Upload
others
View
5
Download
1
Embed Size (px)
Citation preview
Research ArticleSmart HVAC Control in IoT Energy Consumption Minimizationwith User Comfort Constraints
Jordi Serra David Pubill Angelos Antonopoulos and Christos Verikoukis
Centre Tecnologic de Telecomunicacions de Catalunya (CTTC) 08860 Castelldefels Spain
Correspondence should be addressed to Jordi Serra jordiserracttces
Received 10 April 2014 Accepted 16 May 2014 Published 18 June 2014
Academic Editor Xudong Zhu
Copyright copy 2014 Jordi Serra et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
Smart grid is one of the main applications of the Internet of Things (IoT) paradigm Within this context this paper addressesthe efficient energy consumption management of heating ventilation and air conditioning (HVAC) systems in smart grids withvariable energy price To that end first we propose an energy scheduling method that minimizes the energy consumption cost fora particular time interval taking into account the energy price and a set of comfort constraints that is a range of temperaturesaccording to userrsquos preferences for a given room Then we propose an energy scheduler where the user may select to relax thetemperature constraints to save more energy Moreover thanks to the IoT paradigm the user may interact remotely with theHVAC control system In particular the user may decide remotely the temperature of comfort while the temperature and energyconsumption information is sent through Internet and displayed at the end userrsquos device The proposed algorithms have beenimplemented in a real testbed highlighting the potential gains that can be achieved in terms of both energy and cost
1 Introduction
The Internet ofThings (IoT) paves the way for the connectionof sensors actuators and other objects to the Internet per-mitting the perception of the world as well as the interactionwith it in an unprecedented manner In addition IoT willfoster a huge number of new applications for exampleenvironmental monitoring healthcare and efficient manage-ment of energy in smart homes [1] potentially generatingimportant economic benefits [2] Actually the US NationalIntelligence Council considers IoT as one of the six disruptivecivil technologieswith potential impact onUSnational power[3] As a result the concept of IoT in terms of architecturalaspects protocol stacks applications and conceptual visionshas recently started to be studied [4ndash7]
Smart grid is considered as one of the main IoT applica-tions and it has attracted a great interest during the last fewyears [1 8 9] The smart grid is envisioned as the evolutionof the current energy grid which faces important challengessuch as blackouts caused by peaks of energy demand thatexceed the energy grid capacity [10] A proposed approachto alleviate this problem is to incentivize the consumers todefer or reschedule their energy consumption to different
time intervals with lower expected power demand Theseincentives are based on smart (or dynamic) pricing tariffs thatconsider a variable energy price [11] For instance in real-timepricing (RTP) tariffs the price of the energy will be higherat certain time periods where the energy consumption isexpected to be higher for example during the afternoon or incold daysOther types of smart pricing tariffs are critical-peakpricing (CPP) or time-of-use pricing (ToUP) [11ndash13] Energyscheduling algorithms are the state-of-the-art methods tomanage the energy consumption of loads within a smartpricing framework [11 12 14ndash16] These techniques assumea specific smart pricing tariff and various time periodsFor each of these time intervals the scheduler determinesthe operational power of each appliance to minimize theenergy consumption cost It is worth mentioning that theappliances that can be controlled by the energy scheduler canbe categorized into three classes (i) nonshiftable which donot admit any change on their consumption profile (ii) time-shiftable which tolerate postponing their operation but nottheir consumption profile and (iii) power-shiftable whoseoperational power can be changed
Regarding the power-shiftable loads heating ventilationand air conditioning (HVAC) modules are considered as
Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 161874 11 pageshttpdxdoiorg1011552014161874
2 The Scientific World Journal
the most energy demanding appliances in home buildings[17 18] According to studies they represent the 43 ofresidential energy consumption in the USA and the 61in UK and Canada [18] Apparently the significant energyconsumption of the HVAC systems along with their directinfluence on the userrsquos well-being highlights the necessityfor effective HVAC management algorithms that reducethe power consumption in the home buildings taking intoaccount the end-userrsquos comfort
In this paper we propose two HVAC energy schedulingmethods in an IoT framework where the users are ableto interact remotely with the HVAC control system Inparticular the users may retrieve information about thetemperature and the energy consumption at various spotsof the building under control while they are also able toremotely configure the temperature in given places Ourcontribution can be summarized as follows
(i) We propose a dynamic energy scheduler with com-fort constraints (DES-CC) which considers both thesmart pricing tariffs and the userrsquos comfort in orderto select the most energy efficient configuration ofHVACs that satisfies the userrsquos needs We formulatean optimization problem of HVAC control by pre-dicting the temperature that a given set of HVACmodules would cause in different locations We resultin a boolean quadratic optimization problem whichalthough not convex can be solved via an exhaustivesearch when the number of variables (ie HVACmodules in our case) is low In case that a large num-ber of HVAC modules are considered semidefiniterelaxation techniques can be applied [19]
(ii) Taking into account the energy efficiency priority wepropose a dynamic energy scheduler with comfortconstraints relaxation (DES-CCR) where the userrelaxes their comfort constraints allowing a higherdegree of flexibility for the system to further reducethe energy consumption In this case the problemis reformulated and the userrsquos comfort (ie temper-ature) is set as a penalty in the objective functioninstead of constraint
(iii) We have designed and developed a real testbed toevaluate the performance of the proposed algorithmsdemonstrating the potential financial and energygains that can be achieved
The remainder of the paper is organized as followsSection 2 provides a brief review of the related work in thisfield Section 3 describes the general network architectureand the system model under study Section 4 introducesthe two HVAC schedulers in a smart pricing and comfortconstraint context Section 5 provides the description ofthe testbed and the experimental results Finally Section 6concludes the paper
2 Related Work
The energy cost management of HVAC systems has recentlyattracted the research attention In [20] the energy cost is
studied as a function of the parameters that control the airand water subsystems and an evolutionary programmingmethod is proposed to save energy Moreover in [21] adynamic threshold controls the energy consumption and itvaries according to the user satisfaction which also dependson a thermal model However neither [20] nor [21] explicitlyconsider a dynamic pricing cost In [22] smart pricing isconsidered in the energy cost optimization but the usercomfort is not explicitly incorporated in the algorithm asthe authors consider that the HVAC is turned onoff whenthe indoor temperature is outside the margin of comfortRecently in [13 18] both energy scheduling of HVAC undersmart pricing and the user comfort are taken into accountIn [13] Nguyen et al propose the construction of a lookuptable (LUT) of room temperatures that depends on (i) thepast temperatures (ii) the outdoor temperature and (iii) theHVAC power The authors claim that the LUT is built duringa training period (that takes place only once) and permits toassess the temperature of comfort for a given operation of theHVAC energy scheduler However this heuristic approachseems hardly applicable in general scenarios In [18] a linearenergy cost function is considered although quadratic ortwo-step piecewise linear functions are more common inpractice [11] while userrsquos comfort is measured only at aspecific location It is also worth noting that none of theaforementioned works considers an IoT framework
Unlike [20ndash22] in our proposed energy schedulingmeth-ods both the smart pricing tariffs and the user comfort aretaken into account Moreover the temperature of comfortis measured at several building positions by different sensornodes that form a wireless sensor network (WSN) thusproviding a more accurate measure of comfort comparedto [18] Furthermore compared to [13] we adopt a moreanalytical and less heuristic model to assess the user comfortin the HVAC energy cost optimization In particular ourmodel considers the time varying nature of the real thermalconditions without requiring a training period Moreoverour model adaptively updates the past temperature measure-ments for each time period whereas the model in [13] is onlycarried out once to construct the LUT for particular indoorand outdoor conditions Finally unlike most of the abovereferences our methods are validated in a real scenario
3 Network Model
31 General Architecture Figure 1 presents the overall archi-tecture which is used to evaluate the proposed energyschedulers in an IoT context It consists of the followingelements
(i) a set of HVAC modules
(ii) a set of actuators that control the HVAC modules
(iii) a WSN which sends measurements of temperatureand energy consumption to a gateway
(iv) a gateway (GW) that incorporates the proposedenergy scheduling methods and connects the localnetwork to the Internet That is it contains a web
The Scientific World Journal 3
Gatewayenergy scheduler
web serverIP device
Wireless sensornetwork
Temperature sensors(Z1 motes)
Actuators
HVAC modules
Internet
Figure 1 Overall architecture of the proposed HVAC energy scheduler in the IoT context
Server
DatabaseEnergy
scheduler
Actuators
IP device
InternetWSN
Gatewayweb server
Data measurementUser requirementsEnergy scheduler decision
Figure 2 System model of the gateway
server and a database to store data received at the GWfrom the WSN or the internet
(v) an embedded IP device (eg tablet or smartphone)with an interface to interact with the HVAC energyscheduler It also displays both the temperature andthe energy consumption in the building measured bythe WSN
The functionality and flowof information of the proposedarchitecture is explained as follows The temperature ismeasured at several locations bymeans of theWSNThen themeasurements are periodically sent to the gateway where theenergy scheduling algorithm is implemented This algorithmselects the combination of the active HVAC modules thatminimizes the energy cost for given comfort constraints andenergy price during a particular time periodThese decisionsare sent through shell commands to programmable surgeprotectors (actuators) which actuate theHVACmodulesTheHVAC modules modify the room temperature according tothe decisions taken by the energy scheduler
Moreover the gateway hosts a database to store the mea-surements of temperature and energy consumption Thesemeasurements can be accessed by a remote Internet user
More specifically they are displayed at the userrsquos IP deviceas the gateway implements a web server which managesthe communication between the remote user and the localdatabase This is illustrated in more detail in Figure 2 wherethe connections between the most relevant blocks are shownFurthermore users are allowed to interact with the energyscheduler through their IP devices by setting the upper andlower bounds of the temperature of comfort
32 SystemModel Thesystemmodel for the proposedHVACenergy scheduler is depicted in more detail in Figure 3 Inparticular the energy scheduler is implemented within thegateway and it interacts with the following modules First aWSN composed of 119872 sensor nodes 119878
119894 1 ⩽ 119894 ⩽ 119872 which
sense the temperature and transmit the measurements to theenergy scheduler through aWSN sink node Second a set of119870HVACmodules that are controlled by the energy schedulerthrough a set of actuators119860
119896 1 ⩽ 119896 ⩽ 119870Moreover the inputs
that the energy scheduler requires are described as follows
(i) Themeasurements taken by theWSNnodes For eachtime interval 119873 measurements are taken by eachnode when a given configuration of HVAC modules
4 The Scientific World Journal
Uplink wiredUplink wirelessDownlink wiredExternal inputs
middot middot middot
SinkHVAC1 HVACK
OnoffOnoff
AKA1
Tmini
Tmaxi C(L(sj))
TmjMTm
j1S1 SM
Gateway
Figure 3 Detailed system model of the HVAC energy scheduler
is turned on These measurements are denoted by119879119898119895
119894(119899) (as illustrated by the black curve in Figure 4)
where 1 ⩽ 119894 ⩽ 119872 denotes the 119894th node and 1 ⩽ 119895 ⩽ 2119870
is the 119895th combination of HVACs turned on or off(ii) The energy cost function 119862(119871(s
119895)) which is specified
by the energy provider and depends on the smartpricing tariff According to [11] 119862(119871(s
119895)) can be
modeled as a quadratic function that is
119862 (119871 (s119895)) = 119901
1(119871 (s119895))2
+ 1199012119871 (s119895) + 1199013 (1)
where 1199011 1199012 and 119901
3are parameters that the provider
can dynamically vary in timeMoreover119871(s119895)denotes
the userrsquos energy consumption for the s119895combination
of HVACmodules turned on In order to describe theexpression for 119871(s
119895) let us define P isin R119870times2
119870
a matrixthat contains in its 119895th column the energy consump-tion of each HVAC module for the 119895th combinationof modules switched onoff For instance for 119870 = 3the matrix P is
P = (
0 1199091
0 0 11990911199091
0 1199091
0 0 1199092
0 1199092
0 11990921199092
0 0 0 1199093
0 119909311990931199093
) (2)
Moreover s119895is a vector of all zeros except in the 119895th
position that has value 1 Therefore Ps119895selects the
energy consumption related to the 119895th combinationof HVACs switched on and 1119879(Ps
119895) is the energy
consumption related to that combination where 1 isa vector of ones of length 2
119870 Therefore the totalenergy consumption for a given time period denotedby 119871(s
119895) can be expressed as
119871 (s119895) = 1198710+ 1119879 (Ps
119895) (3)
Tem
pera
ture
Time
2K predictedcurves
Past time intervalNsamples N samples
Next time interval
Current time
n + N
Tmaxi
Tmini
Tmji
Tpji
n
Figure 4 Prediction of temperature a fundamental step of theenergy scheduler to assess comfort in the future time interval
where 1198710
is the accumulated consumption and1119879(Ps
119895) is the consumption in the current time inter-
val decision
(iii) The constraints of temperature of comfort providedby the user For the energy scheduler proposed inSection 41 this corresponds to the minimum andmaximum allowed temperatures at the 119894th locationof the room which are denoted by 119879
min119894
and 119879max119894
with 1 ⩽ 119894 ⩽ 119872 respectively since we assumethat the user may specify the desired comfort at 119872different locations For the energy scheduler proposedin Section 42 the comfort is specified by the objectivetemperatures 119879
119906119894 1 ⩽ 119894 ⩽ 119872 which the user would
like to attain at the different119872 locations of the roomthough they may allow some relaxation in order tofurther reduce the energy consumption
To further clarify the operation of the proposed schemelet us shed light on the temporal behavior of the energyschedulers and the role of the temperature constraints onit Recall that the energy scheduler works in a time intervalbasis At the end of each time interval (ldquocurrent timerdquo inFigure 4) the energy scheduler must make a new decisionThat is it must decide which HVAC modules denoted byHVACk in Figure 3 will be active during the next timeinterval In order to make this decision the energy schedulershould predict which would be the temperature provokedby each configuration of HVACs As there are 119870 HVACmodules and we assume that they are either turned on oroff this corresponds to predict 2119870 curves of temperature asit is illustrated in Figure 4 These predicted temperatures aredenoted by 119879119901
119895
119894(119899) where 119894 and 119895 have the same meaning
as for the case of 119879119898119895119894(119899) explained above Finally on one
hand the DES-CC selects the configuration of HVACs thatminimizes the energy consumption cost 119862(119871(s
119895)) within the
bounds of comfort that is 119879min119894
⩽ 119879119901119895
119894(119899) ⩽ 119879
max119894
while theDES-CCR selects the HVAC configuration that optimizes thetradeoff between being closer to the comfort temperatures119879119906119894
and saving energy This selection is executed by the
The Scientific World Journal 5
actuators that control the HVACmodules which are denotedby 119860119896in Figure 3
4 HVAC Energy Scheduling
41 Dynamic Energy Scheduler with Comfort Constraints(DES-CC) Next we present the first of the two proposedHVAC energy schedulers To that end this section is dividedinto four parts First we formulate the energy scheduleras a constrained optimization problem Second recall thatfor each time interval the energy scheduler must decidethe combination of active HVACs to minimize the energyconsumption cost and fulfill the constraints of temperatureof comfort To assess these constraints the temperatureprovoked in the next time interval by each configuration ofHVACs turned on or off should be predicted Thereby thesecond part deals with a thermal model that paves the way topredict the future temperatures The third part specifies howto estimate the parameters of the prediction model thanks tothe measurements of temperature of the past time intervalFinally in the fourth part we summarize the proposed DES-CC algorithm
411 Formulation of DES-CC as a Constrained OptimizationProblem Theenergy scheduler works in a time interval basisWhen 119873 samples of temperature have been collected fromthe WSN at the119872 controlled locations the energy schedulermakes a new decision with respect to the state of the HVACsNamely for the next time interval the energy schedulerselects the optimal configuration of active HVACs Thisconfiguration on the one hand must minimize the energyconsumption cost while on the other hand it must respectthe comfort constraints that is it should lead to predictedtemperatures within the bounds of comfort According to thedefinitions of the system model this optimization problemmay be formulated mathematically as
minimizes119895isin012
119870times1
119862 (119871 (s119895))
subject to 119879min119894
⩽ 119879119901min119894
(s119895) 119894 isin [1119872]
119879max119894
⩾ 119879119901max119894
(s119895) 119894 isin [1119872]
1119879s119895= 1
(4)
where 119879119901min119894
(s119895) and 119879119901
max119894
(s119895) are the minimum and maxi-
mum predicted temperatures respectively at the 119894th locationfor the 119895th combination of HVAC modules turned on Giventhe definition of 119862(119871(s
119895)) in (1) as a quadratic function
the optimization problem (4) has the form of a quadraticprogramming but that the optimization variable is booleanHence it is a boolean quadratic programming problem Theproblems of this class are nonconvex and in general they canbe solved either by a fast method that finds a local solutionor by a slower method that finds the global solution Inour framework the number of HVAC modules 119870 is low ormoderate and the latter approach is preferred for examplethe branch and bound method [23] can be used In order
to proceed we need to model the predicted temperatures119879119901
min119894
(s119895) and 119879119901
max119894
(s119895)
412 Model to Predict the Temperatures of the Future TimeInterval Regarding the predicted temperatures 119879119901
min119894
(s119895)
and 119879119901max119894
(s119895) they can be expressed as
119879119901min119894
(s119895) =
∘q119879119894s119895
119879119901max119894
(s119895) = q119879119894s119895
(5)
where ∘q119894and q
119894are vectors that contain the minimum and
maximum predicted temperatures respectively for each ofthe possible combinations of operating HVAC modules Tofurther clarify let us define 119879119901119895
119894(119899) 2 le 119899 le 119873 the predicted
temperature at the 119899 time instant at the 119894th sensor for the 119895thcombination of HVAC modules turned on where 1 ⩽ 119894 ⩽ 119872
and 1 ⩽ 119895 ⩽ 2119870 Moreover let 119879119901119895
119894(119899min119895) and 119879119901
119895
119894(119899max119895)
be theminimum andmaximum temperatures among119879119901119895119894(119899)
2 le 119899 le 119873 Then ∘q119894and q
119894can be expressed as
∘q119879119894= [119879119901
1
119894(119899min1) 119879119901
2119870
119894(119899min
2119870)]
q119879119894= [119879119901
1
119894(119899max1) 119879119901
2119870
119894(119899max
2119870)]
(6)
In order to proceed a model for the predicted tempera-tures is necessary Intuitively the current temperature is cor-related with the past temperature and a given combination ofHVACs turned on causes a change in temperature Moreoverthe temperature dynamics are rather linear (at least locally)as it will be shown below Therefore the following model isproposed for the temperature prediction
119879119901119895
119894(119899) = 119886
119895
119894119879119901119895
119894(119899 minus 1) + 120574
119895
119894 2 le 119899 le 119873 (7)
where 119886119895119894and 120574119895119894model the relation with the past temperature
and the change of temperature provoked by the 119895th combina-tion of HVACs turned on respectively Observe that in thisexpression 119886119895
119894and 120574
119895
119894are unknown and must be estimated
from the past measurements
413 Estimation of the PredictionModel Parameters In orderto estimate 119886
119895
119894and 120574
119895
119894in (7) we assume that the past
measurements follow a model like (7) corrupted by noise
119879119898119895
119894(119899) = 119886
119895
119894119879119898119895
119894(119899 minus 1) + 120574
119895
119894+ 119908119895
119894(119899) 2 le 119899 le 119873 (8)
Note that the evolution of the temperature is consideredto be linear in (7) and (8) This is a valid assumption at leastfor short periods as the real experiments that we will presentin Section 5 will highlight
For the estimation of 119886119895
119894and 120574
119895
119894 two situations are
considered In the first one all the HVAC modules areswitched off and as a consequence only 119886119895
119894must be estimated
To that end least squares (LS) estimator is considered as noprobabilistic assumptions regarding the data are neededThis
6 The Scientific World Journal
Process For each time interval do(1) Input(a) The measurements of temperature of the previous interval(b) Value of 119901
11199012and 119901
3in the energy cost function (1)
(c) The userrsquos temperature of comfort constraints at eachlocation that is 119879min
119894 119879
max119894
119894 = 1 119872 in (4)(2) Estimation Step
Estimate 119886119895119894and 120574
119895
119894in the prediction model (7) using (9)
to (12) and the temperature measurements from the pasttime interval
(3) Prediction Step(a) Substitute the estimation of 119886119895
119894and 120574
119895
119894into the prediction model (7)
(b) Iterate this model to populate the vectors of minimumand maximum predicted temperatures (6)
(4)Optimization Step(a) Substitute the predicted temperatures (5) and the
quadratic cost function (1) into the optimization problem (4)(b) Solve (4) using Branch and bound method [23]
(5)OutputThe configuration of HVACs turned onoff that optimizes (4)
Algorithm 1 Dynamic energy scheduler with comfort constraints (DES-CC)
estimator minimizes the LS error criterion though it is notoptimal in general [24] Given (8) the LS estimation of 119886119895
119894
denoted by 119886119895119894is given by
119886119895
119894= xy (9)
where the symbol denotes the pseudoinverse operatorwhich is defined as x = (x119879x)minus1x119879 and we define x =
[119879119898119895
119894(1) 119879119898
119895
119894(119873 minus 1)]
119879
and y = [119879119898119895
119894(2) 119879119898
119895
119894(119873)]119879
The second situation is that some of the HVACmodules wereswitched on In this case an LS estimation is considered aswell Namely let us denote by 120574
119895
119894| 119886119895
119894the estimation of
120574119895
119894conditioned to the knowledge of a past estimation of 119886119895
119894
denoted by 119886119895
119894 Then the LS estimation for 120574119895
119894| 119886119895
119894yields
120574119895
119894| 119886119895
119894= 1z (10)
where 1 is a vector of ones of length119873 minus 1 and z is given by
z = [119879119898119895
119894(2) 119879119898
119895
119894(119873)]119879
minus 119886119895
119894[119879119898119895
119894(1) 119879119898
119895
119894(119873 minus 1)]
119879
(11)
Finally given the estimation of 120574119895119894in (10) we can update
the estimation of 119886119895119894as
119886119895
119894| 120574119895
119894= xy (12)
where y = [119879119898119895
119894(2) minus 120574
119895
119894 119879119898
119895
119894(119873) minus 120574
119895
119894]119879
414 Summary of the DES-CC Energy Scheduler At thispoint all the terms in the optimization problem understudy that is (4) are specified The procedure to implementthe proposed energy scheduler for each time interval issummarized in Algorithm 1
42 Dynamic Energy Scheduler with Comfort ConstraintsRelaxation (DES-CCR) Despite its effectiveness and its obvi-ous advantages the proposed energy scheduling algorithm iscompletely focused on the temperature constraints neglect-ing the price aspects of the problem More specificallyalthough there could be time periods where the energyprice increases the energy scheduler switches on the samecombination of HVAC modules in order to respect thetemperature constraints However in such cases users mightcompromise their comfort preferences to decrease the energyconsumption In order to allow the user to have more flexi-bility in the energy consumption a new energy scheduler willbe presented in this section This flexibility is implementedin terms of relaxing the temperature constraints to furtherreduce the energy consumption
This new energy scheduler is formulated so that theuser temperature constraints in (4) are skipped and they areincorporated as a penalty term in the objective functionConsequently the new optimization problem can be writtenas
minimizes119895isin012
119870times1
120579
119862 (119871 (s119895))
120572+ (1 minus 120579)
times
sum119872
119894=1sum119873
119899=2
10038171003817100381710038171003817q119879119894(119899)s119895minus 119879119906119894
10038171003817100381710038171003817
2
120573
(13)
where119862(119871(s119895)) is the energy cost function defined in (1)The
vector q119879119894(119899) is defined as
q119879119894(119899) = [119879119901
1
119894(119899) 119879119901
2119870
119894(119899)] (14)
and recall that 119879119901119895119894(119899) is the predicted temperature at the 119894th
location for the 119895th combination of HVACs modules turned
The Scientific World Journal 7
Process For each time interval do(1) Input(a) The measurements of temperature of the previous interval(b) Value of 119901
1 1199012and 119901
3of the energy cost function 119862(119871(119904
119895)) defined in (1)
(c) The userrsquos temperature of comfort at each location that is 119879119906119894 119894 = 1 119872 in (13)
(d) The parameter 120579 isin (0 1) controlling the comfort relaxation in (13)(2) Estimation Step
Estimate 119886119895119894and 120574
119895
119894in the prediction model (7) using (9)
to (12) and the temperature measurements from the past time interval(3) Prediction Step(a) Substitute the estimation of 119886119895
119894and 120574
119895
119894into the prediction model (7)
(b) For 119894 = 1 to119872For 119899 = 2 to119873Compute and store the vector of predicted temperaturesq119879119894(119899) in (14) using (7)
End For 119899End For 119894
(4)Optimization Step(a) Compute the cost function in (13) using the vectors in
the step 3(b) and the inputs in (1)(b) Solve the optimization problem in (13) using Branch and bound method [23]
(5)OutputThe configuration of HVACs turned onoff that optimizes (13)
Algorithm 2 Dynamic energy scheduler with comfort constraints relaxation (DES-CCR)
on or off see (7)Moreover120572 and120573 are normalizing constantsto adjust the values of the two terms in (13) Indeed we settheir value as
120572 = 119862 (119871 (s2119870))
120573 = maxs119895isin01119870times1
119872
sum
119894=1
119873
sum
119899=2
10038171003817100381710038171003817q119879119894(119899)s119895minus 119879119906119894
10038171003817100381710038171003817
2
(15)
where 119862(119871(s2119870)) is the cost for all the HVACmodules turned
on The term 119879119906119894
is the desired temperature that the userwould like tomaintain at the 119894th location of the room Clearlyour reformulation balances the two optimization problemsthat is the energy cost minimization and the user comfortmaximization Note that the user comfort is defined as anEuclidean norm but it can eventually be redefined withanother distance measurement
Finally 120579 isin (0 1) is defined by the user according to theirpreferences For example in the extreme case where 120579 = 0the demand response algorithm will not consider any priceand it will directly control the HVAC modules so that thedesired temperature is reached On the contrary when 120579 = 1theHVACmodules will always remain off In this context theusers should set the 120579 value according to their preferences andexperience The DES-CCR energy scheduler is summarizedin Algorithm 2
5 Experimental Results
In order to emulate the complete communication in anIoT framework we have designed and developed a customtestbed that integrates the described architecture In our
Figure 5 Z1 WSN mote
experiments we focus on a heating system although theproposed algorithms apply in general HVAC systems In thissection we describe the testbed platform and the experi-mental scenario we define a baseline thermostat model andfinally we present the experimental results of our proposedalgorithms
51 Testbed Description and Experimental Setup The testbedhas been deployed in a 50m2 room within our researchcenter facilities as it is depicted in Figure 5 In our particularscenario we consider three HVAC modules (ie 119870 = 3)and two temperature sensor nodes (ie 119872 = 2) The HVACmodules are distributed around the room while the sensornodes are placed in the middle of the room monitoring thetemperature and sending it to the sink mote every 30 second
8 The Scientific World Journal
Micro-USB
Ceramic embedded antennaUFL connector for external antenna
temperature sensor3-Axis accelerometer+
2 times Phidgets sensor ports
Figure 6 Overall platform detail
(ie 119905119898= 30 s) In addition the samples received from each
sensor are stored in a buffer at the control center and ouralgorithm applies every 10 samples (ie119873 = 10)
Regarding the employed technology the WSN nodes areZ1 motes by Zolertia (Figure 6) They are equipped with asecond generationMSP430F2617 low power microcontrollerwhich features a 16-bit RISC CPU 16MHz clock speed abuilt-in clock factory calibration an 8KB RAM and a 92KBflash memory They also include the CC2420 transceiverwhich is IEEE 802154 compliant operating at 24GHz fre-quency bandwith a data rate of 250 kbpsThe sensors supportContiki OS [25] an open-source operating system for the IoTwhich connects tiny low-cost low-power microcontrollersto the Internet and supports IPv6 through 6LowPAN It isworth noting that each mote can operate as either a sourceor a sink node In particular source nodes carry a TMP102temperature sensor to monitor the target field while the sinknode receives and forwards themeasured data to the gateway
The gateway (an Ubuntu OS machine with MATLAB)implements the proposed algorithms and it is able to processthe collected data Furthermore it connects the WSN to theInternet and acts as an application server using Nodejs andSencha Touch In particular Nodejs is a platform built onChromersquos JavaScript runtime for fast and scalable networkapplications Figure 7 shows a screenshot of the web applica-tion built on Nodejs which enables the user to interact withthe energy scheduler through Internet Regarding SenchaTouch it is a high-performance HTML5 mobile applicationframework which enables developers to build powerfulapplications for various operating systems including iOS andAndroidThe actuators are programmable sockets which canbe controlled remotely thanks to their IP addresses Thesespecial sockets are a set of programmable local area networksurge protectors (EG-PMS-LAN) by Energenie which areconnected via Ethernet to the gateway Finally the HVACmodules are domestic heaters with a maximum power con-sumption of 2000W
