Abstract
we present a methodology of estimating dynamic link flows and origin-destination matrices using lower polling frequency probe vehicle data (e.g. one point every 30-60s). Link travel time is first obtained from map-matched probe points using a method of proportional allocation. A derived speed-density function is then fitted for different types of roads. A Bayesian method that carefully incorporates prior information is used to estimate dynamic link flows from link travel speed. A bi-level generalized least-square (GLS) estimator is formulated so as to estimate dynamic OD matrices from estimated link flows. A traffic simulator in VISSIM is developed for a median size urban network using an open data set. The results validate the advantages of the proposed method for lower polling frequency probe vehicle data.
