An advanced computing architecture for large-scale network O-D estimation

Abstract

Existing studies for road network O-D estimation from link counts remain exploratory in nature, mostly developed on the assumption of having reliable prior O-D matrices and an accurate dynamic traffic assignment model. The computational requirements for use in large-scale networks have never been addressed either. This research presents a mathematical model and its computing architecture that allow for real-time estimation of dynamic O-D matrices in large-scale networks. The proposed model employs only link flow counts and dynamic screenline flows, and makes no assumption on drivers' route choice behavior. For a large network, the proposed model attacks the complex estimation issue in two stages: decomposing the entire network into several subnetworks for parallel computing in the first stage, followed by the update of key parameters with specially-designed screenline flows in the second stage. The preliminary results have shown the promising properties of the proposed method.