An adaptive spatiotemporal dynamic graph convolutional network for traffic prediction

Area of Science:

• Intelligent Transportation Systems
• Data Science
• Network Analysis

Background:

• Traffic prediction is crucial for urban planning and intelligent transportation systems.
• Existing methods struggle with complex spatiotemporal dynamics and fail to capture intrinsic feature couplings.
• Predefined static adjacency matrices and separate feature processing limit accuracy in current traffic prediction models.

Purpose of the Study:

• To propose an adaptive spatiotemporal dynamic graph convolutional network (AST-DGCN) for enhanced traffic prediction.
• To address the limitations of existing methods in capturing dynamic spatiotemporal patterns and feature interdependencies.
• To improve the accuracy and robustness of traffic forecasting.

Main Methods:

• An encoder-decoder architecture is employed, utilizing node embedding for high-dimensional feature extraction.
• Time-evolving adaptive graphs are generated using self-attention mechanisms.
• Dynamic graphs are integrated with gated recurrent units for joint spatiotemporal dependency modeling, incorporating a dual-layer residual correction module.

Main Results:

• The AST-DGCN model demonstrated significant performance advantages over baseline methods on four public traffic datasets.
• The model achieved superior results across key evaluation metrics: root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE).
• Experimental validation confirmed the model's enhanced predictive capabilities and competitive advantages in traffic forecasting.

Conclusions:

• The proposed AST-DGCN effectively models complex spatiotemporal dependencies in traffic networks.
• The adaptive graph generation and residual correction modules significantly enhance prediction accuracy.
• AST-DGCN offers a superior approach for intelligent transportation systems, improving dynamic road network optimization and urban travel planning.