level energy consumption
in residential buildings
on whole house power measurements
Suman Giri, Mario Bergés
Department of Civil and Environmental Engineering,
Carnegie Mellon University, Pittsburgh PA 15217
One way to promote energy efficient behavior
is to provide
them feedback in the form of
level breakdown of
total energy c
onsumption, which is achievable through methods now
known as non
intrusive load monitoring (NILM)
A typical NILM system collects overall voltage
and/or current measurements for a home at a limited number of locations in the power distribution system,
and uses machine learning and signal processing techniques to infer the behavior of individual appliances
from these aggregate measurements, and provides an estimate of their energy consumption.
level energy feedback
esult in savings of up to 12% [2
In the past 20 years considerable advances have been made in the NILM domain
(a good review can be
but a complete and feasible solution
eludes the mainstream market.
has been towards using algorithms that learn based on unsupervised methods [
The use of algorithms like Hidden Markov Models (HMMs) and its derivatives including efficient
inference algorithms have been proposed, but a thorough test of the
efficacy of these methods on real life
data has yet to be carried out.
Although there exist a limited number of publicly available datasets of voltage and current measurements
for homes (e.g., REDD [
, the lack of labels for events
(i.e., appliance st
make it difficult
for researchers to test their event
based algorithms on them, and these datasets sometimes do not include
level energy breakdown resulting in ambiguities
y, another dataset
BLUED has been released [
] which provides detailed labels of
events including individual power consumptions of all monitored devices. In this paper, we test
approaches based on graphical models [
4 & 5
the BLUED dataset to
observe the actual level of
accuracy and device granularity achievable through these means.
n doing so, we aim to make a conclusive
statement about the practicality of using probability based methods as opposed to event
based methods as a
solution for NI
 G. Hart, “Nonintrusive appliance load monitoring,”
Proceedings of the IEEE, vol. 80, no. 12, pp. 1870
 K. Ehrhardt
Martinez, K. Donnelly, J. A. “Skip” Laitner, D. York, J. Talbot, and K. Friedrich,
“Advanced Metering Initiatives and Residential Feedback Programs: A Meta
Review for Economy
Electricity Savings.” American Council for an Energy
Efficient Economy, Washington, D.C., E105, Jun.
 M. Zeifman and K. Roth, “Nonintrusive appliance
load monitoring: Review and outlook,” in 2011 IEEE
International Conference on Consumer Electronics (ICCE), 2011, pp. 239
H. Kim, M. Marwah, M. Arlitt, G. Lyon, and J. Han. “Unsupervised disaggregation of low frequency
power measurements”. In
Proceedings of the SIAM Conference on Data Mining, 2011.
 J. Z. Kolter and T. Jaakkola
. Approximate Inference in Additive Factorial HMMs with Application to
Energy Disaggregation. In International Conference on Artificial Intelligence and Statistics, pages 1472
1482, LaPalma, Canary Islands, 2012.
 Parson O, Ghosh S, Weal M, Rogers A. N
intrusive Load Monitoring using Prior Models of General
Appliance Types. In: 26th AAAI Conference on Artificial Intelligence. Toronto, Canada. 2012.
 Zico Kolter and Matthew J. Johnson. REDD: A public data set for energy disaggregation research. In
proceedings of the SustKDD workshop on Data Mining Applications in Sustainability, 2011.
 K. Anderson, A. Ocneanu, D. Benitez, D. Carlson, A. Rowe, and M. Berges, "BLUED: A Fully Labeled
Public Dataset for Event
Intrusive Load Monitoring Rese
arch," in Proceedings of the 2nd KDD
Workshop on Data Mining Applications in Sustainability (SustKDD), Beijing, China.