PDG2Seq: Periodic Dynamic Graph to Sequence Model for Traffic Flow Prediction.

Area of Science:

• Intelligent transportation systems
• Data science
• Machine learning

Background:

• Current traffic flow prediction models often overlook dynamic correlations and periodic features in traffic data.
• Existing methods struggle to accurately capture future traffic trend changes due to reliance on static historical data.

Purpose of the Study:

• To propose a novel Periodic Dynamic Graph to Sequence Model (PDG2Seq) for improving traffic flow prediction accuracy.
• To address the limitations of current models by incorporating dynamic correlations and periodic traffic data features.

Main Methods:

• Developed a Periodic Dynamic Graph to Sequence Model (PDG2Seq) comprising a Periodic Feature Selection Module (PFSM) and a Periodic Dynamic Graph Convolutional Gated Recurrent Unit (PDCGRU).
• PFSM extracts periodic features, while PDCGRU utilizes these and dynamic traffic features to generate a Periodic Dynamic Graph for spatio-temporal feature extraction.
• The decoding phase employs periodic features for predicting future traffic trends.

Main Results:

• PDG2Seq demonstrated superior performance over state-of-the-art baselines in traffic flow prediction.
• The model effectively extracts spatio-temporal features from dynamic, real-time traffic data.
• Experiments on four large-scale datasets validated the model's accuracy and effectiveness.

Conclusions:

• The proposed PDG2Seq model significantly enhances traffic flow prediction accuracy by integrating periodic and dynamic features.
• This approach offers a more robust solution for intelligent traffic management systems.
• The model's ability to capture future trends marks a significant advancement in the field.