-
Inverting HVAC for Energy Eicient Thermal Comfort inPopulous
Emerging Countries
Khadija HafeezComputer Science, LUMS
Lahore, [email protected]
Yasra ChandioComputer Science, LUMS
Lahore, [email protected]
Abu BakarComputer Science, LUMS
Lahore, [email protected]
Ayesha AliDepartment of Economics, LUMS
Lahore, [email protected]
Aan A. SyedINNEXIV and FAST NUIslamabad, Pakistan
[email protected]
Tariq M. JadoonElectrical Engineering, LUMS
Lahore, [email protected]
Muhammad Hamad AlizaiComputer Science, LUMS
Lahore, [email protected]
ABSTRACTEmerging countries predominantly rely on room-level air
condi-tioning units (window ACs, space heaters, ceiling fans) for
thermalcomfort. ese distributed units have manual, decentralized
con-trol leading to suboptimal energy usage for two reasons:
excessivesetpoints by individuals, and inability to interleave
dierent con-ditioning units for maximal energy savings. We propose
a novelinverted HVAC approach: cheaply retroing these
distributedunits with “on-o” control and providing centralized
control aug-mented with room and environmental sensors. Our binary
controlapproach exploits an understanding of device consumption
char-acteristics at on/o and factors this into the control
algorithms tominimize consumption. We implement this approach
asHawadaarin a prototype 180 2 room to evaluate its ecacy over a
7-monthperiod experiencing both hot and cold climates. We collect
enoughevidence to plausibly scale this evaluation, demonstrating
country-wide benets: with just 20% market penetration, Hawadaar
cansave up to 6% of electricity per capita in residential and
commercialsectors — resulting in a substantial countrywide
impact.
CCS CONCEPTS•Computer systems organization→ Sensors and
actuators;
KEYWORDSthermal comfort, energy eciency, inverted HVACACM
Reference format:Khadija Hafeez, Yasra Chandio, Abu Bakar, Ayesha
Ali, Aan A. Syed, TariqM. Jadoon, andMuhammad Hamad Alizai. 2017.
Inverting HVAC for Energy
Permission to make digital or hard copies of all or part of this
work for personal orclassroom use is granted without fee provided
that copies are not made or distributedfor prot or commercial
advantage and that copies bear this notice and the full citationon
the rst page. Copyrights for components of this work owned by
others than ACMmust be honored. Abstracting with credit is permied.
To copy otherwise, or republish,to post on servers or to
redistribute to lists, requires prior specic permission and/or
afee. Request permissions from [email protected] ’17,
Del, Netherlands© 2017 ACM. 978-1-4503-5544-5/17/11. . .$15.00DOI:
10.1145/3137133.3137137
Ecient ermal Comfort in Populous Emerging Countries. In
Proceedingsof BuildSys ’17, Del, Netherlands, November 8–9, 2017,
10 pages.DOI: 10.1145/3137133.3137137
1 INTRODUCTIONWe are in an era of global warming, with most
studies indicatingexcessive use of energy being its leading cause.
By most accounts,the greatest contribution to global energy
expenditure and theensuing green house gasses will come from
emerging Asian coun-tries [12, 25], such as China, India, Pakistan,
and Bangladesh that incombination account for nearly half of the
world population [13].With their economic upsurge, the energy
demands of these coun-tries are rapidly rising; thermal comfort of
built spaces makinga signicant proportion. It is projected that
there will be a ten-fold increase in the world consumption of
energy for cooling by2050 [20]. us, for example, China alone is
expected to surpassthe USA by 2020 as the world’s biggest consumer
of electricity forair conditioning estimated at a trillion kWh
[11].
With the late uptake of ecient thermal comfort systems
(likeHVAC) in these regions, the majority of buildings still employ
room-level units — such as window or split ACs, space heaters,
ceiling,and sometimes ventilation fans (see Figure 3 for a typical
room) —for thermal comfort. ese buildings will be maintained for
severaldecades with at least 80% to last beyond 2050 [19, 26],
consideringhousing needs, economic constraints, as well as heritage
protec-tion. e challenge is compounded with the continental
climatein most countries in the region, which is characterized by
extremetemperature variations, both daily and seasonally.
Consequently,this extreme climate also results in excessive energy
usage.
Anecdotally, room occupants set these distributively
controllableunits to exceed appropriate temperature creating the
kind of indoorenvironment in which occupants wear sweaters and use
blankets inJuly [10]. We validate this observation by a survey of
temperaturereadings shown in Figure 1; the data validating that a
distributiveapproach to temperature seing results in ineciency. ese
ag-gressive setpoints stem from i) a psychological reaction to
outside
-
BuildSys ’17, November 8–9, 2017, Del, Netherlands K. Hafeez et
al.
0 2 4 6 8 10 12 14 16 18 20Sample number
15
20
25
30
Tem
pera
ture
(°C)
(a) Winters
0 2 4 6 8 10 12 14 16 18 20Sample number
15
20
25
30
(b) Summers
Figure 1: Temperature measurements from randomly sam-pled rooms
in the administrative and residential complexesof our university
with room-level conditioning units: Tem-peratures oen kept too high
in winters or too low in sum-merswith neither the incentive nor the
capability to achievethermal comfort at low energy budget. Our goal
is to pushthese red dots into the shaded region (recommended
range).Energy saving is a reward.
temperatures (which can be extreme), ii) possibly inappropriate
ACsizing for the room resulting in insucient comfort at a
person'slocation in the room, or iii) a lack of per device thermal
control(space heater without thermostat). Furthermore, these
excessiveseings also fail to meet conditioning standards [4], thus
beingdetrimental for the health of the individual.
We propose a novel approach to solve this geographically
uniqueproblem; distributed room-level conditioning governed by a
standards-compliant control abstraction, the centralized
building-level ther-mostat. is is achieved by adding “smartness” to
existing devicesto maintain thermal comfort whilst saving energy.
is approachis interesting as an inversion to the HVAC approach to
managingthermal comfort: a centralized conditioning unit, the AHU,
thatdistributes air to zones controlled by individual
thermostats.
We present Hawadaar1, as an implementation of this invertedHVAC
approach to evaluate its ecacy. Hawadaar has threenovel aspects
that set it apart from the existing literature. First, itsability
to interleave several modes of achieving thermal comfort —such as
cooling using an AC, an evaporative cooler, or through
aircirculation using a fan — with an objective to minimize
electricityconsumption. Second, its adaptive two-position control
strategythat intelligently orchestrates these units, accounting for
thermalimpacting factors such as internal and external thermal
loads, roominsulation, and device characteristics (transients and
short cycling).ird, its ability to handle a wide variety of heating
and coolingdevices to deliver thermal comfort across a wide range
of weatherconditions 24/7 all year around.