52 Baseline Model To evaluate the proposed algorithmsand highlight the potential energy and cost gains that can
Figure 7 Google Chrome screenshot of the web application
be achieved we adopt the traditional thermostat model asthe baseline reference scenario In this model the aim is tomaintain the average temperature of the room between acertain temperature range (ie [119879min119879max]) predefined bythe user To that end when the sensed temperature is above119879max at the end of a time interval all the heaters are switchedoff while the heaters are switched on when the temperaturefalls below the 119879min threshold
Figure 8 illustrates the average measured temperatureinside the room where the heaters are controlled by thethermostat In this particular case we consider 119879min =
21∘C and 119879max = 23
∘C As it can be seen in the figurethe thermostat algorithm is able to maintain the averagetemperature of the room between the desiredmargins during16 hours However it is worth noting that despite its properbehavior the particular model is not cost efficient as allHVAC modules work simultaneously consuming a totalpower consumption of 6000W In the following sectionswe evaluate our proposed methods demonstrating that theycan reduce the electrical cost with respect to the baselineapproach
The Scientific World Journal 9
21
22
23
24
HVA
C sta
tus
Average temperatureHVAC status
Day time
Tem
pera
ture
(∘C) On
Off
Thermostat-baseline model
18 20 22 24 2 4 6 8 10
Figure 8 Experimental evaluation of the thermostat model
53 Experimental Evaluation of the DES-CC in (4) Severalreal experiments have been carried out to assess the perfor-mance of the DES-CC algorithm proposed in (4) In thiscase the pricing parameters in (1) are 119901
1= 0003 and 119901
2=
1199013= 0 euros which are possible values according to [11] The
temperature bounds in (4) have been set to 119879min1
= 119879min2
=
21∘C and 119879
max1
= 119879max2
= 23∘C respectively
Figure 9 plots the variation of both the measured and theestimated temperature (using (8)ndash(12)) in the room duringour experiments As it can be noticed the error between theestimated and the real temperature is negligible somethingthat proves the accuracy of the proposed estimationmodel Inaddition the DES-CC guarantees the proper operation of thesystem as the temperature varies between the desired rangeof 21∘C and 23
∘C most of the time with very few exceptionsdue to prediction errors In the same figure it can be alsoseen that the temperature remains closer to the lower partof the permitted range (ie 21∘C) since the outcome of theproposed method provides a combination of switched onheaters that minimizes the energy consumption satisfyinga minimum acceptable temperature Indeed compared tothe temperature variation in the baseline scenario (Figure 8)DES-CC maintains the temperature more stable and in thelower part of the allowable region intuitively implying lowercost
Figure 10 depicts the financial operation cost gains thatcan be achieved by DES-CC compared to the baseline ther-mostat approach As it can be observed the proposed energyscheduler significantly reduces the energy cost leading to atotal save of 719 eurosmonth
54 Experimental Evaluation of the DES-CCR in (13) A setof experiments have been carried out for the evaluation ofthe DES-CCR in (13) Let us recall that DES-CCR relaxesthe temperature constraints by including the constraints asa penalized term in the objective function As a resultcompared to DES-CC this method is more flexible withrespect to real time pricing tariffs More specifically DES-CC seeks a combination that minimizes the energy cost withrespect to a minimum allowable temperature Consequently
WSN mote 2
WSN mote 1
18 20 22 24 2 4 6 8 1018192021222324
Day time
18 20 22 24 2 4 6 8 10Day time
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18192021222324
Tem
pera
ture
(∘C)
Figure 9 Real and estimated temperature using DES-CC
0 2 4 6 8 10 12 14 160
0005
001
0015
002
0025
003
0035
004
Hours
Ener
gy co
st (E
uros
)
Energy cost per hour of the proposed methodEnergy cost per hour of the thermostat
Figure 10 Energy consumption cost comparison between thermo-stat and DES-CC
although the energy cost may change during time the heatercombination selected byDES-CC is the same due to the stricttemperature constraint On the other hand DES-CCR allowsthe user to further reduce the energy consumption at the costof being outside the range of temperature of comfort In thiscase to highlight the flexibility of DES-CCR we have set a
10 The Scientific World Journal
0 50 100 150 200 25016
18
20
22
24
Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
0 50 100 150 200 25016
18
20
22
24
Sample measures
Tem
pera
ture
(∘C)
Figure 11 Real and estimated temperature using DES-CCR (120579 =
02)
periodically variable value of 1199011 which alternates between
1199011
= 0009 and 1199011
= 0003 euros every thirty minutesMoreover the desired temperature has been set to 119879
119906119894=
22∘CFigures 11 and 12 depict the temperature variation in
two different cases where the users give low (120579 = 02)and high (120579 = 05) priority respectively to reduce of theenergy consumption In particular in Figure 11 (120579 = 02)the achieved temperature is very close to the desired 119879
119906119894 On
the other hand in Figure 12 we assume 120579 = 05 which is amore adapted value to the pricing policy as it correspondsto a user that permits a relaxation of the difference betweenthe real and the desired temperature to reduce the energycost This fact implies higher energy consumption in lowcost zones and lower energy consumption in high costperiods sacrificing though the userrsquos comfort Therefore theexperiments confirm that the real temperature is close to 119879
119906119894
when the energy cost is lower (ie between samples 60 and120) while there is a noticeable temperature drop whichcorresponds to lower energy consumption
6 Concluding Remarks
This paper has dealt with the energy consumption manage-ment of HVACs for a given smart pricing tariff and usersrsquocomfort constraintsMoreover the integrationwithin the IoTframework has been studied To that end we developed areal testbed consisting of (i) heaters (ii) sensor nodes thatmeasure the temperature and (iii) a gateway which providesconnection to the Internet and includes aweb application that
18
19
20
21
22
0 50 100 150 200 250Sample measures
0 50 100 150 200 250Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18
19
20
21
22
Tem
pera
ture
(∘C)
Figure 12 Real and estimated temperature using DES-CCR (120579 =
05)
permits the interaction with the user through InternetMore-over the gateway implements the algorithms that control theenergy consumption Regarding the proposed methods firstwe devised an energy scheduler that optimizes the energycost in a time interval basis for a given energy price tariffand for a given set of temperature of comfort constraintsthat are associated with different locations inside a roomThen we proposed a more flexible energy scheduler whichrelaxes the temperature constraints to further reduce theenergy consumption Namely a new objective function hasbeen considered which consists of a convex combination ofthe energy cost and a penalty term that reflects the comfortThis permits to consider both the case where the user isvery concerned with the comfort and the case where heallows relaxing the comfort constraint to further reduce theenergy consumption Experimental evaluations have beencarried out in an isolated room validating our proposals andhighlighting their potential benefits
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work has been funded by the Energy-to-Smart Grid(E2SG) project httpwwwe2sg-projecteu within theENIAC joint undertaking framework with Grant agreementnumber 296131
The Scientific World Journal 11
References
[1] L Atzori A Iera and G Morabito ldquoThe internet of things asurveyrdquoComputer Networks vol 54 no 15 pp 2787ndash2805 2010
[2] E Fleisch ldquoWhat is the internet of things An economicperspectiverdquo Tech Rep Auto-ID Labs 2010
[3] National Intelligence Council ldquoDisruptive civil technologiesmdashsix technologies with potential impacts on us interests out to2025rdquo Conference Report CR 2008-07 2008
[4] M R Palattella N Accettura X Vilajosana et al ldquoStandardizedprotocol stack for the internet of (important) thingsrdquo IEEECommunications Surveys and Tutorials vol 15 no 3 pp 1389ndash1406 2013
[5] M Zorzi A Gluhak S Lange and A Bassi ldquoFrom todaysINTRAnet of things to a future INTERnet of things a wireless-and mobility-related viewrdquo IEEEWireless Communications vol17 no 6 pp 44ndash51 2010
[6] A Gluhak S Krco M Nati D Pfisterer N Mitton andT Razafindralambo ldquoA survey on facilities for experimentalinternet of things researchrdquo IEEE Communications Magazinevol 49 no 11 pp 58ndash67 2011
[7] S Tozlu M Senel W Mao and A Keshavarzian ldquoWi-Fienabled sensors for internet of things a practical approachrdquoIEEE Communications Magazine vol 50 no 6 pp 134ndash1432012
[8] N Bui A P Castellani P Casari andM Zorzi ldquoThe internet ofenergy a web-enabled smart grid systemrdquo IEEE Network vol26 no 4 pp 39ndash45 2012
[9] X Fang S Misra G Xue and D Yang ldquoSmart gridmdashthe newand improved power grid a surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 14 no 4 pp 944ndash980 2012
[10] G Lu D De and W Song ldquoSmartGridLab a laboratory-basedsmart grid testbedrdquo in Proceedings of the 1st IEEE InternationalConference on Smart Grid Communications (SmartGridComm10) pp 143ndash148 Gaithersburg Md USA October 2010
[11] A-H Mohsenian-Rad V W S Wong J Jatskevich R Schoberand A Leon-Garcia ldquoAutonomous demand-side managementbased on game-theoretic energy consumption scheduling forthe future smart gridrdquo IEEE Transactions on Smart Grid vol1 no 3 pp 320ndash331 2010
[12] Z Zhu S Lambotharan W H Chin and Z Fan ldquoOverviewof demand management in smart grid and enabling wirelesscommunication technologiesrdquo IEEE Wireless Communicationsvol 19 no 3 pp 48ndash56 2012
[13] H T Nguyen D Nguyen and L B Le ldquoHome energy man-agement with generic thermal dynamics and user temperaturepreferencerdquo in Proceedings of the IEEE International Conferenceon Smart Grid Communications (SmartGridComm 13) pp 552ndash557 Vancouver Canada October 2013
[14] Z Zhu J Tang S Lambotharan W H Chin and Z FanldquoAn integer linear programming based optimization for homedemand-side management in smart gridrdquo in Proceedings of theIEEE PES Innovative Smart Grid Technologies (ISGT 12) pp 1ndash5Washington DC USA January 2012
[15] A-H Mohsenian-Rad and A Leon-Garcia ldquoOptimal residen-tial load control with price prediction in real-time electricitypricing environmentsrdquo IEEE Transactions on Smart Grid vol1 no 2 pp 120ndash133 2010
[16] K M Tsui and S C Chan ldquoDemand response optimizationfor smart home scheduling under real-time pricingrdquo IEEETransactions on Smart Grid vol 3 no 4 pp 1812ndash1821 2012
[17] GWood andMNewborough ldquoDynamic energy-consumptionindicators for domestic appliances environment behaviour anddesignrdquo Energy and Buildings vol 35 no 8 pp 821ndash841 2003
[18] M Avci M Erkoc and S S Asfour ldquoResidential HVAC loadcontrol strategy in real-time electricity pricing environmentrdquo inProceedings of the IEEE Energytech pp 1ndash6 Cleveland OhioUSA May 2012
[19] Z Q LuoW KMa A C So Y Ye and S Zhang ldquoSemidefiniterelaxation of quadratic optimization problemsrdquo IEEE SignalProcessing Magazine vol 27 no 3 pp 20ndash34 2010
[20] K F Fong V I Hanby and T T Chow ldquoHVAC systemoptimization for energymanagement by evolutionary program-mingrdquo Energy and Buildings vol 38 no 3 pp 220ndash231 2006
[21] D L Ha F F de Lamotte and Q-H Huynh ldquoReal-timedynamic multilevel optimization for demand-side load man-agementrdquo in Proceedings of the IEEE International Conferenceon Industrial Engineering and Engineering Management (IEEM07) pp 945ndash949 Singapore December 2007
[22] S Noh J Yun and K Kim ldquoAn efficient building air condi-tioning system control under real-time pricingrdquo in Proceedingsof the International Conference on Advanced Power SystemAutomation and Protection (APAP 11) pp 1283ndash1286 BeijingChina October 2011
[23] G L Nemhauser and L A Wolsey Integer and CombinatorialOptimization Wiley-Interscience New York NY USA 1988
[24] S Kay Fundamentals of Statistical Signal Processing EstimationTheory Prentice-Hall New York NY USA 1993
[25] ldquoContiki the open source OS for the internet of thingsrdquohttpwwwcontiki-osorg
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
2 The Scientific World Journal
the most energy demanding appliances in home buildings[17 18] According to studies they represent the 43 ofresidential energy consumption in the USA and the 61in UK and Canada [18] Apparently the significant energyconsumption of the HVAC systems along with their directinfluence on the userrsquos well-being highlights the necessityfor effective HVAC management algorithms that reducethe power consumption in the home buildings taking intoaccount the end-userrsquos comfort
In this paper we propose two HVAC energy schedulingmethods in an IoT framework where the users are ableto interact remotely with the HVAC control system Inparticular the users may retrieve information about thetemperature and the energy consumption at various spotsof the building under control while they are also able toremotely configure the temperature in given places Ourcontribution can be summarized as follows
(i) We propose a dynamic energy scheduler with com-fort constraints (DES-CC) which considers both thesmart pricing tariffs and the userrsquos comfort in orderto select the most energy efficient configuration ofHVACs that satisfies the userrsquos needs We formulatean optimization problem of HVAC control by pre-dicting the temperature that a given set of HVACmodules would cause in different locations We resultin a boolean quadratic optimization problem whichalthough not convex can be solved via an exhaustivesearch when the number of variables (ie HVACmodules in our case) is low In case that a large num-ber of HVAC modules are considered semidefiniterelaxation techniques can be applied [19]
(ii) Taking into account the energy efficiency priority wepropose a dynamic energy scheduler with comfortconstraints relaxation (DES-CCR) where the userrelaxes their comfort constraints allowing a higherdegree of flexibility for the system to further reducethe energy consumption In this case the problemis reformulated and the userrsquos comfort (ie temper-ature) is set as a penalty in the objective functioninstead of constraint
(iii) We have designed and developed a real testbed toevaluate the performance of the proposed algorithmsdemonstrating the potential financial and energygains that can be achieved
The remainder of the paper is organized as followsSection 2 provides a brief review of the related work in thisfield Section 3 describes the general network architectureand the system model under study Section 4 introducesthe two HVAC schedulers in a smart pricing and comfortconstraint context Section 5 provides the description ofthe testbed and the experimental results Finally Section 6concludes the paper
2 Related Work
The energy cost management of HVAC systems has recentlyattracted the research attention In [20] the energy cost is
studied as a function of the parameters that control the airand water subsystems and an evolutionary programmingmethod is proposed to save energy Moreover in [21] adynamic threshold controls the energy consumption and itvaries according to the user satisfaction which also dependson a thermal model However neither [20] nor [21] explicitlyconsider a dynamic pricing cost In [22] smart pricing isconsidered in the energy cost optimization but the usercomfort is not explicitly incorporated in the algorithm asthe authors consider that the HVAC is turned onoff whenthe indoor temperature is outside the margin of comfortRecently in [13 18] both energy scheduling of HVAC undersmart pricing and the user comfort are taken into accountIn [13] Nguyen et al propose the construction of a lookuptable (LUT) of room temperatures that depends on (i) thepast temperatures (ii) the outdoor temperature and (iii) theHVAC power The authors claim that the LUT is built duringa training period (that takes place only once) and permits toassess the temperature of comfort for a given operation of theHVAC energy scheduler However this heuristic approachseems hardly applicable in general scenarios In [18] a linearenergy cost function is considered although quadratic ortwo-step piecewise linear functions are more common inpractice [11] while userrsquos comfort is measured only at aspecific location It is also worth noting that none of theaforementioned works considers an IoT framework
Unlike [20ndash22] in our proposed energy schedulingmeth-ods both the smart pricing tariffs and the user comfort aretaken into account Moreover the temperature of comfortis measured at several building positions by different sensornodes that form a wireless sensor network (WSN) thusproviding a more accurate measure of comfort comparedto [18] Furthermore compared to [13] we adopt a moreanalytical and less heuristic model to assess the user comfortin the HVAC energy cost optimization In particular ourmodel considers the time varying nature of the real thermalconditions without requiring a training period Moreoverour model adaptively updates the past temperature measure-ments for each time period whereas the model in [13] is onlycarried out once to construct the LUT for particular indoorand outdoor conditions Finally unlike most of the abovereferences our methods are validated in a real scenario
3 Network Model
31 General Architecture Figure 1 presents the overall archi-tecture which is used to evaluate the proposed energyschedulers in an IoT context It consists of the followingelements
(i) a set of HVAC modules
(ii) a set of actuators that control the HVAC modules
(iii) a WSN which sends measurements of temperatureand energy consumption to a gateway
(iv) a gateway (GW) that incorporates the proposedenergy scheduling methods and connects the localnetwork to the Internet That is it contains a web
The Scientific World Journal 3
Gatewayenergy scheduler
web serverIP device
Wireless sensornetwork
Temperature sensors(Z1 motes)
Actuators
HVAC modules
Internet
Figure 1 Overall architecture of the proposed HVAC energy scheduler in the IoT context
Server
DatabaseEnergy
scheduler
Actuators
IP device
InternetWSN
Gatewayweb server
Data measurementUser requirementsEnergy scheduler decision
Figure 2 System model of the gateway
server and a database to store data received at the GWfrom the WSN or the internet
(v) an embedded IP device (eg tablet or smartphone)with an interface to interact with the HVAC energyscheduler It also displays both the temperature andthe energy consumption in the building measured bythe WSN
The functionality and flowof information of the proposedarchitecture is explained as follows The temperature ismeasured at several locations bymeans of theWSNThen themeasurements are periodically sent to the gateway where theenergy scheduling algorithm is implemented This algorithmselects the combination of the active HVAC modules thatminimizes the energy cost for given comfort constraints andenergy price during a particular time periodThese decisionsare sent through shell commands to programmable surgeprotectors (actuators) which actuate theHVACmodulesTheHVAC modules modify the room temperature according tothe decisions taken by the energy scheduler
Moreover the gateway hosts a database to store the mea-surements of temperature and energy consumption Thesemeasurements can be accessed by a remote Internet user
More specifically they are displayed at the userrsquos IP deviceas the gateway implements a web server which managesthe communication between the remote user and the localdatabase This is illustrated in more detail in Figure 2 wherethe connections between the most relevant blocks are shownFurthermore users are allowed to interact with the energyscheduler through their IP devices by setting the upper andlower bounds of the temperature of comfort
32 SystemModel Thesystemmodel for the proposedHVACenergy scheduler is depicted in more detail in Figure 3 Inparticular the energy scheduler is implemented within thegateway and it interacts with the following modules First aWSN composed of 119872 sensor nodes 119878
119894 1 ⩽ 119894 ⩽ 119872 which
sense the temperature and transmit the measurements to theenergy scheduler through aWSN sink node Second a set of119870HVACmodules that are controlled by the energy schedulerthrough a set of actuators119860
119896 1 ⩽ 119896 ⩽ 119870Moreover the inputs
that the energy scheduler requires are described as follows
(i) Themeasurements taken by theWSNnodes For eachtime interval 119873 measurements are taken by eachnode when a given configuration of HVAC modules
4 The Scientific World Journal
Uplink wiredUplink wirelessDownlink wiredExternal inputs
middot middot middot
SinkHVAC1 HVACK
OnoffOnoff
AKA1
Tmini
Tmaxi C(L(sj))
TmjMTm
j1S1 SM
Gateway
Figure 3 Detailed system model of the HVAC energy scheduler
is turned on These measurements are denoted by119879119898119895
119894(119899) (as illustrated by the black curve in Figure 4)
where 1 ⩽ 119894 ⩽ 119872 denotes the 119894th node and 1 ⩽ 119895 ⩽ 2119870
is the 119895th combination of HVACs turned on or off(ii) The energy cost function 119862(119871(s
119895)) which is specified
by the energy provider and depends on the smartpricing tariff According to [11] 119862(119871(s
119895)) can be
modeled as a quadratic function that is
119862 (119871 (s119895)) = 119901
1(119871 (s119895))2
+ 1199012119871 (s119895) + 1199013 (1)
where 1199011 1199012 and 119901
3are parameters that the provider
can dynamically vary in timeMoreover119871(s119895)denotes
the userrsquos energy consumption for the s119895combination
of HVACmodules turned on In order to describe theexpression for 119871(s
119895) let us define P isin R119870times2
119870
a matrixthat contains in its 119895th column the energy consump-tion of each HVAC module for the 119895th combinationof modules switched onoff For instance for 119870 = 3the matrix P is
P = (
0 1199091
0 0 11990911199091
0 1199091
0 0 1199092
0 1199092
0 11990921199092
0 0 0 1199093
0 119909311990931199093
) (2)
Moreover s119895is a vector of all zeros except in the 119895th
position that has value 1 Therefore Ps119895selects the
energy consumption related to the 119895th combinationof HVACs switched on and 1119879(Ps
119895) is the energy
consumption related to that combination where 1 isa vector of ones of length 2
119870 Therefore the totalenergy consumption for a given time period denotedby 119871(s
119895) can be expressed as
119871 (s119895) = 1198710+ 1119879 (Ps
119895) (3)
Tem
pera
ture
Time
2K predictedcurves
Past time intervalNsamples N samples
Next time interval
Current time
n + N
Tmaxi
Tmini
Tmji
Tpji
n
Figure 4 Prediction of temperature a fundamental step of theenergy scheduler to assess comfort in the future time interval
where 1198710
is the accumulated consumption and1119879(Ps
119895) is the consumption in the current time inter-
val decision
(iii) The constraints of temperature of comfort providedby the user For the energy scheduler proposed inSection 41 this corresponds to the minimum andmaximum allowed temperatures at the 119894th locationof the room which are denoted by 119879
min119894
and 119879max119894
with 1 ⩽ 119894 ⩽ 119872 respectively since we assumethat the user may specify the desired comfort at 119872different locations For the energy scheduler proposedin Section 42 the comfort is specified by the objectivetemperatures 119879
119906119894 1 ⩽ 119894 ⩽ 119872 which the user would
like to attain at the different119872 locations of the roomthough they may allow some relaxation in order tofurther reduce the energy consumption
To further clarify the operation of the proposed schemelet us shed light on the temporal behavior of the energyschedulers and the role of the temperature constraints onit Recall that the energy scheduler works in a time intervalbasis At the end of each time interval (ldquocurrent timerdquo inFigure 4) the energy scheduler must make a new decisionThat is it must decide which HVAC modules denoted byHVACk in Figure 3 will be active during the next timeinterval In order to make this decision the energy schedulershould predict which would be the temperature provokedby each configuration of HVACs As there are 119870 HVACmodules and we assume that they are either turned on oroff this corresponds to predict 2119870 curves of temperature asit is illustrated in Figure 4 These predicted temperatures aredenoted by 119879119901
119895
119894(119899) where 119894 and 119895 have the same meaning
as for the case of 119879119898119895119894(119899) explained above Finally on one
hand the DES-CC selects the configuration of HVACs thatminimizes the energy consumption cost 119862(119871(s
119895)) within the
bounds of comfort that is 119879min119894
⩽ 119879119901119895
119894(119899) ⩽ 119879
max119894
while theDES-CCR selects the HVAC configuration that optimizes thetradeoff between being closer to the comfort temperatures119879119906119894
and saving energy This selection is executed by the
The Scientific World Journal 5
actuators that control the HVACmodules which are denotedby 119860119896in Figure 3
4 HVAC Energy Scheduling
41 Dynamic Energy Scheduler with Comfort Constraints(DES-CC) Next we present the first of the two proposedHVAC energy schedulers To that end this section is dividedinto four parts First we formulate the energy scheduleras a constrained optimization problem Second recall thatfor each time interval the energy scheduler must decidethe combination of active HVACs to minimize the energyconsumption cost and fulfill the constraints of temperatureof comfort To assess these constraints the temperatureprovoked in the next time interval by each configuration ofHVACs turned on or off should be predicted Thereby thesecond part deals with a thermal model that paves the way topredict the future temperatures The third part specifies howto estimate the parameters of the prediction model thanks tothe measurements of temperature of the past time intervalFinally in the fourth part we summarize the proposed DES-CC algorithm
411 Formulation of DES-CC as a Constrained OptimizationProblem Theenergy scheduler works in a time interval basisWhen 119873 samples of temperature have been collected fromthe WSN at the119872 controlled locations the energy schedulermakes a new decision with respect to the state of the HVACsNamely for the next time interval the energy schedulerselects the optimal configuration of active HVACs Thisconfiguration on the one hand must minimize the energyconsumption cost while on the other hand it must respectthe comfort constraints that is it should lead to predictedtemperatures within the bounds of comfort According to thedefinitions of the system model this optimization problemmay be formulated mathematically as
minimizes119895isin012
119870times1
119862 (119871 (s119895))
subject to 119879min119894
⩽ 119879119901min119894
(s119895) 119894 isin [1119872]
119879max119894
⩾ 119879119901max119894
(s119895) 119894 isin [1119872]
1119879s119895= 1
(4)
where 119879119901min119894
(s119895) and 119879119901
max119894
(s119895) are the minimum and maxi-
mum predicted temperatures respectively at the 119894th locationfor the 119895th combination of HVAC modules turned on Giventhe definition of 119862(119871(s
119895)) in (1) as a quadratic function
the optimization problem (4) has the form of a quadraticprogramming but that the optimization variable is booleanHence it is a boolean quadratic programming problem Theproblems of this class are nonconvex and in general they canbe solved either by a fast method that finds a local solutionor by a slower method that finds the global solution Inour framework the number of HVAC modules 119870 is low ormoderate and the latter approach is preferred for examplethe branch and bound method [23] can be used In order
to proceed we need to model the predicted temperatures119879119901
min119894
(s119895) and 119879119901
max119894
(s119895)
412 Model to Predict the Temperatures of the Future TimeInterval Regarding the predicted temperatures 119879119901
min119894
(s119895)
and 119879119901max119894
(s119895) they can be expressed as
119879119901min119894
(s119895) =
∘q119879119894s119895
119879119901max119894
(s119895) = q119879119894s119895
(5)
where ∘q119894and q
119894are vectors that contain the minimum and
maximum predicted temperatures respectively for each ofthe possible combinations of operating HVAC modules Tofurther clarify let us define 119879119901119895
119894(119899) 2 le 119899 le 119873 the predicted
temperature at the 119899 time instant at the 119894th sensor for the 119895thcombination of HVAC modules turned on where 1 ⩽ 119894 ⩽ 119872
and 1 ⩽ 119895 ⩽ 2119870 Moreover let 119879119901119895
119894(119899min119895) and 119879119901
119895
119894(119899max119895)
be theminimum andmaximum temperatures among119879119901119895119894(119899)
2 le 119899 le 119873 Then ∘q119894and q
119894can be expressed as
∘q119879119894= [119879119901
1
119894(119899min1) 119879119901
2119870
119894(119899min
2119870)]
q119879119894= [119879119901
1
119894(119899max1) 119879119901
2119870
119894(119899max
2119870)]
(6)
In order to proceed a model for the predicted tempera-tures is necessary Intuitively the current temperature is cor-related with the past temperature and a given combination ofHVACs turned on causes a change in temperature Moreoverthe temperature dynamics are rather linear (at least locally)as it will be shown below Therefore the following model isproposed for the temperature prediction
119879119901119895
119894(119899) = 119886
119895
119894119879119901119895
119894(119899 minus 1) + 120574
119895
119894 2 le 119899 le 119873 (7)
where 119886119895119894and 120574119895119894model the relation with the past temperature
and the change of temperature provoked by the 119895th combina-tion of HVACs turned on respectively Observe that in thisexpression 119886119895
119894and 120574
119895
119894are unknown and must be estimated
from the past measurements
413 Estimation of the PredictionModel Parameters In orderto estimate 119886
119895
119894and 120574
119895
119894in (7) we assume that the past
measurements follow a model like (7) corrupted by noise
119879119898119895
119894(119899) = 119886
119895
119894119879119898119895
119894(119899 minus 1) + 120574
119895
119894+ 119908119895
119894(119899) 2 le 119899 le 119873 (8)
Note that the evolution of the temperature is consideredto be linear in (7) and (8) This is a valid assumption at leastfor short periods as the real experiments that we will presentin Section 5 will highlight
For the estimation of 119886119895
119894and 120574
119895
119894 two situations are
considered In the first one all the HVAC modules areswitched off and as a consequence only 119886119895
119894must be estimated
To that end least squares (LS) estimator is considered as noprobabilistic assumptions regarding the data are neededThis
6 The Scientific World Journal
Process For each time interval do(1) Input(a) The measurements of temperature of the previous interval(b) Value of 119901
11199012and 119901
3in the energy cost function (1)
(c) The userrsquos temperature of comfort constraints at eachlocation that is 119879min
119894 119879
max119894
119894 = 1 119872 in (4)(2) Estimation Step
Estimate 119886119895119894and 120574
119895
119894in the prediction model (7) using (9)
to (12) and the temperature measurements from the pasttime interval
(3) Prediction Step(a) Substitute the estimation of 119886119895
119894and 120574
119895
119894into the prediction model (7)
(b) Iterate this model to populate the vectors of minimumand maximum predicted temperatures (6)
(4)Optimization Step(a) Substitute the predicted temperatures (5) and the
quadratic cost function (1) into the optimization problem (4)(b) Solve (4) using Branch and bound method [23]
(5)OutputThe configuration of HVACs turned onoff that optimizes (4)
Algorithm 1 Dynamic energy scheduler with comfort constraints (DES-CC)
estimator minimizes the LS error criterion though it is notoptimal in general [24] Given (8) the LS estimation of 119886119895
119894
denoted by 119886119895119894is given by
119886119895
119894= xy (9)
