Energy Disaggregation

Jack Kelly
jack.kelly@imperial.ac.uk

Background video by Guryanov Andrey / shutterstock

Outline

  1. What is Energy Disaggregation?
  2. My PhD work
  3. Plans for a NILM competition

What is Energy Disaggregation?

Aggregate Energy Bill

Itemised Energy Bill

My PhD Work

  • First to apply deep neural nets to energy disaggregation
  • A lead developer on open source disaggregation tool NILMTK
  • Disaggregated energy dataset: UK-DALE
  • NILM Metadata:
    • Used by > 9 datasets
  • Systematic review: Does disaggregated electricity feedback reduce domestic electricity consumption?

Recurrent Neural Nets

Example Output from Deep Neural Net

Autoencoder

Metrics on Unseen Appliances

Plans for a NILM algorithm competition

Why we need a NILM competition

  • Currently impossible to compare NILM algorithms
    • Different datasets, metrics, pre-processing, appliances etc.
    • Hence no way to measure progress or select best algos
  • Competitions work for other machine learning fields

Work plan

Questions?

jack.kelly@imperial.ac.uk

Deep Neural Nets

ImageNet Large Scale Visual Recognition Challenge (ILSVRC)

From: Krizhevsky, Sutskever & Hinton. ImageNet Classification with Deep Convolutional Neural Networks. NIPS (2012)

Image from devblogs.nvidia.com

Krizhevsky et al.'s DNN Results on ImageNet 2012

Krizhevsky, Sutskever & Hinton. ImageNet Classification with Deep Convolutional Neural Networks. NIPS (2012)

Denoising Autoencoders

Bounding Rectangle

Example Output from Deep Neural Net

LSTM

Autoencoder

Rectangles