STAC-Net: A hierarchical framework for modeling and predicting urban traffic flow with uncertainty quantification

Area of Science:

• Transportation Science
• Artificial Intelligence
• Data Science

Background:

• Urban traffic flow prediction is challenged by complex spatiotemporal dependencies and external factors.
• Existing models struggle to accurately quantify traffic flow uncertainty.

Purpose of the Study:

• To propose STAC-Net, a novel framework for urban traffic flow uncertainty modeling and prediction.
• To enhance the accuracy and robustness of traffic flow predictions by capturing intricate spatiotemporal dynamics and uncertainty.

Main Methods:

• Utilizes spatiotemporal graph convolution to model traffic flow features.
• Employs Convolutional Gated Recurrent Units (ConvGRU) for long-term temporal dependencies.
• Incorporates hierarchical self-attention and Neural Processes (NP) for multi-scale feature extraction and uncertainty quantification.

Main Results:

• STAC-Net outperforms baseline methods on METR-LA, PeMS04, and PeMS08 datasets.
• Achieved a 10.5% reduction in Mean Absolute Error (MAE).
• Demonstrated a 12.3% improvement in Root Mean Squared Error (RMSE).

Conclusions:

• The proposed STAC-Net offers a reliable and efficient framework for urban traffic flow prediction.
• Effectively addresses uncertainty in real-world traffic scenarios.
• Provides enhanced decision support for traffic management departments.