where the symbol denotes the pseudoinverse operatorwhich is defined as x = (x119879x)minus1x119879 and we define x =
[119879119898119895
119894(1) 119879119898
119895
119894(119873 minus 1)]
119879
and y = [119879119898119895
119894(2) 119879119898
119895
119894(119873)]119879
The second situation is that some of the HVACmodules wereswitched on In this case an LS estimation is considered aswell Namely let us denote by 120574
119895
119894| 119886119895
119894the estimation of
120574119895
119894conditioned to the knowledge of a past estimation of 119886119895
119894
denoted by 119886119895
119894 Then the LS estimation for 120574119895
119894| 119886119895
119894yields
120574119895
119894| 119886119895
119894= 1z (10)
where 1 is a vector of ones of length119873 minus 1 and z is given by
z = [119879119898119895
119894(2) 119879119898
119895
119894(119873)]119879
minus 119886119895
119894[119879119898119895
119894(1) 119879119898
119895
119894(119873 minus 1)]
119879
(11)
Finally given the estimation of 120574119895119894in (10) we can update
the estimation of 119886119895119894as
119886119895
119894| 120574119895
119894= xy (12)
where y = [119879119898119895
119894(2) minus 120574
119895
119894 119879119898
119895
119894(119873) minus 120574
119895
119894]119879
414 Summary of the DES-CC Energy Scheduler At thispoint all the terms in the optimization problem understudy that is (4) are specified The procedure to implementthe proposed energy scheduler for each time interval issummarized in Algorithm 1
42 Dynamic Energy Scheduler with Comfort ConstraintsRelaxation (DES-CCR) Despite its effectiveness and its obvi-ous advantages the proposed energy scheduling algorithm iscompletely focused on the temperature constraints neglect-ing the price aspects of the problem More specificallyalthough there could be time periods where the energyprice increases the energy scheduler switches on the samecombination of HVAC modules in order to respect thetemperature constraints However in such cases users mightcompromise their comfort preferences to decrease the energyconsumption In order to allow the user to have more flexi-bility in the energy consumption a new energy scheduler willbe presented in this section This flexibility is implementedin terms of relaxing the temperature constraints to furtherreduce the energy consumption
This new energy scheduler is formulated so that theuser temperature constraints in (4) are skipped and they areincorporated as a penalty term in the objective functionConsequently the new optimization problem can be writtenas
minimizes119895isin012
119870times1
120579
119862 (119871 (s119895))
120572+ (1 minus 120579)
times
sum119872
119894=1sum119873
119899=2
10038171003817100381710038171003817q119879119894(119899)s119895minus 119879119906119894
10038171003817100381710038171003817
2
120573
(13)
where119862(119871(s119895)) is the energy cost function defined in (1)The
vector q119879119894(119899) is defined as
q119879119894(119899) = [119879119901
1
119894(119899) 119879119901
2119870
119894(119899)] (14)
and recall that 119879119901119895119894(119899) is the predicted temperature at the 119894th
location for the 119895th combination of HVACs modules turned
The Scientific World Journal 7
Process For each time interval do(1) Input(a) The measurements of temperature of the previous interval(b) Value of 119901
1 1199012and 119901
3of the energy cost function 119862(119871(119904
119895)) defined in (1)
(c) The userrsquos temperature of comfort at each location that is 119879119906119894 119894 = 1 119872 in (13)
(d) The parameter 120579 isin (0 1) controlling the comfort relaxation in (13)(2) Estimation Step
Estimate 119886119895119894and 120574
119895
119894in the prediction model (7) using (9)
to (12) and the temperature measurements from the past time interval(3) Prediction Step(a) Substitute the estimation of 119886119895
119894and 120574
119895
119894into the prediction model (7)
(b) For 119894 = 1 to119872For 119899 = 2 to119873Compute and store the vector of predicted temperaturesq119879119894(119899) in (14) using (7)
End For 119899End For 119894
(4)Optimization Step(a) Compute the cost function in (13) using the vectors in
the step 3(b) and the inputs in (1)(b) Solve the optimization problem in (13) using Branch and bound method [23]
(5)OutputThe configuration of HVACs turned onoff that optimizes (13)
Algorithm 2 Dynamic energy scheduler with comfort constraints relaxation (DES-CCR)
on or off see (7)Moreover120572 and120573 are normalizing constantsto adjust the values of the two terms in (13) Indeed we settheir value as
120572 = 119862 (119871 (s2119870))
120573 = maxs119895isin01119870times1
119872
sum
119894=1
119873
sum
119899=2
10038171003817100381710038171003817q119879119894(119899)s119895minus 119879119906119894
10038171003817100381710038171003817
2
(15)
where 119862(119871(s2119870)) is the cost for all the HVACmodules turned
on The term 119879119906119894
is the desired temperature that the userwould like tomaintain at the 119894th location of the room Clearlyour reformulation balances the two optimization problemsthat is the energy cost minimization and the user comfortmaximization Note that the user comfort is defined as anEuclidean norm but it can eventually be redefined withanother distance measurement
Finally 120579 isin (0 1) is defined by the user according to theirpreferences For example in the extreme case where 120579 = 0the demand response algorithm will not consider any priceand it will directly control the HVAC modules so that thedesired temperature is reached On the contrary when 120579 = 1theHVACmodules will always remain off In this context theusers should set the 120579 value according to their preferences andexperience The DES-CCR energy scheduler is summarizedin Algorithm 2
5 Experimental Results
In order to emulate the complete communication in anIoT framework we have designed and developed a customtestbed that integrates the described architecture In our
Figure 5 Z1 WSN mote
experiments we focus on a heating system although theproposed algorithms apply in general HVAC systems In thissection we describe the testbed platform and the experi-mental scenario we define a baseline thermostat model andfinally we present the experimental results of our proposedalgorithms
51 Testbed Description and Experimental Setup The testbedhas been deployed in a 50m2 room within our researchcenter facilities as it is depicted in Figure 5 In our particularscenario we consider three HVAC modules (ie 119870 = 3)and two temperature sensor nodes (ie 119872 = 2) The HVACmodules are distributed around the room while the sensornodes are placed in the middle of the room monitoring thetemperature and sending it to the sink mote every 30 second
8 The Scientific World Journal
Micro-USB
Ceramic embedded antennaUFL connector for external antenna
temperature sensor3-Axis accelerometer+
2 times Phidgets sensor ports
Figure 6 Overall platform detail
(ie 119905119898= 30 s) In addition the samples received from each
sensor are stored in a buffer at the control center and ouralgorithm applies every 10 samples (ie119873 = 10)
Regarding the employed technology the WSN nodes areZ1 motes by Zolertia (Figure 6) They are equipped with asecond generationMSP430F2617 low power microcontrollerwhich features a 16-bit RISC CPU 16MHz clock speed abuilt-in clock factory calibration an 8KB RAM and a 92KBflash memory They also include the CC2420 transceiverwhich is IEEE 802154 compliant operating at 24GHz fre-quency bandwith a data rate of 250 kbpsThe sensors supportContiki OS [25] an open-source operating system for the IoTwhich connects tiny low-cost low-power microcontrollersto the Internet and supports IPv6 through 6LowPAN It isworth noting that each mote can operate as either a sourceor a sink node In particular source nodes carry a TMP102temperature sensor to monitor the target field while the sinknode receives and forwards themeasured data to the gateway
The gateway (an Ubuntu OS machine with MATLAB)implements the proposed algorithms and it is able to processthe collected data Furthermore it connects the WSN to theInternet and acts as an application server using Nodejs andSencha Touch In particular Nodejs is a platform built onChromersquos JavaScript runtime for fast and scalable networkapplications Figure 7 shows a screenshot of the web applica-tion built on Nodejs which enables the user to interact withthe energy scheduler through Internet Regarding SenchaTouch it is a high-performance HTML5 mobile applicationframework which enables developers to build powerfulapplications for various operating systems including iOS andAndroidThe actuators are programmable sockets which canbe controlled remotely thanks to their IP addresses Thesespecial sockets are a set of programmable local area networksurge protectors (EG-PMS-LAN) by Energenie which areconnected via Ethernet to the gateway Finally the HVACmodules are domestic heaters with a maximum power con-sumption of 2000W
52 Baseline Model To evaluate the proposed algorithmsand highlight the potential energy and cost gains that can
Figure 7 Google Chrome screenshot of the web application
be achieved we adopt the traditional thermostat model asthe baseline reference scenario In this model the aim is tomaintain the average temperature of the room between acertain temperature range (ie [119879min119879max]) predefined bythe user To that end when the sensed temperature is above119879max at the end of a time interval all the heaters are switchedoff while the heaters are switched on when the temperaturefalls below the 119879min threshold
Figure 8 illustrates the average measured temperatureinside the room where the heaters are controlled by thethermostat In this particular case we consider 119879min =
21∘C and 119879max = 23
∘C As it can be seen in the figurethe thermostat algorithm is able to maintain the averagetemperature of the room between the desiredmargins during16 hours However it is worth noting that despite its properbehavior the particular model is not cost efficient as allHVAC modules work simultaneously consuming a totalpower consumption of 6000W In the following sectionswe evaluate our proposed methods demonstrating that theycan reduce the electrical cost with respect to the baselineapproach
The Scientific World Journal 9
21
22
23
24
HVA
C sta
tus
Average temperatureHVAC status
Day time
Tem
pera
ture
(∘C) On
Off
Thermostat-baseline model
18 20 22 24 2 4 6 8 10
Figure 8 Experimental evaluation of the thermostat model
53 Experimental Evaluation of the DES-CC in (4) Severalreal experiments have been carried out to assess the perfor-mance of the DES-CC algorithm proposed in (4) In thiscase the pricing parameters in (1) are 119901
1= 0003 and 119901
2=
1199013= 0 euros which are possible values according to [11] The
temperature bounds in (4) have been set to 119879min1
= 119879min2
=
21∘C and 119879
max1
= 119879max2
= 23∘C respectively
Figure 9 plots the variation of both the measured and theestimated temperature (using (8)ndash(12)) in the room duringour experiments As it can be noticed the error between theestimated and the real temperature is negligible somethingthat proves the accuracy of the proposed estimationmodel Inaddition the DES-CC guarantees the proper operation of thesystem as the temperature varies between the desired rangeof 21∘C and 23
∘C most of the time with very few exceptionsdue to prediction errors In the same figure it can be alsoseen that the temperature remains closer to the lower partof the permitted range (ie 21∘C) since the outcome of theproposed method provides a combination of switched onheaters that minimizes the energy consumption satisfyinga minimum acceptable temperature Indeed compared tothe temperature variation in the baseline scenario (Figure 8)DES-CC maintains the temperature more stable and in thelower part of the allowable region intuitively implying lowercost
Figure 10 depicts the financial operation cost gains thatcan be achieved by DES-CC compared to the baseline ther-mostat approach As it can be observed the proposed energyscheduler significantly reduces the energy cost leading to atotal save of 719 eurosmonth
54 Experimental Evaluation of the DES-CCR in (13) A setof experiments have been carried out for the evaluation ofthe DES-CCR in (13) Let us recall that DES-CCR relaxesthe temperature constraints by including the constraints asa penalized term in the objective function As a resultcompared to DES-CC this method is more flexible withrespect to real time pricing tariffs More specifically DES-CC seeks a combination that minimizes the energy cost withrespect to a minimum allowable temperature Consequently
WSN mote 2
WSN mote 1
18 20 22 24 2 4 6 8 1018192021222324
Day time
18 20 22 24 2 4 6 8 10Day time
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18192021222324
Tem
pera
ture
(∘C)
Figure 9 Real and estimated temperature using DES-CC
0 2 4 6 8 10 12 14 160
0005
001
0015
002
0025
003
0035
004
Hours
Ener
gy co
st (E
uros
)
Energy cost per hour of the proposed methodEnergy cost per hour of the thermostat
Figure 10 Energy consumption cost comparison between thermo-stat and DES-CC
although the energy cost may change during time the heatercombination selected byDES-CC is the same due to the stricttemperature constraint On the other hand DES-CCR allowsthe user to further reduce the energy consumption at the costof being outside the range of temperature of comfort In thiscase to highlight the flexibility of DES-CCR we have set a
10 The Scientific World Journal
0 50 100 150 200 25016
18
20
22
24
Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
0 50 100 150 200 25016
18
20
22
24
Sample measures
Tem
pera
ture
(∘C)
Figure 11 Real and estimated temperature using DES-CCR (120579 =
02)
periodically variable value of 1199011 which alternates between
1199011
= 0009 and 1199011
= 0003 euros every thirty minutesMoreover the desired temperature has been set to 119879
119906119894=
22∘CFigures 11 and 12 depict the temperature variation in
two different cases where the users give low (120579 = 02)and high (120579 = 05) priority respectively to reduce of theenergy consumption In particular in Figure 11 (120579 = 02)the achieved temperature is very close to the desired 119879
119906119894 On
the other hand in Figure 12 we assume 120579 = 05 which is amore adapted value to the pricing policy as it correspondsto a user that permits a relaxation of the difference betweenthe real and the desired temperature to reduce the energycost This fact implies higher energy consumption in lowcost zones and lower energy consumption in high costperiods sacrificing though the userrsquos comfort Therefore theexperiments confirm that the real temperature is close to 119879
119906119894
when the energy cost is lower (ie between samples 60 and120) while there is a noticeable temperature drop whichcorresponds to lower energy consumption
6 Concluding Remarks
This paper has dealt with the energy consumption manage-ment of HVACs for a given smart pricing tariff and usersrsquocomfort constraintsMoreover the integrationwithin the IoTframework has been studied To that end we developed areal testbed consisting of (i) heaters (ii) sensor nodes thatmeasure the temperature and (iii) a gateway which providesconnection to the Internet and includes aweb application that
18
19
20
21
22
0 50 100 150 200 250Sample measures
0 50 100 150 200 250Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18
19
20
21
22
Tem
pera
ture
(∘C)
Figure 12 Real and estimated temperature using DES-CCR (120579 =
05)
permits the interaction with the user through InternetMore-over the gateway implements the algorithms that control theenergy consumption Regarding the proposed methods firstwe devised an energy scheduler that optimizes the energycost in a time interval basis for a given energy price tariffand for a given set of temperature of comfort constraintsthat are associated with different locations inside a roomThen we proposed a more flexible energy scheduler whichrelaxes the temperature constraints to further reduce theenergy consumption Namely a new objective function hasbeen considered which consists of a convex combination ofthe energy cost and a penalty term that reflects the comfortThis permits to consider both the case where the user isvery concerned with the comfort and the case where heallows relaxing the comfort constraint to further reduce theenergy consumption Experimental evaluations have beencarried out in an isolated room validating our proposals andhighlighting their potential benefits
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work has been funded by the Energy-to-Smart Grid(E2SG) project httpwwwe2sg-projecteu within theENIAC joint undertaking framework with Grant agreementnumber 296131
The Scientific World Journal 11
References
[1] L Atzori A Iera and G Morabito ldquoThe internet of things asurveyrdquoComputer Networks vol 54 no 15 pp 2787ndash2805 2010
[2] E Fleisch ldquoWhat is the internet of things An economicperspectiverdquo Tech Rep Auto-ID Labs 2010
[3] National Intelligence Council ldquoDisruptive civil technologiesmdashsix technologies with potential impacts on us interests out to2025rdquo Conference Report CR 2008-07 2008
[4] M R Palattella N Accettura X Vilajosana et al ldquoStandardizedprotocol stack for the internet of (important) thingsrdquo IEEECommunications Surveys and Tutorials vol 15 no 3 pp 1389ndash1406 2013
[5] M Zorzi A Gluhak S Lange and A Bassi ldquoFrom todaysINTRAnet of things to a future INTERnet of things a wireless-and mobility-related viewrdquo IEEEWireless Communications vol17 no 6 pp 44ndash51 2010
[6] A Gluhak S Krco M Nati D Pfisterer N Mitton andT Razafindralambo ldquoA survey on facilities for experimentalinternet of things researchrdquo IEEE Communications Magazinevol 49 no 11 pp 58ndash67 2011
[7] S Tozlu M Senel W Mao and A Keshavarzian ldquoWi-Fienabled sensors for internet of things a practical approachrdquoIEEE Communications Magazine vol 50 no 6 pp 134ndash1432012
[8] N Bui A P Castellani P Casari andM Zorzi ldquoThe internet ofenergy a web-enabled smart grid systemrdquo IEEE Network vol26 no 4 pp 39ndash45 2012
[9] X Fang S Misra G Xue and D Yang ldquoSmart gridmdashthe newand improved power grid a surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 14 no 4 pp 944ndash980 2012
[10] G Lu D De and W Song ldquoSmartGridLab a laboratory-basedsmart grid testbedrdquo in Proceedings of the 1st IEEE InternationalConference on Smart Grid Communications (SmartGridComm10) pp 143ndash148 Gaithersburg Md USA October 2010
[11] A-H Mohsenian-Rad V W S Wong J Jatskevich R Schoberand A Leon-Garcia ldquoAutonomous demand-side managementbased on game-theoretic energy consumption scheduling forthe future smart gridrdquo IEEE Transactions on Smart Grid vol1 no 3 pp 320ndash331 2010
[12] Z Zhu S Lambotharan W H Chin and Z Fan ldquoOverviewof demand management in smart grid and enabling wirelesscommunication technologiesrdquo IEEE Wireless Communicationsvol 19 no 3 pp 48ndash56 2012
[13] H T Nguyen D Nguyen and L B Le ldquoHome energy man-agement with generic thermal dynamics and user temperaturepreferencerdquo in Proceedings of the IEEE International Conferenceon Smart Grid Communications (SmartGridComm 13) pp 552ndash557 Vancouver Canada October 2013
[14] Z Zhu J Tang S Lambotharan W H Chin and Z FanldquoAn integer linear programming based optimization for homedemand-side management in smart gridrdquo in Proceedings of theIEEE PES Innovative Smart Grid Technologies (ISGT 12) pp 1ndash5Washington DC USA January 2012
[15] A-H Mohsenian-Rad and A Leon-Garcia ldquoOptimal residen-tial load control with price prediction in real-time electricitypricing environmentsrdquo IEEE Transactions on Smart Grid vol1 no 2 pp 120ndash133 2010
[16] K M Tsui and S C Chan ldquoDemand response optimizationfor smart home scheduling under real-time pricingrdquo IEEETransactions on Smart Grid vol 3 no 4 pp 1812ndash1821 2012
[17] GWood andMNewborough ldquoDynamic energy-consumptionindicators for domestic appliances environment behaviour anddesignrdquo Energy and Buildings vol 35 no 8 pp 821ndash841 2003
[18] M Avci M Erkoc and S S Asfour ldquoResidential HVAC loadcontrol strategy in real-time electricity pricing environmentrdquo inProceedings of the IEEE Energytech pp 1ndash6 Cleveland OhioUSA May 2012
[19] Z Q LuoW KMa A C So Y Ye and S Zhang ldquoSemidefiniterelaxation of quadratic optimization problemsrdquo IEEE SignalProcessing Magazine vol 27 no 3 pp 20ndash34 2010
[20] K F Fong V I Hanby and T T Chow ldquoHVAC systemoptimization for energymanagement by evolutionary program-mingrdquo Energy and Buildings vol 38 no 3 pp 220ndash231 2006
[21] D L Ha F F de Lamotte and Q-H Huynh ldquoReal-timedynamic multilevel optimization for demand-side load man-agementrdquo in Proceedings of the IEEE International Conferenceon Industrial Engineering and Engineering Management (IEEM07) pp 945ndash949 Singapore December 2007
[22] S Noh J Yun and K Kim ldquoAn efficient building air condi-tioning system control under real-time pricingrdquo in Proceedingsof the International Conference on Advanced Power SystemAutomation and Protection (APAP 11) pp 1283ndash1286 BeijingChina October 2011
[23] G L Nemhauser and L A Wolsey Integer and CombinatorialOptimization Wiley-Interscience New York NY USA 1988
[24] S Kay Fundamentals of Statistical Signal Processing EstimationTheory Prentice-Hall New York NY USA 1993
[25] ldquoContiki the open source OS for the internet of thingsrdquohttpwwwcontiki-osorg
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World Journal 3
Gatewayenergy scheduler
web serverIP device
Wireless sensornetwork
Temperature sensors(Z1 motes)
Actuators
HVAC modules
Internet
Figure 1 Overall architecture of the proposed HVAC energy scheduler in the IoT context
Server
DatabaseEnergy
scheduler
Actuators
IP device
InternetWSN
Gatewayweb server
Data measurementUser requirementsEnergy scheduler decision
Figure 2 System model of the gateway
server and a database to store data received at the GWfrom the WSN or the internet
(v) an embedded IP device (eg tablet or smartphone)with an interface to interact with the HVAC energyscheduler It also displays both the temperature andthe energy consumption in the building measured bythe WSN
The functionality and flowof information of the proposedarchitecture is explained as follows The temperature ismeasured at several locations bymeans of theWSNThen themeasurements are periodically sent to the gateway where theenergy scheduling algorithm is implemented This algorithmselects the combination of the active HVAC modules thatminimizes the energy cost for given comfort constraints andenergy price during a particular time periodThese decisionsare sent through shell commands to programmable surgeprotectors (actuators) which actuate theHVACmodulesTheHVAC modules modify the room temperature according tothe decisions taken by the energy scheduler
Moreover the gateway hosts a database to store the mea-surements of temperature and energy consumption Thesemeasurements can be accessed by a remote Internet user
More specifically they are displayed at the userrsquos IP deviceas the gateway implements a web server which managesthe communication between the remote user and the localdatabase This is illustrated in more detail in Figure 2 wherethe connections between the most relevant blocks are shownFurthermore users are allowed to interact with the energyscheduler through their IP devices by setting the upper andlower bounds of the temperature of comfort
32 SystemModel Thesystemmodel for the proposedHVACenergy scheduler is depicted in more detail in Figure 3 Inparticular the energy scheduler is implemented within thegateway and it interacts with the following modules First aWSN composed of 119872 sensor nodes 119878
119894 1 ⩽ 119894 ⩽ 119872 which
sense the temperature and transmit the measurements to theenergy scheduler through aWSN sink node Second a set of119870HVACmodules that are controlled by the energy schedulerthrough a set of actuators119860
119896 1 ⩽ 119896 ⩽ 119870Moreover the inputs
that the energy scheduler requires are described as follows
(i) Themeasurements taken by theWSNnodes For eachtime interval 119873 measurements are taken by eachnode when a given configuration of HVAC modules
4 The Scientific World Journal
Uplink wiredUplink wirelessDownlink wiredExternal inputs
middot middot middot
SinkHVAC1 HVACK
OnoffOnoff
AKA1
Tmini
Tmaxi C(L(sj))
TmjMTm
j1S1 SM
Gateway
Figure 3 Detailed system model of the HVAC energy scheduler
is turned on These measurements are denoted by119879119898119895
119894(119899) (as illustrated by the black curve in Figure 4)
where 1 ⩽ 119894 ⩽ 119872 denotes the 119894th node and 1 ⩽ 119895 ⩽ 2119870
is the 119895th combination of HVACs turned on or off(ii) The energy cost function 119862(119871(s
119895)) which is specified
by the energy provider and depends on the smartpricing tariff According to [11] 119862(119871(s
119895)) can be
modeled as a quadratic function that is
119862 (119871 (s119895)) = 119901
1(119871 (s119895))2
+ 1199012119871 (s119895) + 1199013 (1)
where 1199011 1199012 and 119901
3are parameters that the provider
can dynamically vary in timeMoreover119871(s119895)denotes
the userrsquos energy consumption for the s119895combination
of HVACmodules turned on In order to describe theexpression for 119871(s
119895) let us define P isin R119870times2
119870
a matrixthat contains in its 119895th column the energy consump-tion of each HVAC module for the 119895th combinationof modules switched onoff For instance for 119870 = 3the matrix P is
P = (
0 1199091
0 0 11990911199091
0 1199091
0 0 1199092
0 1199092
0 11990921199092
0 0 0 1199093
0 119909311990931199093
) (2)
Moreover s119895is a vector of all zeros except in the 119895th
position that has value 1 Therefore Ps119895selects the
energy consumption related to the 119895th combinationof HVACs switched on and 1119879(Ps
119895) is the energy
consumption related to that combination where 1 isa vector of ones of length 2
119870 Therefore the totalenergy consumption for a given time period denotedby 119871(s
119895) can be expressed as
119871 (s119895) = 1198710+ 1119879 (Ps
119895) (3)
Tem
pera
ture
Time
2K predictedcurves
Past time intervalNsamples N samples
Next time interval
Current time
n + N
Tmaxi
Tmini
Tmji
Tpji
n
Figure 4 Prediction of temperature a fundamental step of theenergy scheduler to assess comfort in the future time interval
where 1198710
is the accumulated consumption and1119879(Ps
119895) is the consumption in the current time inter-
val decision
(iii) The constraints of temperature of comfort providedby the user For the energy scheduler proposed inSection 41 this corresponds to the minimum andmaximum allowed temperatures at the 119894th locationof the room which are denoted by 119879
min119894
and 119879max119894
with 1 ⩽ 119894 ⩽ 119872 respectively since we assumethat the user may specify the desired comfort at 119872different locations For the energy scheduler proposedin Section 42 the comfort is specified by the objectivetemperatures 119879
119906119894 1 ⩽ 119894 ⩽ 119872 which the user would
like to attain at the different119872 locations of the roomthough they may allow some relaxation in order tofurther reduce the energy consumption
To further clarify the operation of the proposed schemelet us shed light on the temporal behavior of the energyschedulers and the role of the temperature constraints onit Recall that the energy scheduler works in a time intervalbasis At the end of each time interval (ldquocurrent timerdquo inFigure 4) the energy scheduler must make a new decisionThat is it must decide which HVAC modules denoted byHVACk in Figure 3 will be active during the next timeinterval In order to make this decision the energy schedulershould predict which would be the temperature provokedby each configuration of HVACs As there are 119870 HVACmodules and we assume that they are either turned on oroff this corresponds to predict 2119870 curves of temperature asit is illustrated in Figure 4 These predicted temperatures aredenoted by 119879119901
119895
119894(119899) where 119894 and 119895 have the same meaning
as for the case of 119879119898119895119894(119899) explained above Finally on one
hand the DES-CC selects the configuration of HVACs thatminimizes the energy consumption cost 119862(119871(s
119895)) within the
bounds of comfort that is 119879min119894
⩽ 119879119901119895
119894(119899) ⩽ 119879
max119894
while theDES-CCR selects the HVAC configuration that optimizes thetradeoff between being closer to the comfort temperatures119879119906119894
and saving energy This selection is executed by the
The Scientific World Journal 5
actuators that control the HVACmodules which are denotedby 119860119896in Figure 3
4 HVAC Energy Scheduling
41 Dynamic Energy Scheduler with Comfort Constraints(DES-CC) Next we present the first of the two proposedHVAC energy schedulers To that end this section is dividedinto four parts First we formulate the energy scheduleras a constrained optimization problem Second recall thatfor each time interval the energy scheduler must decidethe combination of active HVACs to minimize the energyconsumption cost and fulfill the constraints of temperatureof comfort To assess these constraints the temperatureprovoked in the next time interval by each configuration ofHVACs turned on or off should be predicted Thereby thesecond part deals with a thermal model that paves the way topredict the future temperatures The third part specifies howto estimate the parameters of the prediction model thanks tothe measurements of temperature of the past time intervalFinally in the fourth part we summarize the proposed DES-CC algorithm
411 Formulation of DES-CC as a Constrained OptimizationProblem Theenergy scheduler works in a time interval basisWhen 119873 samples of temperature have been collected fromthe WSN at the119872 controlled locations the energy schedulermakes a new decision with respect to the state of the HVACsNamely for the next time interval the energy schedulerselects the optimal configuration of active HVACs Thisconfiguration on the one hand must minimize the energyconsumption cost while on the other hand it must respectthe comfort constraints that is it should lead to predictedtemperatures within the bounds of comfort According to thedefinitions of the system model this optimization problemmay be formulated mathematically as
minimizes119895isin012
119870times1
119862 (119871 (s119895))
subject to 119879min119894
⩽ 119879119901min119894
(s119895) 119894 isin [1119872]
119879max119894
⩾ 119879119901max119894
(s119895) 119894 isin [1119872]
1119879s119895= 1
(4)
where 119879119901min119894
(s119895) and 119879119901
max119894
(s119895) are the minimum and maxi-
mum predicted temperatures respectively at the 119894th locationfor the 119895th combination of HVAC modules turned on Giventhe definition of 119862(119871(s
119895)) in (1) as a quadratic function
the optimization problem (4) has the form of a quadraticprogramming but that the optimization variable is booleanHence it is a boolean quadratic programming problem Theproblems of this class are nonconvex and in general they canbe solved either by a fast method that finds a local solutionor by a slower method that finds the global solution Inour framework the number of HVAC modules 119870 is low ormoderate and the latter approach is preferred for examplethe branch and bound method [23] can be used In order
to proceed we need to model the predicted temperatures119879119901
min119894
(s119895) and 119879119901
max119894
(s119895)
412 Model to Predict the Temperatures of the Future TimeInterval Regarding the predicted temperatures 119879119901
min119894
(s119895)
and 119879119901max119894
(s119895) they can be expressed as
119879119901min119894
(s119895) =
∘q119879119894s119895
119879119901max119894
(s119895) = q119879119894s119895
(5)
where ∘q119894and q
119894are vectors that contain the minimum and
maximum predicted temperatures respectively for each ofthe possible combinations of operating HVAC modules Tofurther clarify let us define 119879119901119895
119894(119899) 2 le 119899 le 119873 the predicted
temperature at the 119899 time instant at the 119894th sensor for the 119895thcombination of HVAC modules turned on where 1 ⩽ 119894 ⩽ 119872
and 1 ⩽ 119895 ⩽ 2119870 Moreover let 119879119901119895
119894(119899min119895) and 119879119901
119895
119894(119899max119895)
be theminimum andmaximum temperatures among119879119901119895119894(119899)
2 le 119899 le 119873 Then ∘q119894and q
119894can be expressed as
∘q119879119894= [119879119901
1
119894(119899min1) 119879119901
2119870
119894(119899min
2119870)]
q119879119894= [119879119901
1
119894(119899max1) 119879119901
2119870
119894(119899max
2119870)]
(6)
