Flow-Based and Density-Based Time-Dependent Demand Estimation for Congested Urban Transportation Networks

This paper compares the estimation quality of two time-dependent origin–destination demand estimation methods in congested networks. The first method uses time-dependent link flow rate observations, whereas the second uses time-dependent link density observations. The developed methods adopt a bilevel formulation framework in which the lower-level problem solves for the time-dependent link proportions by using a simulation-based dynamic traffic assignment model. The upper-level problem is in the form of a least squares error minimization program that minimizes the difference between the observations and their corresponding estimated values. The solution algorithm integrates a simulation-based dynamic traffic assignment model and a linear approximation of the least squares error formulation by using a separable programming approach. An iterative solution algorithm is developed to achieve consistency between the estimated demand and the link proportions. A set of experiments conducted with hypothetical and real-world networks examines the performance of the two methods in replicating the observed congestion pattern in these networks. Results show that the density-based demand estimation method is more suitable for demand estimation in congested networks in that the former method captures the flow breakdown and spillback phenomena associated with high congestion.