Uncertainty-aware Network-level traffic speed, flow, and demand prediction

This is the source code of the uncertainty-aware Network-level traffic speed, flow, and demand prediction model. This model extends the Dynamic Graph Convolution (DGC) module proposed initially by Li et al. (2021).

• Visit the DiTTlab demo page below to visualize how the model works and how the predictions are interpreted (may not working before 27-08-2023):

Dittlab Online Demo (click)

• The paper manuscript is still under review.

Requirement

• Python = 3.9
• PyTorch ≥ 2.0.1
• shapely = 1.8.5
• zarr = 2.16.0

Data Preparation

• The used Dutch highway dataset is provided by National Data Warehouse NDW.
• Data examples can be obtained from the DittLab application page: tools-dittlab.
• Run python get_data.py to get the speed and flow data from the NDW server.
• Processed data will be in the datasets folder.
• If there is any difficulty in preparing the dataset, please send us an email: [email protected]. We will share the fully-processed data that is ready to use.

Model Training

• Run the TrainingModels.ipynb to train the model.
• Detailed instructions are provided in the notebook.