In order to proceed a model for the predicted tempera-tures is necessary Intuitively the current temperature is cor-related with the past temperature and a given combination ofHVACs turned on causes a change in temperature Moreoverthe temperature dynamics are rather linear (at least locally)as it will be shown below Therefore the following model isproposed for the temperature prediction
119879119901119895
119894(119899) = 119886
119895
119894119879119901119895
119894(119899 minus 1) + 120574
119895
119894 2 le 119899 le 119873 (7)
where 119886119895119894and 120574119895119894model the relation with the past temperature
and the change of temperature provoked by the 119895th combina-tion of HVACs turned on respectively Observe that in thisexpression 119886119895
119894and 120574
119895
119894are unknown and must be estimated
from the past measurements
413 Estimation of the PredictionModel Parameters In orderto estimate 119886
119895
119894and 120574
119895
119894in (7) we assume that the past
measurements follow a model like (7) corrupted by noise
119879119898119895
119894(119899) = 119886
119895
119894119879119898119895
119894(119899 minus 1) + 120574
119895
119894+ 119908119895
119894(119899) 2 le 119899 le 119873 (8)
Note that the evolution of the temperature is consideredto be linear in (7) and (8) This is a valid assumption at leastfor short periods as the real experiments that we will presentin Section 5 will highlight
For the estimation of 119886119895
119894and 120574
119895
119894 two situations are
considered In the first one all the HVAC modules areswitched off and as a consequence only 119886119895
119894must be estimated
To that end least squares (LS) estimator is considered as noprobabilistic assumptions regarding the data are neededThis
6 The Scientific World Journal
Process For each time interval do(1) Input(a) The measurements of temperature of the previous interval(b) Value of 119901
11199012and 119901
3in the energy cost function (1)
(c) The userrsquos temperature of comfort constraints at eachlocation that is 119879min
119894 119879
max119894
119894 = 1 119872 in (4)(2) Estimation Step
Estimate 119886119895119894and 120574
119895
119894in the prediction model (7) using (9)
to (12) and the temperature measurements from the pasttime interval
(3) Prediction Step(a) Substitute the estimation of 119886119895
119894and 120574
119895
119894into the prediction model (7)
(b) Iterate this model to populate the vectors of minimumand maximum predicted temperatures (6)
(4)Optimization Step(a) Substitute the predicted temperatures (5) and the
quadratic cost function (1) into the optimization problem (4)(b) Solve (4) using Branch and bound method [23]
(5)OutputThe configuration of HVACs turned onoff that optimizes (4)
Algorithm 1 Dynamic energy scheduler with comfort constraints (DES-CC)
estimator minimizes the LS error criterion though it is notoptimal in general [24] Given (8) the LS estimation of 119886119895
119894
denoted by 119886119895119894is given by
119886119895
119894= xy (9)
where the symbol denotes the pseudoinverse operatorwhich is defined as x = (x119879x)minus1x119879 and we define x =
[119879119898119895
119894(1) 119879119898
119895
119894(119873 minus 1)]
119879
and y = [119879119898119895
119894(2) 119879119898
119895
119894(119873)]119879
The second situation is that some of the HVACmodules wereswitched on In this case an LS estimation is considered aswell Namely let us denote by 120574
119895
119894| 119886119895
119894the estimation of
120574119895
119894conditioned to the knowledge of a past estimation of 119886119895
119894
denoted by 119886119895
119894 Then the LS estimation for 120574119895
119894| 119886119895
119894yields
120574119895
119894| 119886119895
119894= 1z (10)
where 1 is a vector of ones of length119873 minus 1 and z is given by
z = [119879119898119895
119894(2) 119879119898
119895
119894(119873)]119879
minus 119886119895
119894[119879119898119895
119894(1) 119879119898
119895
119894(119873 minus 1)]
119879
(11)
Finally given the estimation of 120574119895119894in (10) we can update
the estimation of 119886119895119894as
119886119895
119894| 120574119895
119894= xy (12)
where y = [119879119898119895
119894(2) minus 120574
119895
119894 119879119898
119895
119894(119873) minus 120574
119895
119894]119879
414 Summary of the DES-CC Energy Scheduler At thispoint all the terms in the optimization problem understudy that is (4) are specified The procedure to implementthe proposed energy scheduler for each time interval issummarized in Algorithm 1
42 Dynamic Energy Scheduler with Comfort ConstraintsRelaxation (DES-CCR) Despite its effectiveness and its obvi-ous advantages the proposed energy scheduling algorithm iscompletely focused on the temperature constraints neglect-ing the price aspects of the problem More specificallyalthough there could be time periods where the energyprice increases the energy scheduler switches on the samecombination of HVAC modules in order to respect thetemperature constraints However in such cases users mightcompromise their comfort preferences to decrease the energyconsumption In order to allow the user to have more flexi-bility in the energy consumption a new energy scheduler willbe presented in this section This flexibility is implementedin terms of relaxing the temperature constraints to furtherreduce the energy consumption
This new energy scheduler is formulated so that theuser temperature constraints in (4) are skipped and they areincorporated as a penalty term in the objective functionConsequently the new optimization problem can be writtenas
minimizes119895isin012
119870times1
120579
119862 (119871 (s119895))
120572+ (1 minus 120579)
times
sum119872
119894=1sum119873
119899=2
10038171003817100381710038171003817q119879119894(119899)s119895minus 119879119906119894
10038171003817100381710038171003817
2
120573
(13)
where119862(119871(s119895)) is the energy cost function defined in (1)The
vector q119879119894(119899) is defined as
q119879119894(119899) = [119879119901
1
119894(119899) 119879119901
2119870
119894(119899)] (14)
and recall that 119879119901119895119894(119899) is the predicted temperature at the 119894th
location for the 119895th combination of HVACs modules turned
The Scientific World Journal 7
Process For each time interval do(1) Input(a) The measurements of temperature of the previous interval(b) Value of 119901
1 1199012and 119901
3of the energy cost function 119862(119871(119904
119895)) defined in (1)
(c) The userrsquos temperature of comfort at each location that is 119879119906119894 119894 = 1 119872 in (13)
(d) The parameter 120579 isin (0 1) controlling the comfort relaxation in (13)(2) Estimation Step
Estimate 119886119895119894and 120574
119895
119894in the prediction model (7) using (9)
to (12) and the temperature measurements from the past time interval(3) Prediction Step(a) Substitute the estimation of 119886119895
119894and 120574
119895
119894into the prediction model (7)
(b) For 119894 = 1 to119872For 119899 = 2 to119873Compute and store the vector of predicted temperaturesq119879119894(119899) in (14) using (7)
End For 119899End For 119894
(4)Optimization Step(a) Compute the cost function in (13) using the vectors in
the step 3(b) and the inputs in (1)(b) Solve the optimization problem in (13) using Branch and bound method [23]
(5)OutputThe configuration of HVACs turned onoff that optimizes (13)
Algorithm 2 Dynamic energy scheduler with comfort constraints relaxation (DES-CCR)
on or off see (7)Moreover120572 and120573 are normalizing constantsto adjust the values of the two terms in (13) Indeed we settheir value as
120572 = 119862 (119871 (s2119870))
120573 = maxs119895isin01119870times1
119872
sum
119894=1
119873
sum
119899=2
10038171003817100381710038171003817q119879119894(119899)s119895minus 119879119906119894
10038171003817100381710038171003817
2
(15)
where 119862(119871(s2119870)) is the cost for all the HVACmodules turned
on The term 119879119906119894
is the desired temperature that the userwould like tomaintain at the 119894th location of the room Clearlyour reformulation balances the two optimization problemsthat is the energy cost minimization and the user comfortmaximization Note that the user comfort is defined as anEuclidean norm but it can eventually be redefined withanother distance measurement
Finally 120579 isin (0 1) is defined by the user according to theirpreferences For example in the extreme case where 120579 = 0the demand response algorithm will not consider any priceand it will directly control the HVAC modules so that thedesired temperature is reached On the contrary when 120579 = 1theHVACmodules will always remain off In this context theusers should set the 120579 value according to their preferences andexperience The DES-CCR energy scheduler is summarizedin Algorithm 2
5 Experimental Results
In order to emulate the complete communication in anIoT framework we have designed and developed a customtestbed that integrates the described architecture In our
Figure 5 Z1 WSN mote
experiments we focus on a heating system although theproposed algorithms apply in general HVAC systems In thissection we describe the testbed platform and the experi-mental scenario we define a baseline thermostat model andfinally we present the experimental results of our proposedalgorithms
51 Testbed Description and Experimental Setup The testbedhas been deployed in a 50m2 room within our researchcenter facilities as it is depicted in Figure 5 In our particularscenario we consider three HVAC modules (ie 119870 = 3)and two temperature sensor nodes (ie 119872 = 2) The HVACmodules are distributed around the room while the sensornodes are placed in the middle of the room monitoring thetemperature and sending it to the sink mote every 30 second
8 The Scientific World Journal
Micro-USB
Ceramic embedded antennaUFL connector for external antenna
temperature sensor3-Axis accelerometer+
2 times Phidgets sensor ports
Figure 6 Overall platform detail
(ie 119905119898= 30 s) In addition the samples received from each
sensor are stored in a buffer at the control center and ouralgorithm applies every 10 samples (ie119873 = 10)
Regarding the employed technology the WSN nodes areZ1 motes by Zolertia (Figure 6) They are equipped with asecond generationMSP430F2617 low power microcontrollerwhich features a 16-bit RISC CPU 16MHz clock speed abuilt-in clock factory calibration an 8KB RAM and a 92KBflash memory They also include the CC2420 transceiverwhich is IEEE 802154 compliant operating at 24GHz fre-quency bandwith a data rate of 250 kbpsThe sensors supportContiki OS [25] an open-source operating system for the IoTwhich connects tiny low-cost low-power microcontrollersto the Internet and supports IPv6 through 6LowPAN It isworth noting that each mote can operate as either a sourceor a sink node In particular source nodes carry a TMP102temperature sensor to monitor the target field while the sinknode receives and forwards themeasured data to the gateway
The gateway (an Ubuntu OS machine with MATLAB)implements the proposed algorithms and it is able to processthe collected data Furthermore it connects the WSN to theInternet and acts as an application server using Nodejs andSencha Touch In particular Nodejs is a platform built onChromersquos JavaScript runtime for fast and scalable networkapplications Figure 7 shows a screenshot of the web applica-tion built on Nodejs which enables the user to interact withthe energy scheduler through Internet Regarding SenchaTouch it is a high-performance HTML5 mobile applicationframework which enables developers to build powerfulapplications for various operating systems including iOS andAndroidThe actuators are programmable sockets which canbe controlled remotely thanks to their IP addresses Thesespecial sockets are a set of programmable local area networksurge protectors (EG-PMS-LAN) by Energenie which areconnected via Ethernet to the gateway Finally the HVACmodules are domestic heaters with a maximum power con-sumption of 2000W
52 Baseline Model To evaluate the proposed algorithmsand highlight the potential energy and cost gains that can
Figure 7 Google Chrome screenshot of the web application
be achieved we adopt the traditional thermostat model asthe baseline reference scenario In this model the aim is tomaintain the average temperature of the room between acertain temperature range (ie [119879min119879max]) predefined bythe user To that end when the sensed temperature is above119879max at the end of a time interval all the heaters are switchedoff while the heaters are switched on when the temperaturefalls below the 119879min threshold
Figure 8 illustrates the average measured temperatureinside the room where the heaters are controlled by thethermostat In this particular case we consider 119879min =
21∘C and 119879max = 23
∘C As it can be seen in the figurethe thermostat algorithm is able to maintain the averagetemperature of the room between the desiredmargins during16 hours However it is worth noting that despite its properbehavior the particular model is not cost efficient as allHVAC modules work simultaneously consuming a totalpower consumption of 6000W In the following sectionswe evaluate our proposed methods demonstrating that theycan reduce the electrical cost with respect to the baselineapproach
The Scientific World Journal 9
21
22
23
24
HVA
C sta
tus
Average temperatureHVAC status
Day time
Tem
pera
ture
(∘C) On
Off
Thermostat-baseline model
18 20 22 24 2 4 6 8 10
Figure 8 Experimental evaluation of the thermostat model
53 Experimental Evaluation of the DES-CC in (4) Severalreal experiments have been carried out to assess the perfor-mance of the DES-CC algorithm proposed in (4) In thiscase the pricing parameters in (1) are 119901
1= 0003 and 119901
2=
1199013= 0 euros which are possible values according to [11] The
temperature bounds in (4) have been set to 119879min1
= 119879min2
=
21∘C and 119879
max1
= 119879max2
= 23∘C respectively
Figure 9 plots the variation of both the measured and theestimated temperature (using (8)ndash(12)) in the room duringour experiments As it can be noticed the error between theestimated and the real temperature is negligible somethingthat proves the accuracy of the proposed estimationmodel Inaddition the DES-CC guarantees the proper operation of thesystem as the temperature varies between the desired rangeof 21∘C and 23
∘C most of the time with very few exceptionsdue to prediction errors In the same figure it can be alsoseen that the temperature remains closer to the lower partof the permitted range (ie 21∘C) since the outcome of theproposed method provides a combination of switched onheaters that minimizes the energy consumption satisfyinga minimum acceptable temperature Indeed compared tothe temperature variation in the baseline scenario (Figure 8)DES-CC maintains the temperature more stable and in thelower part of the allowable region intuitively implying lowercost
Figure 10 depicts the financial operation cost gains thatcan be achieved by DES-CC compared to the baseline ther-mostat approach As it can be observed the proposed energyscheduler significantly reduces the energy cost leading to atotal save of 719 eurosmonth
54 Experimental Evaluation of the DES-CCR in (13) A setof experiments have been carried out for the evaluation ofthe DES-CCR in (13) Let us recall that DES-CCR relaxesthe temperature constraints by including the constraints asa penalized term in the objective function As a resultcompared to DES-CC this method is more flexible withrespect to real time pricing tariffs More specifically DES-CC seeks a combination that minimizes the energy cost withrespect to a minimum allowable temperature Consequently
WSN mote 2
WSN mote 1
18 20 22 24 2 4 6 8 1018192021222324
Day time
18 20 22 24 2 4 6 8 10Day time
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18192021222324
Tem
pera
ture
(∘C)
Figure 9 Real and estimated temperature using DES-CC
0 2 4 6 8 10 12 14 160
0005
001
0015
002
0025
003
0035
004
Hours
Ener
gy co
st (E
uros
)
Energy cost per hour of the proposed methodEnergy cost per hour of the thermostat
Figure 10 Energy consumption cost comparison between thermo-stat and DES-CC
although the energy cost may change during time the heatercombination selected byDES-CC is the same due to the stricttemperature constraint On the other hand DES-CCR allowsthe user to further reduce the energy consumption at the costof being outside the range of temperature of comfort In thiscase to highlight the flexibility of DES-CCR we have set a
10 The Scientific World Journal
0 50 100 150 200 25016
18
20
22
24
Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
0 50 100 150 200 25016
18
20
22
24
Sample measures
Tem
pera
ture
(∘C)
Figure 11 Real and estimated temperature using DES-CCR (120579 =
02)
periodically variable value of 1199011 which alternates between
1199011
= 0009 and 1199011
= 0003 euros every thirty minutesMoreover the desired temperature has been set to 119879
119906119894=
22∘CFigures 11 and 12 depict the temperature variation in
two different cases where the users give low (120579 = 02)and high (120579 = 05) priority respectively to reduce of theenergy consumption In particular in Figure 11 (120579 = 02)the achieved temperature is very close to the desired 119879
119906119894 On
the other hand in Figure 12 we assume 120579 = 05 which is amore adapted value to the pricing policy as it correspondsto a user that permits a relaxation of the difference betweenthe real and the desired temperature to reduce the energycost This fact implies higher energy consumption in lowcost zones and lower energy consumption in high costperiods sacrificing though the userrsquos comfort Therefore theexperiments confirm that the real temperature is close to 119879
119906119894
when the energy cost is lower (ie between samples 60 and120) while there is a noticeable temperature drop whichcorresponds to lower energy consumption
6 Concluding Remarks
This paper has dealt with the energy consumption manage-ment of HVACs for a given smart pricing tariff and usersrsquocomfort constraintsMoreover the integrationwithin the IoTframework has been studied To that end we developed areal testbed consisting of (i) heaters (ii) sensor nodes thatmeasure the temperature and (iii) a gateway which providesconnection to the Internet and includes aweb application that
18
19
20
21
22
0 50 100 150 200 250Sample measures
0 50 100 150 200 250Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18
19
20
21
22
Tem
pera
ture
(∘C)
Figure 12 Real and estimated temperature using DES-CCR (120579 =
05)
permits the interaction with the user through InternetMore-over the gateway implements the algorithms that control theenergy consumption Regarding the proposed methods firstwe devised an energy scheduler that optimizes the energycost in a time interval basis for a given energy price tariffand for a given set of temperature of comfort constraintsthat are associated with different locations inside a roomThen we proposed a more flexible energy scheduler whichrelaxes the temperature constraints to further reduce theenergy consumption Namely a new objective function hasbeen considered which consists of a convex combination ofthe energy cost and a penalty term that reflects the comfortThis permits to consider both the case where the user isvery concerned with the comfort and the case where heallows relaxing the comfort constraint to further reduce theenergy consumption Experimental evaluations have beencarried out in an isolated room validating our proposals andhighlighting their potential benefits
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work has been funded by the Energy-to-Smart Grid(E2SG) project httpwwwe2sg-projecteu within theENIAC joint undertaking framework with Grant agreementnumber 296131
The Scientific World Journal 11
References
[1] L Atzori A Iera and G Morabito ldquoThe internet of things asurveyrdquoComputer Networks vol 54 no 15 pp 2787ndash2805 2010
[2] E Fleisch ldquoWhat is the internet of things An economicperspectiverdquo Tech Rep Auto-ID Labs 2010
[3] National Intelligence Council ldquoDisruptive civil technologiesmdashsix technologies with potential impacts on us interests out to2025rdquo Conference Report CR 2008-07 2008
[4] M R Palattella N Accettura X Vilajosana et al ldquoStandardizedprotocol stack for the internet of (important) thingsrdquo IEEECommunications Surveys and Tutorials vol 15 no 3 pp 1389ndash1406 2013
[5] M Zorzi A Gluhak S Lange and A Bassi ldquoFrom todaysINTRAnet of things to a future INTERnet of things a wireless-and mobility-related viewrdquo IEEEWireless Communications vol17 no 6 pp 44ndash51 2010
[6] A Gluhak S Krco M Nati D Pfisterer N Mitton andT Razafindralambo ldquoA survey on facilities for experimentalinternet of things researchrdquo IEEE Communications Magazinevol 49 no 11 pp 58ndash67 2011
[7] S Tozlu M Senel W Mao and A Keshavarzian ldquoWi-Fienabled sensors for internet of things a practical approachrdquoIEEE Communications Magazine vol 50 no 6 pp 134ndash1432012
[8] N Bui A P Castellani P Casari andM Zorzi ldquoThe internet ofenergy a web-enabled smart grid systemrdquo IEEE Network vol26 no 4 pp 39ndash45 2012
[9] X Fang S Misra G Xue and D Yang ldquoSmart gridmdashthe newand improved power grid a surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 14 no 4 pp 944ndash980 2012
[10] G Lu D De and W Song ldquoSmartGridLab a laboratory-basedsmart grid testbedrdquo in Proceedings of the 1st IEEE InternationalConference on Smart Grid Communications (SmartGridComm10) pp 143ndash148 Gaithersburg Md USA October 2010
[11] A-H Mohsenian-Rad V W S Wong J Jatskevich R Schoberand A Leon-Garcia ldquoAutonomous demand-side managementbased on game-theoretic energy consumption scheduling forthe future smart gridrdquo IEEE Transactions on Smart Grid vol1 no 3 pp 320ndash331 2010
[12] Z Zhu S Lambotharan W H Chin and Z Fan ldquoOverviewof demand management in smart grid and enabling wirelesscommunication technologiesrdquo IEEE Wireless Communicationsvol 19 no 3 pp 48ndash56 2012
[13] H T Nguyen D Nguyen and L B Le ldquoHome energy man-agement with generic thermal dynamics and user temperaturepreferencerdquo in Proceedings of the IEEE International Conferenceon Smart Grid Communications (SmartGridComm 13) pp 552ndash557 Vancouver Canada October 2013
[14] Z Zhu J Tang S Lambotharan W H Chin and Z FanldquoAn integer linear programming based optimization for homedemand-side management in smart gridrdquo in Proceedings of theIEEE PES Innovative Smart Grid Technologies (ISGT 12) pp 1ndash5Washington DC USA January 2012
[15] A-H Mohsenian-Rad and A Leon-Garcia ldquoOptimal residen-tial load control with price prediction in real-time electricitypricing environmentsrdquo IEEE Transactions on Smart Grid vol1 no 2 pp 120ndash133 2010
[16] K M Tsui and S C Chan ldquoDemand response optimizationfor smart home scheduling under real-time pricingrdquo IEEETransactions on Smart Grid vol 3 no 4 pp 1812ndash1821 2012
[17] GWood andMNewborough ldquoDynamic energy-consumptionindicators for domestic appliances environment behaviour anddesignrdquo Energy and Buildings vol 35 no 8 pp 821ndash841 2003
[18] M Avci M Erkoc and S S Asfour ldquoResidential HVAC loadcontrol strategy in real-time electricity pricing environmentrdquo inProceedings of the IEEE Energytech pp 1ndash6 Cleveland OhioUSA May 2012
[19] Z Q LuoW KMa A C So Y Ye and S Zhang ldquoSemidefiniterelaxation of quadratic optimization problemsrdquo IEEE SignalProcessing Magazine vol 27 no 3 pp 20ndash34 2010
[20] K F Fong V I Hanby and T T Chow ldquoHVAC systemoptimization for energymanagement by evolutionary program-mingrdquo Energy and Buildings vol 38 no 3 pp 220ndash231 2006
[21] D L Ha F F de Lamotte and Q-H Huynh ldquoReal-timedynamic multilevel optimization for demand-side load man-agementrdquo in Proceedings of the IEEE International Conferenceon Industrial Engineering and Engineering Management (IEEM07) pp 945ndash949 Singapore December 2007
[22] S Noh J Yun and K Kim ldquoAn efficient building air condi-tioning system control under real-time pricingrdquo in Proceedingsof the International Conference on Advanced Power SystemAutomation and Protection (APAP 11) pp 1283ndash1286 BeijingChina October 2011
[23] G L Nemhauser and L A Wolsey Integer and CombinatorialOptimization Wiley-Interscience New York NY USA 1988
[24] S Kay Fundamentals of Statistical Signal Processing EstimationTheory Prentice-Hall New York NY USA 1993
[25] ldquoContiki the open source OS for the internet of thingsrdquohttpwwwcontiki-osorg
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
4 The Scientific World Journal
Uplink wiredUplink wirelessDownlink wiredExternal inputs
middot middot middot
SinkHVAC1 HVACK
OnoffOnoff
AKA1
Tmini
Tmaxi C(L(sj))
TmjMTm
j1S1 SM
Gateway
Figure 3 Detailed system model of the HVAC energy scheduler
is turned on These measurements are denoted by119879119898119895
119894(119899) (as illustrated by the black curve in Figure 4)
where 1 ⩽ 119894 ⩽ 119872 denotes the 119894th node and 1 ⩽ 119895 ⩽ 2119870
is the 119895th combination of HVACs turned on or off(ii) The energy cost function 119862(119871(s
119895)) which is specified
by the energy provider and depends on the smartpricing tariff According to [11] 119862(119871(s
119895)) can be
modeled as a quadratic function that is
119862 (119871 (s119895)) = 119901
1(119871 (s119895))2
+ 1199012119871 (s119895) + 1199013 (1)
where 1199011 1199012 and 119901
3are parameters that the provider
can dynamically vary in timeMoreover119871(s119895)denotes
the userrsquos energy consumption for the s119895combination
of HVACmodules turned on In order to describe theexpression for 119871(s
119895) let us define P isin R119870times2
119870
a matrixthat contains in its 119895th column the energy consump-tion of each HVAC module for the 119895th combinationof modules switched onoff For instance for 119870 = 3the matrix P is
P = (
0 1199091
0 0 11990911199091
0 1199091
0 0 1199092
0 1199092
0 11990921199092
0 0 0 1199093
0 119909311990931199093
) (2)
Moreover s119895is a vector of all zeros except in the 119895th
position that has value 1 Therefore Ps119895selects the
energy consumption related to the 119895th combinationof HVACs switched on and 1119879(Ps
119895) is the energy
consumption related to that combination where 1 isa vector of ones of length 2
119870 Therefore the totalenergy consumption for a given time period denotedby 119871(s
119895) can be expressed as
119871 (s119895) = 1198710+ 1119879 (Ps
119895) (3)
Tem
pera
ture
Time
2K predictedcurves
Past time intervalNsamples N samples
Next time interval
Current time
n + N
Tmaxi
Tmini
Tmji
Tpji
n
Figure 4 Prediction of temperature a fundamental step of theenergy scheduler to assess comfort in the future time interval
where 1198710
is the accumulated consumption and1119879(Ps
119895) is the consumption in the current time inter-
val decision
(iii) The constraints of temperature of comfort providedby the user For the energy scheduler proposed inSection 41 this corresponds to the minimum andmaximum allowed temperatures at the 119894th locationof the room which are denoted by 119879
min119894
and 119879max119894
with 1 ⩽ 119894 ⩽ 119872 respectively since we assumethat the user may specify the desired comfort at 119872different locations For the energy scheduler proposedin Section 42 the comfort is specified by the objectivetemperatures 119879
119906119894 1 ⩽ 119894 ⩽ 119872 which the user would
like to attain at the different119872 locations of the roomthough they may allow some relaxation in order tofurther reduce the energy consumption
To further clarify the operation of the proposed schemelet us shed light on the temporal behavior of the energyschedulers and the role of the temperature constraints onit Recall that the energy scheduler works in a time intervalbasis At the end of each time interval (ldquocurrent timerdquo inFigure 4) the energy scheduler must make a new decisionThat is it must decide which HVAC modules denoted byHVACk in Figure 3 will be active during the next timeinterval In order to make this decision the energy schedulershould predict which would be the temperature provokedby each configuration of HVACs As there are 119870 HVACmodules and we assume that they are either turned on oroff this corresponds to predict 2119870 curves of temperature asit is illustrated in Figure 4 These predicted temperatures aredenoted by 119879119901
119895
119894(119899) where 119894 and 119895 have the same meaning
as for the case of 119879119898119895119894(119899) explained above Finally on one
hand the DES-CC selects the configuration of HVACs thatminimizes the energy consumption cost 119862(119871(s
119895)) within the
bounds of comfort that is 119879min119894
⩽ 119879119901119895
119894(119899) ⩽ 119879
max119894
while theDES-CCR selects the HVAC configuration that optimizes thetradeoff between being closer to the comfort temperatures119879119906119894
and saving energy This selection is executed by the
The Scientific World Journal 5
actuators that control the HVACmodules which are denotedby 119860119896in Figure 3
4 HVAC Energy Scheduling
41 Dynamic Energy Scheduler with Comfort Constraints(DES-CC) Next we present the first of the two proposedHVAC energy schedulers To that end this section is dividedinto four parts First we formulate the energy scheduleras a constrained optimization problem Second recall thatfor each time interval the energy scheduler must decidethe combination of active HVACs to minimize the energyconsumption cost and fulfill the constraints of temperatureof comfort To assess these constraints the temperatureprovoked in the next time interval by each configuration ofHVACs turned on or off should be predicted Thereby thesecond part deals with a thermal model that paves the way topredict the future temperatures The third part specifies howto estimate the parameters of the prediction model thanks tothe measurements of temperature of the past time intervalFinally in the fourth part we summarize the proposed DES-CC algorithm
411 Formulation of DES-CC as a Constrained OptimizationProblem Theenergy scheduler works in a time interval basisWhen 119873 samples of temperature have been collected fromthe WSN at the119872 controlled locations the energy schedulermakes a new decision with respect to the state of the HVACsNamely for the next time interval the energy schedulerselects the optimal configuration of active HVACs Thisconfiguration on the one hand must minimize the energyconsumption cost while on the other hand it must respectthe comfort constraints that is it should lead to predictedtemperatures within the bounds of comfort According to thedefinitions of the system model this optimization problemmay be formulated mathematically as
minimizes119895isin012
119870times1
119862 (119871 (s119895))
subject to 119879min119894
⩽ 119879119901min119894
(s119895) 119894 isin [1119872]
119879max119894
⩾ 119879119901max119894
(s119895) 119894 isin [1119872]
1119879s119895= 1
(4)
where 119879119901min119894
(s119895) and 119879119901
max119894
(s119895) are the minimum and maxi-
mum predicted temperatures respectively at the 119894th locationfor the 119895th combination of HVAC modules turned on Giventhe definition of 119862(119871(s
119895)) in (1) as a quadratic function
the optimization problem (4) has the form of a quadraticprogramming but that the optimization variable is booleanHence it is a boolean quadratic programming problem Theproblems of this class are nonconvex and in general they canbe solved either by a fast method that finds a local solutionor by a slower method that finds the global solution Inour framework the number of HVAC modules 119870 is low ormoderate and the latter approach is preferred for examplethe branch and bound method [23] can be used In order
to proceed we need to model the predicted temperatures119879119901
min119894
(s119895) and 119879119901
max119894
(s119895)
412 Model to Predict the Temperatures of the Future TimeInterval Regarding the predicted temperatures 119879119901
min119894
(s119895)
and 119879119901max119894
(s119895) they can be expressed as
119879119901min119894
(s119895) =
∘q119879119894s119895
119879119901max119894
(s119895) = q119879119894s119895
(5)
where ∘q119894and q
119894are vectors that contain the minimum and
maximum predicted temperatures respectively for each ofthe possible combinations of operating HVAC modules Tofurther clarify let us define 119879119901119895
119894(119899) 2 le 119899 le 119873 the predicted
temperature at the 119899 time instant at the 119894th sensor for the 119895thcombination of HVAC modules turned on where 1 ⩽ 119894 ⩽ 119872
and 1 ⩽ 119895 ⩽ 2119870 Moreover let 119879119901119895
119894(119899min119895) and 119879119901
119895
119894(119899max119895)
be theminimum andmaximum temperatures among119879119901119895119894(119899)
2 le 119899 le 119873 Then ∘q119894and q
119894can be expressed as
∘q119879119894= [119879119901
1
