ANALYSIS OF LARGE SCALE URBAN TRANSPORTATION NETWORK USING URBAN TRAFFIC DATA

Abstract

The thesis analyzes traffic data for estimation and forecasting in the large-scale urban transportation network, including effective and in-depth data processing, multi-source data fusion, consideration of topological structure, and recognition of heterogeneous patterns. A low-rank minimization technique is first proposed to impute missing traffic data by considering the non-linear spatial and temporal correlations. The time-dependent traffic volume and vehicle fleet composition in a grid network are inferred with Gaussian process regressions, follow by the spatial-temporal estimation of vehicular emissions. The prediction of short-term link traffic speed is performed with the GraphSAGE models in urban road networks. Besides, the thesis also investigates the propagation of traffic congestion by first generating congested area with Markov random field, and therefore modeling the congestion probability with a collaborative filtering method. Case studies show that the proposed methods achieves high performance in terms of accuracy.