Exploiting dynamic spatio-temporal graph convolutional neural networks for citywide traffic flows prediction.

Area of Science:

• Urban Computing
• Artificial Intelligence
• Deep Learning

Background:

• Predicting crowd flows is crucial for urban planning and management.
• Accurately measuring complex spatial-temporal dependencies and external factors remains a challenge.

Purpose of the Study:

• To propose a unified dynamic deep spatio-temporal neural network model (DHSTNet) for simultaneous crowd flow prediction across city regions.
• To enhance the DHSTNet model with a Graph Convolutional Network (GCN) and Long Short-Term Memory (LSTM) for improved spatial pattern and temporal feature capture.

Main Methods:

• Developed DHSTNet with four components: recent, daily, weekly, and external branches.
• Integrated GCN with LSTM (GCN-DHSTNet) to capture spatial patterns and short-term temporal features.
• Utilized a fully connected neural network to fuse spatio-temporal features and external properties.

Main Results:

• The GCN-DHSTNet model demonstrated superior performance in predicting traffic crowd flows.
• Achieved approximately 7.9%-27.2% better RMSE and 11.2%-11.9% better MAPE compared to the AAtt-DHSTNet method.
• Outperformed other state-of-the-art methods in real-world traffic dataset evaluations.

Conclusions:

• The GCN-DHSTNet model effectively captures spatio-temporal dependencies and external influences for accurate crowd flow prediction.
• The proposed model offers enhanced generalization and scalability for urban computing applications.