team leveraged supercomputers at Argonne. The supercomputers were used to map traffic patterns of data obtained from 11,160 sensors for nearly a year along the large California highway system. The data was then used to devise a model to predict traffic at lightning fast speeds. Within a microseconds, the model can view data in the past hour and make predictions of traffic in the next hour with great accuracy.
A collaborated work of Mathematics and Computer Science division at Argonne, and the Argonne Leadership Computing Facility, the team attained exceptional results for traffic forecasting. The results were recently published in Transportation Research Record: Journal of the Transportation Research Board. “The machine learning and supercomputing capabilities used in this work allow to address really large problem,” stated a computer scientist at Mathematics and Computer Science. The amount of data involved in a project of this scale requires an equally large computing resource to address it.