Quasi-sparsity Based Origin-Destination Demand Estimation

Abstract

A good knowledge of the Origin-Destination (OD) demand matrix has been always important in various transportation applications, including simulation studies, transportation planning, traffic operations and control, and etc. For a large real network, the OD demand matrix may have certain quasi-sparsity property, i.e., the majority of the OD pairs have small demands while only a small portion of OD pairs have large demands. Inspired by Compressed Sensing technique, this dissertation proposes a Quasi-Sparsity Origin-Destination (QSOD) framework to explore such quasi-sparsity property of large-scale OD demand matrices. Three QSOD models (the fixed-mapping QSOD model, the bi-level QSOD model, and the distributionally robust QSOD model) are established under such QSOD framework. The results theoretically and numerically demonstrate that under certain conditions the estimated OD demands will share the same quasi-sparsity with the prior OD demands, and the estimated demands of most OD pairs (of a large-size network) will be equal to their prior values or zeros (or a very small value). Such findings provide important practical insights for OD estimation: one may only require the prior OD demands can capture the relative magnitude of the true OD demands of the network, which makes it much easier to prepare prior OD matrix in practice. The comparison between QSOD models and other existing OS estimation studies, and the integration of multi-sourced data for OD estimation under the QSOD framework are also discussed in this study.