Jiakang Lu, Tamim Sookoor, Vijay Srinivasan, Ge Gao, Brian Holben, John Stankovic, Eric Field, Kamin Whitehouse
SenSys’10
The Smart Thermostat: Using Occupancy Sensors to Save
Energy in Homes
AbstractMotivationChallengesIntroductionDesignExperiment SetupEvaluationConclusion
Outline
AbstractMotivationChallengesIntroductionDesignExperiment SetupEvaluationConclusion
Outline
Heating, ventilation and cooling(HVAC) is the largest source of residential energy consumption
Using cheap($5 each) and simple sensors(motion , door sensors) to automatically sense the occupancy and sleep patterns in a home
Automatically turn on or off the HVAC systemAchieve 28% energy saving on average, at a
cost of approximately $25 in sensorsCommercially-available baseline approach
that use similar sensors saves only 6.8% energy on average
Abstract
HVAC is the single largest contributor to a home’s energy bills and carbon emissionAccounting for 43% of residential energy
consumption in the U.S. and 61% in Canada and U.K., which have colder climates
Recent studies show that households with programmable thermostats have higher energy consumption on average than those with manual controlsUsers program them incorrectly or disable
them altogether
Motivation
Quickly and reliably determine when occupants leave the home or go to sleepMotion sensor are notoriously poor occupancy
sensors, which often turn lights off when a room is still occupied
When to turn HVAC system onToo early or too late, both waste energy
Challenges
AbstractMotivationChallengesIntroductionDesignExperiment SetupEvaluationConclusion
Outline
A 2-stage heat pump anda third stag electric heater
Stage 2 is the most efficientstage, but with longer response time
Stage 3 has the fastestresponse time but high energy cost
Stage 1 operates at a lower power level, it is more effective at maintaining a constant temperature
HVAC stage
ProblemsOccupants leave home shortly after 9AM, but
the system continues heating until 10AMShallow setback ( typically 5 degrees Celsius)Comfort loss
The risk of comfort loss causes people to reduce their use of setback schedules
Programmable Thermostat
Uses motion sensors, door sensors, or card key access systems to turn the HVAC on and off
4 out of 8 households actually increase energy usage by up to 10%
Leave at 9:30AM turn off at 10:30AMShallow setbackInefficient stage of heating
Reactive Thermostat
Fast reaction algorithm uses a probabilistic model to process the sensor data to estimate the occupancy
PreheatingDeep setback( about 10 degrees Celsius)
Smart Thermostat
AbstractMotivationChallengesIntroductionDesignExperiment SetupEvaluationConclusion
Outline
Passive infrared(PIR) motion sensors in rooms
Magnetic reed switches on entry waysApproximately $5 each
Hardware
Target preheat time tArrival time aTulum dataset[13], which was created by
monitoring the occupants of a home for approximately one month
a < t heat with stage 3a > t heat with stage 2optimum preheating time 18:06
Preheating
Tlast the time elapsed since the last sensor firing
It changes to idle state when it’s idle time is over a threshold
It changes to active state when it detect an event
“Sleep” from 10PM to 10AM
Reactive State Machine
Estimate the probability of 3 states of the homeAway, Active, Sleep
HMM transitions to a new state every5 minutes
xt is a vector of 3 features of sensor datatime of day at 4-hour granularitytotal number of sensor firings in time interval dTbinary features to indicate presence of front
door, bedroom, bathroom, kitchen, and living room sensor firings in dT
Training the model
Hidden Markov Model(HMM)
AbstractMotivationChallengesIntroductionDesignExperiment SetupEvaluationConclusion
Outline
Empirical data traces from 8 instrumented homes
Occupant surveys of 41 homes(4 weeks)Two public smart home dataset
Tulum, KasterenOne motion sensor in each roomOne door sensor on each entryway to the
home, and some inner doors
Collecting Occupancy data
Daily interviews with the residents to clarify ambiguous or questionable data
Not perfectly accuratePrevious studies have used approaches
ranging from self reports to video camera recordings
None of these schemes for creating ground truth are expected to be perfect
Ground Truth
Empirical data traces from 8 instrumented homes
Occupant surveys of 41 homes(4 weeks)Two public smart home dataset
Tulum, KasterenEach individual wrote down their sleep, wake,
leave, and arrive times every dayRetirees, students, professionals, young
professionals, and families
Collecting Occupancy data
Empirical data traces from 8 instrumented homes
Occupant surveys of 41 homes(4 weeks)Two public smart home dataset
Tulum, KasterenUse only the leave, arrival, and sleep event
labels
Collecting Occupancy data
Performance can be affected by “outdoor temperature”, “air leakage”, and “house insulation”
Evaluate different thermostat algorithms under different housing conditions and climates
Weather data are from the local airport weather station that provides hourly data
EnergyPlus Simulator
AbstractMotivationChallengesIntroductionDesignExperiment SetupEvaluationConclusion
Outline
Threshold ↑, slow reaction, treat inactive as active
Threshold ↓, fast reaction, treat active as inactive
HMM 88% accuracyReact5 78% accuracy
HMM vs. Reactive Algorithm
Run 14 days in January and July using the climate in Charlottesville, VA to evaluate both cooling and heating.
Deep setbacks to 10°C for heating and 40°for cooling
Evaluation
Reactive thermostat wastes energy due to frequent reactions
Reactive saves 2.9kWh(6.8%), misses 60 mins on average
Smart saves 11.8kWh(27.9%), misses 48 mins on average
Evaluation
Using data from “home B”Fast reaction, Deep setback, PreheatingAdd deep setback slightly increase miss timeSave 34% energy, improve miss time by 51
mins
Effect of Each Component
Sensitivity to Number of SensorsAlmost negligibleSelected vs. All energy 28.9% vs. 23.6% miss time 54 mins v.s 48 minsSelected set performs better!!!???
Periodic vs. Aperiodic life styleSave more energy and miss less time for
periodic life style
Sensitivity to Occupancy Patterns
As climate becomes warmer from MN to TX, it approaches the optimal scheme
Deep setbacks are beneficial during the day warm region, peak loads at mid-day cold region, night
Cold region consumes much more energy Lowering same amount of set point 5-8 degrees will only reduce the energy by a fraction
Sensitivity to Climate Zones
Smart thermostat that senses occupancy statistics in a home to save energy through improved control of the HVAC system
Very low cost less than $25 per home save 28% HVAC energyAlthough it just saves $15 per month for a
family, it has nationwide savings about $15 billion annually, and prevent 1.12 billion tons of pollutants from being released into the air.
Conclusion
Q&A
Thanks~