Bayesian mixture model for estimating travel time distribution based on low frequency probe vehicle data

Abstract

This dissertation aims to estimate travel time based on low and unknown sampling rates of probe vehicles and has two main objectives: i) enhance understanding of travel time properties and ii) develop a novel estimation model of travel time distribution. First, we explored the characteristics of travel time according to traffic state changes and congested traffic. The systematic relation between spot and section measurements is theoretically and empirically examined

as well, the characteristics of travel time variability are empirically observed. Then, we developed a novel estimation model of travel time distribution based on the Bayesian mixture model. To reduce the estimation error due to low and unknown sampling rate of probe vehicles, we used prior information in the Bayesian approach based on the characteristics of travel time regularity. To evaluate the accuracy of the methodology, we used empirical data from expressways and arterial roads. The Bayesian mixture model outperforms different estimation approaches.