Multi dynamic temporal representation graph convolutional network for traffic flow prediction

Area of Science:

• Intelligent Transportation Systems (ITS)
• Traffic Flow Prediction
• Graph Neural Networks

Background:

• Accurate traffic flow prediction is crucial for Intelligent Transportation Systems (ITS) to manage road congestion.
• Existing methods struggle to capture the complex spatiotemporal dynamics inherent in traffic data.
• There is a need for advanced models that can better represent these dynamic features for improved prediction accuracy.

Purpose of the Study:

• To propose a novel Multi Dynamic Temporal Representation Graph Convolutional Network (MDTRGCN) for enhanced traffic flow prediction.
• To address the limitations of current approaches in exploiting dynamic spatiotemporal traffic features.
• To provide interpretable insights into the dynamic spatial relationships within traffic networks.

Main Methods:

• Developed a dynamic graph construction method to learn time-space dependencies between road segments.
• Implemented a dynamic graph convolution module for aggregating node information via a dynamic adjacency matrix.
• Introduced a multiaspect fusion module combining traffic volume and speed features.
• Proposed a temporal representation module for inferring masked traffic data subsequences.

Main Results:

• The MDTRGCN achieved state-of-the-art performance on real-world traffic datasets.
• The model demonstrated superior accuracy in traffic flow prediction compared to existing methods.
• Experimental results validated the model's ability to capture complex spatiotemporal traffic patterns.

Conclusions:

• The proposed MDTRGCN effectively models dynamic spatiotemporal traffic features for accurate predictions.
• The method offers significant improvements in traffic flow forecasting for ITS applications.
• The model provides interpretable insights into road segment spatial relationships, aiding traffic management.