Buildings & Energy ! Buildings are major energy consumers 76% of - - PowerPoint PPT Presentation

▶
buildings energy
SMART_READER_LITE
LIVE PREVIEW

Buildings & Energy ! Buildings are major energy consumers 76% of - - PowerPoint PPT Presentation

Feb 07, 2024 558 likes •1.05k views

E MPIRICAL C HARACTERIZATION AND M ODELING OF E LECTRICAL L OADS IN S MART H OMES Sean Barker , Sandeep Kalra, David Irwin, and Prashant Shenoy University of Massachusetts Amherst Department of Computer Science Buildings & Energy ! Buildings

slide-1
SLIDE 1

Department of Computer Science

EMPIRICAL CHARACTERIZATION AND MODELING OF ELECTRICAL LOADS

IN SMART HOMES

Sean Barker, Sandeep Kalra, David Irwin, and Prashant Shenoy

University of Massachusetts Amherst

slide-2
SLIDE 2

Sean Barker (sbarker@cs.umass.edu)

Buildings & Energy

! Buildings are major energy consumers

  • 76% of US electricity, 48% of energy

! Potential for “smart” buildings

  • Reduce energy usage, increase efficiency
  • Demand-side energy management (DSEM)

2

slide-3
SLIDE 3

Sean Barker (sbarker@cs.umass.edu)

Smart Buildings

! Need energy data for effective DSEM ! Lots of data from smart meters

  • Ongoing deployments by utilities

! Use data to optimize energy use

  • E.g., peak load reduction

3

Peak Usage Off-Peak

Shiftable Load

slide-4
SLIDE 4

Sean Barker (sbarker@cs.umass.edu)

Optimization Challenges

! Optimization requires understanding energy use ! Sense-analyze-control for smart buildings

  • Smart meters provide sensing
  • Modeling key to analysis and control

! Model and predict building energy usage

  • Both aggregate and individual loads

4

! Build models of electrical loads in homes

slide-5
SLIDE 5

! Model loads as ‘on-off’ devices

  • Load is either on (with fixed power) or off
  • E.g., a light bulb

! Simple extension: multiple discrete ‘on’ states

  • e.g., [REDD]

Sean Barker (sbarker@cs.umass.edu)

Prior Work: Simple Models

5

50 100 150 200 250 1 2 3 4

Power (W) Time (min)

light

slide-6
SLIDE 6

Sean Barker (sbarker@cs.umass.edu)

Modeling Challenges

! Problem: Modern devices exhibit complex usage ! Not easily described by ‘on-off’ models ! Need better models to capture these behaviors

6

! Goal: Better load models for complex devices

washing machine

slide-7
SLIDE 7

Sean Barker (sbarker@cs.umass.edu)

Outline

! Motivation ! Features of Electrical Loads ! Modeling Household Devices ! Applications of Models ! Conclusions

7

slide-8
SLIDE 8

Sean Barker (sbarker@cs.umass.edu)

Features of Electrical Loads

! Approach: empirically characterize individual loads to distill common characteristics ! Data: 100+ devices, 2+ years of data

  • 1 sec data resolution

! Divide loads into classes based on their electrical usage properties

8

slide-9
SLIDE 9

Sean Barker (sbarker@cs.umass.edu)

Resistive Loads

! Voltage and current waveforms aligned ! Devices with heating elements

  • Lights
  • Toaster
  • Coffeepot
  • Oven
  • Space heater

9

slide-10
SLIDE 10

Sean Barker (sbarker@cs.umass.edu)

Resistive Load Features

10

50 100 150 200 250 1 2 3 4

Power (W) Time (min)

light

200 400 600 800 1000 1200 1400 1600 1 2 3 4

Power (W) Time (min)

toaster

200 400 600 800 1000

1 2 3 4 5 6 7 8 9 10

Power (W) Time (min)

coffee maker

light coffee maker toaster

slide-11
SLIDE 11

Sean Barker (sbarker@cs.umass.edu)