119894(119899min1) 119879119901
2119870
119894(119899min
2119870)]
q119879119894= [119879119901
1
119894(119899max1) 119879119901
2119870
119894(119899max
2119870)]
(6)
In order to proceed a model for the predicted tempera-tures is necessary Intuitively the current temperature is cor-related with the past temperature and a given combination ofHVACs turned on causes a change in temperature Moreoverthe temperature dynamics are rather linear (at least locally)as it will be shown below Therefore the following model isproposed for the temperature prediction
119879119901119895
119894(119899) = 119886
119895
119894119879119901119895
119894(119899 minus 1) + 120574
119895
119894 2 le 119899 le 119873 (7)
where 119886119895119894and 120574119895119894model the relation with the past temperature
and the change of temperature provoked by the 119895th combina-tion of HVACs turned on respectively Observe that in thisexpression 119886119895
119894and 120574
119895
119894are unknown and must be estimated
from the past measurements
413 Estimation of the PredictionModel Parameters In orderto estimate 119886
119895
119894and 120574
119895
119894in (7) we assume that the past
measurements follow a model like (7) corrupted by noise
119879119898119895
119894(119899) = 119886
119895
119894119879119898119895
119894(119899 minus 1) + 120574
119895
119894+ 119908119895
119894(119899) 2 le 119899 le 119873 (8)
Note that the evolution of the temperature is consideredto be linear in (7) and (8) This is a valid assumption at leastfor short periods as the real experiments that we will presentin Section 5 will highlight
For the estimation of 119886119895
119894and 120574
119895
119894 two situations are
considered In the first one all the HVAC modules areswitched off and as a consequence only 119886119895
119894must be estimated
To that end least squares (LS) estimator is considered as noprobabilistic assumptions regarding the data are neededThis
6 The Scientific World Journal
Process For each time interval do(1) Input(a) The measurements of temperature of the previous interval(b) Value of 119901
11199012and 119901
3in the energy cost function (1)
(c) The userrsquos temperature of comfort constraints at eachlocation that is 119879min
119894 119879
max119894
119894 = 1 119872 in (4)(2) Estimation Step
Estimate 119886119895119894and 120574
119895
119894in the prediction model (7) using (9)
to (12) and the temperature measurements from the pasttime interval
(3) Prediction Step(a) Substitute the estimation of 119886119895
119894and 120574
119895
119894into the prediction model (7)
(b) Iterate this model to populate the vectors of minimumand maximum predicted temperatures (6)
(4)Optimization Step(a) Substitute the predicted temperatures (5) and the
quadratic cost function (1) into the optimization problem (4)(b) Solve (4) using Branch and bound method [23]
(5)OutputThe configuration of HVACs turned onoff that optimizes (4)
Algorithm 1 Dynamic energy scheduler with comfort constraints (DES-CC)
estimator minimizes the LS error criterion though it is notoptimal in general [24] Given (8) the LS estimation of 119886119895
119894
denoted by 119886119895119894is given by
119886119895
119894= xy (9)
where the symbol denotes the pseudoinverse operatorwhich is defined as x = (x119879x)minus1x119879 and we define x =
[119879119898119895
119894(1) 119879119898
119895
119894(119873 minus 1)]
119879
and y = [119879119898119895
119894(2) 119879119898
119895
119894(119873)]119879
The second situation is that some of the HVACmodules wereswitched on In this case an LS estimation is considered aswell Namely let us denote by 120574
119895
119894| 119886119895
119894the estimation of
120574119895
119894conditioned to the knowledge of a past estimation of 119886119895
119894
denoted by 119886119895
119894 Then the LS estimation for 120574119895
119894| 119886119895
119894yields
120574119895
119894| 119886119895
119894= 1z (10)
where 1 is a vector of ones of length119873 minus 1 and z is given by
z = [119879119898119895
119894(2) 119879119898
119895
119894(119873)]119879
minus 119886119895
119894[119879119898119895
119894(1) 119879119898
119895
119894(119873 minus 1)]
119879
(11)
Finally given the estimation of 120574119895119894in (10) we can update
the estimation of 119886119895119894as
119886119895
119894| 120574119895
119894= xy (12)
where y = [119879119898119895
119894(2) minus 120574
119895
119894 119879119898
119895
119894(119873) minus 120574
119895
119894]119879
414 Summary of the DES-CC Energy Scheduler At thispoint all the terms in the optimization problem understudy that is (4) are specified The procedure to implementthe proposed energy scheduler for each time interval issummarized in Algorithm 1
42 Dynamic Energy Scheduler with Comfort ConstraintsRelaxation (DES-CCR) Despite its effectiveness and its obvi-ous advantages the proposed energy scheduling algorithm iscompletely focused on the temperature constraints neglect-ing the price aspects of the problem More specificallyalthough there could be time periods where the energyprice increases the energy scheduler switches on the samecombination of HVAC modules in order to respect thetemperature constraints However in such cases users mightcompromise their comfort preferences to decrease the energyconsumption In order to allow the user to have more flexi-bility in the energy consumption a new energy scheduler willbe presented in this section This flexibility is implementedin terms of relaxing the temperature constraints to furtherreduce the energy consumption
This new energy scheduler is formulated so that theuser temperature constraints in (4) are skipped and they areincorporated as a penalty term in the objective functionConsequently the new optimization problem can be writtenas
minimizes119895isin012
119870times1
120579
119862 (119871 (s119895))
120572+ (1 minus 120579)
times
sum119872
119894=1sum119873
119899=2
10038171003817100381710038171003817q119879119894(119899)s119895minus 119879119906119894
10038171003817100381710038171003817
2
120573
(13)
where119862(119871(s119895)) is the energy cost function defined in (1)The
vector q119879119894(119899) is defined as
q119879119894(119899) = [119879119901
1
119894(119899) 119879119901
2119870
119894(119899)] (14)
and recall that 119879119901119895119894(119899) is the predicted temperature at the 119894th
location for the 119895th combination of HVACs modules turned
The Scientific World Journal 7
Process For each time interval do(1) Input(a) The measurements of temperature of the previous interval(b) Value of 119901
1 1199012and 119901
3of the energy cost function 119862(119871(119904
119895)) defined in (1)
(c) The userrsquos temperature of comfort at each location that is 119879119906119894 119894 = 1 119872 in (13)
(d) The parameter 120579 isin (0 1) controlling the comfort relaxation in (13)(2) Estimation Step
Estimate 119886119895119894and 120574
119895
119894in the prediction model (7) using (9)
to (12) and the temperature measurements from the past time interval(3) Prediction Step(a) Substitute the estimation of 119886119895
119894and 120574
119895
119894into the prediction model (7)
(b) For 119894 = 1 to119872For 119899 = 2 to119873Compute and store the vector of predicted temperaturesq119879119894(119899) in (14) using (7)
End For 119899End For 119894
(4)Optimization Step(a) Compute the cost function in (13) using the vectors in
the step 3(b) and the inputs in (1)(b) Solve the optimization problem in (13) using Branch and bound method [23]
(5)OutputThe configuration of HVACs turned onoff that optimizes (13)
Algorithm 2 Dynamic energy scheduler with comfort constraints relaxation (DES-CCR)
on or off see (7)Moreover120572 and120573 are normalizing constantsto adjust the values of the two terms in (13) Indeed we settheir value as
120572 = 119862 (119871 (s2119870))
120573 = maxs119895isin01119870times1
119872
sum
119894=1
119873
sum
119899=2
10038171003817100381710038171003817q119879119894(119899)s119895minus 119879119906119894
10038171003817100381710038171003817
2
(15)
where 119862(119871(s2119870)) is the cost for all the HVACmodules turned
on The term 119879119906119894
is the desired temperature that the userwould like tomaintain at the 119894th location of the room Clearlyour reformulation balances the two optimization problemsthat is the energy cost minimization and the user comfortmaximization Note that the user comfort is defined as anEuclidean norm but it can eventually be redefined withanother distance measurement
Finally 120579 isin (0 1) is defined by the user according to theirpreferences For example in the extreme case where 120579 = 0the demand response algorithm will not consider any priceand it will directly control the HVAC modules so that thedesired temperature is reached On the contrary when 120579 = 1theHVACmodules will always remain off In this context theusers should set the 120579 value according to their preferences andexperience The DES-CCR energy scheduler is summarizedin Algorithm 2
5 Experimental Results
In order to emulate the complete communication in anIoT framework we have designed and developed a customtestbed that integrates the described architecture In our
Figure 5 Z1 WSN mote
experiments we focus on a heating system although theproposed algorithms apply in general HVAC systems In thissection we describe the testbed platform and the experi-mental scenario we define a baseline thermostat model andfinally we present the experimental results of our proposedalgorithms
51 Testbed Description and Experimental Setup The testbedhas been deployed in a 50m2 room within our researchcenter facilities as it is depicted in Figure 5 In our particularscenario we consider three HVAC modules (ie 119870 = 3)and two temperature sensor nodes (ie 119872 = 2) The HVACmodules are distributed around the room while the sensornodes are placed in the middle of the room monitoring thetemperature and sending it to the sink mote every 30 second
8 The Scientific World Journal
Micro-USB
Ceramic embedded antennaUFL connector for external antenna
temperature sensor3-Axis accelerometer+
2 times Phidgets sensor ports
Figure 6 Overall platform detail
(ie 119905119898= 30 s) In addition the samples received from each
sensor are stored in a buffer at the control center and ouralgorithm applies every 10 samples (ie119873 = 10)
Regarding the employed technology the WSN nodes areZ1 motes by Zolertia (Figure 6) They are equipped with asecond generationMSP430F2617 low power microcontrollerwhich features a 16-bit RISC CPU 16MHz clock speed abuilt-in clock factory calibration an 8KB RAM and a 92KBflash memory They also include the CC2420 transceiverwhich is IEEE 802154 compliant operating at 24GHz fre-quency bandwith a data rate of 250 kbpsThe sensors supportContiki OS [25] an open-source operating system for the IoTwhich connects tiny low-cost low-power microcontrollersto the Internet and supports IPv6 through 6LowPAN It isworth noting that each mote can operate as either a sourceor a sink node In particular source nodes carry a TMP102temperature sensor to monitor the target field while the sinknode receives and forwards themeasured data to the gateway
The gateway (an Ubuntu OS machine with MATLAB)implements the proposed algorithms and it is able to processthe collected data Furthermore it connects the WSN to theInternet and acts as an application server using Nodejs andSencha Touch In particular Nodejs is a platform built onChromersquos JavaScript runtime for fast and scalable networkapplications Figure 7 shows a screenshot of the web applica-tion built on Nodejs which enables the user to interact withthe energy scheduler through Internet Regarding SenchaTouch it is a high-performance HTML5 mobile applicationframework which enables developers to build powerfulapplications for various operating systems including iOS andAndroidThe actuators are programmable sockets which canbe controlled remotely thanks to their IP addresses Thesespecial sockets are a set of programmable local area networksurge protectors (EG-PMS-LAN) by Energenie which areconnected via Ethernet to the gateway Finally the HVACmodules are domestic heaters with a maximum power con-sumption of 2000W
52 Baseline Model To evaluate the proposed algorithmsand highlight the potential energy and cost gains that can
Figure 7 Google Chrome screenshot of the web application
be achieved we adopt the traditional thermostat model asthe baseline reference scenario In this model the aim is tomaintain the average temperature of the room between acertain temperature range (ie [119879min119879max]) predefined bythe user To that end when the sensed temperature is above119879max at the end of a time interval all the heaters are switchedoff while the heaters are switched on when the temperaturefalls below the 119879min threshold
Figure 8 illustrates the average measured temperatureinside the room where the heaters are controlled by thethermostat In this particular case we consider 119879min =
21∘C and 119879max = 23
∘C As it can be seen in the figurethe thermostat algorithm is able to maintain the averagetemperature of the room between the desiredmargins during16 hours However it is worth noting that despite its properbehavior the particular model is not cost efficient as allHVAC modules work simultaneously consuming a totalpower consumption of 6000W In the following sectionswe evaluate our proposed methods demonstrating that theycan reduce the electrical cost with respect to the baselineapproach
The Scientific World Journal 9
21
22
23
24
HVA
C sta
tus
Average temperatureHVAC status
Day time
Tem
pera
ture
(∘C) On
Off
Thermostat-baseline model
18 20 22 24 2 4 6 8 10
Figure 8 Experimental evaluation of the thermostat model
53 Experimental Evaluation of the DES-CC in (4) Severalreal experiments have been carried out to assess the perfor-mance of the DES-CC algorithm proposed in (4) In thiscase the pricing parameters in (1) are 119901
1= 0003 and 119901
2=
1199013= 0 euros which are possible values according to [11] The
temperature bounds in (4) have been set to 119879min1
= 119879min2
=
21∘C and 119879
max1
= 119879max2
= 23∘C respectively
Figure 9 plots the variation of both the measured and theestimated temperature (using (8)ndash(12)) in the room duringour experiments As it can be noticed the error between theestimated and the real temperature is negligible somethingthat proves the accuracy of the proposed estimationmodel Inaddition the DES-CC guarantees the proper operation of thesystem as the temperature varies between the desired rangeof 21∘C and 23
∘C most of the time with very few exceptionsdue to prediction errors In the same figure it can be alsoseen that the temperature remains closer to the lower partof the permitted range (ie 21∘C) since the outcome of theproposed method provides a combination of switched onheaters that minimizes the energy consumption satisfyinga minimum acceptable temperature Indeed compared tothe temperature variation in the baseline scenario (Figure 8)DES-CC maintains the temperature more stable and in thelower part of the allowable region intuitively implying lowercost
Figure 10 depicts the financial operation cost gains thatcan be achieved by DES-CC compared to the baseline ther-mostat approach As it can be observed the proposed energyscheduler significantly reduces the energy cost leading to atotal save of 719 eurosmonth
54 Experimental Evaluation of the DES-CCR in (13) A setof experiments have been carried out for the evaluation ofthe DES-CCR in (13) Let us recall that DES-CCR relaxesthe temperature constraints by including the constraints asa penalized term in the objective function As a resultcompared to DES-CC this method is more flexible withrespect to real time pricing tariffs More specifically DES-CC seeks a combination that minimizes the energy cost withrespect to a minimum allowable temperature Consequently
WSN mote 2
WSN mote 1
18 20 22 24 2 4 6 8 1018192021222324
Day time
18 20 22 24 2 4 6 8 10Day time
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18192021222324
Tem
pera
ture
(∘C)
Figure 9 Real and estimated temperature using DES-CC
0 2 4 6 8 10 12 14 160
0005
001
0015
002
0025
003
0035
004
Hours
Ener
gy co
st (E
uros
)
Energy cost per hour of the proposed methodEnergy cost per hour of the thermostat
Figure 10 Energy consumption cost comparison between thermo-stat and DES-CC
although the energy cost may change during time the heatercombination selected byDES-CC is the same due to the stricttemperature constraint On the other hand DES-CCR allowsthe user to further reduce the energy consumption at the costof being outside the range of temperature of comfort In thiscase to highlight the flexibility of DES-CCR we have set a
10 The Scientific World Journal
0 50 100 150 200 25016
18
20
22
24
Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
0 50 100 150 200 25016
18
20
22
24
Sample measures
Tem
pera
ture
(∘C)
Figure 11 Real and estimated temperature using DES-CCR (120579 =
02)
periodically variable value of 1199011 which alternates between
1199011
= 0009 and 1199011
= 0003 euros every thirty minutesMoreover the desired temperature has been set to 119879
119906119894=
22∘CFigures 11 and 12 depict the temperature variation in
two different cases where the users give low (120579 = 02)and high (120579 = 05) priority respectively to reduce of theenergy consumption In particular in Figure 11 (120579 = 02)the achieved temperature is very close to the desired 119879
119906119894 On
the other hand in Figure 12 we assume 120579 = 05 which is amore adapted value to the pricing policy as it correspondsto a user that permits a relaxation of the difference betweenthe real and the desired temperature to reduce the energycost This fact implies higher energy consumption in lowcost zones and lower energy consumption in high costperiods sacrificing though the userrsquos comfort Therefore theexperiments confirm that the real temperature is close to 119879
119906119894
when the energy cost is lower (ie between samples 60 and120) while there is a noticeable temperature drop whichcorresponds to lower energy consumption
6 Concluding Remarks
This paper has dealt with the energy consumption manage-ment of HVACs for a given smart pricing tariff and usersrsquocomfort constraintsMoreover the integrationwithin the IoTframework has been studied To that end we developed areal testbed consisting of (i) heaters (ii) sensor nodes thatmeasure the temperature and (iii) a gateway which providesconnection to the Internet and includes aweb application that
18
19
20
21
22
0 50 100 150 200 250Sample measures
0 50 100 150 200 250Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18
19
20
21
22
Tem
pera
ture
(∘C)
Figure 12 Real and estimated temperature using DES-CCR (120579 =
05)
permits the interaction with the user through InternetMore-over the gateway implements the algorithms that control theenergy consumption Regarding the proposed methods firstwe devised an energy scheduler that optimizes the energycost in a time interval basis for a given energy price tariffand for a given set of temperature of comfort constraintsthat are associated with different locations inside a roomThen we proposed a more flexible energy scheduler whichrelaxes the temperature constraints to further reduce theenergy consumption Namely a new objective function hasbeen considered which consists of a convex combination ofthe energy cost and a penalty term that reflects the comfortThis permits to consider both the case where the user isvery concerned with the comfort and the case where heallows relaxing the comfort constraint to further reduce theenergy consumption Experimental evaluations have beencarried out in an isolated room validating our proposals andhighlighting their potential benefits
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work has been funded by the Energy-to-Smart Grid(E2SG) project httpwwwe2sg-projecteu within theENIAC joint undertaking framework with Grant agreementnumber 296131
The Scientific World Journal 11
References
[1] L Atzori A Iera and G Morabito ldquoThe internet of things asurveyrdquoComputer Networks vol 54 no 15 pp 2787ndash2805 2010
[2] E Fleisch ldquoWhat is the internet of things An economicperspectiverdquo Tech Rep Auto-ID Labs 2010
[3] National Intelligence Council ldquoDisruptive civil technologiesmdashsix technologies with potential impacts on us interests out to2025rdquo Conference Report CR 2008-07 2008
[4] M R Palattella N Accettura X Vilajosana et al ldquoStandardizedprotocol stack for the internet of (important) thingsrdquo IEEECommunications Surveys and Tutorials vol 15 no 3 pp 1389ndash1406 2013
[5] M Zorzi A Gluhak S Lange and A Bassi ldquoFrom todaysINTRAnet of things to a future INTERnet of things a wireless-and mobility-related viewrdquo IEEEWireless Communications vol17 no 6 pp 44ndash51 2010
[6] A Gluhak S Krco M Nati D Pfisterer N Mitton andT Razafindralambo ldquoA survey on facilities for experimentalinternet of things researchrdquo IEEE Communications Magazinevol 49 no 11 pp 58ndash67 2011
[7] S Tozlu M Senel W Mao and A Keshavarzian ldquoWi-Fienabled sensors for internet of things a practical approachrdquoIEEE Communications Magazine vol 50 no 6 pp 134ndash1432012
[8] N Bui A P Castellani P Casari andM Zorzi ldquoThe internet ofenergy a web-enabled smart grid systemrdquo IEEE Network vol26 no 4 pp 39ndash45 2012
[9] X Fang S Misra G Xue and D Yang ldquoSmart gridmdashthe newand improved power grid a surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 14 no 4 pp 944ndash980 2012
[10] G Lu D De and W Song ldquoSmartGridLab a laboratory-basedsmart grid testbedrdquo in Proceedings of the 1st IEEE InternationalConference on Smart Grid Communications (SmartGridComm10) pp 143ndash148 Gaithersburg Md USA October 2010
[11] A-H Mohsenian-Rad V W S Wong J Jatskevich R Schoberand A Leon-Garcia ldquoAutonomous demand-side managementbased on game-theoretic energy consumption scheduling forthe future smart gridrdquo IEEE Transactions on Smart Grid vol1 no 3 pp 320ndash331 2010
[12] Z Zhu S Lambotharan W H Chin and Z Fan ldquoOverviewof demand management in smart grid and enabling wirelesscommunication technologiesrdquo IEEE Wireless Communicationsvol 19 no 3 pp 48ndash56 2012
[13] H T Nguyen D Nguyen and L B Le ldquoHome energy man-agement with generic thermal dynamics and user temperaturepreferencerdquo in Proceedings of the IEEE International Conferenceon Smart Grid Communications (SmartGridComm 13) pp 552ndash557 Vancouver Canada October 2013
[14] Z Zhu J Tang S Lambotharan W H Chin and Z FanldquoAn integer linear programming based optimization for homedemand-side management in smart gridrdquo in Proceedings of theIEEE PES Innovative Smart Grid Technologies (ISGT 12) pp 1ndash5Washington DC USA January 2012
[15] A-H Mohsenian-Rad and A Leon-Garcia ldquoOptimal residen-tial load control with price prediction in real-time electricitypricing environmentsrdquo IEEE Transactions on Smart Grid vol1 no 2 pp 120ndash133 2010
[16] K M Tsui and S C Chan ldquoDemand response optimizationfor smart home scheduling under real-time pricingrdquo IEEETransactions on Smart Grid vol 3 no 4 pp 1812ndash1821 2012
[17] GWood andMNewborough ldquoDynamic energy-consumptionindicators for domestic appliances environment behaviour anddesignrdquo Energy and Buildings vol 35 no 8 pp 821ndash841 2003
[18] M Avci M Erkoc and S S Asfour ldquoResidential HVAC loadcontrol strategy in real-time electricity pricing environmentrdquo inProceedings of the IEEE Energytech pp 1ndash6 Cleveland OhioUSA May 2012
[19] Z Q LuoW KMa A C So Y Ye and S Zhang ldquoSemidefiniterelaxation of quadratic optimization problemsrdquo IEEE SignalProcessing Magazine vol 27 no 3 pp 20ndash34 2010
[20] K F Fong V I Hanby and T T Chow ldquoHVAC systemoptimization for energymanagement by evolutionary program-mingrdquo Energy and Buildings vol 38 no 3 pp 220ndash231 2006
[21] D L Ha F F de Lamotte and Q-H Huynh ldquoReal-timedynamic multilevel optimization for demand-side load man-agementrdquo in Proceedings of the IEEE International Conferenceon Industrial Engineering and Engineering Management (IEEM07) pp 945ndash949 Singapore December 2007
[22] S Noh J Yun and K Kim ldquoAn efficient building air condi-tioning system control under real-time pricingrdquo in Proceedingsof the International Conference on Advanced Power SystemAutomation and Protection (APAP 11) pp 1283ndash1286 BeijingChina October 2011
[23] G L Nemhauser and L A Wolsey Integer and CombinatorialOptimization Wiley-Interscience New York NY USA 1988
[24] S Kay Fundamentals of Statistical Signal Processing EstimationTheory Prentice-Hall New York NY USA 1993
[25] ldquoContiki the open source OS for the internet of thingsrdquohttpwwwcontiki-osorg
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World Journal 5
actuators that control the HVACmodules which are denotedby 119860119896in Figure 3
4 HVAC Energy Scheduling
41 Dynamic Energy Scheduler with Comfort Constraints(DES-CC) Next we present the first of the two proposedHVAC energy schedulers To that end this section is dividedinto four parts First we formulate the energy scheduleras a constrained optimization problem Second recall thatfor each time interval the energy scheduler must decidethe combination of active HVACs to minimize the energyconsumption cost and fulfill the constraints of temperatureof comfort To assess these constraints the temperatureprovoked in the next time interval by each configuration ofHVACs turned on or off should be predicted Thereby thesecond part deals with a thermal model that paves the way topredict the future temperatures The third part specifies howto estimate the parameters of the prediction model thanks tothe measurements of temperature of the past time intervalFinally in the fourth part we summarize the proposed DES-CC algorithm
411 Formulation of DES-CC as a Constrained OptimizationProblem Theenergy scheduler works in a time interval basisWhen 119873 samples of temperature have been collected fromthe WSN at the119872 controlled locations the energy schedulermakes a new decision with respect to the state of the HVACsNamely for the next time interval the energy schedulerselects the optimal configuration of active HVACs Thisconfiguration on the one hand must minimize the energyconsumption cost while on the other hand it must respectthe comfort constraints that is it should lead to predictedtemperatures within the bounds of comfort According to thedefinitions of the system model this optimization problemmay be formulated mathematically as
minimizes119895isin012
119870times1
119862 (119871 (s119895))
subject to 119879min119894
⩽ 119879119901min119894
(s119895) 119894 isin [1119872]
119879max119894
⩾ 119879119901max119894
(s119895) 119894 isin [1119872]
1119879s119895= 1
(4)
where 119879119901min119894
(s119895) and 119879119901
max119894
(s119895) are the minimum and maxi-
mum predicted temperatures respectively at the 119894th locationfor the 119895th combination of HVAC modules turned on Giventhe definition of 119862(119871(s
119895)) in (1) as a quadratic function
the optimization problem (4) has the form of a quadraticprogramming but that the optimization variable is booleanHence it is a boolean quadratic programming problem Theproblems of this class are nonconvex and in general they canbe solved either by a fast method that finds a local solutionor by a slower method that finds the global solution Inour framework the number of HVAC modules 119870 is low ormoderate and the latter approach is preferred for examplethe branch and bound method [23] can be used In order
to proceed we need to model the predicted temperatures119879119901
min119894
(s119895) and 119879119901
max119894
(s119895)
412 Model to Predict the Temperatures of the Future TimeInterval Regarding the predicted temperatures 119879119901
min119894
(s119895)
and 119879119901max119894
(s119895) they can be expressed as
119879119901min119894
(s119895) =
∘q119879119894s119895
119879119901max119894
(s119895) = q119879119894s119895
(5)
where ∘q119894and q
119894are vectors that contain the minimum and
maximum predicted temperatures respectively for each ofthe possible combinations of operating HVAC modules Tofurther clarify let us define 119879119901119895
119894(119899) 2 le 119899 le 119873 the predicted
temperature at the 119899 time instant at the 119894th sensor for the 119895thcombination of HVAC modules turned on where 1 ⩽ 119894 ⩽ 119872
and 1 ⩽ 119895 ⩽ 2119870 Moreover let 119879119901119895
119894(119899min119895) and 119879119901
119895
119894(119899max119895)
be theminimum andmaximum temperatures among119879119901119895119894(119899)
2 le 119899 le 119873 Then ∘q119894and q
119894can be expressed as
∘q119879119894= [119879119901
1
119894(119899min1) 119879119901
2119870
119894(119899min
2119870)]
q119879119894= [119879119901
1
119894(119899max1) 119879119901
2119870
119894(119899max
2119870)]
(6)
In order to proceed a model for the predicted tempera-tures is necessary Intuitively the current temperature is cor-related with the past temperature and a given combination ofHVACs turned on causes a change in temperature Moreoverthe temperature dynamics are rather linear (at least locally)as it will be shown below Therefore the following model isproposed for the temperature prediction
119879119901119895
119894(119899) = 119886
119895
119894119879119901119895
119894(119899 minus 1) + 120574
119895
119894 2 le 119899 le 119873 (7)
where 119886119895119894and 120574119895119894model the relation with the past temperature
and the change of temperature provoked by the 119895th combina-tion of HVACs turned on respectively Observe that in thisexpression 119886119895
119894and 120574
119895
119894are unknown and must be estimated
from the past measurements
413 Estimation of the PredictionModel Parameters In orderto estimate 119886
119895
119894and 120574
119895
119894in (7) we assume that the past
measurements follow a model like (7) corrupted by noise
119879119898119895
119894(119899) = 119886
119895
119894119879119898119895
119894(119899 minus 1) + 120574
119895
119894+ 119908119895
119894(119899) 2 le 119899 le 119873 (8)
Note that the evolution of the temperature is consideredto be linear in (7) and (8) This is a valid assumption at leastfor short periods as the real experiments that we will presentin Section 5 will highlight
For the estimation of 119886119895
119894and 120574
119895
119894 two situations are
considered In the first one all the HVAC modules areswitched off and as a consequence only 119886119895
119894must be estimated
To that end least squares (LS) estimator is considered as noprobabilistic assumptions regarding the data are neededThis
6 The Scientific World Journal
Process For each time interval do(1) Input(a) The measurements of temperature of the previous interval(b) Value of 119901
11199012and 119901
3in the energy cost function (1)
(c) The userrsquos temperature of comfort constraints at eachlocation that is 119879min
119894 119879
max119894
119894 = 1 119872 in (4)(2) Estimation Step
Estimate 119886119895119894and 120574
119895
119894in the prediction model (7) using (9)
to (12) and the temperature measurements from the pasttime interval
(3) Prediction Step(a) Substitute the estimation of 119886119895
119894and 120574
119895
119894into the prediction model (7)
(b) Iterate this model to populate the vectors of minimumand maximum predicted temperatures (6)
(4)Optimization Step(a) Substitute the predicted temperatures (5) and the
quadratic cost function (1) into the optimization problem (4)(b) Solve (4) using Branch and bound method [23]
(5)OutputThe configuration of HVACs turned onoff that optimizes (4)
Algorithm 1 Dynamic energy scheduler with comfort constraints (DES-CC)
estimator minimizes the LS error criterion though it is notoptimal in general [24] Given (8) the LS estimation of 119886119895
119894
denoted by 119886119895119894is given by
119886119895
119894= xy (9)
where the symbol denotes the pseudoinverse operatorwhich is defined as x = (x119879x)minus1x119879 and we define x =
[119879119898119895
119894(1) 119879119898
119895
119894(119873 minus 1)]
119879
and y = [119879119898119895
119894(2) 119879119898
119895
119894(119873)]119879
The second situation is that some of the HVACmodules wereswitched on In this case an LS estimation is considered aswell Namely let us denote by 120574
119895
119894| 119886119895
119894the estimation of
120574119895
119894conditioned to the knowledge of a past estimation of 119886119895
119894
denoted by 119886119895
119894 Then the LS estimation for 120574119895
119894| 119886119895
119894yields
120574119895
119894| 119886119895
119894= 1z (10)
where 1 is a vector of ones of length119873 minus 1 and z is given by
z = [119879119898119895
119894(2) 119879119898
119895
119894(119873)]119879
minus 119886119895
119894[119879119898119895
119894(1) 119879119898
119895
119894(119873 minus 1)]
119879
