Traffic congestion is a complex phenomenon triggered by a combination of multiple interacting factors. One of the main factors is the disturbances caused by individual vehicles, which cannot be identified in aggregate traffic data. Advances in vehicle wireless communications present new opportunities to measure traffic perturbations at the individual vehicle level. The key question is whether it is possible to find the relationship between these perturbations and shockwave formation and utilize this knowledge to improve the identification and prediction of congestion formation. Accordingly, this paper explores the use of three machine learning techniques, logistic regression, random forests, and neural networks, for short-term traffic congestion prediction using vehicle trajectories available through connected vehicles technology. Vehicle trajectories provided by the Next Generation SIMulation (NGSIM) program were utilized in this study. Two types of predictive models were developed in this study: (1) offline models which are calibrated based on historical data and are updated (re-trained) whenever significant changes occur in the system, such as changes/updates to the infrastructure, and (2) online models which are calibrated using historical data and updated regularly using real-time information on prevailing traffic conditions obtained through V2V/V2I communications. Results show that the accuracy of the models built in this study to predict the congested traffic state can reach 97%. The models presented can be used in various potential applications including improving road safety by warning drivers of upcoming traffic slowdowns and improving mobility through integration with traffic control systems.

Modeling travel time reliability requires characterizing travel time distributions. Two key statistics commonly used to describe a distribution are the mean deviation and the standard deviation, with one depicting the central tendency and the other describing the dispersion. Although the mean travel time is easier to measure and predict, the corresponding standard deviation is usually hard to obtain because of the insufficiency of individual trip data. Building on seminal insight that goes back to Herman and Prigogine's kinetic theory, this study explores a robust characterization of travel time variability that provides for a near-linear relation between the standard deviation of travel time per unit distance and the corresponding mean value. Simulation-generated vehicle trajectory data from three road networks are used to explore this relationship. On the basis of multiscale and multilevel analysis, large amounts of data show that these two quantities are highly positively correlated; that is, the dispersion of the distribution of individual travel time per unit distance increases with increasing value of the mean travel time per unit distance. Furthermore, regression models and statistical testing indicate that this relation is linear or near-linear. The relation is also validated by Global Positioning System probe data from the Seattle, Washington, area. This relation provides a robust basis for predicting the standard deviation per unit distance when the mean value is known and thus for characterizing the reliability of travel in a network in strategic and operational studies.

This paper provides a discussion of network traffic flow stability and network capacity including existing definitions and measurement methods in the literature. To characterize network traffic flow, three phases are defined as free flow (F), inhomogenously congested (IC), and homogenously congested. It is shown that in large non-homogenous networks, the F → IC transition is usually a hysteretic phase transition. This paper examines the effects of adaptive driving on network capacity and traffic instability in a simulated network model of the Chicago metropolitan area as a large-scale highly congested complex network with exit flows. Results show that when the number of adaptive drivers increases, the general trend for network capacity is also increasing with some fluctuations due to randomness of traffic and overlapping effects of route choices. It is also found that adaptive driving may increase average network flow but may not necessarily improve network output, especially when the population of adaptive drivers is large and no gridlock exists. Results also suggest that with a larger population of adaptive drivers, hysteresis and gridlock are less likely to form. However, it is shown that bifurcation and multivaluedness can still occur even when the entire population of drivers is adaptive.

This paper presents a spatio-temporal trajectory clustering method for vehicle trajectories in transportation networks to identify heterogeneous trip patterns and explore underlying network assignment mechanisms. The proposed algorithm ST-TOPOSCAN is designed to consider both temporal and spatial information in trajectories. We adopt the time-dependent shortest-path distance measurement and take advantage of topological relations of a predefined network to discover the shared sub-paths among trajectories and construct the clusters. The proposed algorithm is implemented with a trajectory dataset obtained in the Chicago area. The results confirm the method's ability to extract and generate spatio-temporal (sub-)trajectory clusters and identify trip patterns. Extensive numerical experiments verify the method's performance and computational efficiency. Through spatio-temporal data mining, this paper contributes to exploring traffic system dynamics and advancing state-of-the-art spatio-temporal clustering for vehicle trajectories.