Linear attention based spatiotemporal multi graph GCN for traffic flow prediction.

Area of Science:

• Artificial Intelligence
• Transportation Engineering
• Data Science

Background:

• Intelligent Transportation Systems (ITSs) are crucial for urban traffic management.
• Traffic flow prediction is key to reducing congestion and optimizing route planning.
• Existing models face challenges in handling large-scale data and complex dependencies.

Purpose of the Study:

• To introduce a novel deep learning model, LASTGCN, for accurate traffic flow prediction.
• To enhance traffic prediction by integrating meteorological factors and spatial-temporal correlations.
• To achieve computational efficiency for mid-term and potentially real-time traffic management.

Main Methods:

• Developed the Linear Attention Based Spatial-Temporal Multi-Graph Convolutional Neural Network (LASTGCN).
• Incorporated a Multifactor Fusion Unit (MFF-unit) for meteorological data integration.
• Utilized a Receptance Weighted Key Value (RWKV) block with linear attention for historical data processing.

Main Results:

• LASTGCN demonstrated superior accuracy and robustness compared to state-of-the-art methods on real-world highway datasets.
• The model showed particular strength in long-term traffic flow predictions.
• Integration of external factors like weather conditions significantly boosted predictive accuracy.

Conclusions:

• LASTGCN offers an efficient and accurate solution for traffic flow prediction in ITSs.
• The model's design is suitable for mid-term traffic management and adaptable for real-time applications.
• Dynamic integration of multifactorial data enhances the predictive capabilities of traffic management systems.