Attention-Based Spatial-Temporal Convolution Gated Recurrent Unit for Traffic Flow Forecasting.

Area of Science:

• Transportation Science
• Artificial Intelligence
• Data Science

Background:

• Accurate traffic flow forecasting is vital for urban planning and traffic management.
• Complex spatial-temporal dependencies in traffic data pose significant challenges.
• Existing models often overlook long-periodic aspects, limiting forecasting accuracy.

Purpose of the Study:

• To propose a novel model, ASTCG, for enhanced traffic flow forecasting.
• To address the limitations of existing methods by incorporating long-periodic data aspects.
• To improve the accuracy of traffic flow predictions for better urban management.

Main Methods:

• Developed the Attention-Based Spatial-Temporal Convolution Gated Recurrent Unit (ASTCG) model.
• Designed a multi-input module to process near-neighbor, daily-periodic, and weekly-periodic traffic data.
• Integrated Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), and attention mechanisms within the STA-ConvGru module to capture spatial-temporal dependencies.

Main Results:

• The ASTCG model demonstrated superior performance compared to state-of-the-art methods.
• Experiments on real-world datasets validated the model's effectiveness in capturing complex traffic patterns.
• The multi-input module successfully leveraged periodic data for improved time dependence capture.

Conclusions:

• The proposed ASTCG model offers a significant advancement in traffic flow forecasting.
• ASTCG effectively addresses the limitations of previous models by considering long-periodic traffic data.
• This model provides a more accurate and reliable solution for urban traffic management and planning.