Our work, thus, has the following signicant
contributions.Inverted HVAC Approach: We present an approach that
em-
ploys a centralized control abstraction to eciently
air-conditionexisting buildings, lacking HVAC, on a per-room basis.
is ap-proach to implementing thermal comfort is novel and
especiallypertinent to the socioeconomic and climatic constraints
of emerg-ing countries. We elaborate this approach and its
IoT-inspiredarchitecture in Section 2.
1local slang for ventilated and (air) conditioned space.
Wireless Control Channel
Wireless Feedback Channel
Sensor Measurements
Distributed Device Control
User
Control
Indoor Sensor
Temp. / Humidity
Outdoor
Sensor
Actuator
Centralized
Thermostat
Individually Conditioned Rooms
Figure 2: Inverted HVAC architecture: IoT retrotting
fordistributed conditioning via a centralized control abstrac-tion,
to improve thermal comfort and energy eciency oflegacy
buildings.
Control Algorithms: We introduce intelligent control algo-rithms
for the centralized thermostat. ese algorithms are basedon
empirical and theoretical understanding of the (conditioning)device
constraints (Sections 3), as well as factors impacting
thermalconditioning, such as modes of heat transfer, room
insulation, andperceived thermal comfort (Section 4).
System Evaluation, micro and macro scale: We perform ex-tensive
micro and macro-scale experiments on Hawadaar proto-type to
demonstrate a) in Section 5, the eciency of our algorithmsin
achieving setpoints (
-
Hawadaar BuildSys ’17, November 8–9, 2017, Del, Netherlands
1
2
3
4
7
56
1. Window AC2. Ventilation Fan3. Ceiling Fan4. Electric Heater5.
Actuators6. Indoor Sensor7. Outdoor Sensor
Figure 3: Hawadaar Deployment setup: A typical oceroom (180 2)
with legacy devices (1-4) and IoT retrotting(5-7). e roof and three
walls are exposed to elements.
on a per room basis by a user (e.g., the building owner),
through asmartphone app, to either deliver a setpoint or
personalized comfort— based on ASHRAE’s personal comfort metric
(PMV) [4] — withinrespective, standards-compliant comfort bounds.
Further indoorsensors can be added to improve the per room sensing
reliabilityand coverage; however, we emphasize on minimizing
retroingcosts assuming that the indoor sensor is deployed at a
pertinentlocation where the required thermal objective has to be
achieved.Additionally, an outdoor sensor is required for estimating
a room’sheat transfer coecient (see Section 4.3) for intelligent
actuation.For global access and refactoring into other applications
(e.g. smartgrids), the CT can be hosted in the cloud where an
appropriate APIlayer can expose its control and measurements to
user applications.
2.2 A prototype implementation of HawadaarWe now present a
practical realization of the inverted HVAC —Hawadaar — to
demonstrate and then evaluate the ecacy ofthis approach. Figure 3
shows our deployment in a 180 2 roomequippedwithmultiple air
conditioning units (fan, AC, space heater).is room is
representative of closed spaces in the developing world,where all
or a subset of these appliances are present. Our room-level
evaluation can thus be extrapolated to homes, apartments,hotels,
oce buildings, etc. Furthermore, this room has three exter-nal
facing walls and roof exposed to the elements, representing
achallenging scenario for conditioning the room.
We reduce the retroing complexity by choosing to control
eachdevice with a COTS3 smart-plug [32], that cost as low as $2
whenordered in bulk. e existing device sockets are inserted into
theseplugs4. For convenience, we choose Z-wave based plugs [32]
andtemperature and humidity sensors [33]. We emulate our
centralizedthermostat using a Z-wave dongle aached to a RPi with
all ourcontrol algorithms implemented on it. us, depending upon
themagnitude of supply order, the current retroing cost per
roomwith three conditioning units could be as low as $10
includingsmart-plugs, sensor, and the amortized cost of CT.
3commercial o-the-shelf4AC might additionally require a relay in
between to cater for high surge currents.
Figure 4: Air conditioning units under consideration: A) sin-gle
unit window AC with 1-ton cooling capacity; B) ceilingfan with
sweep size of 36 inches; C) electric space heaterwith three halogen
elements.
We choose an arbitrary location for the sensor, i.e, the
wallopposite to the AC when doing setpoint based control,
similarlyon the desk next to the ctional occupant in Figure 3 for
PMVbased control. is sensor, being wireless, can be placed at
anyappropriate location and, thus, the above evaluation is
sucient.ese (one-per-room) sensors report their values once per
minuteto the CT.
We note that our plug-based retroing limits our control tosimple
on-o for each device, and can introduce issues with regardsto
device safety if the control algorithm does not have
appropriatehysteresis. Entirely for this purpose, we evaluate
device constraintsin the next section and use them to tune our
algorithms in Section 4to prevent energy waste as well as
wear-and-tear.
3 DEVICES AND CONSTRAINTSWe rst highlight thermal
characteristics of air conditioning devicesunder consideration,
then briey review what device constraintsare relevant and how they
impact control, and subsequently derivethese constraints
empirically.
3.1 Air conditioning units under considerationFigure 4 depicts
the three devices currently employed byHawadaarfor distributed
conditioning. With a 1-ton cooling capacity, thesingle-unit
windowACprovides convective cooling through a single-stage
heatpump. In our setup, we position the setpoint of theAC's onboard
thermostat to its minimum value, thus allowingHawadaar to
independently control the AC (without altering thecontrol
circuitory) for setpoints above this minimum value. eelectric
heater implements radiant heating through its three halo-gen
elements, each consuming 400W. When tested under coolingand heating
loads greater than 10◦C, the minimum and maximumroom temperatures
achieved through the AC and heater are 18◦Cand 26◦C, respectively.
Both these values fall well outside the com-fort zones for the
respective seasons (see Section 5.2.2), thereby,already
highlighting the potential of energy conservation throughduty
cycling; more so in the case of an electric heater without
ther-mostat. e ceiling fan has a sweep size of 36 inches and can
beregulated at ve dierent speeds, with a maximum air velocity
of3.77ms−1. We set the fan speed to its maximum for on-o controlby
Hawadaar. While the fan only provides air movement, the ACand
heater impact both temperature and humidity.
-
BuildSys ’17, November 8–9, 2017, Del, Netherlands K. Hafeez et
al.
2500
5000
Window ACP
E
Transient stateSteady stateDevice On
1000
1300
P
ower
[wat
ts]
Electric Heater
0 10 20 30 40 50 60Time[secs]
0
100
Ceiling Fan
Figure 5: Transient vs steady state power consumption.
