Code for paper "Evaluation of Graph Neural Networks for Urban Drainage Metamodeling: Key Components and Transferability Analysis"

Repository of the code for the GNN based metamodel of SWMM. This code is linked to the paper "Evaluation of Graph Neural Networks for Urban Drainage Metamodeling: Key Components and Transferability Analysis" by Alexander Garzón, Zoran Kapelan, Jeroen Langeveld, and Riccardo Taormina.

This repository contains the improved code for developing machine learning metamodels of SWMM, and evaluating them under multiple transfer learning settings.

The two case studies are the drainage systems of Loenen and Tuindorp (a section of Utrecht). Both are located in The Netherlands.

The code is designed to work with SWMM simulations of storm water systems. The code is based on PyTorch and PyTorch Geometric.

The repository contains:
Python scripts (.py): Main code for training models, data processing, and utilities
Jupyter notebooks (.ipynb): Interactive notebooks for database creation and model development
YAML files (.yaml): Configuration files for experiments and hyperparameter sweeps
Markdown (.md): Documentation (README, REPRODUCIBILITY guide)
PNG images (.png): Figures for documentation
Supporting files: .txt (requirements), .toml (project config), .cff (citation), .lock (dependencies)

This work is supported by the TU Delft AI Labs programme.