Multicomponent Spatial-Temporal Graph Attention Convolution Networks for Traffic Prediction with Spatially Sparse...

Area of Science:

• Artificial Intelligence
• Transportation Engineering
• Data Science

Background:

• Traffic prediction is crucial for transportation management but is challenging due to complex spatial-temporal dependencies.
• Existing methods struggle with spatially sparse traffic data, failing to capture sufficient spatial correlation and temporal periodicity.

Purpose of the Study:

• To propose a novel deep learning framework, MSTGACN, for effective traffic prediction with spatially sparse data.
• To address the limitations of current methods in handling sparse traffic information.

Main Methods:

• Developed Multi-component Spatial-Temporal Graph Attention Convolutional Networks (MSTGACN).
• MSTGACN integrates three components to model periodic information, utilizing dilated causal convolution, graph convolution, and graph attention layers.
• Applied the framework to predict traffic flow and speed using real-world sparse traffic datasets.

Main Results:

• MSTGACN demonstrates superior performance in traffic prediction with spatially sparse data.
• The model successfully captures spatial correlation and temporal periodicity even with limited data points.
• Experimental validation on METR-LA, PeMS-BAY, and PeMSD7-sparse datasets confirms the method's effectiveness.

Conclusions:

• The proposed MSTGACN framework offers a significant advancement in traffic prediction for sparse data scenarios.
• This approach enhances the reliability of traffic flow and speed forecasting in challenging data conditions.
• MSTGACN provides a robust solution for intelligent transportation systems dealing with incomplete traffic data.