Revealing heterogeneous spatiotemporal traffic flow patterns of urban road network via tensor decomposition-based clustering approach

Understanding the complex heterogeneity of traffic flow for the road network is of great importance to relieving traffic congestion and designing traffic control strategies. This study proposes a 3-stage pattern extraction approach to discover the heterogeneous spatiotemporal traffic flow patterns from the enormous observations: The non-negative tensor decomposition technique is utilized to extract the features of traffic flow in the within-one-day time, day-to-day and spatial segment modes; The spectral clustering is performed in the low-dimensional feature space, and the Gaussian similarity matrix is modified by the spatial proximity coefficient to integrate the spatial compactness into the clustering algorithm; The large-scale road network is partitioned into the independent and compact subnetworks with similar traffic congestion levels, and the macroscopic traffic flow relationships are fitted for each pattern. Research results indicate that the statistical distribution of traffic flow for each pattern is diverse; the heterogeneity of traffic patterns is verified by the low-scatter diagrams of macroscopic traffic flow relation for each subnetwork; the congested subnetworks are detected for both the peak hours and non-peak hours; the structures of subnetworks are highly concordant with the urban functional regions, varying with the different periods.