Prediction of Network Loads Based on Origin-Destination Synthesis from Observed Link Volumes

Abstract

An iterative approach to predicting network loads is presented. It is based on the assumption that some assignment approach first provides a ball-park estimate of network loads and therefore can be used as a prior estimate in an iterative Bayesian procedure. The procedure then uses as an intermediate step a Bayesian update to synthesize an origin-destination (OD) trip matrix from the measured link loads. It allows for different degrees of belief in the prior estimates of ODs as well as in different aspects of these prior estimates. On the basis of differences between the measured link loads and those based on OD estimates, an iterative procedure is proposed so that the updated synthesized ODs approximately produce the measured loads. Because the eventual objective of the developed model is to predict traffic loads in the case of unexpected traffic congestion or a traffic incident, an application of this approach is presented within this scenario.