3.2 Relevant constraints and their impactHow accurately can
Hawadaar maintain a setpoint? is dependson how aggressively can it
actuate the respective air conditioningdevices. For example, in
cooling, two-position control is achievedby switching a device on
at temperatures exceeding Ton and o attemperatures below Tof f ,
and ideally Ton should be identical toTof f . However, such a
strict thermal objective is not achievable dueto deadband
requirement to prevent repeated on-o cycles, as wellas two device
specic constraints: transient power (the increasedpower consumption
at startup) and short cycling (actuating it fasterthan the specied
rate). Hawadaarmust be cognizant of these hardconstraints;
otherwise, the former could increase the overall energyconsumption
of the system and the laer could potentially damageor reduce the
life span of a device. us, to calibrate Hawadaar’scontrol
algorithm, we study both these characteristics in our setupand
calculate the smallest safe interval tsaf e , the minimum dura-tion
between switching the device o and then on. Alternatively,one could
pick a large enough tsaf e to make the algorithm agnosticto such
device constraints but this could compromise thermal com-fort [5].
e idea is to set tsaf e to the larger of two intervals, i.e.,max
(ttp , tsc ), where ttp and tsc refer to the respective
transientpower and short cycling constraints.
3.3 Deriving hard device constraintsFigure 5 shows the power
consumption of devices in transientand steady state. To avoid the
penalty of transient power due tofrequent switching, the minimum
duration (i.e., ttp ) for which adevice must remain o is dened by
the interval over which theexcess transient power equals the power
consumed if the devicehad not been switched o. us, we need to
switch o for at leastlong enough that this saved steady state power
compensates forthe high transient power at the next startup. A
longer o durationwill indeed translate into power gains. In Figure
5, based on thearea under the curve, we compute ttp simply as ttp =
EP , whereE is the excess transient energy and P is the steady
state powerconsumption. With regard to short cycling, there are no
theoreticalrestrictions on the switching frequency of electric
heaters and fans.Whereas, for window ACs, manufacturers specify a
minimum o-duration of three minutes [23, 24]. Table 3 enumerates
tsaf e foreach device that serves as a hard input constraint on the
maximumswitching frequency of the control algorithm. However, as
we
Table 1: Device Constraints: ttp is transient power and tsc
isshort cycling constraint. tsaf e is the larger of these two.
Devicettp(sec)
tsc(sec)
tsaf e =max(ttp ,tsc )
AC 1.62 180 180Heater 0.63 - 0.63Ceiling Fan 4.3 - 4.3
notice in Section 5.2.1, this switching frequency is not
actuatedeven for very narrow comfort bounds.
4 THE CENTRALIZED THERMOSTATWith device constraints clearly
dened, we are now ready to builda control abstraction, i.e. the
centralized thermostat. e CT allowsto congure per room thermal
seings through a mobile-app, toeither maintain a setpoint or
deliver personalized comfort (PMV),within standards-compliant
tolerance. Firstly we explain why a re-active control strategy is
employed in Hawadaar instead of modelpredictive control (MPC) [5].
We then describe our control algo-rithm and its heat-transfer
prediction mechanism, which is neededfor the dynamic adaption of
comfort bounds (i.e., the tolerancerange) based on thermal
load.
4.1 Why two-position control?e CT runs an intelligent control
algorithm that performs adap-tive two-position control: dynamically
adjusts the so constraint(comfort bounds) around a xed setpoint to
satisfy the hard deviceconstraints. We use this simple reactive
control strategy insteadof a complex model predictive approach for
three key reasons:First, unlike centralized HVAC taking tens of
minutes to take eect,these room-level conditioning units aect human
comfort imme-diately [18]. Second, room insulation levels in older
buildings aresuboptimal, thereby requiring an immediate response to
deliveradequate thermal comfort. Finally, a reactive control
strategy isinherently sensitive to changing thermal loads; for
example, airconditioning will run for longer if there are more
occupants thanusual in the room. is eliminates the need for complex
thermalload estimations, further simplifying our pilot system
design.
4.2 Algorithm: adaptive two-position controlWe rst describe the
algorithm in the context of maintaining asetpoint and later extend
our description to delivering personalizedcomfort. We do not make
any assumptions regarding the relativehumidity and use heat index
(a.k.a. “apparent temperature” or “feelslike”) as our temperature
metric. us, a setpoint is dened in termsof heat index; not the
ambient temperature. Hence, before makinga control decision, the
temperature value is rst converted into itscorresponding heat index
(HI), which is calculated as follows:
HI (T ,R) = c1 + c2T + c3R + c4TR + c5T2
+ c6R2 + c7T
2R + c8TR2 + c9T
2R2(1)
Where T is the air temperature, R is relative humidity, and
cnare Rothfusz regression constants [28]. For brevity, we use
thegeneral term “temperature” (instead of heat index) to simplify
thedescription of our algorithm.
-
Hawadaar BuildSys ’17, November 8–9, 2017, Del, Netherlands
Algorithm 1: Setpoint based adaptive two-position control.Input:
desrired setpoint (Ts ), tolerance
1 while True do2 Tin ← read sensor()3 Ton ← Ts + tolerance24 Tof
f ← Ts − tolerance25 if Tin > Ton then6 switch on(AC)7 else if
Tin ≤ Tof f then8 predicted temperature← Ttof f +tsaf e9 if
predicted temperature ≤ Ton then
10 switch o(AC)11 update(prediction parameters)12
reset(tolerance)13 else14 extend tolerance(tolerance)
4.2.1 Setpoint algorithm. To begin, the algorithm needs two
in-puts, the required setpoint Ts and comfort bounds (so
constraint);the laer denes two control positionsTof f andTon
symmetricallyaround Ts . us, with a cooling device, in order to
achieve Ts asthe average temperature, we must satisfy Ttof f +tsaf
e ≤ Ton , i.e.once the device is turned o, the temperature aer the
minimumsafe o duration tsaf e (hard constraint) will not exceed Ton
. ismay, depending upon the thermal load, require the algorithm
tosymmetrically extend the tolerance range on both sides of Ts ,
asdescribed in Algorithm 1. For this, the algorithm needs to
predict attime tof f (cf. Secton 4.3) the temperature at time tof f
+ tsaf e in or-der to decide whether to turn o the device or extend
the tolerancerange to meet the hard constraint. With a minor
adjustment, i.e.by swappingTof f withTon , this algorithm is also
used to actuate aheating device.
4.2.2 PMV algorithm. e same baseline algorithm is used
fordelivering personalized comfort with the following two
extensions:First, the thermal constraint is not the heat index but
PMV describedas follows:
PMV = f (M,Ta ,Tr ,v, Pa , Icl ) (2)
Where,M is the metabolic rate of the occupant (assumed 70W
/m2for an oce worker); Ta is the air temperature; Tr is the
meanradiant temperature (set equal toTa ); v is the relative air
velocity inm/s−1; Pa is the relative humidity; and Icl is the
clothing insulationfactor of the occupant (set to 0.6 clo assuming
a usual oce dresscode of a long sleeved shirt with trousers). In
our work, we assumesome of these parameters to calculate PMV and
realize that inpractice these are dicult to accurately ascertain;
however, thisdoes not aect the delity of the results. Second, while
the heatindex based setpoint only utilizes the heater and AC,
additionalfactors in equation 2, such as air movement, also allow
us to use theceiling fan for maintaining a desired PMV level. us,
as describedin Algorithm 2, we program the CT to prioritize the use
of low-energy fan, and only turn on the AC when air circulation
alonecannot keep the PMV within the required comfort bounds.
