A combined model for short-term traffic flow prediction based on variational modal decomposition and deep learning

Area of Science:

• Artificial Intelligence
• Transportation Engineering
• Data Science

Background:

• Deep Learning models face challenges in traffic flow prediction due to inherent nonlinearity, instability, and volatility.
• Accurate short-term traffic flow prediction is crucial for intelligent transportation systems.

Purpose of the Study:

• To propose a novel combined prediction model, VMD-GAT-MGTCN, to enhance short-term traffic flow prediction accuracy.
• To address the challenges posed by traffic flow uncertainty and volatility.

Main Methods:

• Variational Modal Decomposition (VMD) is used to decompose traffic flow data into modal components.
• A spatio-temporal feature model integrating Graph Attention Network (GAT) and Multi-Gated Attention Time Convolutional Network (MGTCN) is developed.
• Predicted modal components are stacked to generate final traffic flow predictions.

Main Results:

• The proposed VMD-GAT-MGTCN model demonstrates superior prediction accuracy and effectiveness compared to baseline and other models.
• Variational Modal Decomposition significantly enhances the prediction performance of the VMD-GAT-MGTCN model.
• The model achieves good prediction results, particularly in traffic flow mutation regions.

Conclusions:

• The VMD-GAT-MGTCN model offers a robust solution for accurate short-term traffic flow prediction.
• The integration of VMD, GAT, and MGTCN effectively captures spatio-temporal features and handles traffic flow dynamics.
• This approach shows significant potential for improving intelligent transportation systems.