研究目的

To develop a novel short-term traf?c ?ow prediction approach based on dynamic tensor completion (DTC) that can fully utilize the rich information in traf?c data, including temporal variabilities, spatial characteristics, and multimode periodicity, to improve prediction accuracy, especially when dealing with incomplete traf?c data.

研究成果

The proposed Dynamic Tensor Completion (DTC) method effectively utilizes multi-mode periodicity, spatial information, and temporal variations of traf?c ?ow by representing traf?c information as a dynamic tensor pattern. It outperforms conventional vector-based and matrix-based prediction approaches in terms of prediction accuracy and is capable of accurately forecasting traf?c ?ow with missing data. The method's efficiency and applicability to real-time prediction requirements are validated through experiments on benchmark datasets.

研究不足

The study acknowledges the challenge of accurately predicting traf?c ?ow when it is sharply changed, indicating a limitation in handling abrupt variations in traf?c conditions. Additionally, the scalability of the method for large transportation networks and the potential need for non-linear factorization improvements are noted as areas for future research.