Non-instrusive Load Monitoring with Fully Convolution networks

Code for the method in this paper. Dataset: https://datashare.is.ed.ac.uk/handle/10283/3647

Running the code

Python 3 is used. First set the environment variable IDEAL_DATA_DIR to the directory where you wish to store intermediate data, models, and predictions.

Next, run the script csv_to_hdf5_converter.py, with an argument giving the path to the IDEAL dataset. This will parse the data from the CSV files are store in HDF5 format.

python csv_to_hdf5_converter.py --dataset_path <path to dataset>

Next, preprocess the data, filling short gaps and merging sensors:

python generate_cleaned_nilm_data.py

Generate windows of data to use for training and testing:

python generate_s2s_dataset.py

Train and predict with the Fully Convolutional Network:

python fully_conv_separate_valid.py

Train and predict with the Sequence-to-Point baseline:

python pointnet_large_separate_valid.py

Calculate results with the Jupyter Notebook 'fcn-evaluation.ipynb`