Analyses of Start-Stop Waves in Congested Freeway Traffic

Freeway traffic was observed over multiple days and was found to display certain regular features. Oscillations arose only in queues; they had periods of several minutes; and their amplitudes stabilized as they propagated upstream. They propagated at a nearly constant speed of about 20 to 24 kilometers per hour, independent of the location within the queues and the flow measured there; this was observed for a number of locations and for queued flows ranging from about 850 to 2,000 vehicles per hour per lane. The effects of the oscillations were not felt downstream of the bottleneck. Thus, the only effect on upstream traffic was that a queue’s tail meandered over time by small amounts. (For the long queues studied here, the tails deviated by no more than about 16 vehicle spacings, as compared with predictions that ignored the oscillations). Notably, the character of queued traffic at fixed locations did not change with time, despite the oscillations; i.e., traffic did not decay. There were changes over space, however. New oscillations formed in moderately dense queues near ramp interchanges and then grew to their full amplitudes while propagating upstream, even though the range of wave speeds was narrow. The formations of these new oscillations were strongly correlated with vehicle lane-changing. It thus appears that the oscillations were triggered by random vehicle lane-changing in moderately dense queues more than by car-following effects. But this pattern of formation and growth was less evident in a very dense queue (caused by an incident), although frequent lane-changing occurred near the interchanges. Finally, kinematic wave theory was found to describe the propagation of the oscillatory (i.e., start-stop waves) to within small errors. For distances approaching one kilometer, and for two-hour periods, the theory predicted the locations of vehicles to within about 5 vehicle spacings. Further analysis showed that some of these small discrepancies are explained by differences in car-following behavior across drivers.