MD-GCN: A Multi-Scale Temporal Dual Graph Convolution Network for Traffic Flow Prediction.

Area of Science:

• Traffic Engineering
• Artificial Intelligence
• Data Science

Background:

• Traffic flow prediction is crucial for urban planning and management.
• Extracting complex spatial correlations and temporal dynamics in traffic data presents significant challenges for traditional methods.
• Existing forecasting models struggle to capture intricate spatial relationships and evolving temporal patterns.

Purpose of the Study:

• To enhance the accuracy of traffic flow forecasting.
• To address the limitations of traditional methods in capturing spatial-temporal dependencies.
• To propose a novel deep learning model for improved traffic prediction.

Main Methods:

• A multi-scale temporal dual graph convolution network (MD-GCN) was developed.
• Gated temporal convolution with channel attention and inception structure was used for multi-scale temporal dependence extraction.
• A dual graph convolution module, incorporating GraphSAGE and MGCN, was designed to capture local and global spatial correlations.

Main Results:

• The proposed MD-GCN model demonstrated superior performance in traffic flow prediction tasks.
• Extensive experiments on public traffic datasets validated the effectiveness of the developed algorithm.
• The model successfully addressed the challenges of multi-scale temporal feature and spatial correlation extraction.

Conclusions:

• The MD-GCN model offers a significant advancement in spatial-temporal traffic flow prediction.
• The integration of multi-scale temporal convolution and dual graph convolution effectively captures complex traffic dynamics.
• This approach provides a more accurate and reliable solution for traffic management and planning.