Development of a Practical Computer Tool for Dynamic Origin-Destination Matrices Estimation       

Contracting Agency: Caltrans/UC Berkeley                                                     

Principal Investigator: Dr. Yafeng Yin

A considerable number of theoretical studies have been conducted to estimate time-dependent O-D matrices. However, the models produced by these investigations, particularly the ones for general networks are still in its infancy and call for significant improvements. Furthermore, most of these studies are limited to academic exercises and the techniques have hardly been calibrated, tested and applied in the actual world. This project attempts to bridge the gaps between practice and theory in O-D estimation. We plan to:

1)      Develop the methodologies for deriving time-dependent O-D matrices for linear networks and implement them in a computer tool (a linear network is a stretch of highway with multiple entries and exits, where there would be no route choices involved). The tool will work with data readily available from the typical closed-loop signal control system deployed along arterials, including traffic counts, left-turn movements and signal status, to estimate O-D matrices in a time interval of one minute or finer. The tool will also be able to combine complementary information, such as travel time, turning movements on selected intersections from video detection systems or sampled O-D information from AVI technologies to improve the accuracy of the O-D estimation.

2)      Explore the possibilities of developing a model to estimate time-dependent O-D matrices for general networks, where drivers may have alternative routes to choose between their origins and destinations. If the model is practically useful, a prototype tool will be developed for the purpose.

The research is led by Prof. Samer Madanat from University of California at Berkeley, joined by Prof. Michael Zhang from University of California at Davis, and Prof. Yafeng Yin from University of Florida. The tasks that University of Florida is taking the lead include:

• Development of a suite of model formulations and solution algorithms for dynamic O-D estimation for linear networks, primarily addressing the issues of incomplete information, measurement errors and partial O-D information;

• Calibration and verification of the above developed estimators.