Deep multi-view information-powered vessel traffic flow prediction for intelligent transportation management

作者：Huanhuan Li, Yu Zhang, Yan Li, Jasmine Siu Lee Lam, Christian Matthews, Zaili Yang · 发表于：Transportation Research Part E Logistics and Transportation Review · 年份：2025 · DOI：10.1016/j.tre.2025.104072 · 被引用次数：20 · 研究领域：Traffic Prediction and Management Techniques、Air Quality Monitoring and Forecasting、Transportation Planning and Optimization

• Develop a new hierarchical methodology for vessel traffic flow (VTF) prediction. • Extract the periodic and temporal features of VTF data effectively. • Realise collaborative prediction among multiple channels via semantic features. • Design a new loss function from both global and local perspectives. • Compare the proposed model’s performance with eleven state-of-the-art methods. Vessel traffic flow (VTF) prediction, essential for intelligent transportation management, is derived from the statistical analysis of longitude and latitude information from Automatic Identification System (AIS) data. Traditional deep learning approaches have struggled to effectively capture the intricate and dynamic characteristics inherent in VTF data. To address these challenges, this paper proposes a new prediction model called a Multi-view Periodic-Temporal Network with Semantic Representation (i.e., MPTNSR), which leverages three perspectives: periodic, temporal, and semantic. VTF typically conceals the periodic and temporal characteristics during its evolution. A Convolutional Neural Network and Bidirectional Long Short-Term Memory (CNN-BiLSTM) model, constructed from periodic and temporal views, effectively captures this information. However, real-world scenarios frequently involve predicting VTF for multiple target regions simultaneously, where correlations between VTF changes in different areas are significant. The semantic view seeks to extract relationships across different channels b...