Road traffic state prediction based on a graph embedding recurrent neural network under the SCATS

Area of Science:

• Intelligent Transportation Systems
• Traffic Flow Dynamics
• Machine Learning for Transportation

Background:

• Accurate road traffic state prediction is crucial for intelligent transportation systems.
• Complex spatiotemporal dependencies in traffic flow data hinder prediction accuracy, particularly in systems like Sydney's coordinated adaptive traffic system.
• Existing methods struggle to capture the intricate dynamics of traffic networks.

Purpose of the Study:

• To propose a novel framework for enhancing road traffic state prediction accuracy.
• To address the challenges posed by complex spatiotemporal characteristics in traffic flow data.
• To introduce a graph-based approach for traffic flow prediction.

Main Methods:

• Representing the traffic road network as a graph structure.
• Developing a Graph Embedding Recurrent Neural Network (GERNN) framework.
• Utilizing a real-world dataset for numerical testing and comparison.

Main Results:

• The proposed GERNN framework demonstrates improved performance in traffic state prediction.
• GERNN effectively handles the complex spatiotemporal characteristics of traffic flow data.
• Numerical tests show GERNN outperforming existing prediction methods.

Conclusions:

• The Graph Embedding Recurrent Neural Network (GERNN) is a promising approach for accurate road traffic state prediction.
• GERNN offers a viable solution for intelligent transportation systems facing complex traffic dynamics.
• The framework's effectiveness is validated through real-world data analysis.