Internally-consistent estimation of dynamic network origin-destination flows from intelligent transportation systems data using bi-level optimization

Deployment of Intelligent Transportation Systems (ITS) is providing researchers and practitioners with an unprecedented amount of valuable on-line and archived traffic data. To date, ITS data have been used primarily to support real-time operational applications, while other potential uses of these data have been largely ignored. In this research, the effort to extract knowledge from the on-line or archived data gathered by Advanced Transportation Management Systems (ATMS) is focused on the estimation of dynamic origin-destination (OD) flows using optimization methods. In addition to their use for planning purposes, time-dependent OD flows can be used as an input to Dynamic Traffic Assignment (DTA) systems. However, gathering OD demand flow information directly by conducting surveys is very costly and time consuming. To estimate the OD flows, a methodology is developed to minimize an overall measure of the deviation of estimated link-flows from the time-varying link-flow observations, subject to a set of constraints. The set of constraints could include nonnegativity constraints, initial condition constraints, cordon line counts and the user’s route-choice behavior or traffic assignment rules. The traffic assignment solution, itself, is often obtained by optimizing an objective function. This objective function can explicitly be included in the constraints of the main or upper minimization problem. This formulation results in a bi-level optimization or theoretical game problem. In this dissertation, the upper-level problem is formulated alternatively as linear and non-linear optimization problems. To solve the lower-level traffic assignment problem, a DTA simulation program, namely DYNASMART-P, is used to find the equilibrium flows. The suggested algorithm iterates between the upperlevel and the lower-level optimization problems for a pre-specified number of times or until convergence in terms of the estimated OD flows or the simulated link flows is achieved. To integrate the a priori information on OD demand flows with the information extracted from the link flow observations, adoption of the Bayesian inference method is proposed. If such information on OD flows is available, Bayesian inference treats the old information as the target values to update the estimated OD flows from the sample of the link flow observations.