Expressway traffic flow prediction based on MF-TAN and STSA.

Area of Science:

• Intelligent Transportation Systems
• Data Science
• Network Analysis

Background:

• Traffic congestion poses significant challenges to transportation efficiency and cost.
• Traditional traffic flow prediction models struggle with dynamic data correlations and static network structures.
• Accurate traffic prediction is crucial for real-time traffic management and travel guidance.

Purpose of the Study:

• To develop an advanced traffic flow prediction model that addresses limitations of traditional methods.
• To improve the accuracy and adaptability of expressway traffic flow forecasting.
• To leverage multi-feature data and spatial-temporal dependencies for enhanced prediction.

Main Methods:

• Data preprocessing to create a comprehensive dataset.
• Incorporation of multi-feature temporal attention to model dynamic correlations (speed, flow, saturation).
• Application of a spatial-temporal adaptive fusion graph convolutional network for capturing periodic similarities and dependencies.

Main Results:

• The proposed MFSTAPFGCN model demonstrated superior performance compared to traditional baseline models.
• Comparative experiments using real-world Electronic Toll Collection (ETC) data validated the model's effectiveness.
• Ablation experiments confirmed the significant contribution of individual modules within the MFSTAPFGCN framework.

Conclusions:

• The MFSTAPFGCN model offers a significant advancement in expressway traffic flow prediction.
• The model's ability to integrate multi-feature correlations and adaptive spatial-temporal dynamics leads to higher accuracy.
• This research provides a robust tool for intelligent transportation systems, aiding in traffic congestion mitigation.