Algorithm 2: PMV based two-position control.Input: comfort
range, set of on devices (ON)
1 while True do2 calculate(PMV) // see equation 23 if PMV >
upper bound then4 if ON= ∅ then5 v ← AIR VELOCITYf an6
calculate(PMV)7 if PMV ∈ comfort range then8 ON← ON ] fan //
switch-on fan9 else
10 v ← AIR VELOCITYAC11 ON← ON ] AC // switch-on AC12 else if
fan ∈ ON then13 ON← ON \ fan // switch-off fan14 v ← AIR
VELOCITYAC15 ON← ON ] AC16 else if PMV ≤ lower bound then17 ON ← ∅
// switch-off all devices
4.3 Predicting the room’s heat transfer ratee room’s heat
transfer depends upon both external and internalthermal loads. e
external thermal loads result in heat transferthrough the building
envelope from external elements such as thesun, the earth, and the
outside environment. While, internal ther-mal loads come from heat
generated within the room by people,lighting, equipment etc. As
discussed in the previous section, inorder to satisfy the hard
constraint, the CT needs to predict thetemperatureTtof f +tsaf e .
e prediction model should thus accountfor both external and
internal thermal loads whilst being simpleand self calibrating. We
note that the maximum duration of thisprediction, in our setup, is
just three minutes imposed by the AC.is, by the way, also provides
a maximum theoretical duration ofdiscomfort. Since we dynamically
update our prediction parame-ters, as discussed below, our system
can repeatedly x predictionerrors.
Instead of developing a complex thermal model for the room,we
use realtime sensor measurements to glean the heat transfercoecient
using Netwon’s law of cooling as a rst approximation.A similar
strategy has also been employed in a personalized comfortsystem
[18]. According to the law, “the rate of heat loss of a bodyis
proportional to the dierence in temperatures between the bodyand
its surroundings”. us, given an outside temperatureTout androom
temperatureTin , the rate of thermal energy loss for the roomis
proportional to the temperature dierence:
dT
dt= −k (Tin −Tout ) (3)
We set T = Ttof f +tsaf e , Tin = Tof f and t = tsaf e and solve
theabove equation to estimate a room’s heat transfer coecient
(k):
k =ln
(Ttof f +tsaf e −Tout )(Tof f −Tout )tsaf e
(4)
However, we repeatedly update k just before Ton (i.e., when all
thedevices are o), aiming for an aggregate account of both
externaland internal thermal loads using a single heat transfer
coecient
-
BuildSys ’17, November 8–9, 2017, Del, Netherlands K. Hafeez et
al.
0 4 8 12 16 20Tin Tout
2
1
0
1
2
Pred
ictio
n Er
ror (
°C)
WinterSummer
Figure 6: Prediction accuracy: our simple model achieveshigh
accuracy with a root-mean-square error of 0.18◦C.
(k). e sensed Ttof f +tsaf e in round n is used to calculate the
kvalue to predict Ttof f +tsaf e in round n + 1 as follows:
Ttof f +tsaf e = Tout + (Tof f −Tout )e−ktsaf e (5)Figure 6
depicts the accuracy of our prediction mechanism for
numerous samples over a wide range of indoor and outdoor
tem-perature dierences. We can easily conclude that this model
issuciently accurate for a reactive control strategy with a
maxi-mum required prediction length of just three minutes.
5 EVALUATING THE DEPLOYMENTOur pilot deployment seeks answers to
two fundamental questionsregarding a thermal comfort solution: i)
How comfortable is it? andii) what are its energy benets? To answer
the rst question, weevaluate the minimum temperature tolerance
(best eort service)required to operate Hawadaar. is best eort
service is also rel-evant to satisfy high comfort requirements of a
demanding userand to stress test Algorithm 1. To answer the second
question, wecompare energy consumption in multiple seings. For
example,when varying tolerance around a xed setpoint; when
exceedingappropriate temperature seings, based on our anecdotal
obser-vation substantiated in Figure 1; and by interleaving devices
ofvariable energy consumption, as in Algorithm 2.
Although our deployment setup is in place since November
2016,here we only report results from experiments during two
challeng-ing weather spells occurring between Jan. 25 - Jan 28,
2017, andMay 1 - May 15, 2017, respectively. roughout these
experiments,the room occupation varied between 0 (at night) to at
most 3 (duringthe day) occupants.
5.1 ermal comfort: the best eort serviceis part of the
evaluation corresponds to the adaptive control thatminimizes the
tolerance range (|Tof f −Ton |) whilst satisfying harddevice
constraints. us, as described in Algorithm 1, the o-timeof a device
is xed at tsaf e while the on-time (duty cycle) is a func-tion of
thermal load at a certain instant of time. We expect thealgorithm
to minimize its tolerance range, given the device con-straint (tsaf
e ), such that
Tof f +Ton2 is always approximately equal
to the required setpoint (Ts ). In other words, we want our
systemto achieve two goals: (i) dynamically adapt the tolerance
range andthus the duty cycle based on the thermal load, and (ii)
deliver an av-erage temperature — in terms of heat index — that
does not deviatefrom the required setpoint. An inaccurate
temperature predictionat Tof f could potentially lead to such
deviations. Figure 7 veries
Off
OnAC
222426 Heat
index
(Tout) 12PM 2PM 4PMTime[Hour]
35
40
Outs
ide
Tem
pera
ture
--
------
-----(
°C)--
------
-----
Figure 7: e best eort service: Hawadaar successfullyadapts the
thresholds of its two-positions symmetrically os-cillating around
the setpoint. e heat index reects theroom temperature.
the delity of our algorithm and its temperature prediction
undervarying but high external thermal loads, when the setpoint is
keptnear the middle (24◦C) of ASHRAE’s specied comfort range
forsummers. We can clearly see that Hawadaar achieves its goalsby
successfully adapting the tolerance range oscillating around
thesetpoint. Even at such high thermal loads, Hawadaar is able
tomaintain a temperature tolerance of just ±0.4◦C (see Figure 7
for3PM onwards). Since the AC has the longest tsaf e , these
resultsrepresent the worst case of this best eort service.
5.2 antifying energy benets of Hawadaareare twomain aspects of
Hawadaar's implementation that bringabout its energy benets. First,
its standards-compliant thermalcomfort resulting in aggressive duty
cycling of single units. Second,when available, its ability to
prioritize low power devices. issection evaluates these two aspects
of energy eciency in multiplethermal seings.
Comparing energy savings in an experimental seing is notstraight
forward, as each day has its own parameter variationsthat aect how
much energy should be spent to cool or warm aroom. To draw logical
conclusions, we still try to make approximatecomparisons between
days with similar average air temperaturewhile minimizing the
variance among internal thermal loads.
