LOG-LINEAR MODELS FOR THE ESTIMATION OF ORIGIN-DESTINATION MATRICES FROM TRAFFIC COUNTS: AN APPROXIMATION
Mgh Bell · 1984
Origin-destination matrices play a fundamental role in transport planning and analysis as well as in traffic management and control. However, their estimation by direct means such as interviews, licence plate surveys and aerial photography is both time consuming and expensive, as well as being unsuitable for on-line applications where the matrix is to be frequently updated. Therefore indirect procedures based on traffic counts and various forms of route assignment information have received much attention recently. These procedures are reviewed in the paper. In particular, a family of log-linear models has proved popular although iterative calibration of the models is required. This paper suggests a non-iterative procedure for obtaining an approximate log-linear model solution. It is suggested that this approximation is particularly suitable for on-line applications. Certain statistical properties of the approximate solution are discussed, and a relationship to Bayesian estimation is shown. The closeness of the approximation is examined empirically and is found to depend strongly on the quality of the prior information. (Author/TRRL)