Quantifying Input Uncertainty in Traffic Assignment Models

Traffic assignment methods distribute Origin-Destination (OD) flows throughout the links of a given network according to procedures related to specific deterministic or stochastic modeling assumptions. In this paper, we propose a methodology that enhances the information provided from traffic assignment models, in terms of delivering stochastic estimates for traffic flows on links. Stochastic variability is associated to the initial uncertainty related to the OD matrix used as input into a given assignment method, and therefore the proposed methodology is not constrained by the choice of the assignment model. The methodology is based on Bayesian estimation methods which provide a suitable working framework for generating multiple OD matrices from the corresponding predictive distribution of a given statistical model. Predictive inference for link flows is then straightforward to implement, either by assigning summarized OD information or by performing multiple assignments. Interesting applications arise in a natural way from the proposed methodology, as is the identification and evaluation of critical links by means of probability estimates. A real-world application is presented for the road network of the northern, Dutch-speaking region of Flanders in Belgium, under the assumption of a deterministic user equilibrium model.