Study on Dynamic Origin-Destination Estimation of Large-Scale Road Networks

Abstract

Dynamic Origin-Destination (OD) demands are essential input for on-line traffic control，dynamic traffic-assignment simulation and management systems. Although various models of dynamic Origin-Destination estimation have been presented, estimation error will rapidly increase as scale of road networks expands. This article puts forward a method for the dynamic Origin-Destination estimation of a large-scale road network on the basis of geographic information system (GIS). The proposed model which only depends on measurable time series of mainline from traffic counts can automatically realize dynamic Origin-Destination estimation in some small region by reconstructing OD matrixes. The model will be embedded in intelligent traffic systems and provide essential data for real traffic dynamic management. It is a good method to realize dynamic OD estimation of large-scale road network. 1. INTRDUCTION Dynamic Origin-Destination (OD) demands are essential input for advanced traffic management system and on-line traffic control. There have been increasing demands for dynamic OD information. It is difficult to obtain dynamic OD information because the number of parameters of OD information to be estimated is far greater than available information. Therefore, it is challenging to estimate OD demand. In the past several decades, transportation researchers have developed various methods to estimate OD matrixes of road networks with high accuracy on the basis of time-varying traffic volume, prior dynamic OD information and so on. In existing literatures, recent studies for dynamic OD estimation can be classified into two kinds: assignment-based and non-assignment-based approaches (Lin and Chang, 2007). The former is based on the assumption that a reliable prior time-varying OD set and an accurate dynamic traffic assignment model are available. Considering that it is difficult to obtain a reliable time-varying OD set, non-assignment-based approaches have been put forward to reduce the dependency of prior time-varying OD set. We focus on non-assignment-based approaches and follow the research line of the approaches.