Temporal representation learning enhanced dynamic adversarial graph convolutional network for traffic flow prediction

Area of Science:

• Intelligent Transportation Systems
• Data Science
• Machine Learning

Background:

• Accurate traffic flow prediction is vital for urban traffic management and intelligent transportation.
• Existing methods struggle with complex traffic patterns and periodicity, limiting prediction precision.
• This necessitates advanced models to capture intricate spatiotemporal traffic dynamics.

Purpose of the Study:

• To develop a novel model for high-precision traffic flow prediction.
• To address limitations in capturing complex patterns and periodic characteristics of traffic flow.
• To improve urban traffic guidance and regulation through enhanced forecasting.

Main Methods:

• Proposing a temporal representation learning enhanced dynamic adversarial graph convolutional network (TRL-DAG).
• Utilizing masked reconstruction for pre-training to extract temporal representations from historical traffic data.
• Implementing a dynamic graph generation network and an adversarial graph convolutional framework for dynamic spatiotemporal correlation and loss optimization.

Main Results:

• TRL-DAG demonstrated superior performance in traffic flow prediction compared to state-of-the-art methods.
• The model effectively captures dynamic spatiotemporal correlations by integrating current and historical traffic states.
• Adversarial training reduced the trend discrepancy between predicted and actual traffic flow values.

Conclusions:

• The proposed TRL-DAG model significantly enhances traffic flow prediction accuracy.
• The integration of temporal representation learning and dynamic adversarial graph convolution is effective.
• TRL-DAG offers a promising solution for intelligent transportation management and regulation.