5.2.1 When varying tolerance around setpoint. We want to seeif
changing the tolerance level around a xed setpoint aects theenergy
eciency of the system. We use two tolerance ranges:±0.5◦C and ±1◦C,
thus satisfying category A (2◦C) of ASHRAE’scomfort requirement in
terms of temperature deviation [4]. Wekeep the setpoint
approximately at the middle of the comfort range:22.5◦C in winters
and 24◦C in summers, as depicted in Figure 8. Forthis part of the
evaluation, the indoor sensor is placed on the wallopposite to the
window sill, where the AC is installed (cf. Figure 3).We make the
following key observations:
e nature of the heat transfer mechanism results in
dierentthermal behaviors. e rise and fall of room temperature is
gradualin winters (cf. Figures 8(a) and 8(b)) but rapid in summers
(cf.Figures 8(c) and Figures 8(d)). In other words, the period —
onecomplete on-o cycle — is larger for the heater in comparisonwith
the AC as a consequence of the heat transfer mechanism. Forthe AC,
cooling occurs through air convection. is can be short-lived aer
the AC is switched o as the air absorbs heat from thesurrounding
objects including external walls rapidly. For the heater,radiation
is the physical mechanism of heat transfer, both in the
-
Hawadaar BuildSys ’17, November 8–9, 2017, Del, Netherlands
OffOn
Heater
202224 Heat
index
(Tout) 9AM 1PM 5PM 9PM 1AM 5AM 9AMTime[Hour]
10
15
20
Outs
ide
Tem
pera
ture
--
------
-----(
°C)--
------
-----
0 4----------minutes----------22
22.18----------minutes----------
22.2
(a) Winter, January 25th (±0.5◦C tolerance)
Off
OnHeater
202224 Heat
index
(Tout) 9AM 1PM 5PM 9PM 1AM 5AM 9AMTime[Hour]
10
15
20
Outs
ide
Tem
pera
ture
--
------
-----(
°C)--
------
-----
(b) Winter, January 27th (±1◦C tolerance)
Off
OnAC
222426 Heat
index
(Tout) 9AM 1PM 5PM 9PM 1AM 5AM 9AMTime[Hour]
25
30
35
40
Outs
ide
Tem
pera
ture
--
------
-----(
°C)--
------
-----
3PM 3:30PM 4PM
24
(c) Summer, May 8th (±0.5◦C tolerance)
Off
OnAC
222426 Heat
index
(Tout) 9AM 1PM 5PM 9PM 1AM 5AM 9AMTime[Hour]
25
30
35
40
Outs
ide
Tem
pera
ture
--
------
-----(
°C)--
------
-----
(d) Summer, May 10th (±1◦C tolerance)
Figure 8: Setpoint evaluation: Results are shown for two
tolerance levels. e heat index reects the room temperature.
ethermal behavior in winters and summers is dierent due to the
nature of heat transfer mechanisms: convection for AC andradiation
for heater. e latter yields higher energy savings with wider
temperature tolerance around the setpoint.
Table 2: Daily energy savings of Hawadaar through aggres-sive
duty cycling of single units. Savings fromAC are not ap-plicable in
this particular case aswe expect theAC's onboardthermostat to
achieve similar results for the same tempera-ture settings.
Tolerance On-time (hours/day) Energy Consumption (kWh/day)
Energy Savings (kWh/day)Heater AC Heater AC Heater AC±0.5
14.42(60%) 9.3(38%) 17.5 31.35 11.6 −±1 11.26(47%) 9.27(38%) 13.6
31.15 15.5 −
form of visible and non-visible light, and eects all the
surroundingobjects not just the air. is process of heating is
slower but longerlasting aer the heater is switched o as the room
air continues toabsorb heat from heated objects maintaining a
residual warm airtemperature.
e nature of heat transfer has a contrasting impact on
energyeciency ofHawadaar over dierent tolerance levels. Table 2
summa-rizes the energy consumption and total on-time of devices per
day.In winters, we can see that the on-time of the heater is
reduced byapproximately 3 hours when the tolerance increases from
±0.5◦Cto ±1◦C. In contrast, in summers the AC on-time per day is
similarfor both tolerances. ere are two reasons: First, the “slow
start”of radiative heating makes it less ecient for frequent
switching,as this increases the number of times the cold halogen
elementwill have to be reheated (see magnier in Figure 8(a)).
Second, asdescribed above, the ability of radiative heating to
sustain the airtemperature for longer durations given a larger
tolerance range.
As an aside, duty cycling of the heater at ±0.5◦C daily saves
afurther 11.6 kWh (40%) compared with continuous operation.
SinceACs have onboard thermostats that already duty cycle,
assumingsimilar setpoints and tolerance, such savings are not
applicable.
5.2.2 When exceeding appropriate temperature seings. We nowwant
to highlight the energy benets of Hawadaar due to its abilityto
implement centralized policies, thus prohibiting excessive use
of
0 -1 -2 -3Deviation from Optimal Setpoint (°C)
0369
1215182124
On-ti
me
(hou
rs/d
ay)
11%30%
54%Excessive setpointHawadaar
Figure 9: Optimal vs excessive: e standards-compliantHawadaar
saves energy by prohibiting excessive setpoints.
conditioning units. is refers back to our anecdotal observation
inSection 1. Figure 9 shows how the AC on-time increases when
thesetpoint veers from optimal (24◦C) to excessive (23, 22, and
21◦C).We can see that even a single degree deviation from the
optimalsetpoint increases the on-time by 11%, which translates to
≈5 kWhper day. Similar observations, in the context of centralized
HVAC,have been reported in [22]. ese results clearly reiterate the
needfor an Hawadaar-like solution to implement centralized
policiesin regions, where the excessive and unhealthy use of
conditioningunits puts tremendous load on their stressed power
grids.
5.2.3 By interleaving devices of variable energy consumption.
Wenow turn our focus onHawadaar’s energy eciencywhen thermalcomfort
is dened in terms of PMV. e idea is to evaluate theenergy benet of
interleaving low-power ceiling fans. We measurethis impact for
ASHRAE’s recommended range of PMV (-0.5 < PMV< 0.5) during
two noticeably dierent weather conditions occurringat dierent times
of the day in hot summers.
Day time results. Figure 10 shows results for maintaining
PMVwith (Figures 10(b)) and without (Figures 10(a)) a fan during
hot day-time. We can observe that, under challenging weather
conditions,the use of a fan has limited impact on the operational
time of the
-
BuildSys ’17, November 8–9, 2017, Del, Netherlands K. Hafeez et
al.
