AVERAGE VELOCITY OF WAVES PROPAGATING THROUGH CONGESTED FREEWAY TRAFFIC

摘要

This paper employed cross correlation to estimate wave velocities between several successive detector stations during congested periods and over a large data set. Given homogeneous vehicles and drivers, LWR predicts that for a convex flow-density relationship all waves should propagate upstream during congested periods. The analysis first employed cumulative arrivals -- a function of flow -- and then the local traffic speed at the detector stations. It was shown that the flow-based analysis yield mixed results, with many measurements being consistent with earlier research, but many more measurements falling outside the typical range of measured wave velocities from the literature. But vehicles are not homogeneous and it was also shown that flow and occupancy depend on effective vehicle length as well as the local traffic speed. Because the vehicle lengths travel downstream with the vehicles, this information will hinder attempts to extract wave velocities propagating against the flow of traffic when using flow-based measures. This fact is evident in fig. 2 and 3. Trucks are restricted from lane 2, thus the standard deviation of vehicle lengths is smaller and less information travels with the vehicles. As a result, comparing across lanes for all of the plots in these figures, lane 2 had the highest percentage of observations that fell within the 15 km/h to 25 km/h window. To reduce the influence of the confounding vehicle lengths, the analysis was repeated over the exact same samples using local traffic speed measurements. Because drivers are constrained by downstream vehicles in congestion, the traffic speed trends are much less dependent on specific vehicle or driver characteristics. It was confirmed that LWR does a better job predicting the evolution of average traffic speed over time and space, with over 90 percent of the samples having an average wave velocity propagating upstream in the range of 15 km/h to 25 km/h. Of course these results represent the aggregate performance over one-hour samples and individual waves could differ significantly from this range, e.g., it is likely that other sources of noise remain in the data, such as lane change maneuvers disrupting the propagation of waves.