WiSED

WiSED is a research implementation for discovering compact, open-form partial differential equations (PDEs) from noisy spatiotemporal fields. The release contains the framework used for the manuscript, four controlled PDE benchmarks, and two engineering validations: spherical-particle ion diffusion and coupled SOC-EIS impedance-manifold dynamics.

Repository contents

• data/: benchmark-field generators. Large release data are installed from the companion dataset repository.
• engineering_validation/: runnable SIB and SOC-EIS experiments.
• models/, src/, utils/: WiSED implementation.
• scripts/install_release_data.py: installs benchmark and engineering data from the companion dataset repository.
• REPRODUCIBILITY.md: retained experimental settings and output locations.

Installation

git clone https://github.com/HUSTOROP/Hust-WiSED.git
git clone https://github.com/HUSTOROP/Hust-WiSED-Dataset.git
cd Hust-WiSED
python -m venv .venv

Activate the virtual environment, install the pinned dependencies, and install the companion data:

python -m pip install -r requirements.txt
cd ../Hust-WiSED-Dataset
git lfs install
git lfs pull
cd ../Hust-WiSED
python scripts/install_release_data.py --dataset-root ../Hust-WiSED-Dataset

On Windows PowerShell, activate the environment with .venv\\Scripts\\Activate.ps1; on macOS or Linux, use source .venv/bin/activate.

Reproducing the engineering examples

Run commands from the repository root. Use cpu when CUDA is unavailable.

# SIB: raw spherical-particle concentration field, 3% field noise
python -m engineering_validation.sib_diffusion_raw_field.run_sib_diffusion_discovery \
  --device cpu --seed 0 --noise-level 0.03 --use-cached-dataset

# SOC-EIS: B11 held-out battery experiment
python -m engineering_validation.soc_eis_impedance_manifold.run_soc_eis_b11_discovery \
  --device cpu --seed 0

Data provenance

The SOC-EIS validation uses the public dataset of Mustafa et al. from Mendeley Data (version 2, DOI: 10.17632/cb887gkmxw.2). The data article is Mustafa et al., Data in Brief 57, 110947 (2024), DOI: 10.1016/j.dib.2024.110947. The original authors and the CC BY 4.0 terms must be retained when reusing the measurements. See the companion repository for the data README and license boundaries.