HIERARCHICAL BAYESIAN ESTIMATION OF FREEWAY TRAVEL TIME DISTRIBUTIONS WITH SMALL SAMPLES OF NON-IDENTICAL PROBES IN SUCCESSIVE DAYS

摘要

This study formulates a hierarchical Bayesian mixture model for estimating freeway sectiontravel time distributions from small probe samples collected over multiple days. Two normalcomponents are used to capture the heterogeneity in experienced travel times and modelvarious distributional shapes which are generally known to be skewed or multimodal.Historical travel time data collected during different intervals under similar traffic conditionsare used to construct the prior for model parameters via the hierarchical Bayesian paradigm.The posterior distributions can be continuously updated when new probe data becomeavailable. Based on this model mechanism, travel time is able to be predicted under differentlevels of data availability. A simulation study with 3% sample rate shows that the travel timedistributions can be well approximated using the proposed model.