NILM Transfer Learning

Python Requirements

Create a conda environment

Create a conda environment

• conda env create -f environment.yml

Activate the environment

• conda activate nilmtk-env

Install NILMTK with the changes made

• conda install nilmtk-3.5-py_0.tar.bz2 (might need to build package if using not using liinux - the dir with everything you need is the nilmtk folder)

Install Packages (in the environment)

• Install tensorflow
  • pip3 install tensorflow==2.5.0
• Install PyWavellets
  • pip3 install PyWavelets==1.1.1

Datasets

The datasets should be placed outside the repository in a folder called datasets. The folder structure should be:

• datasets

  • ukdale
    • ukdale.h5
  • refit
    • refit.h5

• Download UKDale H5 - https://data.ukedc.rl.ac.uk/browse/edc/efficiency/residential/EnergyConsumption/Domestic/UK-DALE-2017/UK-DALE-FULL-disaggregated/ukdale.h5.zip

• Download REFIT CSV -

• Convert REFIT to H5 using the NILMTK Converter

Side Note

In the transfer learning process you need to change the fridge_frezzer name in the refit base_results to fridge.

Authors

• Rafael Teixeira - rgtzths

License

This project is licensed under the MIT License - see the LICENSE file for details

Citation

If you use this code please site our work: Teixeira, Rafael & Antunes, Mário & Gomes, Diogo. (2021). Using Deep Learning and Knowledge Transfer to Disaggregate Energy Consumption. 1-7. 10.1109/ICWAPR54887.2021.9736149.