nilmtk: Non-Intrusive Load Monitoring Toolkit

Non-Intrusive Load Monitoring (NILM) is the process of estimating the energy consumed by individual appliances given just a whole-house power meter reading. In other words, it produces an (estimated) itemised energy bill from just a single, whole-house power meter.

NILM is sometimes called:

• "non-intrusive appliance load monitoring (NALM or NIALM)"
• "[electriciy | energy | smart meter] disaggregation"

Below is an illustration¹ of what NILM, in general, can do.

NILMTK Documentation

http://nilmtk.github.io/nilmtk/

Academic paper on NILMTK

Batra, N., Kelly, J., Parson, O., Dutta, H., Knottenbelt, W., Rogers, A., Singh, A., Srivastava, M. (2014). NILMTK: An Open Source Toolkit for Non-intrusive Load Monitoring. In Fifth International Conference on Future Energy Systems (ACM e-Energy). Cambridge, UK. arXiv:1404.3878 DOI:10.1145/2602044.2602051

(the DOI isn't active yet but will be soon)

Bibtex:

@inproceedings{NILMTK,
   title = {{NILMTK: An Open Source Toolkit for Non-intrusive Load Monitoring}},
   year = {2014},
   author = {Batra, Nipun and Kelly, Jack and Parson, Oliver and Dutta, Haimonti and Knottenbelt, William and Rogers, Alex  and Singh, Amarjeet and Srivastava, Mani},
   booktitle = {Fifth International Conference on Future Energy Systems (ACM e-Energy)},
   address = {Cambridge, UK},
   archivePrefix = {arXiv},
   arxivId = {1404.3878},
   doi = {10.1145/2602044.2602051},
   eprint = {1404.3878}
}

Current state of the project

The project is in its early stages.

Please note that NILMTK is currently a research tool. It is not yet ready for use by end-users, although we certainly hope that NILMTK will be capable of doing 'plug and play' disaggregation in the future.

At present, NILMTK can:

• import data from the following data sets: REDD, Pecan Street, AMPds, iAWE, SMART* and UKPD
• store imported data in a common in-memory format, which can also be exported to a standard on-disk format
• compute diagnostic statistics for the raw data (uptime, dropout rate, find gaps etc)
• pre-process the data (downsample, fill gaps, mask appliance data with gaps from the mains data, select only contiguous blocks, normalise power etc)
• compute usage statistics (distribution of appliance activity per day, week or month, distribution of on-power, proportion of total energy per appliance, on-durations etc)
• provide a common input and output interface to NILM algorithms
• disaggregate data using two supervised benchmark algorithms: combinatorial optimisation and factorial hidden Markov model (these are not competitive with the current state-of-the-art. More sophisticated NILM algorithms will be added to NILMTK later this year.)
• compute a range of NILM performance metrics (confusion matrices, error in assigned energy, F1 score, fraction of energy assigned correctly etc).
• work has started on a disaggregation web interface. Here's a demo.

Installing

If you just want to use the code without modifying it then:

python setup.py install

(you may have to run as sudo)

If you want to get involved in development then:

python setup.py develop

Software Dependencies

1. Pandas
2. matplotlib
3. numpy => 1.8
4. scikit-learn > 0.13

Getting started

Loading a supported dataset is simple. For example, to load REDD:

from nilmtk.dataset import REDD
redd = REDD()
redd.load('/data/REDD/low_freq/')

Please see the examples folder for further information on getting started.

We have started writing a user guide although it is by no means complete yet. A full user guide will be written in February.

We also have API documentation. This documentation covers almost all functions in NILMTK but needs some tidying up which will be done over February.

Further info

Please see the nilmtk wiki for more details.

Development plans for April and May 2014:

• To make the code as easy as possible to use and to maintain, it will be undergoing some refactoring in April and May 2014.
• Improve the documentation
• Write more unit tests
• easy installation via pip
• Build a semantic wiki for storing information relevant to NILM (e.g. a database of appliances) - NILMTK will integrate with the wiki wherever possible, especially regarding appliance metadata.

Notes

1. The image is from the following paper and since the main author is contributing to nilmtk, so no permission issues. The reference is: Nipun Batra, Haimonti Dutta, Amarjeet Singh, “INDiC: Improved Non-Intrusive load monitoring using load Division and Calibration”, to appear at the 12th International Conference on Machine Learning and Applications (ICMLA’13) will be held in Miami, Florida, USA, December 4 – December 7, 2013 Preprint. IPython notebook