A dynamic linear model for the estimation of time-varying origin-destination matrices from link counts

Abstract

In this paper, we propose a dynamic linear model (DLM) for the estimation of timevarying OD matrices from link counts. Our main research objective is to investigate whether we can overcome the non-identifiability problem, which occurs in static models, by dynamically modeling the time evolution of OD flows. We establish the conditions under which mean OD flows may be estimated and carry out computational studies on two benchmark transportation networks from the literature. In both cases the DLM converged to the unobserved mean OD flows when given sufficient observations of traffic link volumes despite the use of uninformative prior OD matrices. These results indicate that the dynamic modeling of OD flows is a promising research direction for the estimation of transportation demand.