non-intrusive appliance load monitoring: promise and practice · non-intrusive appliance load...

17
© Fraunhofer USA Nonintrusive Appliance Load Monitoring (NIALM): Promise and Practice Michael Zeifman, Ph.D. and Kurt Roth, Ph.D. March 1 st , 2012 Building America Stakeholder Meeting

Upload: others

Post on 27-Jul-2020

9 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

Non‐intrusive Appliance Load Monitoring (NIALM): Promise and PracticeMichael Zeifman, Ph.D. and Kurt Roth, Ph.D.

March 1st, 2012

Building America Stakeholder Meeting

Page 2: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

What Is NIALM?

Non‐Intrusive Appliance Load Monitoring A.k.a. Non‐Intrusive Load Monitoring

Main breaker/circuit level Data acquisition (hardware) and disaggregation algorithms (software)

2

07:00 07:30 08:00 08:30 09:00 09:300

200

400

600

800

1000

1200

1400

Time

Pow

er, W

07:00 07:30 08:00 08:30 09:00 09:300

200400

Pow

er, W

Coffee Maker

07:00 07:30 08:00 08:30 09:00 09:300

5001000

Pow

er, W

Refrigerator

07:00 07:30 08:00 08:30 09:00 09:300

200400

TimeP

ower

, W

TV

Page 3: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

NIALM: Interest

Significant growth in U.S. granted patents

Both large (Belkin, GE, IBM, Intel,) and small (4home, PlotWatt, Enetics, Navetas) companies

3

1989-1995 1996-2002 2003-2006 2007-20110

0.5

1

1.5

2

2.5

3

3.5

Year

Pat

ents

/yea

r

Page 4: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

NIALM: Motivation

Energy saving potential Electricity constitutes >70% of residential primary energy consumption Whole‐house electricity consumption reduction of 5‐15% 

More specific, actionable feedback has greater savings!

Potential for further savings for fault detection and diagnostics 

Smart Grid – unique business opportunity Large smart meter deployment underway  Data acquisition specs for NIALM can be incorporated into smart meters Non‐intrusive “manual” alternative to home automation

Time‐of‐use electricity pricing – valuing Demand Response  Potential transition to time‐dependent pricing in Europe and North America

4

Page 5: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

The HEM Barrier: Complexity of HEM use and deployment

Complexity of HEM systems impedes their acceptance and use by consumers due to: Deployment complexity Complexity of use Complexity of information presented

NIALM if fully developed can potentially close this gap Deployment: May not require professional installation Complexity of use: Detect appliances and track their presence in time Complexity of information presented: Provide specific information for detected appliances (e.g., state, current 

power draw, historical data and comparisons)  Tailor energy‐saving tips to each household  Increase credibility (accuracy) of information and tips

Page 6: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

How Does NIALM Work?

6

Data Acquisition Appliance Signal Features

Disaggregation Algorithm

NIALM

Page 7: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

Data for NIALM: Sampling Rate

7

Typical daily number of switching events in a household: up to ~ 103‐104 (Baranski 2004)

0.1 Hz – minimum sampling rate

22:35 22:40 22:45 22:50 22:55 23:00 23:05500

1000

1500

Pow

er, W

0 0.05 0.1 0.15 0.2-0.1

-0.05

0

0.05

0.1

Time, s

Cur

rent

, A

Inexpensive hardware:   1 Hz

More expensive hardware: > 5 kHz

Page 8: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

Variability of signatures

-540 -520 -500 -480 -460 -440 -420 -400 -380 -3600

50

100

150

200

250

300

350

-P, W

Freq

uenc

y

Coffee maker CFL 0 2000 4000 6000 8000 10000

-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

“Signatures”

Current waveformPower

Conventional NIALM: Challenges 

No distinct signatures (overlap, missing/ corrupted data, multi‐state‐, permanent‐, and variable power devices)

Matching signatures with appliances

Low

frequency

High 

frequency

Page 9: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

Basic Method (MIT)

Commercial Product:

SPEED (Single Point End‐use Energy Disaggregation) by Enetics

Requires professional hardware installation

Capable of NIALM of large appliances

Requires training

$1,300

Hart (1992)

Page 10: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

Advanced method (Patel) Commercial Products:

In development by Belkin(since 2010)

