Power Quality State Estimation for Distribution Grids based on Physics-Aware Neural Networks - Harmonic State Estimation [Code Repository]

Code for the concept paper presenting physics-aware neural networks for power quality state estimation.

The full paper is available here: DOI: 10.3390/en17215452

The training, test and validation data sets and model weights are available here: Zenodo

Citing

If you use any code or data provided in this repository, please use the following citation or visit the full paper link above for more citation options and provide us with a copy of your contribution. Mack P, de Koster M, Lehnen P, Waffenschmidt E, Stadler I. Power Quality State Estimation for Distribution Grids Based on Physics-Aware Neural Networks—Harmonic State Estimation. Energies. 2024; 17(21):5452. https://doi.org/10.3390/en17215452 All authors affiliated with TH Köln - Cologned University of Applied Sciences, Germany.

Replication of results

To replicate the results of the paper follow these steps:

• clone the linked repository
• Download the following files for the CIGRE low voltage distribution system:
  • y_mats_per_frequency.pic (Admittance matrices per frequency in a range from 50Hz to 1000Hz in 50Hz steps)
  • y_train.pic (Training set)
  • y_test.pic (Test set)
  • y_validation.pic (Validation set)
  • Move those files to pqse_concept_pann/data/cigrelv/
• or download the following files for the IEEE33 bus system:
  • ieee33_data.pic (Includes training, test and validation set after preprocessing)
  • Move those files to pqse_concept_pann/data/ieee33/
• Optionally download the weight files, in that case set load_weights to True in experiments.py.
  • weight files are placed in the respective grid folder (data/ieee33/MODELNAME/weights or data/cigrelv/MODELNAME/weights)
• If no pre-trained weights are used, you can train the model yourself.

The code provides methods for reading in and looking at the original data sets. Transformations such as scaling and conversion to other complex representations are done at a later stage for the CIGRE grid and already complete for the IEEE33 grid.

experiments.py provides examples of how to use the model and all experiments used for result and plot generation. You can change the grid and modify the most important training parameters on top of the file.

Harmonic injections

Harmonic injections were modeled using spectra injecting different amounts of harmonic currents at various frequencies. The harmonic injections scale based on load or generation profiles for a full year and the configured rated loads.

Training set CIGRE low voltage distribution system:

• "2015 Mercedes B-Class charged at 15.4A, 240V"
  • source: https://avt.inl.gov/sites/default/files/pdf/fsev/SteadyStateVehicleChargingFactSheet_2015MercedesBClass.pdf
  • nodes: 16, 37, 42 (0-indexed)
• "2015 Nissan Leaf charged at 16A, 208V"
  • source: https://avt.inl.gov/sites/default/files/pdf/fsev/SteadyStateLoadCharacterization2015Leaf.pdf
  • nodes: 17, 41 (0-indexed)
• "PV Inverter 33kW"
  • source: R. F. Arritt and R. C. Dugan, ‘Validation of Harmonic Models for PV Inverters: PV-MOD Milestone 2.8.2’, Electric Power Research Institute (EPRI), Washington, D.C. (United States), 2.8.2, Sep. 2022. doi: 10.2172/1894588
  • nodes: 9, 32, 35 (0-indexed)

Training set IEEE33:

• randomly sampled within 3x the allowed deviations according to EN 50160 at nodes 13, 17, 20, 23, 28 (0-indexed)

Test/Validation set CIGRE low voltage distribution system:

• "2014 BMW i3 charged at 15.5A, 240V", source: https://avt.inl.gov/sites/default/files/pdf/fsev/SteadyStateLoadCharacterization2014BMWi3.pdf
  • nodes: 16, 17, 37, 41, 42 (0-indexed)
• "PV Inverter 33kW", source: doi: 10.2172/1894588
  • nodes: 9, 32, 35 (0-indexed)

Test/Validation set IEEE33:

• "2014 BMW i3 charged at 15.5A, 240V", source: https://avt.inl.gov/sites/default/files/pdf/fsev/SteadyStateLoadCharacterization2014BMWi3.pdf
  • nodes: 13, 17, 20, 23, 28 (0-indexed)

The harmonic spectra of electric vehicles used in this work were measured in the EVs@Scale Next-Gen Profiles, US Dept. of Energy.