(11)
Finally given the estimation of 120574119895119894in (10) we can update
the estimation of 119886119895119894as
119886119895
119894| 120574119895
119894= xy (12)
where y = [119879119898119895
119894(2) minus 120574
119895
119894 119879119898
119895
119894(119873) minus 120574
119895
119894]119879
414 Summary of the DES-CC Energy Scheduler At thispoint all the terms in the optimization problem understudy that is (4) are specified The procedure to implementthe proposed energy scheduler for each time interval issummarized in Algorithm 1
42 Dynamic Energy Scheduler with Comfort ConstraintsRelaxation (DES-CCR) Despite its effectiveness and its obvi-ous advantages the proposed energy scheduling algorithm iscompletely focused on the temperature constraints neglect-ing the price aspects of the problem More specificallyalthough there could be time periods where the energyprice increases the energy scheduler switches on the samecombination of HVAC modules in order to respect thetemperature constraints However in such cases users mightcompromise their comfort preferences to decrease the energyconsumption In order to allow the user to have more flexi-bility in the energy consumption a new energy scheduler willbe presented in this section This flexibility is implementedin terms of relaxing the temperature constraints to furtherreduce the energy consumption
This new energy scheduler is formulated so that theuser temperature constraints in (4) are skipped and they areincorporated as a penalty term in the objective functionConsequently the new optimization problem can be writtenas
minimizes119895isin012
119870times1
120579
119862 (119871 (s119895))
120572+ (1 minus 120579)
times
sum119872
119894=1sum119873
119899=2
10038171003817100381710038171003817q119879119894(119899)s119895minus 119879119906119894
10038171003817100381710038171003817
2
120573
(13)
where119862(119871(s119895)) is the energy cost function defined in (1)The
vector q119879119894(119899) is defined as
q119879119894(119899) = [119879119901
1
119894(119899) 119879119901
2119870
119894(119899)] (14)
and recall that 119879119901119895119894(119899) is the predicted temperature at the 119894th
location for the 119895th combination of HVACs modules turned
The Scientific World Journal 7
Process For each time interval do(1) Input(a) The measurements of temperature of the previous interval(b) Value of 119901
1 1199012and 119901
3of the energy cost function 119862(119871(119904
119895)) defined in (1)
(c) The userrsquos temperature of comfort at each location that is 119879119906119894 119894 = 1 119872 in (13)
(d) The parameter 120579 isin (0 1) controlling the comfort relaxation in (13)(2) Estimation Step
Estimate 119886119895119894and 120574
119895
119894in the prediction model (7) using (9)
to (12) and the temperature measurements from the past time interval(3) Prediction Step(a) Substitute the estimation of 119886119895
119894and 120574
119895
119894into the prediction model (7)
(b) For 119894 = 1 to119872For 119899 = 2 to119873Compute and store the vector of predicted temperaturesq119879119894(119899) in (14) using (7)
End For 119899End For 119894
(4)Optimization Step(a) Compute the cost function in (13) using the vectors in
the step 3(b) and the inputs in (1)(b) Solve the optimization problem in (13) using Branch and bound method [23]
(5)OutputThe configuration of HVACs turned onoff that optimizes (13)
Algorithm 2 Dynamic energy scheduler with comfort constraints relaxation (DES-CCR)
on or off see (7)Moreover120572 and120573 are normalizing constantsto adjust the values of the two terms in (13) Indeed we settheir value as
120572 = 119862 (119871 (s2119870))
120573 = maxs119895isin01119870times1
119872
sum
119894=1
119873
sum
119899=2
10038171003817100381710038171003817q119879119894(119899)s119895minus 119879119906119894
10038171003817100381710038171003817
2
(15)
where 119862(119871(s2119870)) is the cost for all the HVACmodules turned
on The term 119879119906119894
is the desired temperature that the userwould like tomaintain at the 119894th location of the room Clearlyour reformulation balances the two optimization problemsthat is the energy cost minimization and the user comfortmaximization Note that the user comfort is defined as anEuclidean norm but it can eventually be redefined withanother distance measurement
Finally 120579 isin (0 1) is defined by the user according to theirpreferences For example in the extreme case where 120579 = 0the demand response algorithm will not consider any priceand it will directly control the HVAC modules so that thedesired temperature is reached On the contrary when 120579 = 1theHVACmodules will always remain off In this context theusers should set the 120579 value according to their preferences andexperience The DES-CCR energy scheduler is summarizedin Algorithm 2
5 Experimental Results
In order to emulate the complete communication in anIoT framework we have designed and developed a customtestbed that integrates the described architecture In our
Figure 5 Z1 WSN mote
experiments we focus on a heating system although theproposed algorithms apply in general HVAC systems In thissection we describe the testbed platform and the experi-mental scenario we define a baseline thermostat model andfinally we present the experimental results of our proposedalgorithms
51 Testbed Description and Experimental Setup The testbedhas been deployed in a 50m2 room within our researchcenter facilities as it is depicted in Figure 5 In our particularscenario we consider three HVAC modules (ie 119870 = 3)and two temperature sensor nodes (ie 119872 = 2) The HVACmodules are distributed around the room while the sensornodes are placed in the middle of the room monitoring thetemperature and sending it to the sink mote every 30 second
8 The Scientific World Journal
Micro-USB
Ceramic embedded antennaUFL connector for external antenna
temperature sensor3-Axis accelerometer+
2 times Phidgets sensor ports
Figure 6 Overall platform detail
(ie 119905119898= 30 s) In addition the samples received from each
sensor are stored in a buffer at the control center and ouralgorithm applies every 10 samples (ie119873 = 10)
Regarding the employed technology the WSN nodes areZ1 motes by Zolertia (Figure 6) They are equipped with asecond generationMSP430F2617 low power microcontrollerwhich features a 16-bit RISC CPU 16MHz clock speed abuilt-in clock factory calibration an 8KB RAM and a 92KBflash memory They also include the CC2420 transceiverwhich is IEEE 802154 compliant operating at 24GHz fre-quency bandwith a data rate of 250 kbpsThe sensors supportContiki OS [25] an open-source operating system for the IoTwhich connects tiny low-cost low-power microcontrollersto the Internet and supports IPv6 through 6LowPAN It isworth noting that each mote can operate as either a sourceor a sink node In particular source nodes carry a TMP102temperature sensor to monitor the target field while the sinknode receives and forwards themeasured data to the gateway
The gateway (an Ubuntu OS machine with MATLAB)implements the proposed algorithms and it is able to processthe collected data Furthermore it connects the WSN to theInternet and acts as an application server using Nodejs andSencha Touch In particular Nodejs is a platform built onChromersquos JavaScript runtime for fast and scalable networkapplications Figure 7 shows a screenshot of the web applica-tion built on Nodejs which enables the user to interact withthe energy scheduler through Internet Regarding SenchaTouch it is a high-performance HTML5 mobile applicationframework which enables developers to build powerfulapplications for various operating systems including iOS andAndroidThe actuators are programmable sockets which canbe controlled remotely thanks to their IP addresses Thesespecial sockets are a set of programmable local area networksurge protectors (EG-PMS-LAN) by Energenie which areconnected via Ethernet to the gateway Finally the HVACmodules are domestic heaters with a maximum power con-sumption of 2000W
52 Baseline Model To evaluate the proposed algorithmsand highlight the potential energy and cost gains that can
Figure 7 Google Chrome screenshot of the web application
be achieved we adopt the traditional thermostat model asthe baseline reference scenario In this model the aim is tomaintain the average temperature of the room between acertain temperature range (ie [119879min119879max]) predefined bythe user To that end when the sensed temperature is above119879max at the end of a time interval all the heaters are switchedoff while the heaters are switched on when the temperaturefalls below the 119879min threshold
Figure 8 illustrates the average measured temperatureinside the room where the heaters are controlled by thethermostat In this particular case we consider 119879min =
21∘C and 119879max = 23
∘C As it can be seen in the figurethe thermostat algorithm is able to maintain the averagetemperature of the room between the desiredmargins during16 hours However it is worth noting that despite its properbehavior the particular model is not cost efficient as allHVAC modules work simultaneously consuming a totalpower consumption of 6000W In the following sectionswe evaluate our proposed methods demonstrating that theycan reduce the electrical cost with respect to the baselineapproach
The Scientific World Journal 9
21
22
23
24
HVA
C sta
tus
Average temperatureHVAC status
Day time
Tem
pera
ture
(∘C) On
Off
Thermostat-baseline model
18 20 22 24 2 4 6 8 10
Figure 8 Experimental evaluation of the thermostat model
53 Experimental Evaluation of the DES-CC in (4) Severalreal experiments have been carried out to assess the perfor-mance of the DES-CC algorithm proposed in (4) In thiscase the pricing parameters in (1) are 119901
1= 0003 and 119901
2=
1199013= 0 euros which are possible values according to [11] The
temperature bounds in (4) have been set to 119879min1
= 119879min2
=
21∘C and 119879
max1
= 119879max2
= 23∘C respectively
Figure 9 plots the variation of both the measured and theestimated temperature (using (8)ndash(12)) in the room duringour experiments As it can be noticed the error between theestimated and the real temperature is negligible somethingthat proves the accuracy of the proposed estimationmodel Inaddition the DES-CC guarantees the proper operation of thesystem as the temperature varies between the desired rangeof 21∘C and 23
∘C most of the time with very few exceptionsdue to prediction errors In the same figure it can be alsoseen that the temperature remains closer to the lower partof the permitted range (ie 21∘C) since the outcome of theproposed method provides a combination of switched onheaters that minimizes the energy consumption satisfyinga minimum acceptable temperature Indeed compared tothe temperature variation in the baseline scenario (Figure 8)DES-CC maintains the temperature more stable and in thelower part of the allowable region intuitively implying lowercost
Figure 10 depicts the financial operation cost gains thatcan be achieved by DES-CC compared to the baseline ther-mostat approach As it can be observed the proposed energyscheduler significantly reduces the energy cost leading to atotal save of 719 eurosmonth
54 Experimental Evaluation of the DES-CCR in (13) A setof experiments have been carried out for the evaluation ofthe DES-CCR in (13) Let us recall that DES-CCR relaxesthe temperature constraints by including the constraints asa penalized term in the objective function As a resultcompared to DES-CC this method is more flexible withrespect to real time pricing tariffs More specifically DES-CC seeks a combination that minimizes the energy cost withrespect to a minimum allowable temperature Consequently
WSN mote 2
WSN mote 1
18 20 22 24 2 4 6 8 1018192021222324
Day time
18 20 22 24 2 4 6 8 10Day time
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18192021222324
Tem
pera
ture
(∘C)
Figure 9 Real and estimated temperature using DES-CC
0 2 4 6 8 10 12 14 160
0005
001
0015
002
0025
003
0035
004
Hours
Ener
gy co
st (E
uros
)
Energy cost per hour of the proposed methodEnergy cost per hour of the thermostat
Figure 10 Energy consumption cost comparison between thermo-stat and DES-CC
although the energy cost may change during time the heatercombination selected byDES-CC is the same due to the stricttemperature constraint On the other hand DES-CCR allowsthe user to further reduce the energy consumption at the costof being outside the range of temperature of comfort In thiscase to highlight the flexibility of DES-CCR we have set a
10 The Scientific World Journal
0 50 100 150 200 25016
18
20
22
24
Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
0 50 100 150 200 25016
18
20
22
24
Sample measures
Tem
pera
ture
(∘C)
Figure 11 Real and estimated temperature using DES-CCR (120579 =
02)
periodically variable value of 1199011 which alternates between
1199011
= 0009 and 1199011
= 0003 euros every thirty minutesMoreover the desired temperature has been set to 119879
119906119894=
22∘CFigures 11 and 12 depict the temperature variation in
two different cases where the users give low (120579 = 02)and high (120579 = 05) priority respectively to reduce of theenergy consumption In particular in Figure 11 (120579 = 02)the achieved temperature is very close to the desired 119879
119906119894 On
the other hand in Figure 12 we assume 120579 = 05 which is amore adapted value to the pricing policy as it correspondsto a user that permits a relaxation of the difference betweenthe real and the desired temperature to reduce the energycost This fact implies higher energy consumption in lowcost zones and lower energy consumption in high costperiods sacrificing though the userrsquos comfort Therefore theexperiments confirm that the real temperature is close to 119879
119906119894
when the energy cost is lower (ie between samples 60 and120) while there is a noticeable temperature drop whichcorresponds to lower energy consumption
6 Concluding Remarks
This paper has dealt with the energy consumption manage-ment of HVACs for a given smart pricing tariff and usersrsquocomfort constraintsMoreover the integrationwithin the IoTframework has been studied To that end we developed areal testbed consisting of (i) heaters (ii) sensor nodes thatmeasure the temperature and (iii) a gateway which providesconnection to the Internet and includes aweb application that
18
19
20
21
22
0 50 100 150 200 250Sample measures
0 50 100 150 200 250Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18
19
20
21
22
Tem
pera
ture
(∘C)
Figure 12 Real and estimated temperature using DES-CCR (120579 =
05)
permits the interaction with the user through InternetMore-over the gateway implements the algorithms that control theenergy consumption Regarding the proposed methods firstwe devised an energy scheduler that optimizes the energycost in a time interval basis for a given energy price tariffand for a given set of temperature of comfort constraintsthat are associated with different locations inside a roomThen we proposed a more flexible energy scheduler whichrelaxes the temperature constraints to further reduce theenergy consumption Namely a new objective function hasbeen considered which consists of a convex combination ofthe energy cost and a penalty term that reflects the comfortThis permits to consider both the case where the user isvery concerned with the comfort and the case where heallows relaxing the comfort constraint to further reduce theenergy consumption Experimental evaluations have beencarried out in an isolated room validating our proposals andhighlighting their potential benefits
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work has been funded by the Energy-to-Smart Grid(E2SG) project httpwwwe2sg-projecteu within theENIAC joint undertaking framework with Grant agreementnumber 296131
The Scientific World Journal 11
References
[1] L Atzori A Iera and G Morabito ldquoThe internet of things asurveyrdquoComputer Networks vol 54 no 15 pp 2787ndash2805 2010
[2] E Fleisch ldquoWhat is the internet of things An economicperspectiverdquo Tech Rep Auto-ID Labs 2010
[3] National Intelligence Council ldquoDisruptive civil technologiesmdashsix technologies with potential impacts on us interests out to2025rdquo Conference Report CR 2008-07 2008
[4] M R Palattella N Accettura X Vilajosana et al ldquoStandardizedprotocol stack for the internet of (important) thingsrdquo IEEECommunications Surveys and Tutorials vol 15 no 3 pp 1389ndash1406 2013
[5] M Zorzi A Gluhak S Lange and A Bassi ldquoFrom todaysINTRAnet of things to a future INTERnet of things a wireless-and mobility-related viewrdquo IEEEWireless Communications vol17 no 6 pp 44ndash51 2010
[6] A Gluhak S Krco M Nati D Pfisterer N Mitton andT Razafindralambo ldquoA survey on facilities for experimentalinternet of things researchrdquo IEEE Communications Magazinevol 49 no 11 pp 58ndash67 2011
[7] S Tozlu M Senel W Mao and A Keshavarzian ldquoWi-Fienabled sensors for internet of things a practical approachrdquoIEEE Communications Magazine vol 50 no 6 pp 134ndash1432012
[8] N Bui A P Castellani P Casari andM Zorzi ldquoThe internet ofenergy a web-enabled smart grid systemrdquo IEEE Network vol26 no 4 pp 39ndash45 2012
[9] X Fang S Misra G Xue and D Yang ldquoSmart gridmdashthe newand improved power grid a surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 14 no 4 pp 944ndash980 2012
[10] G Lu D De and W Song ldquoSmartGridLab a laboratory-basedsmart grid testbedrdquo in Proceedings of the 1st IEEE InternationalConference on Smart Grid Communications (SmartGridComm10) pp 143ndash148 Gaithersburg Md USA October 2010
[11] A-H Mohsenian-Rad V W S Wong J Jatskevich R Schoberand A Leon-Garcia ldquoAutonomous demand-side managementbased on game-theoretic energy consumption scheduling forthe future smart gridrdquo IEEE Transactions on Smart Grid vol1 no 3 pp 320ndash331 2010
[12] Z Zhu S Lambotharan W H Chin and Z Fan ldquoOverviewof demand management in smart grid and enabling wirelesscommunication technologiesrdquo IEEE Wireless Communicationsvol 19 no 3 pp 48ndash56 2012
[13] H T Nguyen D Nguyen and L B Le ldquoHome energy man-agement with generic thermal dynamics and user temperaturepreferencerdquo in Proceedings of the IEEE International Conferenceon Smart Grid Communications (SmartGridComm 13) pp 552ndash557 Vancouver Canada October 2013
[14] Z Zhu J Tang S Lambotharan W H Chin and Z FanldquoAn integer linear programming based optimization for homedemand-side management in smart gridrdquo in Proceedings of theIEEE PES Innovative Smart Grid Technologies (ISGT 12) pp 1ndash5Washington DC USA January 2012
[15] A-H Mohsenian-Rad and A Leon-Garcia ldquoOptimal residen-tial load control with price prediction in real-time electricitypricing environmentsrdquo IEEE Transactions on Smart Grid vol1 no 2 pp 120ndash133 2010
[16] K M Tsui and S C Chan ldquoDemand response optimizationfor smart home scheduling under real-time pricingrdquo IEEETransactions on Smart Grid vol 3 no 4 pp 1812ndash1821 2012
[17] GWood andMNewborough ldquoDynamic energy-consumptionindicators for domestic appliances environment behaviour anddesignrdquo Energy and Buildings vol 35 no 8 pp 821ndash841 2003
[18] M Avci M Erkoc and S S Asfour ldquoResidential HVAC loadcontrol strategy in real-time electricity pricing environmentrdquo inProceedings of the IEEE Energytech pp 1ndash6 Cleveland OhioUSA May 2012
[19] Z Q LuoW KMa A C So Y Ye and S Zhang ldquoSemidefiniterelaxation of quadratic optimization problemsrdquo IEEE SignalProcessing Magazine vol 27 no 3 pp 20ndash34 2010
[20] K F Fong V I Hanby and T T Chow ldquoHVAC systemoptimization for energymanagement by evolutionary program-mingrdquo Energy and Buildings vol 38 no 3 pp 220ndash231 2006
[21] D L Ha F F de Lamotte and Q-H Huynh ldquoReal-timedynamic multilevel optimization for demand-side load man-agementrdquo in Proceedings of the IEEE International Conferenceon Industrial Engineering and Engineering Management (IEEM07) pp 945ndash949 Singapore December 2007
[22] S Noh J Yun and K Kim ldquoAn efficient building air condi-tioning system control under real-time pricingrdquo in Proceedingsof the International Conference on Advanced Power SystemAutomation and Protection (APAP 11) pp 1283ndash1286 BeijingChina October 2011
[23] G L Nemhauser and L A Wolsey Integer and CombinatorialOptimization Wiley-Interscience New York NY USA 1988
[24] S Kay Fundamentals of Statistical Signal Processing EstimationTheory Prentice-Hall New York NY USA 1993
[25] ldquoContiki the open source OS for the internet of thingsrdquohttpwwwcontiki-osorg
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
6 The Scientific World Journal
Process For each time interval do(1) Input(a) The measurements of temperature of the previous interval(b) Value of 119901
11199012and 119901
3in the energy cost function (1)
(c) The userrsquos temperature of comfort constraints at eachlocation that is 119879min
119894 119879
max119894
119894 = 1 119872 in (4)(2) Estimation Step
Estimate 119886119895119894and 120574
119895
119894in the prediction model (7) using (9)
to (12) and the temperature measurements from the pasttime interval
(3) Prediction Step(a) Substitute the estimation of 119886119895
119894and 120574
119895
119894into the prediction model (7)
(b) Iterate this model to populate the vectors of minimumand maximum predicted temperatures (6)
(4)Optimization Step(a) Substitute the predicted temperatures (5) and the
quadratic cost function (1) into the optimization problem (4)(b) Solve (4) using Branch and bound method [23]
(5)OutputThe configuration of HVACs turned onoff that optimizes (4)
Algorithm 1 Dynamic energy scheduler with comfort constraints (DES-CC)
estimator minimizes the LS error criterion though it is notoptimal in general [24] Given (8) the LS estimation of 119886119895
119894
denoted by 119886119895119894is given by
119886119895
119894= xy (9)
where the symbol denotes the pseudoinverse operatorwhich is defined as x = (x119879x)minus1x119879 and we define x =
[119879119898119895
119894(1) 119879119898
119895
119894(119873 minus 1)]
119879
and y = [119879119898119895
119894(2) 119879119898
119895
119894(119873)]119879
The second situation is that some of the HVACmodules wereswitched on In this case an LS estimation is considered aswell Namely let us denote by 120574
119895
119894| 119886119895
119894the estimation of
120574119895
119894conditioned to the knowledge of a past estimation of 119886119895
119894
denoted by 119886119895
119894 Then the LS estimation for 120574119895
119894| 119886119895
119894yields
120574119895
119894| 119886119895
119894= 1z (10)
where 1 is a vector of ones of length119873 minus 1 and z is given by
z = [119879119898119895
119894(2) 119879119898
119895
119894(119873)]119879
minus 119886119895
119894[119879119898119895
119894(1) 119879119898
119895
119894(119873 minus 1)]
119879
(11)
Finally given the estimation of 120574119895119894in (10) we can update
the estimation of 119886119895119894as
119886119895
119894| 120574119895
119894= xy (12)
where y = [119879119898119895
119894(2) minus 120574
119895
119894 119879119898
119895
119894(119873) minus 120574
119895
119894]119879
414 Summary of the DES-CC Energy Scheduler At thispoint all the terms in the optimization problem understudy that is (4) are specified The procedure to implementthe proposed energy scheduler for each time interval issummarized in Algorithm 1
42 Dynamic Energy Scheduler with Comfort ConstraintsRelaxation (DES-CCR) Despite its effectiveness and its obvi-ous advantages the proposed energy scheduling algorithm iscompletely focused on the temperature constraints neglect-ing the price aspects of the problem More specificallyalthough there could be time periods where the energyprice increases the energy scheduler switches on the samecombination of HVAC modules in order to respect thetemperature constraints However in such cases users mightcompromise their comfort preferences to decrease the energyconsumption In order to allow the user to have more flexi-bility in the energy consumption a new energy scheduler willbe presented in this section This flexibility is implementedin terms of relaxing the temperature constraints to furtherreduce the energy consumption
This new energy scheduler is formulated so that theuser temperature constraints in (4) are skipped and they areincorporated as a penalty term in the objective functionConsequently the new optimization problem can be writtenas
minimizes119895isin012
119870times1
120579
119862 (119871 (s119895))
120572+ (1 minus 120579)
times
sum119872
119894=1sum119873
119899=2
10038171003817100381710038171003817q119879119894(119899)s119895minus 119879119906119894
10038171003817100381710038171003817
2
120573
(13)
where119862(119871(s119895)) is the energy cost function defined in (1)The
vector q119879119894(119899) is defined as
q119879119894(119899) = [119879119901
1
119894(119899) 119879119901
2119870
119894(119899)] (14)
and recall that 119879119901119895119894(119899) is the predicted temperature at the 119894th
location for the 119895th combination of HVACs modules turned
The Scientific World Journal 7
Process For each time interval do(1) Input(a) The measurements of temperature of the previous interval(b) Value of 119901
1 1199012and 119901
3of the energy cost function 119862(119871(119904
119895)) defined in (1)
(c) The userrsquos temperature of comfort at each location that is 119879119906119894 119894 = 1 119872 in (13)
(d) The parameter 120579 isin (0 1) controlling the comfort relaxation in (13)(2) Estimation Step
Estimate 119886119895119894and 120574
119895
119894in the prediction model (7) using (9)
to (12) and the temperature measurements from the past time interval(3) Prediction Step(a) Substitute the estimation of 119886119895
119894and 120574
119895
119894into the prediction model (7)
(b) For 119894 = 1 to119872For 119899 = 2 to119873Compute and store the vector of predicted temperaturesq119879119894(119899) in (14) using (7)
End For 119899End For 119894
(4)Optimization Step(a) Compute the cost function in (13) using the vectors in
the step 3(b) and the inputs in (1)(b) Solve the optimization problem in (13) using Branch and bound method [23]
(5)OutputThe configuration of HVACs turned onoff that optimizes (13)
Algorithm 2 Dynamic energy scheduler with comfort constraints relaxation (DES-CCR)
on or off see (7)Moreover120572 and120573 are normalizing constantsto adjust the values of the two terms in (13) Indeed we settheir value as
120572 = 119862 (119871 (s2119870))
120573 = maxs119895isin01119870times1
119872
sum
119894=1
119873
sum
119899=2
10038171003817100381710038171003817q119879119894(119899)s119895minus 119879119906119894
10038171003817100381710038171003817
2
(15)
where 119862(119871(s2119870)) is the cost for all the HVACmodules turned
on The term 119879119906119894
is the desired temperature that the userwould like tomaintain at the 119894th location of the room Clearlyour reformulation balances the two optimization problemsthat is the energy cost minimization and the user comfortmaximization Note that the user comfort is defined as anEuclidean norm but it can eventually be redefined withanother distance measurement
Finally 120579 isin (0 1) is defined by the user according to theirpreferences For example in the extreme case where 120579 = 0the demand response algorithm will not consider any priceand it will directly control the HVAC modules so that thedesired temperature is reached On the contrary when 120579 = 1theHVACmodules will always remain off In this context theusers should set the 120579 value according to their preferences andexperience The DES-CCR energy scheduler is summarizedin Algorithm 2
5 Experimental Results
In order to emulate the complete communication in anIoT framework we have designed and developed a customtestbed that integrates the described architecture In our
Figure 5 Z1 WSN mote
experiments we focus on a heating system although theproposed algorithms apply in general HVAC systems In thissection we describe the testbed platform and the experi-mental scenario we define a baseline thermostat model andfinally we present the experimental results of our proposedalgorithms
51 Testbed Description and Experimental Setup The testbedhas been deployed in a 50m2 room within our researchcenter facilities as it is depicted in Figure 5 In our particularscenario we consider three HVAC modules (ie 119870 = 3)and two temperature sensor nodes (ie 119872 = 2) The HVACmodules are distributed around the room while the sensornodes are placed in the middle of the room monitoring thetemperature and sending it to the sink mote every 30 second
8 The Scientific World Journal
Micro-USB
Ceramic embedded antennaUFL connector for external antenna
temperature sensor3-Axis accelerometer+
2 times Phidgets sensor ports
Figure 6 Overall platform detail
(ie 119905119898= 30 s) In addition the samples received from each
sensor are stored in a buffer at the control center and ouralgorithm applies every 10 samples (ie119873 = 10)
Regarding the employed technology the WSN nodes areZ1 motes by Zolertia (Figure 6) They are equipped with asecond generationMSP430F2617 low power microcontrollerwhich features a 16-bit RISC CPU 16MHz clock speed abuilt-in clock factory calibration an 8KB RAM and a 92KBflash memory They also include the CC2420 transceiverwhich is IEEE 802154 compliant operating at 24GHz fre-quency bandwith a data rate of 250 kbpsThe sensors supportContiki OS [25] an open-source operating system for the IoTwhich connects tiny low-cost low-power microcontrollersto the Internet and supports IPv6 through 6LowPAN It isworth noting that each mote can operate as either a sourceor a sink node In particular source nodes carry a TMP102temperature sensor to monitor the target field while the sinknode receives and forwards themeasured data to the gateway
The gateway (an Ubuntu OS machine with MATLAB)implements the proposed algorithms and it is able to processthe collected data Furthermore it connects the WSN to theInternet and acts as an application server using Nodejs andSencha Touch In particular Nodejs is a platform built onChromersquos JavaScript runtime for fast and scalable networkapplications Figure 7 shows a screenshot of the web applica-tion built on Nodejs which enables the user to interact withthe energy scheduler through Internet Regarding SenchaTouch it is a high-performance HTML5 mobile applicationframework which enables developers to build powerfulapplications for various operating systems including iOS andAndroidThe actuators are programmable sockets which canbe controlled remotely thanks to their IP addresses Thesespecial sockets are a set of programmable local area networksurge protectors (EG-PMS-LAN) by Energenie which areconnected via Ethernet to the gateway Finally the HVACmodules are domestic heaters with a maximum power con-sumption of 2000W
52 Baseline Model To evaluate the proposed algorithmsand highlight the potential energy and cost gains that can
Figure 7 Google Chrome screenshot of the web application
be achieved we adopt the traditional thermostat model asthe baseline reference scenario In this model the aim is tomaintain the average temperature of the room between acertain temperature range (ie [119879min119879max]) predefined bythe user To that end when the sensed temperature is above119879max at the end of a time interval all the heaters are switchedoff while the heaters are switched on when the temperaturefalls below the 119879min threshold
Figure 8 illustrates the average measured temperatureinside the room where the heaters are controlled by thethermostat In this particular case we consider 119879min =
21∘C and 119879max = 23
∘C As it can be seen in the figurethe thermostat algorithm is able to maintain the averagetemperature of the room between the desiredmargins during16 hours However it is worth noting that despite its properbehavior the particular model is not cost efficient as allHVAC modules work simultaneously consuming a totalpower consumption of 6000W In the following sectionswe evaluate our proposed methods demonstrating that theycan reduce the electrical cost with respect to the baselineapproach
The Scientific World Journal 9
21
22
23
24
HVA
C sta
tus
Average temperatureHVAC status
Day time
Tem
pera
ture
(∘C) On
Off
Thermostat-baseline model
18 20 22 24 2 4 6 8 10
Figure 8 Experimental evaluation of the thermostat model
53 Experimental Evaluation of the DES-CC in (4) Severalreal experiments have been carried out to assess the perfor-mance of the DES-CC algorithm proposed in (4) In thiscase the pricing parameters in (1) are 119901
1= 0003 and 119901
2=
1199013= 0 euros which are possible values according to [11] The
temperature bounds in (4) have been set to 119879min1
= 119879min2
=
21∘C and 119879
max1
= 119879max2
= 23∘C respectively
