This article is concerned with the development of eco-driving metrics for instrumented vehicles in a longitudinal study environment. Motivations for developing such metrics include an ability to distill driving style effects on fuel use from other confounding factors, to contrast and benchmark driving styles for a cohort of drivers and to ascertain the effects of information and/or incentives on fuel use both in the short and long term. High resolution (1 Hz) trip data were collected for a local sample of 35 drivers over a period of 2 years, yielding over 20 million second by second observations. To account for the difference in vehicle type choice, a standard vehicle was used to model fuel consumption based on instantaneous vehicle activity. Difference in route choice was accounted for using speed-bin dependent metrics. Two metrics were developed: a trip-based measure called the fuel efficiency score (FES), and a difference in fuel use metric that uses the second by second observations called the fuel use difference (FUD). FES varies from 20 to 100 while FUD covers positive and negative percentage differences from a speed-bin dependent mean value. Both measures passed the test of consistency so that, at the driver level, both revealed no temporal trend in the scores from month to month across a period of 2 years. Moreover, the FES metric passed the heterogeneity test. It was able to identify four distinct clusters of driving styles.

Point-based traffic sensors, such as microwave radar and acoustic sensors, provide the valuable capability of sampling the entire traffic stream. However, full network coverage with point sensors requires a significant initial capital investment and ongoing maintenance expenditures. Probe-based sensors can cover an extensive roadway network at a much lower cost because roadway-based field equipment is not required. Decisions regarding the relative level of point sensor- versus probe-based deployment for traffic monitoring involve evaluating the trade-off between the value of comprehensive detection versus total system costs. An essential step in evaluating this trade-off involves directly comparing collocated point sensor and probe vehicle systems to understand how the derived traffic stream measures from the two approaches differ. This study compared 5-min speeds from microwave radar and acoustic sensors with link speeds from Global Positioning System (GPS) probes for both directions at five freeway locations. Systematic differences were found at one location. Floating car GPS runs were performed to confirm that the systematic error lay in the point speeds. The speed differences at all sites were normally distributed, with three locations indicating a mean speed difference greater than 5 mph. Nonsystematic speed differences were identified; the difference was more than 1.5 standard deviations lower than the mean difference. This difference may indicate inherent inaccuracies in reported GPS speeds under heavy congestion, including instances of time lag in recovering from congested speeds.

No abstract

Oversaturated speed, flow, and density relationships are of key importance to studies of freeway operations. The Highway Capacity Manual (HCM) oversaturated model, which is defined by a linear transition from the flow and the density at capacity to a zero flow at jam density in the flow–density space, provides a reasonable representation of this relationship but does not provide an unbiased representation for all freeway facilities with different road conditions or driver behavior. This study proposes a method for fitting the HCM model to oversaturated flow and density. Fifteen-minute aggregated flow rate and speed data were collected in 2010 from Traffic.com fixed-location sensors at three sites on North Carolina urban freeways. Density was calculated as the flow rate divided by the speed. The fitted models for these sites were compared with the default HCM model. A set of thresholds was defined to identify eligible sensor observations that represented the steady-state congested traffic conditions. The results revealed that data observations during inclement weather, lane closures, or incidents biased the model-fitting results and therefore needed to be filtered out. The steady-state congestion data identified in the manner proposed in this study fit well with the HCM-based linear flow–density oversaturated model. This method avoids possible bias caused by capacity and jam density differences between the default HCM model and the site-specific models; therefore, the fitted models represented the actual traffic characteristics relationships better than the default HCM models did. Fitting a site-specific HCM-based model is recommended for sites with sufficient speed and flow data.