Alternative Approaches for Real-Time Estimation and Prediction of Time-Dependent Origin–Destination Flows

TLDR

Both approaches can be formulated as state‑space models, with the first extending the authors’ prior work. This paper examines two different approaches for real‑time estimation and prediction of time‑dependent Origin–Destination flows. The first model represents the state vector as deviations in O–D flows, while the second defines it using deviations of departure rates and destination shares, and both are evaluated on real traffic data. Approximations render the real‑time estimation computationally tractable with minimal loss in accuracy, and the departure‑rate/share formulation improves prediction quality at the cost of slightly less accurate filtered estimates.

Abstract

This paper examines two different approaches for real-time estimation/prediction of time-dependent Origin–Destination (O–D) flows. Both approaches lend themselves to formulation as state-space models. The first approach is an extension of previous work by the authors. The key idea in this approach is to define the state-vector in terms of deviations in O–D flows instead of the O–D flows themselves. We demonstrate that approximations to this model make the real-time estimation process computationally more tractable with little deterioration in quality of estimates. In the second approach, the state vector is defined in terms of deviations of departure rates from each origin and the shares headed to each destination. This approach attempts to capture the differential variation of departure rates and shares over time. Performance of the proposed models is evaluated using actual traffic data from different sources. Preliminary results indicate that the filtering procedure is robust and that, compared to the original model, a formulation based on departure rates and shares yields better predictions with some loss of accuracy in filtered estimates.