Figure 9 plots the variation of both the measured and theestimated temperature (using (8)ndash(12)) in the room duringour experiments As it can be noticed the error between theestimated and the real temperature is negligible somethingthat proves the accuracy of the proposed estimationmodel Inaddition the DES-CC guarantees the proper operation of thesystem as the temperature varies between the desired rangeof 21∘C and 23
∘C most of the time with very few exceptionsdue to prediction errors In the same figure it can be alsoseen that the temperature remains closer to the lower partof the permitted range (ie 21∘C) since the outcome of theproposed method provides a combination of switched onheaters that minimizes the energy consumption satisfyinga minimum acceptable temperature Indeed compared tothe temperature variation in the baseline scenario (Figure 8)DES-CC maintains the temperature more stable and in thelower part of the allowable region intuitively implying lowercost
Figure 10 depicts the financial operation cost gains thatcan be achieved by DES-CC compared to the baseline ther-mostat approach As it can be observed the proposed energyscheduler significantly reduces the energy cost leading to atotal save of 719 eurosmonth
54 Experimental Evaluation of the DES-CCR in (13) A setof experiments have been carried out for the evaluation ofthe DES-CCR in (13) Let us recall that DES-CCR relaxesthe temperature constraints by including the constraints asa penalized term in the objective function As a resultcompared to DES-CC this method is more flexible withrespect to real time pricing tariffs More specifically DES-CC seeks a combination that minimizes the energy cost withrespect to a minimum allowable temperature Consequently
WSN mote 2
WSN mote 1
18 20 22 24 2 4 6 8 1018192021222324
Day time
18 20 22 24 2 4 6 8 10Day time
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18192021222324
Tem
pera
ture
(∘C)
Figure 9 Real and estimated temperature using DES-CC
0 2 4 6 8 10 12 14 160
0005
001
0015
002
0025
003
0035
004
Hours
Ener
gy co
st (E
uros
)
Energy cost per hour of the proposed methodEnergy cost per hour of the thermostat
Figure 10 Energy consumption cost comparison between thermo-stat and DES-CC
although the energy cost may change during time the heatercombination selected byDES-CC is the same due to the stricttemperature constraint On the other hand DES-CCR allowsthe user to further reduce the energy consumption at the costof being outside the range of temperature of comfort In thiscase to highlight the flexibility of DES-CCR we have set a
10 The Scientific World Journal
0 50 100 150 200 25016
18
20
22
24
Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
0 50 100 150 200 25016
18
20
22
24
Sample measures
Tem
pera
ture
(∘C)
Figure 11 Real and estimated temperature using DES-CCR (120579 =
02)
periodically variable value of 1199011 which alternates between
1199011
= 0009 and 1199011
= 0003 euros every thirty minutesMoreover the desired temperature has been set to 119879
119906119894=
22∘CFigures 11 and 12 depict the temperature variation in
two different cases where the users give low (120579 = 02)and high (120579 = 05) priority respectively to reduce of theenergy consumption In particular in Figure 11 (120579 = 02)the achieved temperature is very close to the desired 119879
119906119894 On
the other hand in Figure 12 we assume 120579 = 05 which is amore adapted value to the pricing policy as it correspondsto a user that permits a relaxation of the difference betweenthe real and the desired temperature to reduce the energycost This fact implies higher energy consumption in lowcost zones and lower energy consumption in high costperiods sacrificing though the userrsquos comfort Therefore theexperiments confirm that the real temperature is close to 119879
119906119894
when the energy cost is lower (ie between samples 60 and120) while there is a noticeable temperature drop whichcorresponds to lower energy consumption
6 Concluding Remarks
This paper has dealt with the energy consumption manage-ment of HVACs for a given smart pricing tariff and usersrsquocomfort constraintsMoreover the integrationwithin the IoTframework has been studied To that end we developed areal testbed consisting of (i) heaters (ii) sensor nodes thatmeasure the temperature and (iii) a gateway which providesconnection to the Internet and includes aweb application that
18
19
20
21
22
0 50 100 150 200 250Sample measures
0 50 100 150 200 250Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18
19
20
21
22
Tem
pera
ture
(∘C)
Figure 12 Real and estimated temperature using DES-CCR (120579 =
05)
permits the interaction with the user through InternetMore-over the gateway implements the algorithms that control theenergy consumption Regarding the proposed methods firstwe devised an energy scheduler that optimizes the energycost in a time interval basis for a given energy price tariffand for a given set of temperature of comfort constraintsthat are associated with different locations inside a roomThen we proposed a more flexible energy scheduler whichrelaxes the temperature constraints to further reduce theenergy consumption Namely a new objective function hasbeen considered which consists of a convex combination ofthe energy cost and a penalty term that reflects the comfortThis permits to consider both the case where the user isvery concerned with the comfort and the case where heallows relaxing the comfort constraint to further reduce theenergy consumption Experimental evaluations have beencarried out in an isolated room validating our proposals andhighlighting their potential benefits
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work has been funded by the Energy-to-Smart Grid(E2SG) project httpwwwe2sg-projecteu within theENIAC joint undertaking framework with Grant agreementnumber 296131
The Scientific World Journal 11
References
[1] L Atzori A Iera and G Morabito ldquoThe internet of things asurveyrdquoComputer Networks vol 54 no 15 pp 2787ndash2805 2010
[2] E Fleisch ldquoWhat is the internet of things An economicperspectiverdquo Tech Rep Auto-ID Labs 2010
[3] National Intelligence Council ldquoDisruptive civil technologiesmdashsix technologies with potential impacts on us interests out to2025rdquo Conference Report CR 2008-07 2008
[4] M R Palattella N Accettura X Vilajosana et al ldquoStandardizedprotocol stack for the internet of (important) thingsrdquo IEEECommunications Surveys and Tutorials vol 15 no 3 pp 1389ndash1406 2013
[5] M Zorzi A Gluhak S Lange and A Bassi ldquoFrom todaysINTRAnet of things to a future INTERnet of things a wireless-and mobility-related viewrdquo IEEEWireless Communications vol17 no 6 pp 44ndash51 2010
[6] A Gluhak S Krco M Nati D Pfisterer N Mitton andT Razafindralambo ldquoA survey on facilities for experimentalinternet of things researchrdquo IEEE Communications Magazinevol 49 no 11 pp 58ndash67 2011
[7] S Tozlu M Senel W Mao and A Keshavarzian ldquoWi-Fienabled sensors for internet of things a practical approachrdquoIEEE Communications Magazine vol 50 no 6 pp 134ndash1432012
[8] N Bui A P Castellani P Casari andM Zorzi ldquoThe internet ofenergy a web-enabled smart grid systemrdquo IEEE Network vol26 no 4 pp 39ndash45 2012
[9] X Fang S Misra G Xue and D Yang ldquoSmart gridmdashthe newand improved power grid a surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 14 no 4 pp 944ndash980 2012
[10] G Lu D De and W Song ldquoSmartGridLab a laboratory-basedsmart grid testbedrdquo in Proceedings of the 1st IEEE InternationalConference on Smart Grid Communications (SmartGridComm10) pp 143ndash148 Gaithersburg Md USA October 2010
[11] A-H Mohsenian-Rad V W S Wong J Jatskevich R Schoberand A Leon-Garcia ldquoAutonomous demand-side managementbased on game-theoretic energy consumption scheduling forthe future smart gridrdquo IEEE Transactions on Smart Grid vol1 no 3 pp 320ndash331 2010
[12] Z Zhu S Lambotharan W H Chin and Z Fan ldquoOverviewof demand management in smart grid and enabling wirelesscommunication technologiesrdquo IEEE Wireless Communicationsvol 19 no 3 pp 48ndash56 2012
[13] H T Nguyen D Nguyen and L B Le ldquoHome energy man-agement with generic thermal dynamics and user temperaturepreferencerdquo in Proceedings of the IEEE International Conferenceon Smart Grid Communications (SmartGridComm 13) pp 552ndash557 Vancouver Canada October 2013
[14] Z Zhu J Tang S Lambotharan W H Chin and Z FanldquoAn integer linear programming based optimization for homedemand-side management in smart gridrdquo in Proceedings of theIEEE PES Innovative Smart Grid Technologies (ISGT 12) pp 1ndash5Washington DC USA January 2012
[15] A-H Mohsenian-Rad and A Leon-Garcia ldquoOptimal residen-tial load control with price prediction in real-time electricitypricing environmentsrdquo IEEE Transactions on Smart Grid vol1 no 2 pp 120ndash133 2010
[16] K M Tsui and S C Chan ldquoDemand response optimizationfor smart home scheduling under real-time pricingrdquo IEEETransactions on Smart Grid vol 3 no 4 pp 1812ndash1821 2012
[17] GWood andMNewborough ldquoDynamic energy-consumptionindicators for domestic appliances environment behaviour anddesignrdquo Energy and Buildings vol 35 no 8 pp 821ndash841 2003
[18] M Avci M Erkoc and S S Asfour ldquoResidential HVAC loadcontrol strategy in real-time electricity pricing environmentrdquo inProceedings of the IEEE Energytech pp 1ndash6 Cleveland OhioUSA May 2012
[19] Z Q LuoW KMa A C So Y Ye and S Zhang ldquoSemidefiniterelaxation of quadratic optimization problemsrdquo IEEE SignalProcessing Magazine vol 27 no 3 pp 20ndash34 2010
[20] K F Fong V I Hanby and T T Chow ldquoHVAC systemoptimization for energymanagement by evolutionary program-mingrdquo Energy and Buildings vol 38 no 3 pp 220ndash231 2006
[21] D L Ha F F de Lamotte and Q-H Huynh ldquoReal-timedynamic multilevel optimization for demand-side load man-agementrdquo in Proceedings of the IEEE International Conferenceon Industrial Engineering and Engineering Management (IEEM07) pp 945ndash949 Singapore December 2007
[22] S Noh J Yun and K Kim ldquoAn efficient building air condi-tioning system control under real-time pricingrdquo in Proceedingsof the International Conference on Advanced Power SystemAutomation and Protection (APAP 11) pp 1283ndash1286 BeijingChina October 2011
[23] G L Nemhauser and L A Wolsey Integer and CombinatorialOptimization Wiley-Interscience New York NY USA 1988
[24] S Kay Fundamentals of Statistical Signal Processing EstimationTheory Prentice-Hall New York NY USA 1993
[25] ldquoContiki the open source OS for the internet of thingsrdquohttpwwwcontiki-osorg
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World Journal 7
Process For each time interval do(1) Input(a) The measurements of temperature of the previous interval(b) Value of 119901
1 1199012and 119901
3of the energy cost function 119862(119871(119904
119895)) defined in (1)
(c) The userrsquos temperature of comfort at each location that is 119879119906119894 119894 = 1 119872 in (13)
(d) The parameter 120579 isin (0 1) controlling the comfort relaxation in (13)(2) Estimation Step
Estimate 119886119895119894and 120574
119895
119894in the prediction model (7) using (9)
to (12) and the temperature measurements from the past time interval(3) Prediction Step(a) Substitute the estimation of 119886119895
119894and 120574
119895
119894into the prediction model (7)
(b) For 119894 = 1 to119872For 119899 = 2 to119873Compute and store the vector of predicted temperaturesq119879119894(119899) in (14) using (7)
End For 119899End For 119894
(4)Optimization Step(a) Compute the cost function in (13) using the vectors in
the step 3(b) and the inputs in (1)(b) Solve the optimization problem in (13) using Branch and bound method [23]
(5)OutputThe configuration of HVACs turned onoff that optimizes (13)
Algorithm 2 Dynamic energy scheduler with comfort constraints relaxation (DES-CCR)
on or off see (7)Moreover120572 and120573 are normalizing constantsto adjust the values of the two terms in (13) Indeed we settheir value as
120572 = 119862 (119871 (s2119870))
120573 = maxs119895isin01119870times1
119872
sum
119894=1
119873
sum
119899=2
10038171003817100381710038171003817q119879119894(119899)s119895minus 119879119906119894
10038171003817100381710038171003817
2
(15)
where 119862(119871(s2119870)) is the cost for all the HVACmodules turned
on The term 119879119906119894
is the desired temperature that the userwould like tomaintain at the 119894th location of the room Clearlyour reformulation balances the two optimization problemsthat is the energy cost minimization and the user comfortmaximization Note that the user comfort is defined as anEuclidean norm but it can eventually be redefined withanother distance measurement
Finally 120579 isin (0 1) is defined by the user according to theirpreferences For example in the extreme case where 120579 = 0the demand response algorithm will not consider any priceand it will directly control the HVAC modules so that thedesired temperature is reached On the contrary when 120579 = 1theHVACmodules will always remain off In this context theusers should set the 120579 value according to their preferences andexperience The DES-CCR energy scheduler is summarizedin Algorithm 2
5 Experimental Results
In order to emulate the complete communication in anIoT framework we have designed and developed a customtestbed that integrates the described architecture In our
Figure 5 Z1 WSN mote
experiments we focus on a heating system although theproposed algorithms apply in general HVAC systems In thissection we describe the testbed platform and the experi-mental scenario we define a baseline thermostat model andfinally we present the experimental results of our proposedalgorithms
51 Testbed Description and Experimental Setup The testbedhas been deployed in a 50m2 room within our researchcenter facilities as it is depicted in Figure 5 In our particularscenario we consider three HVAC modules (ie 119870 = 3)and two temperature sensor nodes (ie 119872 = 2) The HVACmodules are distributed around the room while the sensornodes are placed in the middle of the room monitoring thetemperature and sending it to the sink mote every 30 second
8 The Scientific World Journal
Micro-USB
Ceramic embedded antennaUFL connector for external antenna
temperature sensor3-Axis accelerometer+
2 times Phidgets sensor ports
Figure 6 Overall platform detail
(ie 119905119898= 30 s) In addition the samples received from each
sensor are stored in a buffer at the control center and ouralgorithm applies every 10 samples (ie119873 = 10)
Regarding the employed technology the WSN nodes areZ1 motes by Zolertia (Figure 6) They are equipped with asecond generationMSP430F2617 low power microcontrollerwhich features a 16-bit RISC CPU 16MHz clock speed abuilt-in clock factory calibration an 8KB RAM and a 92KBflash memory They also include the CC2420 transceiverwhich is IEEE 802154 compliant operating at 24GHz fre-quency bandwith a data rate of 250 kbpsThe sensors supportContiki OS [25] an open-source operating system for the IoTwhich connects tiny low-cost low-power microcontrollersto the Internet and supports IPv6 through 6LowPAN It isworth noting that each mote can operate as either a sourceor a sink node In particular source nodes carry a TMP102temperature sensor to monitor the target field while the sinknode receives and forwards themeasured data to the gateway
The gateway (an Ubuntu OS machine with MATLAB)implements the proposed algorithms and it is able to processthe collected data Furthermore it connects the WSN to theInternet and acts as an application server using Nodejs andSencha Touch In particular Nodejs is a platform built onChromersquos JavaScript runtime for fast and scalable networkapplications Figure 7 shows a screenshot of the web applica-tion built on Nodejs which enables the user to interact withthe energy scheduler through Internet Regarding SenchaTouch it is a high-performance HTML5 mobile applicationframework which enables developers to build powerfulapplications for various operating systems including iOS andAndroidThe actuators are programmable sockets which canbe controlled remotely thanks to their IP addresses Thesespecial sockets are a set of programmable local area networksurge protectors (EG-PMS-LAN) by Energenie which areconnected via Ethernet to the gateway Finally the HVACmodules are domestic heaters with a maximum power con-sumption of 2000W
52 Baseline Model To evaluate the proposed algorithmsand highlight the potential energy and cost gains that can
Figure 7 Google Chrome screenshot of the web application
be achieved we adopt the traditional thermostat model asthe baseline reference scenario In this model the aim is tomaintain the average temperature of the room between acertain temperature range (ie [119879min119879max]) predefined bythe user To that end when the sensed temperature is above119879max at the end of a time interval all the heaters are switchedoff while the heaters are switched on when the temperaturefalls below the 119879min threshold
Figure 8 illustrates the average measured temperatureinside the room where the heaters are controlled by thethermostat In this particular case we consider 119879min =
21∘C and 119879max = 23
∘C As it can be seen in the figurethe thermostat algorithm is able to maintain the averagetemperature of the room between the desiredmargins during16 hours However it is worth noting that despite its properbehavior the particular model is not cost efficient as allHVAC modules work simultaneously consuming a totalpower consumption of 6000W In the following sectionswe evaluate our proposed methods demonstrating that theycan reduce the electrical cost with respect to the baselineapproach
The Scientific World Journal 9
21
22
23
24
HVA
C sta
tus
Average temperatureHVAC status
Day time
Tem
pera
ture
(∘C) On
Off
Thermostat-baseline model
18 20 22 24 2 4 6 8 10
Figure 8 Experimental evaluation of the thermostat model
53 Experimental Evaluation of the DES-CC in (4) Severalreal experiments have been carried out to assess the perfor-mance of the DES-CC algorithm proposed in (4) In thiscase the pricing parameters in (1) are 119901
1= 0003 and 119901
2=
1199013= 0 euros which are possible values according to [11] The
temperature bounds in (4) have been set to 119879min1
= 119879min2
=
21∘C and 119879
max1
= 119879max2
= 23∘C respectively
Figure 9 plots the variation of both the measured and theestimated temperature (using (8)ndash(12)) in the room duringour experiments As it can be noticed the error between theestimated and the real temperature is negligible somethingthat proves the accuracy of the proposed estimationmodel Inaddition the DES-CC guarantees the proper operation of thesystem as the temperature varies between the desired rangeof 21∘C and 23
∘C most of the time with very few exceptionsdue to prediction errors In the same figure it can be alsoseen that the temperature remains closer to the lower partof the permitted range (ie 21∘C) since the outcome of theproposed method provides a combination of switched onheaters that minimizes the energy consumption satisfyinga minimum acceptable temperature Indeed compared tothe temperature variation in the baseline scenario (Figure 8)DES-CC maintains the temperature more stable and in thelower part of the allowable region intuitively implying lowercost
Figure 10 depicts the financial operation cost gains thatcan be achieved by DES-CC compared to the baseline ther-mostat approach As it can be observed the proposed energyscheduler significantly reduces the energy cost leading to atotal save of 719 eurosmonth
54 Experimental Evaluation of the DES-CCR in (13) A setof experiments have been carried out for the evaluation ofthe DES-CCR in (13) Let us recall that DES-CCR relaxesthe temperature constraints by including the constraints asa penalized term in the objective function As a resultcompared to DES-CC this method is more flexible withrespect to real time pricing tariffs More specifically DES-CC seeks a combination that minimizes the energy cost withrespect to a minimum allowable temperature Consequently
WSN mote 2
WSN mote 1
18 20 22 24 2 4 6 8 1018192021222324
Day time
18 20 22 24 2 4 6 8 10Day time
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18192021222324
Tem
pera
ture
(∘C)
Figure 9 Real and estimated temperature using DES-CC
0 2 4 6 8 10 12 14 160
0005
001
0015
002
0025
003
0035
004
Hours
Ener
gy co
st (E
uros
)
Energy cost per hour of the proposed methodEnergy cost per hour of the thermostat
Figure 10 Energy consumption cost comparison between thermo-stat and DES-CC
although the energy cost may change during time the heatercombination selected byDES-CC is the same due to the stricttemperature constraint On the other hand DES-CCR allowsthe user to further reduce the energy consumption at the costof being outside the range of temperature of comfort In thiscase to highlight the flexibility of DES-CCR we have set a
10 The Scientific World Journal
0 50 100 150 200 25016
18
20
22
24
Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
0 50 100 150 200 25016
18
20
22
24
Sample measures
Tem
pera
ture
(∘C)
Figure 11 Real and estimated temperature using DES-CCR (120579 =
02)
periodically variable value of 1199011 which alternates between
1199011
= 0009 and 1199011
= 0003 euros every thirty minutesMoreover the desired temperature has been set to 119879
119906119894=
22∘CFigures 11 and 12 depict the temperature variation in
two different cases where the users give low (120579 = 02)and high (120579 = 05) priority respectively to reduce of theenergy consumption In particular in Figure 11 (120579 = 02)the achieved temperature is very close to the desired 119879
119906119894 On
the other hand in Figure 12 we assume 120579 = 05 which is amore adapted value to the pricing policy as it correspondsto a user that permits a relaxation of the difference betweenthe real and the desired temperature to reduce the energycost This fact implies higher energy consumption in lowcost zones and lower energy consumption in high costperiods sacrificing though the userrsquos comfort Therefore theexperiments confirm that the real temperature is close to 119879
119906119894
when the energy cost is lower (ie between samples 60 and120) while there is a noticeable temperature drop whichcorresponds to lower energy consumption
6 Concluding Remarks
This paper has dealt with the energy consumption manage-ment of HVACs for a given smart pricing tariff and usersrsquocomfort constraintsMoreover the integrationwithin the IoTframework has been studied To that end we developed areal testbed consisting of (i) heaters (ii) sensor nodes thatmeasure the temperature and (iii) a gateway which providesconnection to the Internet and includes aweb application that
18
19
20
21
22
0 50 100 150 200 250Sample measures
0 50 100 150 200 250Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18
19
20
21
22
Tem
pera
ture
(∘C)
Figure 12 Real and estimated temperature using DES-CCR (120579 =
05)
permits the interaction with the user through InternetMore-over the gateway implements the algorithms that control theenergy consumption Regarding the proposed methods firstwe devised an energy scheduler that optimizes the energycost in a time interval basis for a given energy price tariffand for a given set of temperature of comfort constraintsthat are associated with different locations inside a roomThen we proposed a more flexible energy scheduler whichrelaxes the temperature constraints to further reduce theenergy consumption Namely a new objective function hasbeen considered which consists of a convex combination ofthe energy cost and a penalty term that reflects the comfortThis permits to consider both the case where the user isvery concerned with the comfort and the case where heallows relaxing the comfort constraint to further reduce theenergy consumption Experimental evaluations have beencarried out in an isolated room validating our proposals andhighlighting their potential benefits
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work has been funded by the Energy-to-Smart Grid(E2SG) project httpwwwe2sg-projecteu within theENIAC joint undertaking framework with Grant agreementnumber 296131
The Scientific World Journal 11
References
[1] L Atzori A Iera and G Morabito ldquoThe internet of things asurveyrdquoComputer Networks vol 54 no 15 pp 2787ndash2805 2010
[2] E Fleisch ldquoWhat is the internet of things An economicperspectiverdquo Tech Rep Auto-ID Labs 2010
[3] National Intelligence Council ldquoDisruptive civil technologiesmdashsix technologies with potential impacts on us interests out to2025rdquo Conference Report CR 2008-07 2008
[4] M R Palattella N Accettura X Vilajosana et al ldquoStandardizedprotocol stack for the internet of (important) thingsrdquo IEEECommunications Surveys and Tutorials vol 15 no 3 pp 1389ndash1406 2013
[5] M Zorzi A Gluhak S Lange and A Bassi ldquoFrom todaysINTRAnet of things to a future INTERnet of things a wireless-and mobility-related viewrdquo IEEEWireless Communications vol17 no 6 pp 44ndash51 2010
[6] A Gluhak S Krco M Nati D Pfisterer N Mitton andT Razafindralambo ldquoA survey on facilities for experimentalinternet of things researchrdquo IEEE Communications Magazinevol 49 no 11 pp 58ndash67 2011
[7] S Tozlu M Senel W Mao and A Keshavarzian ldquoWi-Fienabled sensors for internet of things a practical approachrdquoIEEE Communications Magazine vol 50 no 6 pp 134ndash1432012
[8] N Bui A P Castellani P Casari andM Zorzi ldquoThe internet ofenergy a web-enabled smart grid systemrdquo IEEE Network vol26 no 4 pp 39ndash45 2012
[9] X Fang S Misra G Xue and D Yang ldquoSmart gridmdashthe newand improved power grid a surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 14 no 4 pp 944ndash980 2012
[10] G Lu D De and W Song ldquoSmartGridLab a laboratory-basedsmart grid testbedrdquo in Proceedings of the 1st IEEE InternationalConference on Smart Grid Communications (SmartGridComm10) pp 143ndash148 Gaithersburg Md USA October 2010
[11] A-H Mohsenian-Rad V W S Wong J Jatskevich R Schoberand A Leon-Garcia ldquoAutonomous demand-side managementbased on game-theoretic energy consumption scheduling forthe future smart gridrdquo IEEE Transactions on Smart Grid vol1 no 3 pp 320ndash331 2010
[12] Z Zhu S Lambotharan W H Chin and Z Fan ldquoOverviewof demand management in smart grid and enabling wirelesscommunication technologiesrdquo IEEE Wireless Communicationsvol 19 no 3 pp 48ndash56 2012
[13] H T Nguyen D Nguyen and L B Le ldquoHome energy man-agement with generic thermal dynamics and user temperaturepreferencerdquo in Proceedings of the IEEE International Conferenceon Smart Grid Communications (SmartGridComm 13) pp 552ndash557 Vancouver Canada October 2013
[14] Z Zhu J Tang S Lambotharan W H Chin and Z FanldquoAn integer linear programming based optimization for homedemand-side management in smart gridrdquo in Proceedings of theIEEE PES Innovative Smart Grid Technologies (ISGT 12) pp 1ndash5Washington DC USA January 2012
[15] A-H Mohsenian-Rad and A Leon-Garcia ldquoOptimal residen-tial load control with price prediction in real-time electricitypricing environmentsrdquo IEEE Transactions on Smart Grid vol1 no 2 pp 120ndash133 2010
[16] K M Tsui and S C Chan ldquoDemand response optimizationfor smart home scheduling under real-time pricingrdquo IEEETransactions on Smart Grid vol 3 no 4 pp 1812ndash1821 2012
[17] GWood andMNewborough ldquoDynamic energy-consumptionindicators for domestic appliances environment behaviour anddesignrdquo Energy and Buildings vol 35 no 8 pp 821ndash841 2003
[18] M Avci M Erkoc and S S Asfour ldquoResidential HVAC loadcontrol strategy in real-time electricity pricing environmentrdquo inProceedings of the IEEE Energytech pp 1ndash6 Cleveland OhioUSA May 2012
[19] Z Q LuoW KMa A C So Y Ye and S Zhang ldquoSemidefiniterelaxation of quadratic optimization problemsrdquo IEEE SignalProcessing Magazine vol 27 no 3 pp 20ndash34 2010
[20] K F Fong V I Hanby and T T Chow ldquoHVAC systemoptimization for energymanagement by evolutionary program-mingrdquo Energy and Buildings vol 38 no 3 pp 220ndash231 2006
[21] D L Ha F F de Lamotte and Q-H Huynh ldquoReal-timedynamic multilevel optimization for demand-side load man-agementrdquo in Proceedings of the IEEE International Conferenceon Industrial Engineering and Engineering Management (IEEM07) pp 945ndash949 Singapore December 2007
[22] S Noh J Yun and K Kim ldquoAn efficient building air condi-tioning system control under real-time pricingrdquo in Proceedingsof the International Conference on Advanced Power SystemAutomation and Protection (APAP 11) pp 1283ndash1286 BeijingChina October 2011
[23] G L Nemhauser and L A Wolsey Integer and CombinatorialOptimization Wiley-Interscience New York NY USA 1988
[24] S Kay Fundamentals of Statistical Signal Processing EstimationTheory Prentice-Hall New York NY USA 1993
[25] ldquoContiki the open source OS for the internet of thingsrdquohttpwwwcontiki-osorg
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
8 The Scientific World Journal
Micro-USB
Ceramic embedded antennaUFL connector for external antenna
temperature sensor3-Axis accelerometer+
2 times Phidgets sensor ports
Figure 6 Overall platform detail
(ie 119905119898= 30 s) In addition the samples received from each
sensor are stored in a buffer at the control center and ouralgorithm applies every 10 samples (ie119873 = 10)
Regarding the employed technology the WSN nodes areZ1 motes by Zolertia (Figure 6) They are equipped with asecond generationMSP430F2617 low power microcontrollerwhich features a 16-bit RISC CPU 16MHz clock speed abuilt-in clock factory calibration an 8KB RAM and a 92KBflash memory They also include the CC2420 transceiverwhich is IEEE 802154 compliant operating at 24GHz fre-quency bandwith a data rate of 250 kbpsThe sensors supportContiki OS [25] an open-source operating system for the IoTwhich connects tiny low-cost low-power microcontrollersto the Internet and supports IPv6 through 6LowPAN It isworth noting that each mote can operate as either a sourceor a sink node In particular source nodes carry a TMP102temperature sensor to monitor the target field while the sinknode receives and forwards themeasured data to the gateway
The gateway (an Ubuntu OS machine with MATLAB)implements the proposed algorithms and it is able to processthe collected data Furthermore it connects the WSN to theInternet and acts as an application server using Nodejs andSencha Touch In particular Nodejs is a platform built onChromersquos JavaScript runtime for fast and scalable networkapplications Figure 7 shows a screenshot of the web applica-tion built on Nodejs which enables the user to interact withthe energy scheduler through Internet Regarding SenchaTouch it is a high-performance HTML5 mobile applicationframework which enables developers to build powerfulapplications for various operating systems including iOS andAndroidThe actuators are programmable sockets which canbe controlled remotely thanks to their IP addresses Thesespecial sockets are a set of programmable local area networksurge protectors (EG-PMS-LAN) by Energenie which areconnected via Ethernet to the gateway Finally the HVACmodules are domestic heaters with a maximum power con-sumption of 2000W
52 Baseline Model To evaluate the proposed algorithmsand highlight the potential energy and cost gains that can
Figure 7 Google Chrome screenshot of the web application
be achieved we adopt the traditional thermostat model asthe baseline reference scenario In this model the aim is tomaintain the average temperature of the room between acertain temperature range (ie [119879min119879max]) predefined bythe user To that end when the sensed temperature is above119879max at the end of a time interval all the heaters are switchedoff while the heaters are switched on when the temperaturefalls below the 119879min threshold
Figure 8 illustrates the average measured temperatureinside the room where the heaters are controlled by thethermostat In this particular case we consider 119879min =
21∘C and 119879max = 23
∘C As it can be seen in the figurethe thermostat algorithm is able to maintain the averagetemperature of the room between the desiredmargins during16 hours However it is worth noting that despite its properbehavior the particular model is not cost efficient as allHVAC modules work simultaneously consuming a totalpower consumption of 6000W In the following sectionswe evaluate our proposed methods demonstrating that theycan reduce the electrical cost with respect to the baselineapproach
The Scientific World Journal 9
21
22
23
24
HVA
C sta
tus
Average temperatureHVAC status
Day time
Tem
pera
ture
(∘C) On
Off
Thermostat-baseline model
18 20 22 24 2 4 6 8 10
Figure 8 Experimental evaluation of the thermostat model
53 Experimental Evaluation of the DES-CC in (4) Severalreal experiments have been carried out to assess the perfor-mance of the DES-CC algorithm proposed in (4) In thiscase the pricing parameters in (1) are 119901
