Dynamic graph convolutional networks with Temporal representation learning for traffic flow prediction.

Area of Science:

• Computer Science
• Artificial Intelligence
• Transportation Engineering

Background:

• Graph convolutional networks (GCNs) are increasingly used for traffic prediction.
• Existing GCN methods struggle with limited pattern sharing, static relationships, and capturing complex traffic dynamics.

Purpose of the Study:

• To propose a novel framework, DGCN-TRL, to address limitations in current traffic flow prediction models.
• To improve the accuracy and adaptability of traffic prediction systems.

Main Methods:

• Developed a temporal graph convolution block to process dynamic time series and capture global temporal dependencies.
• Introduced a dynamic graph constructor to identify spatiotemporal correlations and temporal dependencies.
• Implemented a temporal representation learning module using a masked subsequence transformer for pre-training.

Main Results:

• The proposed DGCN-TRL framework demonstrates superior performance compared to existing methods.
• Empirical evaluations on four real-world datasets validate the model's effectiveness.
• The model successfully captures dynamic spatiotemporal relationships and complex traffic trends.

Conclusions:

• DGCN-TRL offers a significant advancement in traffic flow prediction.
• The framework's ability to learn dynamic relationships and temporal patterns leads to enhanced prediction accuracy.
• This approach provides a robust solution for real-world traffic management challenges.