Research on Traffic Flow Prediction at Intersections Based on DT-TCN-Attention

Area of Science:

• Intelligent Transportation Systems
• Traffic Engineering
• Artificial Intelligence in Transportation

Background:

• Traditional road traffic signal systems lack efficiency, particularly during peak hours, due to their non-intelligent nature.
• Existing systems struggle to provide optimal traffic guidance, leading to congestion and delays at intersections.

Purpose of the Study:

• To develop a dynamic traffic signal phase timing optimization strategy for improved intersection efficiency.
• To leverage advanced AI techniques for accurate traffic flow prediction and signal control.

Main Methods:

• Utilized a time convolution network (TCN) to extract complex, non-linear temporal characteristics from intersection traffic data.
• Integrated an attention mechanism to enhance the prediction of traffic flow by analyzing historical time series importance and duration.
• Employed digital twinning for simulating and predicting traffic conditions under various scenarios to inform optimization.

Main Results:

• The proposed model demonstrated a strong ability to learn temporal traffic flow characteristics.
• Achieved high prediction accuracy and effective optimization of traffic signal timing.
• The digital twinning technique facilitated scenario-based optimization, improving overall traffic management.

Conclusions:

• The developed AI-driven traffic signal control strategy significantly improves intersection traffic efficiency.
• The model shows high accuracy, robust optimization capabilities, and broad applicability across different traffic scenarios.
• This approach offers a promising solution for intelligent traffic management and congestion reduction.