A Study of Adjacent Intersection Correlation Based on Temporal Graph Attention Network.

Area of Science:

• Intelligent Transportation Systems
• Graph Neural Networks
• Traffic Engineering

Background:

• Traffic state classification and intersection relevance calculation are critical yet challenging problems in traffic control.
• Existing methods often struggle with accuracy and handling complex traffic dynamics.

Purpose of the Study:

• To propose a novel intersection relevance model using a temporal graph attention network (TGAT).
• To simultaneously address traffic state classification and relevance calculation at intersections.
• To enhance the operational efficiency of road networks through improved traffic management.

Main Methods:

• Utilizing intersection features, interaction times, and initial traffic data labels as inputs.
• Employing the temporal graph attention (TGAT) model for classification and correlation analysis.
• Validating the model's effectiveness through VISSIM simulation experiments.

Main Results:

• The TGAT model achieved higher classification accuracy compared to three traditional models, effectively handling uneven sample distribution.
• Average delay was identified as the most influential factor on intersection status using information gain.
• The TGAT model's correlation output is interpretable, positively correlating with traffic flow.

Conclusions:

• The proposed TGAT model offers a robust solution for simultaneous traffic state classification and intersection relevance calculation.
• The model's correlation mechanism significantly improves road network operational efficiency compared to traditional approaches.
• The TGAT model proves effective and interpretable for advanced traffic control applications.