Quantifying uncertainties in synthetic origin-destination trip matrix estimates

The use of microscopic traffic simulation models in traffic operations, transportation design, and transportation planning has become widespread across the United States because of: (i) rapidly increasing computer power which is required for complex micro-simulations; (ii) the development of sophisticated traffic micro-simulation tools; and (iii) the need by transportation engineers to solve complex problems which do not lend themselves to traditional analysis techniques. The origin-destination trip matrix is a fundamental input to most transportation systems analysis models including micro-simulation models. The matrix reflects the volume of traffic between all origins and destinations in the transportation network. The OD matrix is difficult and often costly to obtain by direct methods such as license plate surveys. Consequently, indirect or synthetic techniques that seek to simulate an OD matrix close to a prior or possibly outdated matrix and which when assigned to the network produces a link flow pattern sufficiently “close” to a set of traffic counts observed on sections of the traffic network are widely used. It is also important that the resulting micro-simulation model operates as close to reality as possible. This requires that the default driver-behavior and other model parameters are adjusted to match those of the specific network and driver population for which the model is being developed. This dissertation develops a genetic algorithm procedure for simultaneously estimating an OD matrix and calibrating microscopic traffic simulation models to local conditions. In particular, the procedure treats the elements of the OD matrix as unknown parameters that must be jointly calibrated along with those of the driver-behavior parameters. The dissertation also demonstrates, through a case study, that an erroneous OD input could have far-reaching negative consequences. This is not surprising because the OD matrix is such a fundamental input to the micro-simulation model. However, the finding highlights the need for a measure that gives some indication of the quality of (or the uncertainties associated with) an OD estimate in the form of, for example, a standard deviation (or a given multiple of it), or the width of a confidence interval. Building on this, the dissertation presents the gapped bootstrap uncertainty estimator - a recently developed statistical technique that is uniquely suited for handling uncertainties in dependent exchangeable data. An application of the gapped bootstrap method to both empirical and simulated data led to fairly conservative estimates of uncertainty that were, on average, larger than those of the traditional and block bootstrap uncertainty estimators. In many Intelligent Transportation Systems (ITS) applications having a slightly conservative estimate of uncertainties is not a major disadvantage