Researchers reduced the time required to create a traffic congestion model by an order of magnitude, from hours to minutes, by using public data from the entire 1,500-square-mile Los Angeles metropolitan area. The TranSEC tool was created at the United States Department of Energy’s Pacific Northwest National Laboratory to assist urban traffic engineers in gaining access to actionable information about traffic patterns in their cities.
Currently, publicly available street-level traffic information is scarce and incomplete. To determine roadway conditions, traffic engineers have traditionally relied on isolated traffic counts, collision statistics, and speed data. The new tool maps street-level traffic flow over time using traffic datasets collected from UBER drivers and other publicly available traffic sensor data. It uses machine learning tools and computing resources available at a national laboratory to create a big picture of city traffic.
A new machine-learning algorithm is poised to help urban transportation analysts relieve bottlenecks and chokepoints that routinely snarl city traffic.
A new machine learning algorithm developed at the United States Department of Energy’s Pacific Northwest National Laboratory will assist urban transportation analysts in identifying and alleviating traffic patterns in their cities. The tool can use machine learning to analyze sparse and incomplete data by connecting segments of information with missing data, providing analysts with near real-time street-level traffic estimates.
“What’s novel here is the estimation of street-level over a large metropolitan area,” said Arif Khan, a PNNL computer scientist who assisted in the development of TranSEC. “Moreover, unlike other models that are limited to a single metro area, our tool is portable and can be used in any urban area where aggregated traffic data is available.”
UBER-fast traffic analysis
The ability to analyze sparse and incomplete information distinguishes TranSEC (which stands for transportation state estimation capability) from other traffic monitoring methods. It connects segments with missing data using machine learning, allowing it to make near real-time street level estimations.
In contrast, the map features on our smartphones can assist us in optimizing our journey through a city landscape by highlighting bottlenecks and suggesting alternate routes. However, smart phone tools are only useful for a single driver attempting to get from point A to point B. City traffic engineers are concerned with how to help all vehicles arrive at their destinations as quickly as possible. Sometimes a route that appears to be efficient for a single driver results in too many vehicles attempting to access a road that was not designed to handle that volume of traffic.
The team reduced the time required to create a traffic congestion model by an order of magnitude, from hours to minutes, by using public data from the entire 1,500-square-mile Los Angeles metropolitan area. The acceleration, achieved with high-performance computing resources at PNNL, enables near-real-time traffic analysis. The research team recently presented that analysis at the Knowledge Discovery and Data Mining (SIGKDD) conference’s virtual Urban Computing Workshop in August 2020, and in September 2020, they sought input from traffic engineers at a TranSEC virtual meeting.
“TranSEC has the potential to initiate a paradigm shift in how traffic professionals monitor and predict system mobility performance,” said Mark Franz, a meeting attendee and research engineer at the University of Maryland, College Park’s Center for Advanced Transportation Technology. “TranSEC bridges the data gaps inherent in legacy data collection methods and has enormous potential.”
Machine learning improves accuracy over time
TranSEC’s machine learning capability means that as more data is collected and processed, it becomes more refined and useful over time. This type of analysis is used to determine how disruptions spread across networks. Given enough data, the machine learning component will be able to predict impacts, allowing traffic engineers to devise corrective measures.
“We use a graph-based model in conjunction with novel sampling methods and optimization engines to learn both the travel times and the routes,” said Arun Sathanur, a PNNL computer scientist and the team’s lead researcher. “The method has a lot of potential for being applied to other modes of transportation, like transit and freight traffic. It can investigate how a traffic condition spreads as an analytic tool.”
Users in a transportation control center can upload real-time data and update TranSEC on a regular basis using PNNL’s data-driven approach. Engineers can use short-term forecasts to help them make decisions about traffic issues. The approach developed by PNNL can also be extended to include weather or other data that affects road conditions.
Computing power for transportation planners nationwide
Just as a driver’s situational awareness influences his or her decisions, TranSEC’s approach provides situational awareness on a system-wide scale to aid in the reduction of urban traffic congestion. “Traffic engineers across the country haven’t had a tool that gives them anywhere near real-time estimation of transportation network states,” said Robert Rallo, PNNL computer scientist and TranSEC project principal investigator. “Being able to predict conditions an hour or more ahead of time would be extremely valuable in determining where the blockages will be.”
TranSEC is scalable, even though running a full-scale city model still necessitates high-performance computing resources. A road network with only major highways and arterials, for example, could be modeled on a powerful desktop computer. “We are working to make TranSEC available to municipalities across the country,” said Katherine Wolf, TranSEC’s project manager.
According to the research team, TranSEC could eventually be used to help program autonomous vehicle routes after further development.