Identifying and Mitigating Congestion Onset (STRIDE Project H5)

This is Phase 2 of Project J3 under the same title, which began in January 2020.

Rush hour on a Miami highway (image source:

Managing the operation of highway networks is a challenging task. The vehicles, which create the congestion and cause the incidents, are operated by motorists over which the traffic management center (TMC) has little or no control. Yet TMC staff are expected to mitigate the impacts of congestion and respond quickly to incidents. While these are reasonable expectations, the challenge is to anticipate congestion and spot the incidents; and sooner is better.

“Waze is great for incidents and Google is great for congestion, but TMC operators do not necessarily have official access to these data streams” said Dr. George List of North Carolina State University (NCSU). “This project is exploring new ways to identify or the occurrence of these events that capitalize on data streams from probes and system detectors. While the final answers are not in yet, the results look promising.”

Dr. List is the lead PI on this project along with his research team, which include Dr. Billy Williams (NCSU); Dr. Michel Hunter (GaTech); and Dr. Mohammed Hadi (FIU).

With probes, for example, downward trends in the speeds of the fastest moving vehicles (actually, upward trends in their travel rates) seem to give a leading indicator that congested conditions are about to arise. The researchers are able to predict the probability that congestion is going to arise in the next 30, 20, 10, 5 minutes. Also, based on the probe data, abrupt changes in the spread between the lowest and highest travel rates seems to be a reliable predictor of disruptive incidents, whether they occur during congested or uncongested conditions.

“The normal flow of traffic gets disrupted, and the faster and slower vehicles both are affected, but the distribution of their speeds (or travel rates) becomes much wider because the motorists cannot as well control their progress through the traffic stream,” Dr. List said. “For historical data, where incidents were recorded, we can spot all the ones that were identified, and some that were not.”

The researchers are also looking at whether the same trends can be detected using system (loop) sensor data.

“Our end goal is to make these detection algorithms perform well enough that TMC operators will want to use them to make their job easier and their actions timelier,” he said.