VEHICULAR TRAFFIC MODELLING, DATA ASSIMILATION, ESTIMATION AND SHORT TERM TRAVEL TIME PREDICTION

Loading...
Thumbnail Image

Publication or External Link

Date

2014

Citation

Abstract

This dissertation deals with the problem of short term travel time prediction. Traffic dynamics models and traffic measurements are in particular the tools in approaching this problem. Effectively, a data-driven traffic modeling approach is adopted. Assimilating key traffic variables (flow, density, and speed) under standard continuum traffic flow models is fairly straight-forward. In current practice, travel time (space integral of pace or inverse of speed) is obtained through trajectory construction methods. However, the inverse problem of estimating speeds based on travel times is generally under-determined. In this dissertation, appropriate dynamic model and solution algorithms are proposed to jointly estimate speeds and travel times. This model essentially paves the way to assimilate travel time data with other traffic measurements. The proposed travel time prediction framework takes into account the fact that in reality neither traffic models nor measurements are flawless. Therefore, optimal state estimation methods to solve the resulting state-space model in real-time are proposed. Alternative optimality criterion such as minimization of the variance of estimate errors and minimization of the maximum (minmax) estimate errors are considered. Practical considerations such as occurrence of missing data, delayed (out of order) arrival of measurements and their impact on solution quality are addressed. Proposed models and algorithms are tested on datasets provided under NGSIM project.

Notes

Rights