The repository at hand contains the python code relative to our BuildSys20' paper
Hafsa Bousbiat, Christoph Klemenjak, and Wilfried Elmenreich. 2020. Exploring Time Series Imaging for Load Disaggregation. In The 7th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation (BuildSys ’20), November 18–20, 2020, Virtual Event, Japan.
Many wide-spread load disaggregation techniques perform a sequence-to-sequence mapping between an input sequence (aggregate power readings) and an output sequence that consists of power readings associated with a particular electrical appliance. We propose to augment this pipeline with a 1D-2D transform: a time series imaging block, as illustrated by the figure below.
The proposed approach, which we refer to as Image-to-Sequence (I2S), was designed to be fully compatible to NILMTK , a NILM toolkit for reproducible experiments. I2S incorporates implementations of GAF, MTF, and RP provided by the PyTS package. We provide below three sample images created from 150 minutes of smart meter data to get a sense of how images obtained from GASF, MTF and RP differ. All three transforms generate symmetric images where the value of a pixel gives insights on the similarity between two instants.
@inproceedings{bousbiat2020imaging,
title={Exploring Time Series Imaging for Load Disaggregation},
author={Bousbiat, Hafsa and Klemenjak, Christoph and Elmenreich, Wilfried},
booktitle={Proceedings of the 7th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation},
year={2020}
}