ST-AFN: a spatial-temporal attention based fusion network for lane-level traffic flow prediction.

Area of Science:

• Intelligent Transportation Systems
• Deep Learning for Traffic Analysis
• Urban Mobility

Background:

• Accurate traffic flow prediction is crucial for smart city applications.
• Existing methods often lack the granular precision required for refined traffic management.
• There is a need for advanced models capable of lane-level traffic forecasting.

Purpose of the Study:

• To propose a novel deep learning model for precise lane-level traffic flow prediction.
• To enhance the accuracy and reliability of traffic flow forecasting on urban ground roads.
• To address the limitations of current traffic prediction methodologies.

Main Methods:

• Development of a spatial-temporal attention-based fusion network (ST-AFN), a seq2seq model.
• Integration of attention mechanism blocks to capture dynamic lane dependencies.
• Utilizing a deep spatial-temporal information matrix and a ground lane selection method.

Main Results:

• The ST-AFN model demonstrated superior accuracy and stability compared to benchmark models.
• Experimental validation using four months of real-world traffic data from Hangzhou, China.
• Successful application of deep learning for lane-level traffic flow prediction on urban ground roads.

Conclusions:

• The proposed ST-AFN model significantly improves lane-level traffic flow prediction accuracy and reliability.
• This research pioneers the application of deep learning for traffic forecasting on urban ground roads.
• The findings offer a foundation for more effective smart city traffic management systems.