Graph Construction for Electrical Power Transmissions Networks (GC4EPTN)

Source repo for Data-Driven Graph Construction of Power Flow Graphs for Electric Power Transmission Networks.

For the online appendix containing additional results, see the online-appendix.pdf file.

Environment

To create a a conda environment, follow the below instructions.

1. To install the conda environment simply run the below command from the root of the repository:

   conda env create --file env.yaml
   

2. Once complete you can activate the conda environment using conda activate gc4etpn and you can deactivate using conda deactivate

3. Next, you will need to install the gc4eptn repo. First make sure to have activated the gc4eptn conda environment. Then, run the following command from the root of the repository:

   pip install -e .
   

4. Finally, to run the GGM algorithms the required R (e.g., GGMncv) packages will need to be installed, do so by running the following command. You should be prompt to select and install server before the install begins.

   python scripts/install_r.py
   

Structure

Below will be a brief description of how this repository is structured.

• notebooks/: Jupyter notebooks for running code and replicating paper results.
• exps/: Storage location for all results produced by notebooks.
• gc4eptn/: This contains all general code for the GGM algorithms, running experiments, plotting, metrics, utilities, etc.
  • gc4eptn.dataloaders: Contains classes for loading the real-time data simulation (RTDS) and MATPOWER data.
  • gc4eptn.ggm: Contains code for running GGM algorithms and their experiments.
  • gc4eptn.pngs: Contains code for running the power network graph score (PNGS) algorithm and experiments for running it with GGMs.
  • gc4eptn.gsp: Contains code for running a basic GPS for graph construction algorithm.
  • gc4eptn.utils: Contains utility code from plotting, to metrics, to normalization, and more.
• datasets/: Contains all the datasets that can be loaded using gc4eptn. Currently only supports the RTDS Kundur’s two-area, four-machine system (as in paper) and two MATPOWER systems with no current data (case9 and case14).
• scripts/: Scripts for managing experiments, installing R packages, and running MATPOWER.

Running the Code

All code execution is done through Jupyter Notebooks. See the notebooks/ directory for the set of notebooks which run various aspects of the code. A brief description of each notebook is given below.

• Primary Notebooks for Replicating Paper Results
  • data-loading-rtds-v5: Display examples of how to load the RTDS data.
  • data-loading-matpower: Displays examples of how to load the MATPOWER data.
  • kernel-analysis: Runs experiments that visualizes various different kernels. Useful to validate kernels before being used for GGM algorithms.
  • ggm: Runs experiments for network graph prediction using various GGM algorithms. Useful for validating just the network graph estimation part before refinement with PNGS.
  • pngs-single-exp: Runs a single experiment for the PNGS algorithm from network estimation using GGMs to graph refinement for flow graph estimation.
  • pngs-multi-exp: Runs a multiple experiments using various different parameters for the PNGS algorithm. Automated version of running different variations of experiments in pngs-single-exp.ipynb.
• Other Secondary Notebooks
  • gsp: Runs experiments for network graph prediction using a GSP algorithm.
  • gsp-syn-test: Tests the GSP algorithm using synthetic data.
  • norm-tests: Tests various normalization effects on RTDS and MATPOWER data. Specifically used to observer effect of feature-wise normalization.
  • pngs-synthetic-test: Test PNGS algorithm assuming fully connected network graph estimation. Allows for testing graph refinement into flow graph when using worst case scenario for network estimation.

Citation

@INPROCEEDINGS{10903228,
  author={Poole, Benjamin and Ratnakumar, Rajan and Madurasinghe, Dulip Tharaka and Kümmerle, Christian and Venayagamoorthy, Ganesh Kumar and Lee, Minwoo},
  booktitle={2024 International Conference on Machine Learning and Applications (ICMLA)}, 
  title={Data-Driven Graph Construction of Power Flow Graphs for Electric Power Transmission Networks}, 
  year={2024},
  pages={346-353},
  doi={10.1109/ICMLA61862.2024.00053}}