Dynamic multiple-graph spatial-temporal synchronous aggregation framework for traffic prediction in intelligent...

Area of Science:

• Intelligent Transportation Systems (ITS)
• Traffic Prediction
• Graph Neural Networks

Background:

• Accurate traffic prediction is vital for optimizing intelligent transportation systems (ITS) and road network efficiency.
• Existing methods struggle to model complex spatial-temporal correlations, especially when incorporating external factors and diverse graph structures.

Purpose of the Study:

• To propose a novel framework, the dynamic multiple-graph spatial-temporal synchronous aggregation framework (DMSTSAF), for enhancing traffic prediction.
• To address limitations in existing models regarding external factor integration and multi-perspective graph construction.

Main Methods:

• DMSTSAF employs a feature augmentation module (FAM) to fuse traffic data with external factors.
• The framework utilizes diverse spatial and temporal graphs and designs synchronous aggregation modules to extract features from multiple perspectives simultaneously.

Main Results:

• DMSTSAF demonstrated significant improvements in traffic prediction accuracy.
• The model achieved performance gains of 3.68-8.54% over state-of-the-art baselines on four real-world datasets.

Conclusions:

• The proposed DMSTSAF effectively models spatial-temporal correlations in traffic data.
• The framework's ability to incorporate external factors and leverage multiple graph perspectives leads to superior traffic prediction performance.