Bayesian estimation of the Origin-Destination matrix using traffic flow dynamics

Origin-destination (O-D) matrices are essential inputs to dynamic traffic assignment and traffic simulation models and important tools to transportation planning. We present a novel approach for static O-D matrix estimation using traffic flow dynamics. A path-based cell transmission model is developed to capture the dynamics of a network and associate link count observations with path demand patterns. We assume that the path demand follows a Poisson distribution with unknown rates and definition of path choice probabilities is not required. A state space model is utilised to associate link densities (observations) with per path densities (state vector). The proposed model is combined with Bayesian inference techniques to develop an efficient methodology for estimating accurate posterior probability density functions of the path demand from which we obtain the O-D demand. This involves numerical techniques such as Markov chain Monte Carlo, that make use of the Metropolis-Hastings update. We illustrate the proposed approach in a network studied in the literature and present results that show the advantage of the proposed methodology compared to existing methods studied in the literature. Finally, we discuss the findings of the analysis that suggest that the path-based CTM combined with the Bayesian principles can be utilised to efficiently approximate unknown path demand patterns and obtain static O-D matrices, along with a future direction for further advances of the methodology.