Resistive Load Features

10

50 100 150 200 250 1 2 3 4

Power (W) Time (min)

light

200 400 600 800 1000 1200 1400 1600 1 2 3 4

Power (W) Time (min)

toaster

200 400 600 800 1000

1 2 3 4 5 6 7 8 9 10

Power (W) Time (min)

coffee maker

light coffee maker toaster

slide-12
SLIDE 12

Sean Barker (sbarker@cs.umass.edu)

Resistive Load Features

10

50 100 150 200 250 1 2 3 4

Power (W) Time (min)

light

200 400 600 800 1000 1200 1400 1600 1 2 3 4

Power (W) Time (min)

toaster

200 400 600 800 1000

1 2 3 4 5 6 7 8 9 10

Power (W) Time (min)

coffee maker

880 900 920 940 960 980 1000

coffee maker zoom

light coffee maker toaster

slide-13
SLIDE 13

Sean Barker (sbarker@cs.umass.edu)

Resistive Load Features

10

! On-off (small loads), on-off with decay (large loads)

50 100 150 200 250 1 2 3 4

Power (W) Time (min)

light

200 400 600 800 1000 1200 1400 1600 1 2 3 4

Power (W) Time (min)

toaster

200 400 600 800 1000

1 2 3 4 5 6 7 8 9 10

Power (W) Time (min)

coffee maker

1420 1430 1440 1450 1460 1470

toaster zoom

light coffee maker toaster

slide-14
SLIDE 14

! Current waveform lags voltage ! Devices with AC motors

  • Refrigerator/freezer compressor
  • Air conditioner
  • Vacuum

Sean Barker (sbarker@cs.umass.edu)

Inductive Loads

11

slide-15
SLIDE 15

Sean Barker (sbarker@cs.umass.edu)

Inductive Load Features

12

200 400 600 800 1000 1200 1400 1600

10 20 30 40 50 60

Power (W) Time (sec)

vacuum cleaner

500 1000 1500 2000 2500

10 20 30 40 50 60

Power (W) Time (min)

Central A/C

50 100 150 200 250 300 350 400 450

5 10 15 20 25 30 35 40 45

Power (W) Time (min)

freezer

100 200 300 400 500 600 700 800

20 40 60 80 100 120

Power (W) Time (min)

refrigerator

vacuum A/C fridge freezer

slide-16
SLIDE 16

Sean Barker (sbarker@cs.umass.edu)

Inductive Load Features

12

200 400 600 800 1000 1200 1400 1600

10 20 30 40 50 60

Power (W) Time (sec)

vacuum cleaner

500 1000 1500 2000 2500

10 20 30 40 50 60

Power (W) Time (min)

Central A/C

50 100 150 200 250 300 350 400 450

5 10 15 20 25 30 35 40 45

Power (W) Time (min)

freezer

100 200 300 400 500 600 700 800

20 40 60 80 100 120

Power (W) Time (min)

refrigerator

vacuum A/C fridge freezer

slide-17
SLIDE 17

Sean Barker (sbarker@cs.umass.edu)

Inductive Load Features

12

200 400 600 800 1000 1200 1400 1600

10 20 30 40 50 60

Power (W) Time (sec)

vacuum cleaner

500 1000 1500 2000 2500

10 20 30 40 50 60

Power (W) Time (min)

Central A/C

50 100 150 200 250 300 350 400 450

5 10 15 20 25 30 35 40 45

Power (W) Time (min)

freezer

100 200 300 400 500 600 700 800

20 40 60 80 100 120

Power (W) Time (min)

refrigerator

vacuum A/C fridge freezer

slide-18
SLIDE 18

Sean Barker (sbarker@cs.umass.edu)

Inductive Load Features

12

200 400 600 800 1000 1200 1400 1600

10 20 30 40 50 60

Power (W) Time (sec)

vacuum cleaner

500 1000 1500 2000 2500

10 20 30 40 50 60

Power (W) Time (min)

Central A/C

50 100 150 200 250 300 350 400 450

5 10 15 20 25 30 35 40 45

Power (W) Time (min)

freezer

100 200 300 400 500 600 700 800

20 40 60 80 100 120

Power (W) Time (min)

refrigerator

vacuum A/C fridge freezer

slide-19
SLIDE 19

Sean Barker (sbarker@cs.umass.edu)