OffOn
Fan
OffOn
AC
101
PMV
(Tout)9AM 1PM 5PM 9PM
Time[Hour]20
25
30
35
40
Outs
ide
Tem
pera
ture
(°C)
(a) PMV without fan (May 2nd)
OffOnFan
OffOnAC
101
PMV
(Tout)9AM 1PM 5PM 9PM
Time[Hour]20
25
30
35
40
Outs
ide
Tem
pera
ture
(°C) 7% reduction in AC usage due to fan
(b) PMV with fan (May 3rd)
Figure 10: PMV evaluation (day): In hot conditions,
inter-leaving a fan has limited impact on AC usage. e fan ex-tends
the duration for which PMV remains within speciedrange without
using AC but results in a further rise in tem-perature, adding to
the cooling load of AC.
AC, for two reasons. Firstly, the ceiling fan pushes down the
hot airthat rises up aer absorbing heat from surrounding objects,
thusincreasing air temperature. Secondly, the use of fan extends
theduration for which PMV remains within specied range due toair
circulation resulting in a further rise in air temperature.
isincreases the cooling load of AC during hot weather
conditions.
Night time results. e impact of the fan is signicantly
pro-nounced under relatively moderate weather conditions at night,
ascan be seen in Figure 11. e prolonged use of ceiling fan
helpsmaintain a desired PMV without the need for a high-power
AC,resulting in ≈30% reduction in AC usage. ese types of
energyoptimizations through aggregate usage of devices are highly
un-likely through manual operation or device specic thermostat;
anddenitely not possible at night with occupants asleep.
Overall, with PMV as comfort metric, we record a 15% reductionin
AC usage per day, that translates into approximately 2.5 kWh
ofenergy savings aer subtracting the power consumed by the fan.
6 ESTIMATING COUNTRYWIDE BENEFITSTo illustrate the energy benets
of Hawadaar on a countrywidescale, we use the results from Section
5, together with estimatesof annual electricity consumption on air
conditioners and electricheaters in residential and commercial
buildings. We are interestedin extrapolating energy savings for an
emerging economy that ispopulous, with a sizeable middle and upper
middle class in urbanareas living in buildings which can be retroed
with Hawadaar,and faces extreme climate with signicant temperatures
variationsin summer and winter months.
OffOn
Fan
OffOn
AC
101
PMV
(Tout)9PM 1AM 5AM 9AM
Time[Hour]20
25
30
35
40
Outs
ide
Tem
pera
ture
(°C)
(a) PMV without fan (May 2nd/3rd)
OffOn
Fan
OffOn
AC
101
PMV
(Tout)9PM 1AM 5AM 9AM
Time[Hour]20
25
30
35
40
Outs
ide
Tem
pera
ture
(°C) 30% reduction in AC usage due to fan
(b) PMV with fan (May 4th/5th)
Figure 11: PMV evaluation (night): In moderate condi-tion, the
AC on-time is signicantly reduced (≈30%) due toHawadaar’s ability
to intelligently interleave the fan.
e main challenge we encounter in this exercise is the lack
ofelectricity consumption estimates in buildings (residential or
com-mercial), at an aggregate level or by end use, in such a
context. Toaddress this, we employ an international benchmarking
approachthat uses plausible inputs from US electricity consumption
surveys,to arrive at conservative estimates of annual electricity
consump-tion in residential and commercial buildings on air
conditioners andelectric heaters in our typical economy. e inputs
and assumptionsunderlying our projected savings are discussed
below.
6.1 Inputs and assumptionse rst input is the annual electricity
consumption by end usein buildings in the US as recorded in the
Residential Energy Con-sumption Survey (RECS) [14] and Commercial
Building Energy Con-sumption Survey (CBECS) [6]. e RECS provides
data on annualelectricity usage in kWh for residences, while the
CBECS providesdata on electricity usage in kWh and oor space in
square feet forcommercial buildings in dierent climate regions of
the US by enduse, such as air conditioning and space heating,
amongst otheruses. e simplifying assumption we make is that US
electricityconsumption data is representative of electricity usage
in emergingeconomies. ere are undoubtedly variations in building
materials,appliance eciency, and engineering systems in the US
relative toany developing country, and we should expect the US
consumptionto be lower when accounting for beer technologies for
coolingand heating. is would lead us to start out with more
conservativeestimates of consumption in emerging economies, and
thus moreconservative estimates of savings. To ensure that
consumptionpaerns due to seasonal variations are similar to our
context, we
-
Hawadaar BuildSys ’17, November 8–9, 2017, Del, Netherlands
Table 3: Inputs for estimating total annual electricity
saving.Average Hot Humid Mixed-Dry/Hot-Dry Mixed-Humid
Residential AC (kWh/household) 2442.00 4077.00 1899.00
1777.00Commercial AC (kWh/f t2) 3.10 5.34 2.10 2.52
Commercial Heating (kWh/f t2) 0.34 0.12 0.15 0.52Population
Number of households 10 millionCommercial Floor Space 1000
million square feet
Penetration Rate 0–50%
take care to focus only on buildings in hot-humid,
mixed-humid,and mixed-dry/hot-dry building regions 5.
e second input is the stock of residential and commercial
build-ings that can benet from Hawadaar. We calculate energy
savingsfor a large emerging economy, with signicant urbanization.
Weassume a total population of 250 million persons, urbanization
ratesof 40%, implying an urban population of 100 million persons.
As-suming that half of the urban population lives in buildings
whereHawadaar can be used, and average household consists of 5
per-sons, we have a beneciary population of 10 million households
orresidential units. To arrive at estimates of commercial oor
space,we use plausible numbers of the density of commercial oor
spaceper capita of various developing countries available from
recentresearch [16], assuming that the stock of commercial oor
space is1000 million square feet.
e third input is the penetration rate of Hawadaar in our
ben-eciary population. In the base case, we assume that the
averagepenetration rates of Hawadaar for air conditioners is 20%.
Forelectric heaters, we assume 20% penetration in commercial
build-ings and 0% in residential units, as we expect natural gas
basedspace heating to be predominant in households due to its
cheapercost; currently not supported by Hawadaar.
Finally, based on our empirical results in Section 5, we
normalizeenergy savings emanating from both the key features of
Hawadaar:(i) aggressive duty cycling of single units and (ii)
interleaving oflow-power devices. In the rst case, we set energy
savings fromAC to 35% (cf. Section 5.2.2): is is the optimistic
case where weassume an average deviation of ≈-2◦C from the optimal
setpointbased on our anecdotal observation. Hawadaar can preempt
suchexcessive seings to claim its energy savings because it is
standards-compliant. In the second case, we assume the energy
savings fromAC to 15% (cf. Section 5.2.3). is is the pessimistic
case wherewe disregard our anecdotal observation and only consider
savingsfromHawadaar’s ability to minimize energy consumption
throughinterleaving low-power devices when using PMV as a
comfortmetric. We strongly believe that this is the bare minimum
benetof deploying Hawadaar. Finally, in both cases the energy
savingsfrom electric heater are set to 40% (cf. Section 5.2.1),
assumingno onboard temperature control units and that only
commercialbuildings use this type of heating.
