A new machine learning algorithm – TranSEC – has been designed to solve the problem of urban traffic congestion. Short for ‘transportation state estimation capability’, TranSEC has been developed at the Pacific Northwest National Laboratory (PNNL), in the U.S. Department of Energy, to help urban traffic engineers get access to actionable information about traffic patterns in their cities.

Unlike other traffic monitoring methods, TranSEC is capable of analysing sparse and incomplete information using machine learning. This ability to overcome the inherent data gaps in legacy data collection methods allows it to make near real-time street level estimations.

The new tool uses traffic datasets collected from UBER drivers and other publicly available traffic sensor data to map street-level traffic flow over time. It creates a big picture of city traffic using machine learning tools and the computing resources available at a national laboratory. While the smart phone map tools only work for an individual driver trying to get from point A to point B, this tool is designed to help all vehicles get to their destinations efficiently by reducing urban traffic congestion.

TranSEC's near real-time display of traffic state estimation in the entire Los Angeles Metro Area at 6 p.m. on a weekday. This display was computed in less than one hour. Green areas indicate traffic is flowing freely and yellow and red areas indicate congestion. (Image courtesy of Arun Sathanur | PNNL)

Using public data from the entire 1,500 sq mile Los Angeles metropolitan area, the team reduced the time needed to create a traffic congestion model by 10 times, down to a few minutes, making near-real-time traffic analysis feasible. The research team recently presented that analysis at the August 2020 virtual Urban Computing Workshop as part of the Knowledge Discovery and Data Mining (SIGKDD) conference, and in September 2020 they sought the input of traffic engineers at a virtual meeting on TranSEC.

The machine learning feature of TranSEC means that as more data is acquired and processed it becomes more refined and useful over time. This kind of analysis is used to understand how disturbances spread across networks. Given enough data, the machine learning element will be able to predict impacts so that traffic engineers can create corrective strategies.

"We use a graph-based model together with novel sampling methods and optimization engines, to learn both the travel times and the routes. The method has significant potential to be expanded to other modes of transportation, such as transit and freight traffic. As an analytic tool, it is capable of investigating how a traffic condition spreads,” said Arun Sathanur, a PNNL computer scientist and a lead researcher on the team.

Eventually, after further development, TranSEC could be used to help program autonomous vehicle routes, according to the research team.