This research addresses the problem of modeling time-dependent traffic flow with real-time traffic sensor data for the purpose of online traffic estimation and prediction to support ATMS/ATIS in an urban transportation network. The fundamental objectives of this study are to formulate and develop a dynamic traffic flow model driven by real-world observations, which is suitable for mesoscopic type dynamic traffic assignment simulation.

A dynamic speed-density relation is identified by incorporating the physical concept in continuum and kinetic models, coupled with the structural formulation of the transfer function model which is used to represent dynamic relationship. The model recognizes the time-lagged response of speed to the influential factors (speed relaxation, speed convection and density anticipation) as well as the potential autocorrelated system noise. The procedures adapted from transfer function theory are presented for the model estimation and speed prediction using the real-time data. Speed prediction is performed by means of minimum mean square error and conditional on the past information.

In the context of real-time dynamic traffic assignment simulation operation, a framework based on the rolling-horizon methodology is proposed for the adaptive calibration of dynamic speed-density relations to reflect more recent traffic trends. To deal with the different time scales in the data observation interval and the traffic simulation interval, an approximation procedure is proposed to derive proper impulse responses for traffic simulation. Short term correction procedures, based on feedback control theory, are formulated to identify discrepancies between simulation and real-world observation in order to adjust speed periodically.

Numerical tests to evaluate the dynamic model are conducted in a standalone manner firstly and then by integrating the model into a real-time DTA system. The overall conclusion from the results is that the proposed dynamic model is preferable in the context of real-time application to the use of conventional static traffic flow models due to its higher responsiveness and accuracy, although many other aspects remain to be investigated in further steps.