Upgrade to Pro — share decks privately, control downloads, hide ads and more …

NILM Workshop 2017 Vendor Talk

9ded99d0d611c62c53b00aaaf4b01072?s=47 hassaku
December 21, 2017

NILM Workshop 2017 Vendor Talk

7th November 2017 @ London



December 21, 2017

More Decks by hassaku

Other Decks in Research


  1. NILM  EU  Workshop  5  minutes  presentation UK  /  JAPAN Research

     Engineer Takashi  Hasuo
  2. System Automatic  Labeling   without  user  intervention

  3. Over  a  thousand  sensors  are  installed  in  partnerships  with  

    companies  including  the  largest  Electricity  Utility  Company  in  Japan. Home  Snapshot Display  items   showing  a   breakdown  of  your   home  energy  costs   Display  items   showing  the   recent  status  of   your  home AC  was  on   for  2  hours   since  9  pm Washer   turned  off   at  7  p.m. Microwave   was  used   twice  at  12   a.m. Power   Usage   Analytics Elderly  /  Relative  /    Patient Family  /  Carer Assisted  Living Services (Japan)
  4. Technical  Advisers Challenges Feature  space No  matter  how  labeled  data

     are  gathered,  it  is  impossible  to  satisfy   requested  high  accuracy  (e.g.  F  score  >  0.9)  of  real  services. Exploration  of  machine  learning  techniques  that  effectively  utilize   unlabelled  data  including  target  householdsʼ’  is  important. We  should  assume  that  the  bias  of  NILM  data  is  large. All  households Target   households Labeled     households (high  cost) Ground  truth  sensor Labeled  households