Traffic forecasting with graph spatial-temporal position recurrent network.

Area of Science:

• Computer Science
• Artificial Intelligence
• Transportation Engineering

Background:

• Rapid vehicle growth in smart cities presents significant traffic forecasting challenges.
• Existing graph-based methods inadequately capture spatial position and neighborhood information.
• Smart city infrastructure demands advanced traffic prediction capabilities.

Purpose of the Study:

• To propose a novel Graph Spatial-Temporal Position Recurrent Network (GSTPRN) architecture.
• To address limitations in existing traffic forecasting models regarding spatial information.
• To enhance the accuracy and efficiency of traffic prediction in smart urban environments.

Main Methods:

• Developed a position graph convolution module utilizing self-attention for spatial dependence.
• Implemented approximate personalized propagation to extend spatial information range.
• Integrated these components with adaptive graph learning into Gated Recurrent Units.

Main Results:

• The GSTPRN architecture effectively captures spatial dependence and neighborhood information.
• Experimental results on benchmark datasets show superior performance compared to state-of-the-art methods.
• Demonstrated significant improvements in traffic forecasting accuracy.

Conclusions:

• The GSTPRN model offers a robust solution for complex traffic forecasting tasks.
• Incorporating spatial position and extended neighborhood information is crucial for accuracy.
• The proposed method advances the field of intelligent transportation systems.