Demand-side Energy Management in Smart Buildings Prashant Shenoy - - PowerPoint PPT Presentation

School of Computer Science

Prashant Shenoy University of Massachusetts GreenMetrics 2013 Keynote

Demand-side Energy Management in Smart Buildings

University of Massachusetts Amherst - School of Computer Science

Motivation: Buildings and their Energy Usage

q Buildings are significant energy consumers

§ 76% of electricity and 48% of total energy in US

2

University of Massachusetts Amherst - School of Computer Science

Building Energy Usage Breakdown

q Residential: lighting: 11%, HVAC: 55% q Office: Lighting: 26%, HVAC: 50%

3

University of Massachusetts Amherst - School of Computer Science

Green Net-Zero Buildings

q Net-zero buildings: zero overall footprint

§ “Green” design: balance generation and consumption § Many new green buildings are net zero

q What about existing buildings?

§ Fact: 80-90% of buildings we will encounter already built

4

University of Massachusetts Amherst - School of Computer Science

How to Smarten and Green Buildings?

q Demand-side Energy Management (DSEM)

§ Manage energy usage by regulating demand

q Use on-site renewables: solar, wind, geo-thermal

§ Fall back to grid only when needed § Reduces carbon footprint and grid load

q Manual: reduce usage, conserve energy

§ e.g., turn off some lights when not needed

q Automated DSEM

§ Sense-Analyze-Control Approach § Grid provides signals, smart-building responds

5

University of Massachusetts Amherst - School of Computer Science

Intelligently Reducing Energy Use

q Intelligent DSEM q Automatically defer elastic loads

§ Control charging of Electric Vehicles § Schedule appliances

q Identify and eliminate waste

§ Align AC thermostat schedules to

• ccupancy patterns

q Manage demand during peak periods

6

University of Massachusetts Amherst - School of Computer Science

Peak Load Reduction: Why and How?

q Peak grid load: disproportionate marginal and environmental costs

§ Peaking power plants: inefficient and “dirty” § Lower peaks: reduced brown-outs

q Peak load reduction techniques

§ Time-shift supply: Use energy storage

• charge at off-peak and use at peak

§ Time-shift demand: schedule loads, shed loads

7

University of Massachusetts Amherst - School of Computer Science

Modeling and Prediction Challenges

q Modeling and prediction is key

§ Need to understand before we can optimize

q Sense-analyze-control approach to smart buildings

§ Modeling, prediction key to analysis and control

q Analyze, model and predict building energy usage q Model individual loads as well as aggregate loads

8

§ How to model electrical loads in a home?

University of Massachusetts Amherst - School of Computer Science

Talk Outline

q Motivation q Background on Electrical Loads q Modeling Electrical Loads q Using the Models q Conclusions

9

University of Massachusetts Amherst - School of Computer Science

Smart Meters and Energy Monitors

q Smart meters can meter fine-grain real-time usage

§ Utility-grade: 1-5 min resolution § Consumer-grade: 1sec resolution

q Meters and sensors can monitor individual loads

§ Per circuit-breaker or per outlet

q Data: 3 homes for 2+ years, >80 breakers, 100+ devices

10

University of Massachusetts Amherst - School of Computer Science

Modeling Electrical Loads

q Prior work: modeling aggregate demand profiles [ISO] q Individual loads: simple on-off models [Hart’89,Kim’12]

§ Device/load can either be on or off § Draws a fixed power when on § Example: light bulb

q Simple extension: multiple discrete ‘on’ states [REDD]

11

50 100 150 200 250 1 2 3 4

Power (W) Time (min)

light

University of Massachusetts Amherst - School of Computer Science

Today’s Electrical Loads

q Today’s devices are significantly more complex

§ exhibit rich, complex variations in power usage

q Question: How can we design better models to capture this behavior?

12

200 400 600 800 1000

5 10 15 20

Power (W) Time (min)

HRV

20 40 60 80 100 120 140 160 180

1 2 3 4 5

Power (W) Time (min)

LCD TV

University of Massachusetts Amherst - School of Computer Science

Electrical Loads: A Primer

q Loads: resistive, inductive, capacitive, nonlinear q Resistive: AC current and voltage waveforms align q Devices with heating (pure resistive) elements

§ Lights, toaster, coffee maker, oven, space heater

13

University of Massachusetts Amherst - School of Computer Science

Electrical Loads

q Inductive loads: current waveform lags voltage

§ Devices with AC Motors: AC, vacuum cleaner, fridge

q Non-linear loads: non-sinusoidal current draw

§ Electronic devices with switch-mode power supplies

• LCD TV, music system, computer, battery chargers

14

University of Massachusetts Amherst - School of Computer Science

Talk Outline

q Motivation q Background on Electrical Loads q Modeling Electrical Loads q Using the Models q Conclusions