Inductive Load Features

12

! AC motor surge current, then stabilization

200 400 600 800 1000 1200 1400 1600

10 20 30 40 50 60

Power (W) Time (sec)

vacuum cleaner

500 1000 1500 2000 2500

10 20 30 40 50 60

Power (W) Time (min)

Central A/C

50 100 150 200 250 300 350 400 450

5 10 15 20 25 30 35 40 45

Power (W) Time (min)

freezer

100 200 300 400 500 600 700 800

20 40 60 80 100 120

Power (W) Time (min)

refrigerator

vacuum A/C fridge freezer

slide-20
SLIDE 20

! Non-sinusoidal current draw ! Electronic devices

  • Switch-mode power supplies
  • Computers
  • Televisions

! Fluorescent lights ! Battery chargers

Sean Barker (sbarker@cs.umass.edu)

Non-Linear Loads

13

slide-21
SLIDE 21

20 40 60 80 100 120 140 160 180

1 2 3 4 5

Power (W) Time (min)

LCD TV

Sean Barker (sbarker@cs.umass.edu)

Non-Linear Load Features

14

5 10 15 20 25 30 35 40 45

5 10 15 20

Power (W) Time (min)

Mac Mini

200 400 600 800 1000

5 10 15 20

Power (W) Time (min)

HRV

200 400 600 800 1000 1200 1400 1600

1 2 3 4 5

Power (W) Time (min)

microwave

LCD TV Mac Mini

heat recovery ventilator (HRV)

microwave

slide-22
SLIDE 22

20 40 60 80 100 120 140 160 180

1 2 3 4 5

Power (W) Time (min)

LCD TV

Sean Barker (sbarker@cs.umass.edu)

Non-Linear Load Features

14

5 10 15 20 25 30 35 40 45

5 10 15 20

Power (W) Time (min)

Mac Mini

200 400 600 800 1000

5 10 15 20

Power (W) Time (min)

HRV

200 400 600 800 1000 1200 1400 1600

1 2 3 4 5

Power (W) Time (min)

microwave

LCD TV Mac Mini

heat recovery ventilator (HRV)

microwave

slide-23
SLIDE 23

20 40 60 80 100 120 140 160 180

1 2 3 4 5

Power (W) Time (min)

LCD TV

Sean Barker (sbarker@cs.umass.edu)

Non-Linear Load Features

14

5 10 15 20 25 30 35 40 45

5 10 15 20

Power (W) Time (min)

Mac Mini

200 400 600 800 1000

5 10 15 20

Power (W) Time (min)

HRV

200 400 600 800 1000 1200 1400 1600

1 2 3 4 5

Power (W) Time (min)

microwave

LCD TV Mac Mini

heat recovery ventilator (HRV)

microwave

slide-24
SLIDE 24

20 40 60 80 100 120 140 160 180

1 2 3 4 5

Power (W) Time (min)

LCD TV

Sean Barker (sbarker@cs.umass.edu)

Non-Linear Load Features

14

5 10 15 20 25 30 35 40 45

5 10 15 20

Power (W) Time (min)

Mac Mini

200 400 600 800 1000

5 10 15 20

Power (W) Time (min)

HRV

200 400 600 800 1000 1200 1400 1600

1 2 3 4 5

Power (W) Time (min)

microwave

LCD TV Mac Mini

heat recovery ventilator (HRV)

microwave

slide-25
SLIDE 25

20 40 60 80 100 120 140 160 180

1 2 3 4 5

Power (W) Time (min)

LCD TV

Sean Barker (sbarker@cs.umass.edu)

Non-Linear Load Features

14

! Unpredictable variations (often from a stable level)

5 10 15 20 25 30 35 40 45

5 10 15 20

Power (W) Time (min)

Mac Mini

200 400 600 800 1000

5 10 15 20

Power (W) Time (min)

