Estimating traffic today for better roads tomorrow

Since 1963, the Committee on Traffic Flow Theory and Characteristics of the Transportation Research Board has been promoting the development, validation, and dissemination of research on traffic flow theory and characteristics. Each year the committee awards the prestigious Greenshields Prize to one paper that demonstrates the spirit of the award’s namesake, Bruce D. Greenshields, basing sound theory on rigorous empirical analysis. The 2017 recipients for the Greenshields Prize were none other than civil and environmental engineering’s (CEE) own Sean Qian, and Qian’s former doctoral student Yiming Gu, who is currently a Senior Research Engineer at the United Technologies Research Center. They received the award for their paper titled Traffic State Estimation for Urban Road Networks Using a Link Queue Model.

Traffic State Estimation (TSE) is the process by which engineers take data from multiple sources and use it to draw inferences about a number of traffic variables, such as speed, density, and flow on roads where no sensors are deployed. The focus of the Greenshields prize is to bring ideas off of the paper and into the real world. “If the theory is just a theory, it’s very hard to use,” explains Qian. “We’re trying to translate theory into practice.”

Most attempts at TSE within road networks have focused on highway networks as opposed to urban networks, due to the limited number of on-ramps, off-ramps, and exits on highways. This smaller pool of data makes it easier to compute accurate estimates in a reasonable time period. However, in this research, Qian’s sights were set a little higher; his goal was to glean accurate estimates for an urban network.

If the theory is just a theory, it’s very hard to use. We’re trying to translate theory into practice.

Sean Qian, Assistant Professor, Civil and Environmental Engineering, Carnegie Mellon University

While it might not be immediately evident, the ability to estimate traffic states—especially in urban environments—is of extreme importance. On urban streets, tools like Google Maps are only able to provide speed, not flow and density, and it is no secret that its inferences of speed have considerable errors. The immediate impact of having a reliable estimate of the traffic state on urban streets is that it allows for real-time traffic management, optimizing the routing of emergency responders in the case of an accident or the rerouting of traffic to adjust for delays. The data and estimates gathered also enable engineers like Qian to understand and predict how traffic is moving within a given space, and how the infrastructure can be better utilized. They can then alter existing directions to better ease the flow of traffic, such as by changing road markings or the timing on lights.

In the past, the sheer number of roads (e.g. links) and intersections in an urban environment has been too large to create accurate estimations, and it would be infeasible to place a sensor on every road and intersection. However, the link queue model Qian used is able to take large amounts of data from numerous preexisting sources—cell phones, GPS devices, probe vehicles—in conjunction with strategically positioned sensors, and perform an efficient computation to output a reasonably good TSE. Within a small portion of the Washington, D.C. area, Qian combined multiple data sets with just two well-placed speed detectors to accurately estimate travel speed to an acceptable error rate within 8.5%.

While Qian’s demonstration of the link queue model has plenty of immediate applications, perhaps the largest impact of these findings is in the long-term outlook of TSE. Using the link queue model, engineers can compute years’ worth of data to create accurate estimates of the future growth and change of traffic characteristics in a given network. “In the long-term, we want to know on average, throughout the entire year, and over multiple years, how traffic increases,” said Qian. This information can be used to inform engineers in designing the next generation of roadways and traffic infrastructure, shaping the street and highways of tomorrow.

The road ahead is not without its speedbumps, however. One of the biggest obstacles facing the future of urban TSE is applying the small-scale demonstration that Qian put into practice across a much wider area. “The main challenge is ensuring the quality and coverage of our data,” he says. “Mathematically, we’re still determining if it’s feasible to run such a model for a very large-scale urban network.” It is that avenue which he hopes to pursue in the future. The Mobility Analytics Data Center (MAC), which Qian directs, is already hard at work developing a centralized data engine for compiling and computing large amounts of data sets from sources all throughout the city of Pittsburgh. “The D.C. network for demonstration in this paper is small, but I’m still very excited about the progress that we’re making,” said Qian. “Now we’re trying to find a real-world experiment here in Pittsburgh with the MAC center, where we already have all the data.”