In development by Intel (since 2009)

Major challenges to overcome:

Distinctive signatures

No‐SMPS appliances

Power‐line communication

Power draw estimation

Cost

Basic idea 1: appliance connected to a socket induces noise (electromagnetic interference) in another socket

Basic idea 2: switch mode power supplies  (SMPS) generate distinctive noises

Patel et al. (2007)

Page 11: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

Fraunhofer Approach to Low‐Frequency Data 

Principles: Appropriate to major appliances Detection and use of transient events to obtain signatures Power signature statistics (not unique) Time on/off statistics (self‐learned) in addition to power signatures Probabilistic algorithm that optimally decodes appliance states Currently applied to on‐off appliances, but can be extended

Advantages: Can work with a 3rd‐party data, potentially including  smart meter data High accuracy for overlapping appliances Robust to missing and/or corrupted data Easier appliance matching – can be based on both power and time of use

Page 12: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

Our Approach to Low‐Frequency Data: Scheme

Data are collected for ~ two weeks Negative changes of power are clustered and matched to positive changes Each cluster can correspond to appliance Cluster signature (power, time) are statistically characterized Obtained statistical distributions can be used to disaggregate new data Clusters need to be matched to real appliances

DataClustering & Distribution Estimations

Distributions of power & time features

Disaggregation algorithm

Page 13: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

Our Approach to Low‐Frequency Data: Some Results 

Real household publicly‐available data (Kolterand Johnson, 2011)

9 appliances, 3 weeks of monitoring

Overall accuracy

Our method: 0.89

Other methods : 0.71

9.95 9.952 9.954 9.956

x 105

0

100

200

300

400

Time,s

Pow

er, W

SubmeteredEstimated startEstimated end

4 4.5 5 5.5 6 6.5

x 104

0

500

1000

1500

2000

2500

Time,s

Pow

er, W

SubmeteredEstimated startEstimated end

DishwasherRefrigerator 

Accuracy in Kim et al. (2011)

Page 14: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

Our Approach to High‐Frequency Data*

Principles: Appropriate to both MELs and major appliances Detection and use of transient events Two or more “orthogonal” methods for waveform characterization Probabilistic fusion of orthogonal methods to enhance accuracy Requires ~ 50‐100 kHz sampling rate

Advantages: Enhanced classification accuracy (high true positive rate, at low false positive 

rate) Can incorporate additional waveform characterization methods to further 

enhance accuracy

*Supported in part by DOE through a subcontract from Lawrence Berkeley National Laboratory (LBNL).

Page 15: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

Our Approach to High‐Frequency Data: Scheme

Method currently requires training (waveform templates for each MEL) After transient is detected, waveform of the change is obtained Two or more “orthogonal” methods implemented to match waveform to template

We use image processing method (Fourier descriptors ≈ harmonics) and multivariate statistical technique (Partial least squares, used in spectroscopy)

Results are fused together to classify the waveform

DataTransient detection

Waveform matching 1

Waveform matching 2

Waveform

Templates

Classification

Page 16: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

Our Approach to High‐Frequency Data: Some Results 

Experiment 1: CFL lamp, CRT TV, halogen lamp, fan (no overlap in waveforms, power draw  ~ 20 – 200 W)

Experiment 2: printer, CRT monitor, CFL lamp, incandescent lamp, fan (noticeable overlap in waveforms, power draw  ~ 20 – 200 W)

Experiment 3: LCD TV, Pioneer receiver, netbook, vacuum cleaner, CFL light, pencil sharpener, hair dryer, night lamp, DVD, incandescent lamp (noticeable overlap in waveforms, power draw  ~ 2 – 2,000 W)

Page 17: Non-Intrusive Appliance Load Monitoring: Promise and Practice · Non-Intrusive Appliance Load Monitoring: Promise and Practice Author: M. Zeifman and K. Roth Subject: This presentation

© Fraunhofer USA

Conclusions

NIALM can enable significant energy savings 

NIALM deployment low‐cost IF embedded in smart meter

NIALM is challenging!

Appreciable research, one product 

Technical Challenges

Variability and overlap of “signatures”

Automatic appliance matching 

Fraunhofer NIALM 

Developed two new methods that outperform most other methods

Current status: proof of concept

Further applied research is required