D-MGDCN-CLSTM: A Traffic Prediction Model Based on Multi-Graph Gated Convolution and Convolutional Long-Short-Term...

Area of Science:

• Intelligent Transportation Systems
• Data Science
• Machine Learning

Background:

• Accurate traffic forecasting is crucial for urban planning and congestion management.
• Partial data loss and the dual nature of traffic data (short-term fluctuations and long-term trends) pose significant challenges for existing models.

Purpose of the Study:

• To develop a robust traffic forecasting model capable of addressing data loss and capturing dual temporal characteristics.
• To enhance the accuracy and reliability of real-time traffic predictions.

Main Methods:

• A novel model, D-MGDCN-CLSTM, integrating Multi-Graph Gated Dilated Convolution (MGDCN) and Convolutional Long Short-Term Memory (ConvLSTM).
• Utilizes the Discrete Wavelet Transform (DWT) for multi-scale decomposition to separate short-term and long-term traffic patterns.
• Employs the DTWN algorithm for imputing missing traffic data.

Main Results:

• The D-MGDCN-CLSTM model demonstrated superior performance compared to 10 other algorithms on the PeMSD7(M) and PeMSD7(L) datasets.
• Achieved average improvements in Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Accuracy (ACC).
• Ablation studies and parameter analysis confirmed the effectiveness of the proposed decompositional approach.

Conclusions:

• The D-MGDCN-CLSTM model effectively handles missing data and captures dual temporal characteristics in traffic forecasting.
• The proposed method offers a significant advancement in real-time traffic prediction accuracy and reliability.