HRV

200 400 600 800 1000 1200 1400 1600

1 2 3 4 5

Power (W) Time (min)

microwave

1380 1400 1420 1440 1460 1480 1500

fluctuation zoom

LCD TV Mac Mini

heat recovery ventilator (HRV)

slide-26
SLIDE 26

Sean Barker (sbarker@cs.umass.edu)

Outline

! Motivation ! Features of Electrical Loads ! Modeling Household Devices ! Applications of Models ! Conclusions

15

slide-27
SLIDE 27

poff = 5W pactive = 240W

Sean Barker (sbarker@cs.umass.edu)

On-Off Model

! Simplest two-state model

  • Active/inactive
  • Static power levels

! Example: non-dimmable light

16 50 100 150 200 250 1 2 3 4

Power (W) Time (min)

light

slide-28
SLIDE 28

! Model continuous changes in inductive/resistive loads ! Exponential decay or logarithmic growth ! Example: fitting a coffee maker (decay)

(pactive, ppeak, λ) = (905, 990, 0.045)

Sean Barker (sbarker@cs.umass.edu)

On-Off Growth/Decay Model

17

p(t) = pactive + (ppeak − pactive)e−λt

880 900 920 940 960 980 1000 1 2 3 4 5 6 7 8 9

Power (W) Time (min)

coffee model (peak) coffee data (peak) 200 400 600 800 1000 1 2 3 4 5 6 7 8 9

Power (W) Time (min)

coffee maker model coffee maker data

slide-29
SLIDE 29

! Deviations from a stable min or max power ! Captures behavior of many non-linear loads ! Model parameters:

  • Active power
  • Maximum ‘spike’ deviation (uniformly distributed)
  • Mean inter-arrival time (exponentially distributed)

Sean Barker (sbarker@cs.umass.edu)

Stable Min-Max Model

18

λ

(pactive, pspike, λ) = (160, 120, 10.82)

20 40 60 80 100 120 140 160 180

2 4 6 8 10 12 14 16

Power (W) Time (min)

TV model

TV model

slide-30
SLIDE 30

! Some non-linear loads lack a stable min or max

  • E.g., microwave

! Model these devices as a random walk

  • Bounded by a max, min power

Sean Barker (sbarker@cs.umass.edu)

Random Range Model

19

1380 1400 1420 1440 1460 1480 1500 0.5 1 1.5 2 2.5 3

Power (W) Time (min) microwave

pmax = 1480 pmin = 1400

microwave

slide-31
SLIDE 31

! Some loads have multiple component loads

  • Refrigerator: compressor, light
  • Dishwasher: motor, heating element

! Model each component as basic model type ! Compositions of component models

  • Parallel
  • Sequential
  • Periodic

Sean Barker (sbarker@cs.umass.edu)

Composite Loads

20

slide-32
SLIDE 32

Sean Barker (sbarker@cs.umass.edu)

Composite Load Example

21

1000 2000 3000 4000 5000 6000

20 40 60 80 100 120 140 Power (W) Time (min) dryer

slide-33
SLIDE 33

Sean Barker (sbarker@cs.umass.edu)

Composite Load Example

21

1920 1930 1940 1950 1960 1970

5 min zoom

1000 2000 3000 4000 5000 6000

20 40 60 80 100 120 140 Power (W) Time (min) dryer

slide-34
SLIDE 34

Sean Barker (sbarker@cs.umass.edu)

Composite Load Example

21

5300 5350 5400 5450 5500 5550 5600 5650 5700

5 min zoom

1000 2000 3000 4000 5000 6000

20 40 60 80 100 120 140 Power (W) Time (min) dryer

slide-35
SLIDE 35

Sean Barker (sbarker@cs.umass.edu)

Composite Load Example

21

Periodic composition

1000 2000 3000 4000 5000 6000

20 40 60 80 100 120 140 Power (W) Time (min) dryer

slide-36
SLIDE 36

Sean Barker (sbarker@cs.umass.edu)