Our inputs to calculate projected savings are summarized inTable
3. Total savings are found by multiplying the per unit
airconditioner and electric heating consumption for dierent
climatezones with savings above, assuming a penetration rate of
20%. Wealso calculate savings for dierent penetration rates. With
these
5ese climate regions were created by the Building America
program and are meantto capture the dierences in climate and
building types in dierent parts of the country.ese regions include
cities such as Houston, Dallas, Pheonix, Memphis, and Atlanta,which
have climate similar to cities such as Mumbai, New Delhi, Karachi,
Lahore,Tehran, Dhaka, Beijing, and Cairo.
Average Hot Humid
Mixed-Dry Hot-Dry
Mixed Humid
Climate Zones
0500
100015002000250030003500400045005000
Milli
ons k
Wh
Optimistic Pessimistic
Commercial HeatingCommercial Air ConditioningResidential Air
Conditioning
(a) By climate zones at xed Hawadaarpenetration rate of 20%.
0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50Penetration
rates
0500
100015002000250030003500400045005000
OptimisticPessimistic
(b) By varying market penetration.
Figure 12: Countrywide estimates of annual energy savings.
inputs, we calculate very conservative estimates of total
savings(cf. Figure 12), which can be considered a lower bound on
actualsavings that may be realized from implementing Hawadaar.
6.2 Savings and environmental impactFigure 12(a) shows the
countrywide estimate of energy savingsby dierent climate zones at a
market penetration of 20%. In theoptimistic scenario, savings range
from 1462 million kWh to 3237million kWh, and in the pessimistic
scenario, savings range from645 million kWh to 1393 million kWh.
Savings are higher for cool-ing in humid zones, given the higher
consumption per householdand per unit area. Similarly, heating
consumption and thus savingsare higher in climate zones with
greater need for heating duringthe year. Given our assumptions
about the population size, resi-dential air conditioning accounts
for ≈85%, while commercial airconditioning and heating together
account for 15% of total esti-mated savings. In Figure 12(b) we
display the eect of changing thepenetration rates on savings for
the buildings located in averageclimate zone. Total savings rise
proportionally with penetrationrates, as we assume a uniform
consumption rate in our population.
To understand the magnitude of total energy savings, we canalso
express them as a fraction of per capita consumption in
theresidential and commercial sectors. Assuming an average per
capitaelectricity consumption of 255 kWh in residential and
commercialsectors6, the estimates in Figure 12(a) imply that
between 1.15% to5.76% of electricity consumption in these sectors
can be saved withpenetration rates of just 20%. As we increase the
penetration ratesfrom 5% to 50%, the percentage savings range from
0.87% to 8.68%in the optimistic scenario and from 0.38% to 3.79% in
the pessimisticscenario. Overall, these projections illustrate that
even with veryconservative assumptions, Hawadaar can have an
economicallymeaningful impact on consumption at a macro level.
7 RELATEDWORKEcient operation of HVAC has been at the forefront
of existingliterature on energy conservation [1, 3, 7, 9, 21, 29,
30] and ther-mal comfort [15, 18, 27, 31] in buildings. Energy
conservation istypically achieved by incorporating occupancy paerns
[1, 7, 21]6e average per capita electricity consumption in lower
middle income countriesranges from 500 to 1000 kwh per capita out
of which approximately 30% is residentialand commercial electricity
consumption [8]. Using the midpoint consumption of 750kwh per
capita we can conclude that residential and commercial sectors
account for255 kwh per capita of electricity consumption. en we can
express our total savingsin per capita terms and nd the percent
saved.
-
BuildSys ’17, November 8–9, 2017, Del, Netherlands K. Hafeez et
al.
and thermal load predictions [5, 9] into HVAC operation
schedules,or by exercising more ne-grained control, such as
room-level airow control [29] and stage selection in a multi-stage
HVAC [30].Aswani et al. [5] employ similar predictive techniques
for the energyecient operation of a room-level AC only, thereby
addressing asubset of problems considered in this paper. Studies
focusing on im-proving thermal comfort of centralized HVAC try to
nd improvedsetpoints based on occupants' feedback [15, 31], or by
augmentingHVAC with personal devices to create micro thermal-zones
arounda user for highly personalized thermal comfort [2, 18, 27].
esepersonal comfort systems nonetheless rely on HVAC to rst
achievea building wide setpoint; this facility is not available in
our setup.
Our goal to achieve thermal comfort at low energy budget
isaligned with existing literature but the nature of challenges
weface is inherently dierent. For example, the type of buildings,
theextreme weather, as well as thermal characteristics and
locationof conditioning units entail us to build more aggressive
controlstrategies, such as the ones employed by Hawadaar. is
paperthus primarily focuses on developing and evaluating these
cen-tralized control strategies. While the current implementation
ofHawadaar is oblivious to occupancy prediction (estimating
thenumber of occupants for determining internal thermal load) due
toits reactive control strategy, occupancy detection (if and when
theroom is occupied) is an orthogonal but well-researched
problemoutside the current scope of this paper. us, we do not
foreseeany inherent challenges in the seamless incorporation of
existingoccupancy detection solutions inHawadaar, to rene its
operationschedules and further reduce energy consumption.
e work that comes closest to our idea of interleaving AC
andceiling fan is the collaboration of NEST with a smart ceiling
fancompany [17]. e key idea is to adjust fan'speed as
temperaturesrise, allowing to increase thermostat setpoint of a
centralized HVACwhile still feeling just as cool. However, we
observed that for ourrange of operating temperature and suboptimal
building insula-tion, simultaneous use of AC and fan results in
higher averagetemperature as the fan forces hot air down.
8 CONCLUSIONSis paper serves as a proof of concept for the
fundamental compo-nents of an unconventional, inverted HVAC
architecture for olderbuildings in emerging countries. As an
alternative tomodernHVAC,we proposed IoT-based retrots to reinforce
legacy air conditioningunits for policy driven actuation. Hawadaar
is a practical realiza-tion of this proposal, demonstrating its
ecacy in achieving a highlevel of thermal comfort at low energy
budget. Our empirical eval-uations, when plausibly scaled to
countrywide estimates highlightthe worthwhile impact of Hawadaar,
providing energy savingsthat directly translate into reduced carbon
emissions in countriesthat rely heavily on burning fossil fuels for
electricity generation.We see a greater value in pursuing this work
further not only towiden its impact, such as by including more
device types, but alsoto improve its implementation and algorithmic
aspects.
REFERENCES[1] Yuvraj Agarwal, Bharathan Balaji, Seemanta Dua,
Rajesh K. Gupta, andomas
Weng. 2011. Duty-cycling buildings aggressively: e next frontier
in HVACcontrol. In IPSN.
[2] Michael P. Andersen, Gabe Fierro, Sam Kumar, Michael Chen,
Leonard Truong,Joyce Kim, Edward A. Arens, Hui Zhang, Paul Raery,
and David E. Culler. 2015.Poster Abstract: Well-Connected
Microzones for Increased Building Eciencyand Occupant Comfort. In
BuildSys.