15

University of Massachusetts Amherst - School of Computer Science

On-Off Model

q On-Off Model: two state model

§ states:

q Captures behavior of small resistive loads q Example: Lights

16

50 100 150 200 250 1 2 3 4

Power (W) Time (min)

light

University of Massachusetts Amherst - School of Computer Science

On-Off Decay Model

q Models inductive and large resistive loads

§ rush of current at startup, then settles to steady power

q On-off decay: on surge, decay to stable, off

§ active, inactive, peak power § rate of exponential decay

17

p(t) = ⇢ pactive + (ppeak − pactive)e−λt , 0 ≤ t < tactive poff , t ≥ tactive

200 400 600 800 1000 1200 1400 1600

10 20 30 40 50 60

Power (W) Time (sec)

vacuum cleaner

100 200 300 400 500 600 700 800

20 40 60 80 100 120

Power (W) Time (min)

refrigerator

200 400 600 800 1000

1 2 3 4 5 6 7 8 9 10

Power (W) Time (min)

coffee maker

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

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

University of Massachusetts Amherst - School of Computer Science

Stable Min-Max Model

18

20 40 60 80 100 120 140 160 180

1 2 3 4 5

Power (W) Time (min)

LCD TV

200 400 600 800 1000

5 10 15 20

Power (W) Time (min)

HRV

5 10 15 20 25 30 35 40 45

5 10 15 20

Power (W) Time (min)

Mac Mini

q Some non-linear loads lack a stable min or max, but are range-bound

§ E.g., microwave

q Model these devices as a random walk

§ Bounded by a max, min power

University of Massachusetts Amherst - School of Computer Science

Range-bound Random 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

University of Massachusetts Amherst - School of Computer Science

Basic Model Types

20

Model Load Examples On-Off Pure Resistive Lights On-off decay Inductive, large resistive motors, large heating elements Stable min-max Non-linear LCD TV Range-bound random Non-linear Microwave

University of Massachusetts Amherst - School of Computer Science

Model Accuracy

q Compare models to actual device power signature q Model as on-off decay versus simple on-off

§ Entire device cycle and first 30 seconds only

q Models more accurate at capturing device behavior

21

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)

University of Massachusetts Amherst - School of Computer Science

Composite Loads

q Many devices consist of multiple basic load types

§ Fridge: compressor, light, water dispenser § Dishwasher: motor, pump, heating element § Central AC: compressor, fan, humidifier

q Composite model: composition of basic models

§ Parallel composition § Serial composition § Cyclic/Periodic : repeating sequence

22

100 200 300 400 500 600

5 10 15 20 25 30 35 40 45 Power (W) Time (min) washing machine

random range, cyclic

• n-off decay,

cyclic stable min random range

• n-off decay, cyclic
• n-off decay

(growth)

University of Massachusetts Amherst - School of Computer Science

Talk Outline

q Motivation q Background on Electrical Loads q Modeling Electrical Loads q Using the Models q Conclusions

23

University of Massachusetts Amherst - School of Computer Science

Using the Models: NILM

q NILM: Non-intrusive Load Monitoring

§ Disaggregate individual loads from aggregate § Premise: individual loads have discernible “features”

q Prior work: edge detection or HMM-based

§ Assume loads are on-off devices

q Better load models => Better NILM

24

University of Massachusetts Amherst - School of Computer Science

Waste Identification

q HVAC: > 40-50% of total usage

§ Controlled by a thermostat

q Thermostats are hard to program “correct” q Correlate occupancy with thermostat schedules

§ Identify waste § Derive “optimal” schedules for each building/home

25

500 1000 1500 2000 2500 3000 1 2 3 4 5 6 7 Time [minutes] power

University of Massachusetts Amherst - School of Computer Science

Concluding Remarks

q Buildings: strong potential for energy efficiency gains

§ 48% of total energy usage

q Sense-analyze-control approach for smartening and greening buildings q Analysis, modeling, prediction (and control) are rich areas with many open problems q Joint work with D. Irwin, S. Barker, S. Kalra

26

University of Massachusetts Amherst - School of Computer Science

Questions?

q Empirical Characterization and Modeling of Electrical Loads in Smart Homes. IGCC’13 q SMART* Data: http://smart.cs.umass.edu q ACM BuildSys 2013 and ACM e-Energy 2014

27