DYNAMIC DEMAND INPUT PREPARATION FOR PLANNING APPLICATIONS

*PhD Defense*
Time
March 16 2009 09:30–11:30
Location
4080 AIR Building
Klayut Jintanakul
TSE PhD Candidate
Abstract

A spectrum of traffic engineering and modern transportation planning problems requires the knowledge of the underlying trip pattern, commonly represented by dynamic Origin-Destination (OD) trip tables. In view of the fact that direct survey of trip pattern is technically problematic and economically infeasible, there have been a great number of methods proposed in the literature for updating the existing OD tables from traffic counts and/or other data sources. Unfortunately, there remain several common theoretical and practical aspects which impact the estimation accuracy and limit the use of these methods from most real-world applications. This dissertation itemizes and examines these critical issues. Then, the dissertation presents the developments, evaluations, and applications of two new frameworks intended to be used with the current and near-future data, respectively.

The first framework offers a systematic and practical procedure for preparing dynamic demand inputs for microscopic traffic simulation under planning applications with an estimation module based solely on traffic counts. Under this framework, the traditional planning model is augmented with a filter traffic simulation step, which captures important spatial-temporal characteristics of route and traffic patterns within a large surrounding network, to improve the flow estimates entering and leaving the final microscopic simulation network. A new bounded dynamic OD estimation model and a solution algorithm for solving a large problem are also proposed.

The second framework utilizes additional information from small probe samples collected over multiple days. There are two steps under this framework. The first step includes a suite of empirical and hierarchical Bayesian models used in estimating time-dependent travel time distributions, destination fractions, and route fractions from probe data. These models provide multi-level posterior parameters and tend to moderate extreme estimates toward the overall mean with the magnitude depending on their precision, thus overcoming several problems due to non-uniform (over time and space) small sampling rates. The second step involves a construction of initial OD tables, an estimation of route-link fractions via a Monte Carlo simulation, and an updating procedure using a new dynamic OD estimation formulation which can also take into account the stochastic properties of the assignment matrix.