Network traffic modeling with vehicle trajectory data

This dissertation is motivated by the growing appearance of traffic data collected by probe vehicles in the context of transportation network modeling and planning problems. In contrast to conventional fixed-sensor data, probe data, especially in the form of vehicle trajectories, offers considerably wider geographic coverage and richer information. By extracting the information inside trajectory data, this dissertation explores the dynamics of traffic flow from the network-wide perspective. Both simulation and real-world observations show that trajectory data, after appropriately processed and aggregated, can help investigate areas which are hard to address with conventional traffic data, in particular, travel time reliability and network accessibility problems. The research also extends the application of trajectory data from offline planning to the online operation and traffic management framework.;More specifically, a robust linear relationship between standard deviation of travel time per unit distance and mean travel time per unit distance are validated at different aggregation levels for different scales of road networks (Irvine, Washington-Baltimore, Chicago, Salt Lake City, New York City) across the United States, using either simulated or real trajectories. A network accessibility measure is proposed, which takes into account of both network topological structure and prevailing traffic condition. It can be easily quantified using information from vehicle trajectory data, and its usefulness is demonstrated through a case study on Chicago network. The bivariate relationships among network reliability, network accessibility, and other commonly known traffic stream variables (e.g. network density, network flow, space-mean speed) are investigated through empirical observations from simulated and real data. The theoretical relationships between these network-wide traffic flow quantities are established by combining the existing models from different aspects. A systematic framework is developed to provide guidelines of incorporating vehicle trajectory data into real-time traffic estimation and prediction systems. Examples of utilizing trajectory data in the real-time environment are demonstrated by a case study on Salt Lake City network.;Overall, this dissertation provides both conceptual and methodological ways of leveraging potential values of vehicle trajectory to model, evaluate, and manage network traffic flows from different perspectives. The research outcomes of this dissertation serve as the basis for recommendations on how trajectory data can be used to further enhance and enable the application of network traffic flow models from both offline planning and online operation perspectives.