Composite Load Example

21

Parallel composition

1000 2000 3000 4000 5000 6000

20 40 60 80 100 120 140 Power (W) Time (min) dryer

slide-37
SLIDE 37

Sean Barker (sbarker@cs.umass.edu)

Composite Load Example

21

1000 2000 3000 4000 5000 6000

20 40 60 80 100 120 140 Power (W) Time (min) dryer

Sequential composition

slide-38
SLIDE 38

Sean Barker (sbarker@cs.umass.edu)

Model Summary

22

Model Type Load Type Example On-Off Resistive Lights On-Off Growth/Decay Inductive, large resistive Motors, large heating elements Stable Min-Max Non-linear Television Random Range Non-linear Microwave Composite Multiple Dishwasher

slide-39
SLIDE 39

Sean Barker (sbarker@cs.umass.edu)

Model Accuracy

! Compare models to actual device power usage ! Example: on-off models vs on-off decay models

23

! Better load models = better accuracy

25 50 75 100 125 150 Coffee maker Toaster Dryer Root Mean Square Error

On-off Decay On-off Decay (first 30 secs) On-off On-off (first 30 secs)

slide-40
SLIDE 40

Sean Barker (sbarker@cs.umass.edu)

Outline

! Motivation ! Features of Electrical Loads ! Modeling Household Devices ! Applications of Models ! Conclusions

24

slide-41
SLIDE 41

Sean Barker (sbarker@cs.umass.edu)

Synthetic Building Traces

! Smart home research needs home energy traces

  • Both aggregate and device-level usage

! Difficult to collect device-level home traces

  • Complete instrumentation expensive, invasive

! Use models in synthetic home generation

25

Device Model Library Device Subset Synthetic Home

slide-42
SLIDE 42

Sean Barker (sbarker@cs.umass.edu)

Synthetic Trace Example

! Replace actual data with model-generated signatures ! Also generated trace using only on-off models ! Example metric: >15W power changes

  • Actual: 5591, Synthetic: 5833, On-off only: 1985

26

1000 2000 3000 4000 5000 6000 7000 8000

2 4 6 8 10 12 14 16

Power (W) Time (hours)

Measured Power Data

1000 2000 3000 4000 5000 6000 7000 8000

2 4 6 8 10 12 14 16

Power (W) Time (hours)

Synthetic Model-based Power Data

! Better models enable better trace generation

slide-43
SLIDE 43

Sean Barker (sbarker@cs.umass.edu)

Event Filters

! Non-linear loads often responsible for most power changes in homes ! Goal: filter out these pseudo-events

27

2000 4000 6000 8000 10000

ActiveHRV WashingMachine Dryer LivingRoom Refrigerator PassiveHRV Microwave CellarOutlets Dishwasher CellarLights BedroomLights KitchenLights CounterOutlets1 MasterOutlets CounterOutlets2 MasterLights GuestHallLights LivingRoomPatio BedroomOutlets

Power Events (>10W) Circuit

49827 5324 3162 2432 393 175 173 115 76 55 55 53 30 22 21 20 15 7 3 49827

HRV = 80%

slide-44
SLIDE 44

Sean Barker (sbarker@cs.umass.edu)

Stable Min-Max Filter

! Scan through data, maintain a stable power ! Filter changes under model-specific threshold

  • I.e., filter non-linear variations
  • Remaining power changes expose other events

! Example: TV plus light

28

! Models can simplify power event identification

200 400 600 800 1000 1200 1400 10 20 30 40 50 60 70 80

Power (W) Time (min)

unfiltered data

200 400 600 800 1000 1200 1400 10 20 30 40 50 60 70 80

Power (W) Time (min)

filtered data

slide-45
SLIDE 45

Sean Barker (sbarker@cs.umass.edu)

Conclusions

! Loads in homes are complex

  • Require complex load models

! Models can be built on features common to device classes

  • Features corresponding to basic model types
  • Complex devices as model compositions

! Better models can improve data analysis

  • Synthetic trace generation
  • Device filters

29

slide-46
SLIDE 46

Data: smart.cs.umass.edu Questions?

sbarker@cs.umass.edu

Department of Computer Science