Next Generation Simulation (NGSIM)
-Improved Simulation of Stop Bar Driver Behavior at Signalized Intersections-
Los Angeles, CA and Atlanta, GA Data Sets

Working Papers

The arrival times of vehicles traveling southbound along a two-lane, bidirectional highway were recorded at eight neighboring locations upstream of a bottleneck caused by an oversaturated traffic signal. Cumulative curves constructed from these observations describe completely and in great detail the evolution of the resulting long queues. These queues formed directly upstream of the signal when the signal's service rate fell below the southbound arrival rates, and never formed away from the bottleneck. The predictability of bottlenecks like the one studied here can be exploited to manage traffic more effectively. The behavior of vehicles within the queue, however, was rather interesting. Although the flow oscillations generated by the traffic signal were damped out within 0.8 km (0.5 mi) of the bottleneck, it was found that other oscillations arose within the queue farther upstream, at varied locations, and then grew in amplitude as they propagated in the upstream direction. Thus, the queue appeared to be stable close to the bottleneck and unstable far away. Oscillations never propagated beyond the upstream end of the queue, however; that is, the unusual phenomena always arose after the onset of queueing and remained confined within the queue. Some of these findings run contrary to current theories of traffic flow. Because the data set collected in this study is unprecedented in scope and detail and so that it may be of use to other researchers, it has been posted on the Internet and is fully described here.

This paper appears in the Transportation Research Record No. 1678 [Year of Publication 1999].

Summary

Introduction and experimental setup

This paper is an attempt to get a better understanding about how queues form and propagate. This may lead to improved methods of managing heavy traffic. The study tries to investigate and understand the important and reproducible features of evolving traffic.

The experiment road stretch (2-lane, bi-directional highway with only one lightly-traveled access road in the road stretch under consideration) was ideally suited to this purpose because it had a downstream traffic signal that generated a bottleneck when vehicle arrival rates rose during the morning peak hour, and because there was very little vehicle overtaking and almost no side traffic. There were no loop detectors and 8 human observers with laptop computers were placed at positions up to 4 miles upstream from the signal. They recorded each vehicle’s arrival time and the vehicle class at their positions. A pilot car which was easily distinguishable from other vehicles cycled through the site during the study periods.

The data collected was collected for the morning peak hour period (from 6:45AM to 9:00AM) for 2 days, and it was corrected for hardware malfunctions (computer keyboard problems), clock sync problems, computer failure (battery fully discharged) and human error (missed observation or unwarranted keystroke).

Analysis and Conclusions

Curves were drawn of vehicle count versus time, N(j,t), where N(j,t) is the cumulative number of vehicles to pass stationary observer j at time t, measured from the first passage of the southbound pilot car.

Observations showed a drop in flow at around 7:04AM, believed to be caused by growth in the conflicting traffic streams, this can be seen in the step-function like nature for j=8. Soon after that, slopes of the upstream N-curves dropped in sequence. Hence, the drop in slope at any point j is simply the flow reduction brought about by the growing queue. Changes in the slopes of the curves can also be seen when a queue from further downstream spilled over. Oscillations in the flow near the tail end of the queue appeared to be softer, and they never affected traffic upstream of the queue.

The authors assert that visual inspection of the N-curves indicates that the queues formed in predictable ways at the most obvious inhomogeneity of the facility, the traffic signal. It was also found that the queues dissipated in fairly predictable ways. The authors also found stop-and-go jams in the queued traffic that were uncorrelated with the traffic signal. This is interesting because in the absence of passing, traffic information is unlikely to overtake vehicles, and conventional theories do not explain what is happening here. Another thing worth mentioning is that even though the instabilities seemed to grow in amplitude far from the bottleneck, the pronounced stop-and-go oscillations were themselves damped out within half a mile distance upstream.