PINNSim: A Simulator for Power System Dynamics based on Physics-Informed Neural Networks

This repository provides the implementation to the paper PINNSim: A Simulator for Power System Dynamics based on Physics-Informed Neural Networks that was submitted to the Power System Computing Conference (PSCC) 2024.

The following provides a brief overview on how to install the pinnsim package and how to get started.

Installation

To setup the python environment, we provide a environment.yml that can be used to create the pinnsim_pscc environment

conda env create -f environment.yml
conda activate pinnsim_pscc

To then install the pinnsim package navigate in a terminal to the folder containing this repository -- your working directory should then contain the pyproject.toml file. By running the following command, pinnsim is installed and can be accessed using import pinnsim. The addition of the flag -e indicates that the repository can be modified without re-running the install command.

python -m pip install -e .

Getting started

To illustrate the main functionalities of the package, we provide a few examples in jupyter_notebooks. They contain references to central files and should help to explore the repository.

The repository has three major parts

• power_system_models and numerics describe the basic modelling of the power system and its components and how to solve the resulting differential equations.
• dataset_functions and learning_functions provide the functionality that is needed to train (learn) the neural network models and generate the require datasets.
• configurations contains a lot of the case specific functions and setups.

The "PINNSim algorithm" (Algorithm 1 in the paper) can be found in pinnsim.numerics.simulators.simulator_distributed.

General settings are stored in the __init__.py file in src.pinnsim. In particular, the WandB settings needed for logging the neural network training online require adjustments.

Reference this work

If you find this work helpful, please cite this work

@article{stiasny_pinnsim_2024,
    title = {{PINNSim}: {A} simulator for power system dynamics based on {Physics}-{Informed} {Neural} {Networks}},
    author = {Stiasny, Jochen and Zhang, Baosen and Chatzivasileiadis, Spyros},
    journal = {Electric Power Systems Research},
    volume = {235},
    pages = {110796},
    doi = {10.1016/j.epsr.2024.110796},
    month = oct,
    year = {2024},
}

License

This project is made available under the MIT License.