[3] Omid Ardakanian, Arka Bhaacharya, and David Culler. 2016.
Non-IntrusiveTechniques for Establishing Occupancy Related Energy
Savings in CommercialBuildings. In BuildSys.
[4] ASHRAE. 2010. ermal Environmental Conditions for Human
Occupancy,Standard 55.
[5] A. Aswani, N. Master, J. Taneja, D. Culler, and C. Tomlin.
2012. ReducingTransient and Steady State Electricity Consumption in
HVAC Using Learning-Based Model-Predictive Control. Proc. IEEE 100,
1 (2012).
[6] CBECS 2012 (Tables B27 and E5). 2012. Commercial Buildings
Energy Consump-tion Survey.
hps://www.eia.gov/consumption/commercial/data/2012
[7] Bharathan Balaji, Jian Xu, Anthony Nwokafor, Rajesh Gupta,
and Yuvraj Agar-wal. 2013. Sentinel: Occupancy Based HVAC Actuation
Using Existing WiFiInfrastructure Within Commercial Buildings. In
SenSys.
[8] World Bank. 2016. World Development Indicators.
hp://data.worldbank.org/data-catalog/world-development-indicators
[9] Alex Beltran and Alberto E. Cerpa. 2014. Optimal HVAC
Building Control withOccupancy Prediction. In BuildSys.
[10] Stan Cox. 2010. Losing Our Cool: Uncomfortable Truths About
Our Air-ConditionedWorld (and Finding New Ways to Get rough the
Summer). e New Press.hp://thenewpress.com/books/losing-our-cool
[11] Stan Cox. 2012. Cooling a Warming Planet: A Global Air
Conditioning Surge. InYale Environment 360. hp://e360.yale.edu
[12] Lucas W. Davis and Paul J. Gertler. 2015. Contribution of
Air ConditioningAdoption to Future Energy Use Under Global Warming.
Proceedings of theNational Academy of Sciences of the United States
of America (2015).
[13] United Nations Population Devision. 2015. Revision of World
PopulationProspects. hps://esa.un.org/unpd/wpp/.
[14] RECS 2009 (Table C E41). 2009. Residential Energy
Consumption Survey.
hps://www.eia.gov/consumption/residential/data/2009/
[15] Varick L. Erickson and Alberto E. Cerpa. 2012. ermovote:
Participatory Sensingfor Ecient Building HVAC Conditioning. In
BuildSys.
[16] Burak Gneralp et al. 2016. Global scenarios of urban
density and its impactson building energy use through 2050. In
Proceedings of the National Academy ofSciences of the United States
of America.
[17] Haiku fans. hps://www.haikuhome.com/[18] Peter Xiang Gao
and S. Keshav. 2013. Optimal Personal Comfort Management
Using SPOT+. In BuildSys.[19] International Energy Angency
(IEA). 2013. Policy Pathways: Modernising Build-
ing Energy Codes.[20] Morna Isaac and Detlef P. van Vuuren.
2009. Modeling global residential sector
energy demand for heating and air conditioning in the context of
climate change.Energy Policy 37, 2 (2009).
[21] Jiakang Lu, Tamim Sookoor, Vijay Srinivasan, Ge Gao, Brian
Holben, JohnStankovic, Eric Field, and Kamin Whitehouse. 2010. e
Smart ermostat:Using Occupancy Sensors to Save Energy in Homes. In
SenSys.
[22] M.M. Manning, M.C. Swinton, F. Szadkowski, J. Gusdorf, and
K. Ruest. 2005. eEects of ermostat Seing on Seasonal Energy
Consumption at the CCHTResearch Facility. Canada Center for Housing
Technology.
[23] Amana (User Manual). Room Air Conditioner & Heat Pump:
Use and CareManual. www.amana.com
[24] Frigidaire (User Manual). All about the use and care of
your room air conditioner.www.frigidaire.com
[25] Michael A. McNeil and Virginie E. Letschert. 2008. Future
Air ConditioningEnergy Consumption in Developing Countries and what
can be done about it:e Potential of Eciency in the Residential
Sector. hp://escholarship.org/uc/item/64f9r6wr
[26] Karsten Neuho, Hermann Amecke, Aleksandra Novikova, and
Kateryna Stel-makh. 2011. ermal Eciency Retrot of Residential
Buildings: e GermanExperience. In CPI Report, Climate Policy
Initiative. hp://hdl.handle.net/10419/65868
[27] Alimohammad Rabbani and S. Keshav. 2016. e SPOT* Personal
ermalComfort System. In BuildSys.
[28] Lans P. Rothfusz. e Heat Index “Equation”. In 1990 National
Weather Service(NWS) Technical Aachment (SR 90-23).
[29] Virginia Smith, Tamim Sookoor, and KaminWhitehouse. 2012.
Modeling Buildingermal Response to HVAC Zoning. SIGBED Rev. 9, 3
(2012).
[30] T. Sookoor and K. Whitehouse. 2013. Roomzoner:
Occupancy-based room-levelzoning of a centralized HVAC system. In
ICCPS.
[31] Daniel A. Winkler, Alex Beltran, Niloufar P. Esfahani, Paul
P. Maglio, and Al-berto E. Cerpa. 2016. FORCES: Feedback and
Control for Occupants to ReneComfort and Energy Savings. In
UbiComp.
[32] wireless female zwave power plug.
hps://qianpeng.en.alibaba.com/[33] zwave temperature humidity
sensor. hps://heimansmart.en.alibaba.com/
https://www.eia.gov/consumption/commercial/data/2012http://data.worldbank.org/data-catalog/world-development-indicatorshttp://data.worldbank.org/data-catalog/world-development-indicatorshttp://thenewpress.com/books/losing-our-coolhttp://e360.yale.eduhttps://www.eia.gov/consumption/residential/data/2009/https://www.eia.gov/consumption/residential/data/2009/https://www.haikuhome.com/www.amana.comwww.frigidaire.comhttp://escholarship.org/uc/item/64f9r6wrhttp://escholarship.org/uc/item/64f9r6wrhttp://hdl.handle.net/10419/65868http://hdl.handle.net/10419/65868https://qianpeng.en.alibaba.com/https://heimansmart.en.alibaba.com/
Abstract1 Introduction2 Hawadaar: The Inverted HVAC2.1 Whats an
Inverted HVAC?2.2 A prototype implementation of Hawadaar
3 Devices and Constraints3.1 Air conditioning units under
consideration3.2 Relevant constraints and their impact3.3 Deriving
hard device constraints
4 The Centralized Thermostat4.1 Why two-position control?4.2
Algorithm: adaptive two-position control4.3 Predicting the room's
heat transfer rate
5 Evaluating the Deployment5.1 Thermal comfort: the best effort
service5.2 Quantifying energy benefits of Hawadaar
6 Estimating Countrywide Benefits6.1 Inputs and assumptions6.2
Savings and environmental impact
7 Related Work8 ConclusionsReferences
