Estimating Freeway Traffic Stream Modal Activities for Air Quality Modeling
Final Report - Summary
Michael Zhang, Associate Professor
Bo Wang, Graduate Student Researcher
Civil and Environmental Engineering
University of California, Davis
Davis, CA 95616
hmzhang@ucdavis.edu
University of California Transportation Center Year 12 (1999-2000)
Overview
The research develops a method that uses data provided by widely deployed point sensors, namely inductive loop detectors, to construct vehicle trajectories of freeway traffic, from which modal activities of traffic streams can be estimated. This method provides a cost-effective way to develop freeway driving cycles used in air quality models and emission adjustment factors for freeways whose traffic flow patterns largely differ from those embodied in the driving cycles, thereby improving the accuracy of emission estimates by those models. It also produces the ground truth for calibrating transportation planning models when accurate speed estimates are desired.
Background
Transportation is responsible for a large proportion of harmful air pollutant, such as carbon monoxide, nitrogen oxides, and volatile organic compounds. Many factors affect motor vehicle emissions. They include vehicle characteristics (type and age), operating mode, driving characteristics, and weather conditions (Horowitz 1982). For a given vehicle technology, emission rates of certain pollutant are closely related to the modal activities of a vehicle in a trip—its operating speed, acceleration rate and time, deceleration rate and time, as well as engine idle time (e.g., Horowitz 1982, Chatterjee et. al. 1997, Holmen and Niemeier 1998).
The emissions of a transportation system can be estimated using a number of emission models, such as MOBILE and EMFAC. The inputs to these models include vehicle-miles-traveled (VMT), average travel speed, and vehicle fleet characteristics. Both VMT and average travel speed can be obtained from either field measurements or transportation planning models. However, it is often found that there are significant discrepancies between model outputs and real world measurements of vehicle emissions (Niemeier 1998, Sperling 1998). One of the major sources of error is attributed to average travel speed (Chatterjee et. al. 1997). For a large-scale transportation system, it is impossible to measure or simulate all detailed vehicle motions for each vehicle. Therefore, procedures are developed to measure tailpipe emissions of particular vehicles in representative trips, called driving cycles, in a region and relate the emission rates with the average travel speed patterns. Significant estimation errors will arise if the underlying speed patterns over various sections of highways differ largely from those of driving cycles. In order to reduce the estimation error caused by mismatch between testing driving cycles and real world driving conditions and by the inadequacy of transportation planning models to estimate average speed in congested networks, one has to find better ways to characterize modal activities from coarsely measured or estimated traffic flow data.
Modal activity refers to the operating mode of a vehicle during a trip: cruise, acceleration, deceleration, and idling. It is related to both space and time, and an accurate description of it requires the knowledge of traffic evolution in time-space. The best way to describe the modal activity of a vehicle is through a time-space diagram, from which the speed, acceleration and deceleration of the vehicle can be obtained. Ideally one should get such a trajectory for each vehicle in a traffic stream such that the modal activities of a traffic stream can be accurately obtained. However, this is impractical if in large-scale applications. Methods of sampling and estimation have to be developed to obtain as much information as one can from limited numbers of measurements about the modal activities of a traffic stream on various types of facilities.
In this research we develop methods that estimate traffic modal activities in a highway network from data gathered by widely deployed sensors—loop detectors. Here we use loop data in a special way. We piece data together over both time and space, and set up a velocity-vector field, from which one can construct vehicle trajectories in the detectable region. We can obtain the velocity vectors at any detector location by reducing the polling interval to as small as a second. However, we cannot place as many detectors as desired on the freeway to obtain an accurate speed distribution along the freeway. Here, we can use interpolation methods to produce the vector field on the entire detectable region, because the flow conditions at adjacent detector locations on a freeway are strongly related if these detectors are not too far apart.
Two interpolation methods are proposed to construct vehicle trajectories from loop data. One is spline interpolation and the other is quadratic polynomial interpolation. The details of these two methods can be found in Zhang and Wang (2001). The two methods are applied to sample data of a segment on freeway I-880. In each lane, there are 11 detectors along the 22,100ft (4.2 miles) length. The distances between any two adjacent detectors vary from 1,000ft (0.19 miles) to 4,600ft (0.87 miles). The data start at time 62,010 second, and the detecting interval is 30 seconds, i.e., loop detectors report measuring results every 30 seconds.
Selected vehicle trajectories constructed from the spline method are shown in Figure 1, and those constructed from the quadratic interpolation method are shown in Figure 2. The speed contours are also shown in both figures. One can see from both figures that the traffic slow-down was captured by both methods. It should be noted, however, that the constructed vehicle trajectories are tightly bunched during the sharp traffic acceleration movement, which is not desirable. Further research is needed to refine the interpolation methods to obtain more realistic vehicle trajectories in sharp acceleration regimes. Moreover, vehicle trajectories from real traffic flow are needed to check the accuracy of the constructed vehicle trajectories from loop data.
Fig. 5. Speed contour and sample vehicle trajectories (by Spline model)
Fig. 6. Speed contour and sample vehicle trajectories (by quadratic interpolation model)
References
Chatterjee, A., T. L. Miller, J. W. Philpot, T. F. Wholley, R. Guensler, D. Hartgen, R. A. Margiotta, and P. R. Stopher. (1997). Improving Transportation Data for Mobile Source Emissions Estimates. NCHRP Report 394, Transportation Research Board, Washington D. C.
Davis, S. T. (1997). \textit{Transportation Energy Data Book}, Edition 17. Report ORNL-5198, Oak Ridge National Laboratory, Oak Ridge, Tennessee.
Eugene V. S., Alexander I. P., 1995. Handbook on Splines for the user.
Gaffney, P. W., 1973. Fortran subroutines for bicubic spline interpolation.
Holmen, B. A. and D. A. Niemeier (1998). Characterizing the effects of driver variability on real-world vehicle emissions. \textit{Transpn. Res. -D}. Vol. 3, No. 2. pp. 117-128.
Horowitz, J. L. (1982). \textit{Air quality analysis for urban transportation planning}. The MIT Press, Cambridge, Massachusetts.
McLeod, Robin J. Y., 1998. Geometry and interpolation of curves and surfaces.
Michael Zhang and Bo Wang, 2001. Construction of vehicle trajectories from loop data, ITS-Davis Project Report.
Niemeier, D. A. (1998). Private communications.
Sakhnovich, L. A., 1997. Interpolation theory and its applications.
Sperling, D. (1998). Private communications.
Skabardonis, A., Petty, K., Noeimi, H. Rydzewski, D. and Varaiya, P. (1996). I-880 field experiment: data-base development and incident delay estimation procedures. Transportation Research Record 1554, pp. 204-212.
Washington, S. and C. Roberts (1998). Development of facility-specific level-of-service based driving cycles. Draft final report to Caltrans.
Papers: