Time-dependent Origin-Destination (OD) demand information is a fundamental input for Dynamic Traffic Assignment (DTA) models to describe and predict time-varying traffic network flow patterns, as well as to generate anticipatory and coordinated control and information supply strategies for intelligent traffic network management. This dissertation addresses a series of critical and challenging issues in estimating and predicting dynamic OD demand for off-line and on-line DTA operation in a large-scale traffic network with various information sources.

Based on an iterative bi-level estimation framework, this dissertation aims to enhance the quality of OD demand estimates by combining available historical static demand information and time-varying traffic measurements into a multi-objective optimization framework that minimizes the overall sum of squared deviations. The multi-day link traffic counts are also utilized to estimate the variation in traffic demand over multiple days. To circumvent the difficulties of obtaining sampling rates in a demand population, this research proposes a novel OD demand estimation formulation to effectively exploit OD demand distribution information provided by emerging Automatic Vehicle Identification (AVI) sensor data, and presents several robust formulations to accommodate possible deviations from idealized conditions in the demand estimation process.

A structural real-time OD demand estimation and prediction model and a polynomial trend filter are developed to systematically model regular demand pattern information, structural deviations and random fluctuations, so as to provide reliable prediction and capture the structural changes in time-varying demand. Based on a Kalman filtering framework, an optimal adaptive updating procedure is further presented to use the real-time demand estimates to obtain a priori estimates of the mean and variance of regular demand patterns. To maintain a representation of the network states which consistent with that of the real-world traffic system in a real-time operational environment, this research proposes a dynamic OD demand optimal adjustment model and efficient sub-optimal feedback controllers to regulate the demand input for the real-time DTA simulator while reducing the adjustment magnitude.