Measuring the performance of surface transportation facilities is becoming an increasingly important activity. Traditionally, transportation agencies have sought to quantify the extent of travel on facilities using measures such as vehicle miles of travel. However, agencies are now seeking to quantify the quality of travel from a motorist’s perspective by measuring congestion in terms of delay and the predictability of travel. These measures of congestion are usually based on estimates of the time it takes travelers to traverse a particular corridor, defined as travel time. Travel time is a very attractive measure in that it is intuitive to citizens outside of the transportation profession (unlike other traditional measures, such as level-of-service). For this reason, many measures of surface transportation performance are based on travel time or the normalized derivation known as travel rate (time/unit distance, such as minutes/mile).
While there is little disagreement that travel time is an ideal measure, there is considerable debate concerning how best to measure travel time. The Virginia Smart Travel Laboratory (STL), a joint research facility of the Virginia Department of Transportation (VDOT) and the University of Virginia’s Center for Transportation Studies, is currently addressing this issue. The STL is directly integrated with VDOT transportation management systems, receiving over 11 million records of traffic data every day. Furthermore, the laboratory incorporates sophisticated simulation capabilities. Combining these resources, the STL has addressed (1) how to measure travel time using today’s data, and (2) how to improve agencies’ ability to measure travel time in the future.
Point mistaken
Point sensors collect the vast majority of traffic data available for performance measurement. Point sensors, such as inductive loop detectors and video image detection systems, generally measure traffic volume, speed and occupancy at a single location (i.e., point). These detectors are usually installed at regular intervals, often every 1/2 mile, along urban freeways. In most cases, data from point sensors are used to estimate travel times by assuming that the point data adequately represents travel conditions in the linear area between sensors. Thus, while travel time is an ideal measure to work with conceptually, in reality it is rarely measured directly in urban regions.
This leads to the concern that these travel-time estimates derived from point sensors may not truly capture the experience of motorists in a region. Furthermore, it may not be advisable to compare the same mobility measures between one region that estimates travel time with point sensors and another region that directly measures travel time using an alternative approach/technology (such as with toll tags).
This problem is best illustrated using data collected on I-66 in Northern Virginia. Travel-time estimates using the “retrospective extrapolation” method (assumes uniform conditions in the vicinity of point sensors) were compared to direct travel-time measures of vehicles generated by manually matching license plates at two points along the corridor. Figure 1 illustrates that the estimation method consistently produces travel times that are significantly below those actually experienced by motorists. For example, trips with a nearly 40-minute duration were estimated to have a travel time of 22 minutes. Research attempts to produce ad-hoc algorithms to correct the travel time (represented by the black dots) were of limited effectiveness.
Why did this occur? A combination of a number of factors is responsible. First, some point sensors were malfunctioning in key locations, a common occurrence with point detectors given the environment in which they function. More importantly, however, the sensor placement was originally set to support traffic operations, not travel time estimation. As a result, the sensors were generally in “pockets” of free-flow traffic—missing the numerous areas of stop-and-go traffic.
A key conclusion that can be drawn from this research, and similar experiences elsewhere, is that if travel time is to become a foundation for performance measures, then it should be directly measured, not estimated using point detectors. As a result, new approaches for direct travel-time measurement need to be investigated.
On location
Traffic data, of course, is not only of use or interest to transportation agencies. In fact, the private sector has collected this type of data for years to support early generation traveler information services (such as radio traffic reports).
Now, as technology and markets mature, a number of private-sector firms are looking to position themselves to sell measured travel-time data to both the public and private sectors.
While there are multiple technologies and approaches available to support such an approach, one with high potential is the use of wireless location technology (WLT) to anonymously track the locations of individual wireless devices. By anonymously tracking a series of positions of wireless devices in vehicles, it is conceptually possible to generate speed and/or travel-time estimates for roadway links. While this idea is appealing, the field deployments of WLT-based traffic monitoring to date in the U.S. have not been entirely successful. In cooperation with the Virginia Transportation Research Council (VTRC), the STL has conducted extensive simulation-based research to learn more about WLT-based traffic monitoring. The results of this research have been published and presented in the open literature in an attempt to help accelerate the development of this technology.
The results of this simulation testing have shown that WLT-based traffic monitoring is viable and have identified critical factors that influence its overall effectiveness. The testing was conducted using a variety of different combinations of system and roadway characteristics, and included case studies of two simulated real-world networks in the Washington, D.C., suburbs: Springfield and Tyson’s Corner. On freeway systems, WLT-based monitoring systems were found to accurately measure travel time. Simulation results for a chronically congested portion of the Washington beltway in the Springfield area are discussed below.
Figures 2a and 2b shows the difference between the true speed on the link and speed estimates generated by simulated point detectors and four variations of simulated WLT-based monitoring. For the WLT-based monitoring, two different methods of matching vehicle positions to roads (geometric map matching including network topology and the multiple hypothesis technique [MHT]) are shown, as well as two different mean times between vehicle position samples (F). Much like the I-66 case presented earlier, point detectors overestimated the speed on the road. The various WLT-based monitoring techniques were able to generate speed estimates that were more representative of the true speeds since they track vehicles as they traverse the entire link, rather than just collecting data at discrete points.
The potential advantages of WLT-based monitoring systems are even more pronounced on arterial roads. It is very difficult to develop arterial travel time and speed estimates using extrapolated point detector data due to the influence of control delay at traffic signals. WLT-based monitoring can explicitly include control delay in condition estimates, potentially enabling more accurate monitoring of arterial road conditions.
Results from a major arterial in the Tyson’s Corner area illustrate some of the potential benefits of using a well-designed probe-based system. In the case of Rte. 123 EB, the point detector based speed estimate was more than 20 mph higher than actual speed. WLT-based methods could bring these errors in speed estimation down to 1 or 2 mph, depending on how the system is designed.
Probe-based monitoring systems offer the opportunity to directly measure travel times experienced by a sample of motorists along a road. The results of the testing showed that WLT-based traffic monitoring methods have the potential to significantly improve the quality of travel-time measures over those generated by point sensors. While these results are promising, field deployments of this technology to date have not been entirely successful in generating the quality of data needed by transportation agencies and more work needs to be done to improve system performance. The STL continues to investigate how these systems should be designed and operated, with the ultimate goal of providing a sound basis for successfully implementing WLT-based systems in the future.
TME
RSS: Roads & Bridges Articles
