Sign In

Communications of the ACM

ACM TechNews

Building a Better Traffic Forecasting Model


View as: Print Mobile App Share: Send by email Share on reddit Share on StumbleUpon Share on Hacker News Share on Tweeter Share on Facebook
A traffic jam in Los Angeles.

Argonne National Laboratory scientists are using artificial intelligence to forecast large-scale traffic patterns.

Credit: Bus & Motorcoach News

Scientists at the U.S. Department of Energy's Argonne National Laboratory (ANL) are using artificial intelligence to forecast large-scale traffic patterns with greater accuracy.

The research team fed ANL supercomputers traffic patterns from roughly a year's worth of data culled from 11,160 sensors along the California highway system.

The team trained a machine learning model on that data to forecast traffic within milliseconds; the model can analyze the past hour of data, and anticipate the next hour of traffic.

The ANL researchers dramatically shortened the time required to train the model, from a week on a top-of-the-line desktop to three hours on a supercomputer.

The model applies graph-based deep learning to predict traffic patterns from historical patterns, while concurrently forecasting traffic speed and flow.

From Argonne National Laboratory
View Full Article

 

Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA


 

No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account
ACM Resources