Freeway Performance Measurement in a Connected Vehicle Environment Utilizing Traffic Disturbance Metrics

Abstract

The introduction of connected vehicles, connected and automated vehicles, and advanced infrastructure sensors will allow the collection of microscopic measures that can be used in combination with macroscopic measures for better estimation of traffic safety and mobility. This dissertation examines the use of microscopic measures in combination with the usually used macroscopic measures for traffic congestion evaluation, traffic state categorization, traffic flow breakdown prediction, and estimation of traffic safety. The considered macroscopic measures are the mean speed, traffic flow rate, and occupancy. The investigated microscopic measures for the stated purpose are: standard deviations of individual vehicle’s speeds, standard deviation of vehicles’ speed, and disturbance metrics. The utilized disturbance metrics to capture the stop-and-go operations are: the number of oscillations and a measure of disturbance durations in terms of the time exposed time–to–collision (TET), which has been used in other studies as a safety surrogate measure. However, this measure of disturbance duration requires the location and speed of both the leading and following vehicles and therefore cannot be measured accurately with low sample sizes of connected vehicles (CV). Thus, this study derived a model to estimate this measure based on speed parameters. The developed model was tested using real-world trajectory data from two locations that were not used in the development of the model. Moreover, the percentage of vehicles in the platoon and the platoon size distribution were evaluated as additional indicators of congestion. The relationship between the platooning and disturbance metrics and the speed parameters were further explored. It is recognized that the parameters required to identify the platoons, such as the time headway, will not be available based on data from low market penetrations of CV. Thus, a model was developed that utilize other measures for the estimation of the platooning measures at lower CV market penetrations. For the purpose of traffic state recognition and prediction, first, the study used a hybrid of two unsupervised clustering techniques to classify traffic states into “breakdown” and “non-breakdown”. The study found that adding the disturbance metrics in data clustering when identifying the traffic states will result in better traffic state recognition and traffic flow breakdown identification by capturing the disturbances in the traffic stream. The categorized traffic state was then used as a binary response to the macroscopic and microscopic measures, as features, to train supervised machine learning techniques for predicting traffic flow breakdown in the following 5-minute interval in real-time operations.