Spatio−Temporal Congestion-Aware Travel Time Prediction Based on Coupled Tensor-SVD

Accurate Travel Time Prediction (TTP) is critical for intelligent transportation systems to optimize route planning and congestion management. However, existing methods often suffer from two key limitations: (1) insufficient awareness of spatio−temporal context due to fragmented traffic data, and (2) low robustness to anomalous traffic data (e.g., sensor errors or extreme congestion), both of which lead to unreliable predictions. To address these challenges, we propose a spatio−temporal congestion-aware TTP method based on Coupled Tensor Singular Value Decomposition (CT-SVD-TTP). First, we construct two third-order tensors to represent the travel time and the corresponding occurrence probability for different road congestion levels on road segments during different time slots. Next, we construct another third-order tensor to characterize the dynamic spatio−temporal congestion effects across road segments, zones, and time slots. Subsequently, we propose a Coupled Tensor Singular Value Decomposition (CT-SVD) method to efficiently recover anomalous data in the original travel time tensor and the corresponding occurrence probability tensor by incorporating the dynamic traffic effects tensor during the coupled tensor decomposition process. Furthermore, we predict travel times of various road segments under different congestion levels during specific time slots by contracting the occurrence probability tensor with the recovered travel time tensor. Extensive experimental results demonstrate that the proposed CT-SVD-TTP method effectively recovers anomalous traffic data and achieves higher prediction accuracy than state-of-the-art methods on commonly used benchmarks.