Reliable Spatiotemporal Traffic Data Reconstruction for Intelligent Transportation Systems: An Adaptive Reweighted Triple Tensor Decomposition Framework

Reliable spatiotemporal traffic data are fundamental to Intelligent Transportation Systems (ITS), where traffic planning and management depend on continuous observations from multiple sensors. However, sensor failures, equipment maintenance, and communication disturbances can cause data loss and degrade the reliability of traffic information used by ITS applications. To address this issue, the study develops an Adaptive Reweighted Triple Tensor Decomposition (ARTTD) framework for reliable spatiotemporal traffic data reconstruction. Traffic features are organized as a third-order tensor with day, time interval, and sensor location. Unlike conventional triple decomposition methods with predefined ranks, ARTTD calculates from an initial candidate rank and evaluates latent component contributions. The nonconvex log-sum regularization, proximal alternating minimization, and adaptive triple-rank pruning are integrated to suppress redundant latent components and identify an effective triple rank. In addition, a Spatial Correlation Rate (SCR) index is designed to characterize sensor information dependencies and construct different sensor configurations. Experiments using real-world License Plate Recognition (LPR) data from Chongqing, China, consider point-wise and line-wise data loss patterns with missing ratios ranging from 10% to 80%. The results reveal relationships among sensor configuration, effective triple rank, and reconstruction accuracy, demonstrating that the ARTTD framework can support reliable traffic data reconstruction under diverse conditions and provides methodological support for data-driven intelligent transportation applications.