1= 0003 and 119901
2=
1199013= 0 euros which are possible values according to [11] The
temperature bounds in (4) have been set to 119879min1
= 119879min2
=
21∘C and 119879
max1
= 119879max2
= 23∘C respectively
Figure 9 plots the variation of both the measured and theestimated temperature (using (8)ndash(12)) in the room duringour experiments As it can be noticed the error between theestimated and the real temperature is negligible somethingthat proves the accuracy of the proposed estimationmodel Inaddition the DES-CC guarantees the proper operation of thesystem as the temperature varies between the desired rangeof 21∘C and 23
∘C most of the time with very few exceptionsdue to prediction errors In the same figure it can be alsoseen that the temperature remains closer to the lower partof the permitted range (ie 21∘C) since the outcome of theproposed method provides a combination of switched onheaters that minimizes the energy consumption satisfyinga minimum acceptable temperature Indeed compared tothe temperature variation in the baseline scenario (Figure 8)DES-CC maintains the temperature more stable and in thelower part of the allowable region intuitively implying lowercost
Figure 10 depicts the financial operation cost gains thatcan be achieved by DES-CC compared to the baseline ther-mostat approach As it can be observed the proposed energyscheduler significantly reduces the energy cost leading to atotal save of 719 eurosmonth
54 Experimental Evaluation of the DES-CCR in (13) A setof experiments have been carried out for the evaluation ofthe DES-CCR in (13) Let us recall that DES-CCR relaxesthe temperature constraints by including the constraints asa penalized term in the objective function As a resultcompared to DES-CC this method is more flexible withrespect to real time pricing tariffs More specifically DES-CC seeks a combination that minimizes the energy cost withrespect to a minimum allowable temperature Consequently
WSN mote 2
WSN mote 1
18 20 22 24 2 4 6 8 1018192021222324
Day time
18 20 22 24 2 4 6 8 10Day time
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18192021222324
Tem
pera
ture
(∘C)
Figure 9 Real and estimated temperature using DES-CC
0 2 4 6 8 10 12 14 160
0005
001
0015
002
0025
003
0035
004
Hours
Ener
gy co
st (E
uros
)
Energy cost per hour of the proposed methodEnergy cost per hour of the thermostat
Figure 10 Energy consumption cost comparison between thermo-stat and DES-CC
although the energy cost may change during time the heatercombination selected byDES-CC is the same due to the stricttemperature constraint On the other hand DES-CCR allowsthe user to further reduce the energy consumption at the costof being outside the range of temperature of comfort In thiscase to highlight the flexibility of DES-CCR we have set a
10 The Scientific World Journal
0 50 100 150 200 25016
18
20
22
24
Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
0 50 100 150 200 25016
18
20
22
24
Sample measures
Tem
pera
ture
(∘C)
Figure 11 Real and estimated temperature using DES-CCR (120579 =
02)
periodically variable value of 1199011 which alternates between
1199011
= 0009 and 1199011
= 0003 euros every thirty minutesMoreover the desired temperature has been set to 119879
119906119894=
22∘CFigures 11 and 12 depict the temperature variation in
two different cases where the users give low (120579 = 02)and high (120579 = 05) priority respectively to reduce of theenergy consumption In particular in Figure 11 (120579 = 02)the achieved temperature is very close to the desired 119879
119906119894 On
the other hand in Figure 12 we assume 120579 = 05 which is amore adapted value to the pricing policy as it correspondsto a user that permits a relaxation of the difference betweenthe real and the desired temperature to reduce the energycost This fact implies higher energy consumption in lowcost zones and lower energy consumption in high costperiods sacrificing though the userrsquos comfort Therefore theexperiments confirm that the real temperature is close to 119879
119906119894
when the energy cost is lower (ie between samples 60 and120) while there is a noticeable temperature drop whichcorresponds to lower energy consumption
6 Concluding Remarks
This paper has dealt with the energy consumption manage-ment of HVACs for a given smart pricing tariff and usersrsquocomfort constraintsMoreover the integrationwithin the IoTframework has been studied To that end we developed areal testbed consisting of (i) heaters (ii) sensor nodes thatmeasure the temperature and (iii) a gateway which providesconnection to the Internet and includes aweb application that
18
19
20
21
22
0 50 100 150 200 250Sample measures
0 50 100 150 200 250Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18
19
20
21
22
Tem
pera
ture
(∘C)
Figure 12 Real and estimated temperature using DES-CCR (120579 =
05)
permits the interaction with the user through InternetMore-over the gateway implements the algorithms that control theenergy consumption Regarding the proposed methods firstwe devised an energy scheduler that optimizes the energycost in a time interval basis for a given energy price tariffand for a given set of temperature of comfort constraintsthat are associated with different locations inside a roomThen we proposed a more flexible energy scheduler whichrelaxes the temperature constraints to further reduce theenergy consumption Namely a new objective function hasbeen considered which consists of a convex combination ofthe energy cost and a penalty term that reflects the comfortThis permits to consider both the case where the user isvery concerned with the comfort and the case where heallows relaxing the comfort constraint to further reduce theenergy consumption Experimental evaluations have beencarried out in an isolated room validating our proposals andhighlighting their potential benefits
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work has been funded by the Energy-to-Smart Grid(E2SG) project httpwwwe2sg-projecteu within theENIAC joint undertaking framework with Grant agreementnumber 296131
The Scientific World Journal 11
References
[1] L Atzori A Iera and G Morabito ldquoThe internet of things asurveyrdquoComputer Networks vol 54 no 15 pp 2787ndash2805 2010
[2] E Fleisch ldquoWhat is the internet of things An economicperspectiverdquo Tech Rep Auto-ID Labs 2010
[3] National Intelligence Council ldquoDisruptive civil technologiesmdashsix technologies with potential impacts on us interests out to2025rdquo Conference Report CR 2008-07 2008
[4] M R Palattella N Accettura X Vilajosana et al ldquoStandardizedprotocol stack for the internet of (important) thingsrdquo IEEECommunications Surveys and Tutorials vol 15 no 3 pp 1389ndash1406 2013
[5] M Zorzi A Gluhak S Lange and A Bassi ldquoFrom todaysINTRAnet of things to a future INTERnet of things a wireless-and mobility-related viewrdquo IEEEWireless Communications vol17 no 6 pp 44ndash51 2010
[6] A Gluhak S Krco M Nati D Pfisterer N Mitton andT Razafindralambo ldquoA survey on facilities for experimentalinternet of things researchrdquo IEEE Communications Magazinevol 49 no 11 pp 58ndash67 2011
[7] S Tozlu M Senel W Mao and A Keshavarzian ldquoWi-Fienabled sensors for internet of things a practical approachrdquoIEEE Communications Magazine vol 50 no 6 pp 134ndash1432012
[8] N Bui A P Castellani P Casari andM Zorzi ldquoThe internet ofenergy a web-enabled smart grid systemrdquo IEEE Network vol26 no 4 pp 39ndash45 2012
[9] X Fang S Misra G Xue and D Yang ldquoSmart gridmdashthe newand improved power grid a surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 14 no 4 pp 944ndash980 2012
[10] G Lu D De and W Song ldquoSmartGridLab a laboratory-basedsmart grid testbedrdquo in Proceedings of the 1st IEEE InternationalConference on Smart Grid Communications (SmartGridComm10) pp 143ndash148 Gaithersburg Md USA October 2010
[11] A-H Mohsenian-Rad V W S Wong J Jatskevich R Schoberand A Leon-Garcia ldquoAutonomous demand-side managementbased on game-theoretic energy consumption scheduling forthe future smart gridrdquo IEEE Transactions on Smart Grid vol1 no 3 pp 320ndash331 2010
[12] Z Zhu S Lambotharan W H Chin and Z Fan ldquoOverviewof demand management in smart grid and enabling wirelesscommunication technologiesrdquo IEEE Wireless Communicationsvol 19 no 3 pp 48ndash56 2012
[13] H T Nguyen D Nguyen and L B Le ldquoHome energy man-agement with generic thermal dynamics and user temperaturepreferencerdquo in Proceedings of the IEEE International Conferenceon Smart Grid Communications (SmartGridComm 13) pp 552ndash557 Vancouver Canada October 2013
[14] Z Zhu J Tang S Lambotharan W H Chin and Z FanldquoAn integer linear programming based optimization for homedemand-side management in smart gridrdquo in Proceedings of theIEEE PES Innovative Smart Grid Technologies (ISGT 12) pp 1ndash5Washington DC USA January 2012
[15] A-H Mohsenian-Rad and A Leon-Garcia ldquoOptimal residen-tial load control with price prediction in real-time electricitypricing environmentsrdquo IEEE Transactions on Smart Grid vol1 no 2 pp 120ndash133 2010
[16] K M Tsui and S C Chan ldquoDemand response optimizationfor smart home scheduling under real-time pricingrdquo IEEETransactions on Smart Grid vol 3 no 4 pp 1812ndash1821 2012
[17] GWood andMNewborough ldquoDynamic energy-consumptionindicators for domestic appliances environment behaviour anddesignrdquo Energy and Buildings vol 35 no 8 pp 821ndash841 2003
[18] M Avci M Erkoc and S S Asfour ldquoResidential HVAC loadcontrol strategy in real-time electricity pricing environmentrdquo inProceedings of the IEEE Energytech pp 1ndash6 Cleveland OhioUSA May 2012
[19] Z Q LuoW KMa A C So Y Ye and S Zhang ldquoSemidefiniterelaxation of quadratic optimization problemsrdquo IEEE SignalProcessing Magazine vol 27 no 3 pp 20ndash34 2010
[20] K F Fong V I Hanby and T T Chow ldquoHVAC systemoptimization for energymanagement by evolutionary program-mingrdquo Energy and Buildings vol 38 no 3 pp 220ndash231 2006
[21] D L Ha F F de Lamotte and Q-H Huynh ldquoReal-timedynamic multilevel optimization for demand-side load man-agementrdquo in Proceedings of the IEEE International Conferenceon Industrial Engineering and Engineering Management (IEEM07) pp 945ndash949 Singapore December 2007
[22] S Noh J Yun and K Kim ldquoAn efficient building air condi-tioning system control under real-time pricingrdquo in Proceedingsof the International Conference on Advanced Power SystemAutomation and Protection (APAP 11) pp 1283ndash1286 BeijingChina October 2011
[23] G L Nemhauser and L A Wolsey Integer and CombinatorialOptimization Wiley-Interscience New York NY USA 1988
[24] S Kay Fundamentals of Statistical Signal Processing EstimationTheory Prentice-Hall New York NY USA 1993
[25] ldquoContiki the open source OS for the internet of thingsrdquohttpwwwcontiki-osorg
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World Journal 9
21
22
23
24
HVA
C sta
tus
Average temperatureHVAC status
Day time
Tem
pera
ture
(∘C) On
Off
Thermostat-baseline model
18 20 22 24 2 4 6 8 10
Figure 8 Experimental evaluation of the thermostat model
53 Experimental Evaluation of the DES-CC in (4) Severalreal experiments have been carried out to assess the perfor-mance of the DES-CC algorithm proposed in (4) In thiscase the pricing parameters in (1) are 119901
1= 0003 and 119901
2=
1199013= 0 euros which are possible values according to [11] The
temperature bounds in (4) have been set to 119879min1
= 119879min2
=
21∘C and 119879
max1
= 119879max2
= 23∘C respectively
Figure 9 plots the variation of both the measured and theestimated temperature (using (8)ndash(12)) in the room duringour experiments As it can be noticed the error between theestimated and the real temperature is negligible somethingthat proves the accuracy of the proposed estimationmodel Inaddition the DES-CC guarantees the proper operation of thesystem as the temperature varies between the desired rangeof 21∘C and 23
∘C most of the time with very few exceptionsdue to prediction errors In the same figure it can be alsoseen that the temperature remains closer to the lower partof the permitted range (ie 21∘C) since the outcome of theproposed method provides a combination of switched onheaters that minimizes the energy consumption satisfyinga minimum acceptable temperature Indeed compared tothe temperature variation in the baseline scenario (Figure 8)DES-CC maintains the temperature more stable and in thelower part of the allowable region intuitively implying lowercost
Figure 10 depicts the financial operation cost gains thatcan be achieved by DES-CC compared to the baseline ther-mostat approach As it can be observed the proposed energyscheduler significantly reduces the energy cost leading to atotal save of 719 eurosmonth
54 Experimental Evaluation of the DES-CCR in (13) A setof experiments have been carried out for the evaluation ofthe DES-CCR in (13) Let us recall that DES-CCR relaxesthe temperature constraints by including the constraints asa penalized term in the objective function As a resultcompared to DES-CC this method is more flexible withrespect to real time pricing tariffs More specifically DES-CC seeks a combination that minimizes the energy cost withrespect to a minimum allowable temperature Consequently
WSN mote 2
WSN mote 1
18 20 22 24 2 4 6 8 1018192021222324
Day time
18 20 22 24 2 4 6 8 10Day time
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18192021222324
Tem
pera
ture
(∘C)
Figure 9 Real and estimated temperature using DES-CC
0 2 4 6 8 10 12 14 160
0005
001
0015
002
0025
003
0035
004
Hours
Ener
gy co
st (E
uros
)
Energy cost per hour of the proposed methodEnergy cost per hour of the thermostat
Figure 10 Energy consumption cost comparison between thermo-stat and DES-CC
although the energy cost may change during time the heatercombination selected byDES-CC is the same due to the stricttemperature constraint On the other hand DES-CCR allowsthe user to further reduce the energy consumption at the costof being outside the range of temperature of comfort In thiscase to highlight the flexibility of DES-CCR we have set a
10 The Scientific World Journal
0 50 100 150 200 25016
18
20
22
24
Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
0 50 100 150 200 25016
18
20
22
24
Sample measures
Tem
pera
ture
(∘C)
Figure 11 Real and estimated temperature using DES-CCR (120579 =
02)
periodically variable value of 1199011 which alternates between
1199011
= 0009 and 1199011
= 0003 euros every thirty minutesMoreover the desired temperature has been set to 119879
119906119894=
22∘CFigures 11 and 12 depict the temperature variation in
two different cases where the users give low (120579 = 02)and high (120579 = 05) priority respectively to reduce of theenergy consumption In particular in Figure 11 (120579 = 02)the achieved temperature is very close to the desired 119879
119906119894 On
the other hand in Figure 12 we assume 120579 = 05 which is amore adapted value to the pricing policy as it correspondsto a user that permits a relaxation of the difference betweenthe real and the desired temperature to reduce the energycost This fact implies higher energy consumption in lowcost zones and lower energy consumption in high costperiods sacrificing though the userrsquos comfort Therefore theexperiments confirm that the real temperature is close to 119879
119906119894
when the energy cost is lower (ie between samples 60 and120) while there is a noticeable temperature drop whichcorresponds to lower energy consumption
6 Concluding Remarks
This paper has dealt with the energy consumption manage-ment of HVACs for a given smart pricing tariff and usersrsquocomfort constraintsMoreover the integrationwithin the IoTframework has been studied To that end we developed areal testbed consisting of (i) heaters (ii) sensor nodes thatmeasure the temperature and (iii) a gateway which providesconnection to the Internet and includes aweb application that
18
19
20
21
22
0 50 100 150 200 250Sample measures
0 50 100 150 200 250Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18
19
20
21
22
Tem
pera
ture
(∘C)
Figure 12 Real and estimated temperature using DES-CCR (120579 =
05)
permits the interaction with the user through InternetMore-over the gateway implements the algorithms that control theenergy consumption Regarding the proposed methods firstwe devised an energy scheduler that optimizes the energycost in a time interval basis for a given energy price tariffand for a given set of temperature of comfort constraintsthat are associated with different locations inside a roomThen we proposed a more flexible energy scheduler whichrelaxes the temperature constraints to further reduce theenergy consumption Namely a new objective function hasbeen considered which consists of a convex combination ofthe energy cost and a penalty term that reflects the comfortThis permits to consider both the case where the user isvery concerned with the comfort and the case where heallows relaxing the comfort constraint to further reduce theenergy consumption Experimental evaluations have beencarried out in an isolated room validating our proposals andhighlighting their potential benefits
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work has been funded by the Energy-to-Smart Grid(E2SG) project httpwwwe2sg-projecteu within theENIAC joint undertaking framework with Grant agreementnumber 296131
The Scientific World Journal 11
References
[1] L Atzori A Iera and G Morabito ldquoThe internet of things asurveyrdquoComputer Networks vol 54 no 15 pp 2787ndash2805 2010
[2] E Fleisch ldquoWhat is the internet of things An economicperspectiverdquo Tech Rep Auto-ID Labs 2010
[3] National Intelligence Council ldquoDisruptive civil technologiesmdashsix technologies with potential impacts on us interests out to2025rdquo Conference Report CR 2008-07 2008
[4] M R Palattella N Accettura X Vilajosana et al ldquoStandardizedprotocol stack for the internet of (important) thingsrdquo IEEECommunications Surveys and Tutorials vol 15 no 3 pp 1389ndash1406 2013
[5] M Zorzi A Gluhak S Lange and A Bassi ldquoFrom todaysINTRAnet of things to a future INTERnet of things a wireless-and mobility-related viewrdquo IEEEWireless Communications vol17 no 6 pp 44ndash51 2010
[6] A Gluhak S Krco M Nati D Pfisterer N Mitton andT Razafindralambo ldquoA survey on facilities for experimentalinternet of things researchrdquo IEEE Communications Magazinevol 49 no 11 pp 58ndash67 2011
[7] S Tozlu M Senel W Mao and A Keshavarzian ldquoWi-Fienabled sensors for internet of things a practical approachrdquoIEEE Communications Magazine vol 50 no 6 pp 134ndash1432012
[8] N Bui A P Castellani P Casari andM Zorzi ldquoThe internet ofenergy a web-enabled smart grid systemrdquo IEEE Network vol26 no 4 pp 39ndash45 2012
[9] X Fang S Misra G Xue and D Yang ldquoSmart gridmdashthe newand improved power grid a surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 14 no 4 pp 944ndash980 2012
[10] G Lu D De and W Song ldquoSmartGridLab a laboratory-basedsmart grid testbedrdquo in Proceedings of the 1st IEEE InternationalConference on Smart Grid Communications (SmartGridComm10) pp 143ndash148 Gaithersburg Md USA October 2010
[11] A-H Mohsenian-Rad V W S Wong J Jatskevich R Schoberand A Leon-Garcia ldquoAutonomous demand-side managementbased on game-theoretic energy consumption scheduling forthe future smart gridrdquo IEEE Transactions on Smart Grid vol1 no 3 pp 320ndash331 2010
[12] Z Zhu S Lambotharan W H Chin and Z Fan ldquoOverviewof demand management in smart grid and enabling wirelesscommunication technologiesrdquo IEEE Wireless Communicationsvol 19 no 3 pp 48ndash56 2012
[13] H T Nguyen D Nguyen and L B Le ldquoHome energy man-agement with generic thermal dynamics and user temperaturepreferencerdquo in Proceedings of the IEEE International Conferenceon Smart Grid Communications (SmartGridComm 13) pp 552ndash557 Vancouver Canada October 2013
[14] Z Zhu J Tang S Lambotharan W H Chin and Z FanldquoAn integer linear programming based optimization for homedemand-side management in smart gridrdquo in Proceedings of theIEEE PES Innovative Smart Grid Technologies (ISGT 12) pp 1ndash5Washington DC USA January 2012
[15] A-H Mohsenian-Rad and A Leon-Garcia ldquoOptimal residen-tial load control with price prediction in real-time electricitypricing environmentsrdquo IEEE Transactions on Smart Grid vol1 no 2 pp 120ndash133 2010
[16] K M Tsui and S C Chan ldquoDemand response optimizationfor smart home scheduling under real-time pricingrdquo IEEETransactions on Smart Grid vol 3 no 4 pp 1812ndash1821 2012
[17] GWood andMNewborough ldquoDynamic energy-consumptionindicators for domestic appliances environment behaviour anddesignrdquo Energy and Buildings vol 35 no 8 pp 821ndash841 2003
[18] M Avci M Erkoc and S S Asfour ldquoResidential HVAC loadcontrol strategy in real-time electricity pricing environmentrdquo inProceedings of the IEEE Energytech pp 1ndash6 Cleveland OhioUSA May 2012
[19] Z Q LuoW KMa A C So Y Ye and S Zhang ldquoSemidefiniterelaxation of quadratic optimization problemsrdquo IEEE SignalProcessing Magazine vol 27 no 3 pp 20ndash34 2010
[20] K F Fong V I Hanby and T T Chow ldquoHVAC systemoptimization for energymanagement by evolutionary program-mingrdquo Energy and Buildings vol 38 no 3 pp 220ndash231 2006
[21] D L Ha F F de Lamotte and Q-H Huynh ldquoReal-timedynamic multilevel optimization for demand-side load man-agementrdquo in Proceedings of the IEEE International Conferenceon Industrial Engineering and Engineering Management (IEEM07) pp 945ndash949 Singapore December 2007
[22] S Noh J Yun and K Kim ldquoAn efficient building air condi-tioning system control under real-time pricingrdquo in Proceedingsof the International Conference on Advanced Power SystemAutomation and Protection (APAP 11) pp 1283ndash1286 BeijingChina October 2011
[23] G L Nemhauser and L A Wolsey Integer and CombinatorialOptimization Wiley-Interscience New York NY USA 1988
[24] S Kay Fundamentals of Statistical Signal Processing EstimationTheory Prentice-Hall New York NY USA 1993
[25] ldquoContiki the open source OS for the internet of thingsrdquohttpwwwcontiki-osorg
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
10 The Scientific World Journal
0 50 100 150 200 25016
18
20
22
24
Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
0 50 100 150 200 25016
18
20
22
24
Sample measures
Tem
pera
ture
(∘C)
Figure 11 Real and estimated temperature using DES-CCR (120579 =
02)
periodically variable value of 1199011 which alternates between
1199011
= 0009 and 1199011
= 0003 euros every thirty minutesMoreover the desired temperature has been set to 119879
119906119894=
22∘CFigures 11 and 12 depict the temperature variation in
two different cases where the users give low (120579 = 02)and high (120579 = 05) priority respectively to reduce of theenergy consumption In particular in Figure 11 (120579 = 02)the achieved temperature is very close to the desired 119879
119906119894 On
the other hand in Figure 12 we assume 120579 = 05 which is amore adapted value to the pricing policy as it correspondsto a user that permits a relaxation of the difference betweenthe real and the desired temperature to reduce the energycost This fact implies higher energy consumption in lowcost zones and lower energy consumption in high costperiods sacrificing though the userrsquos comfort Therefore theexperiments confirm that the real temperature is close to 119879
119906119894
when the energy cost is lower (ie between samples 60 and120) while there is a noticeable temperature drop whichcorresponds to lower energy consumption
6 Concluding Remarks
This paper has dealt with the energy consumption manage-ment of HVACs for a given smart pricing tariff and usersrsquocomfort constraintsMoreover the integrationwithin the IoTframework has been studied To that end we developed areal testbed consisting of (i) heaters (ii) sensor nodes thatmeasure the temperature and (iii) a gateway which providesconnection to the Internet and includes aweb application that
18
19
20
21
22
0 50 100 150 200 250Sample measures
0 50 100 150 200 250Sample measures
WSN mote 2
WSN mote 1
Real temperatureEstimated temperature
Real temperatureEstimated temperature
Tem
pera
ture
(∘C)
18
19
20
21
22
Tem
pera
ture
(∘C)
Figure 12 Real and estimated temperature using DES-CCR (120579 =
05)
permits the interaction with the user through InternetMore-over the gateway implements the algorithms that control theenergy consumption Regarding the proposed methods firstwe devised an energy scheduler that optimizes the energycost in a time interval basis for a given energy price tariffand for a given set of temperature of comfort constraintsthat are associated with different locations inside a roomThen we proposed a more flexible energy scheduler whichrelaxes the temperature constraints to further reduce theenergy consumption Namely a new objective function hasbeen considered which consists of a convex combination ofthe energy cost and a penalty term that reflects the comfortThis permits to consider both the case where the user isvery concerned with the comfort and the case where heallows relaxing the comfort constraint to further reduce theenergy consumption Experimental evaluations have beencarried out in an isolated room validating our proposals andhighlighting their potential benefits
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work has been funded by the Energy-to-Smart Grid(E2SG) project httpwwwe2sg-projecteu within theENIAC joint undertaking framework with Grant agreementnumber 296131
The Scientific World Journal 11
References
[1] L Atzori A Iera and G Morabito ldquoThe internet of things asurveyrdquoComputer Networks vol 54 no 15 pp 2787ndash2805 2010
[2] E Fleisch ldquoWhat is the internet of things An economicperspectiverdquo Tech Rep Auto-ID Labs 2010
[3] National Intelligence Council ldquoDisruptive civil technologiesmdashsix technologies with potential impacts on us interests out to2025rdquo Conference Report CR 2008-07 2008
[4] M R Palattella N Accettura X Vilajosana et al ldquoStandardizedprotocol stack for the internet of (important) thingsrdquo IEEECommunications Surveys and Tutorials vol 15 no 3 pp 1389ndash1406 2013
[5] M Zorzi A Gluhak S Lange and A Bassi ldquoFrom todaysINTRAnet of things to a future INTERnet of things a wireless-and mobility-related viewrdquo IEEEWireless Communications vol17 no 6 pp 44ndash51 2010
[6] A Gluhak S Krco M Nati D Pfisterer N Mitton andT Razafindralambo ldquoA survey on facilities for experimentalinternet of things researchrdquo IEEE Communications Magazinevol 49 no 11 pp 58ndash67 2011
[7] S Tozlu M Senel W Mao and A Keshavarzian ldquoWi-Fienabled sensors for internet of things a practical approachrdquoIEEE Communications Magazine vol 50 no 6 pp 134ndash1432012
[8] N Bui A P Castellani P Casari andM Zorzi ldquoThe internet ofenergy a web-enabled smart grid systemrdquo IEEE Network vol26 no 4 pp 39ndash45 2012
[9] X Fang S Misra G Xue and D Yang ldquoSmart gridmdashthe newand improved power grid a surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 14 no 4 pp 944ndash980 2012
[10] G Lu D De and W Song ldquoSmartGridLab a laboratory-basedsmart grid testbedrdquo in Proceedings of the 1st IEEE InternationalConference on Smart Grid Communications (SmartGridComm10) pp 143ndash148 Gaithersburg Md USA October 2010
[11] A-H Mohsenian-Rad V W S Wong J Jatskevich R Schoberand A Leon-Garcia ldquoAutonomous demand-side managementbased on game-theoretic energy consumption scheduling forthe future smart gridrdquo IEEE Transactions on Smart Grid vol1 no 3 pp 320ndash331 2010
[12] Z Zhu S Lambotharan W H Chin and Z Fan ldquoOverviewof demand management in smart grid and enabling wirelesscommunication technologiesrdquo IEEE Wireless Communicationsvol 19 no 3 pp 48ndash56 2012
[13] H T Nguyen D Nguyen and L B Le ldquoHome energy man-agement with generic thermal dynamics and user temperaturepreferencerdquo in Proceedings of the IEEE International Conferenceon Smart Grid Communications (SmartGridComm 13) pp 552ndash557 Vancouver Canada October 2013
[14] Z Zhu J Tang S Lambotharan W H Chin and Z FanldquoAn integer linear programming based optimization for homedemand-side management in smart gridrdquo in Proceedings of theIEEE PES Innovative Smart Grid Technologies (ISGT 12) pp 1ndash5Washington DC USA January 2012
[15] A-H Mohsenian-Rad and A Leon-Garcia ldquoOptimal residen-tial load control with price prediction in real-time electricitypricing environmentsrdquo IEEE Transactions on Smart Grid vol1 no 2 pp 120ndash133 2010
[16] K M Tsui and S C Chan ldquoDemand response optimizationfor smart home scheduling under real-time pricingrdquo IEEETransactions on Smart Grid vol 3 no 4 pp 1812ndash1821 2012
[17] GWood andMNewborough ldquoDynamic energy-consumptionindicators for domestic appliances environment behaviour anddesignrdquo Energy and Buildings vol 35 no 8 pp 821ndash841 2003
[18] M Avci M Erkoc and S S Asfour ldquoResidential HVAC loadcontrol strategy in real-time electricity pricing environmentrdquo inProceedings of the IEEE Energytech pp 1ndash6 Cleveland OhioUSA May 2012
[19] Z Q LuoW KMa A C So Y Ye and S Zhang ldquoSemidefiniterelaxation of quadratic optimization problemsrdquo IEEE SignalProcessing Magazine vol 27 no 3 pp 20ndash34 2010
[20] K F Fong V I Hanby and T T Chow ldquoHVAC systemoptimization for energymanagement by evolutionary program-mingrdquo Energy and Buildings vol 38 no 3 pp 220ndash231 2006
[21] D L Ha F F de Lamotte and Q-H Huynh ldquoReal-timedynamic multilevel optimization for demand-side load man-agementrdquo in Proceedings of the IEEE International Conferenceon Industrial Engineering and Engineering Management (IEEM07) pp 945ndash949 Singapore December 2007
[22] S Noh J Yun and K Kim ldquoAn efficient building air condi-tioning system control under real-time pricingrdquo in Proceedingsof the International Conference on Advanced Power SystemAutomation and Protection (APAP 11) pp 1283ndash1286 BeijingChina October 2011
[23] G L Nemhauser and L A Wolsey Integer and CombinatorialOptimization Wiley-Interscience New York NY USA 1988
[24] S Kay Fundamentals of Statistical Signal Processing EstimationTheory Prentice-Hall New York NY USA 1993
[25] ldquoContiki the open source OS for the internet of thingsrdquohttpwwwcontiki-osorg
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World Journal 11
References
[1] L Atzori A Iera and G Morabito ldquoThe internet of things asurveyrdquoComputer Networks vol 54 no 15 pp 2787ndash2805 2010
[2] E Fleisch ldquoWhat is the internet of things An economicperspectiverdquo Tech Rep Auto-ID Labs 2010
[3] National Intelligence Council ldquoDisruptive civil technologiesmdashsix technologies with potential impacts on us interests out to2025rdquo Conference Report CR 2008-07 2008
[4] M R Palattella N Accettura X Vilajosana et al ldquoStandardizedprotocol stack for the internet of (important) thingsrdquo IEEECommunications Surveys and Tutorials vol 15 no 3 pp 1389ndash1406 2013
[5] M Zorzi A Gluhak S Lange and A Bassi ldquoFrom todaysINTRAnet of things to a future INTERnet of things a wireless-and mobility-related viewrdquo IEEEWireless Communications vol17 no 6 pp 44ndash51 2010
[6] A Gluhak S Krco M Nati D Pfisterer N Mitton andT Razafindralambo ldquoA survey on facilities for experimentalinternet of things researchrdquo IEEE Communications Magazinevol 49 no 11 pp 58ndash67 2011
[7] S Tozlu M Senel W Mao and A Keshavarzian ldquoWi-Fienabled sensors for internet of things a practical approachrdquoIEEE Communications Magazine vol 50 no 6 pp 134ndash1432012
[8] N Bui A P Castellani P Casari andM Zorzi ldquoThe internet ofenergy a web-enabled smart grid systemrdquo IEEE Network vol26 no 4 pp 39ndash45 2012
[9] X Fang S Misra G Xue and D Yang ldquoSmart gridmdashthe newand improved power grid a surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 14 no 4 pp 944ndash980 2012
[10] G Lu D De and W Song ldquoSmartGridLab a laboratory-basedsmart grid testbedrdquo in Proceedings of the 1st IEEE InternationalConference on Smart Grid Communications (SmartGridComm10) pp 143ndash148 Gaithersburg Md USA October 2010
[11] A-H Mohsenian-Rad V W S Wong J Jatskevich R Schoberand A Leon-Garcia ldquoAutonomous demand-side managementbased on game-theoretic energy consumption scheduling forthe future smart gridrdquo IEEE Transactions on Smart Grid vol1 no 3 pp 320ndash331 2010
[12] Z Zhu S Lambotharan W H Chin and Z Fan ldquoOverviewof demand management in smart grid and enabling wirelesscommunication technologiesrdquo IEEE Wireless Communicationsvol 19 no 3 pp 48ndash56 2012
[13] H T Nguyen D Nguyen and L B Le ldquoHome energy man-agement with generic thermal dynamics and user temperaturepreferencerdquo in Proceedings of the IEEE International Conferenceon Smart Grid Communications (SmartGridComm 13) pp 552ndash557 Vancouver Canada October 2013
[14] Z Zhu J Tang S Lambotharan W H Chin and Z FanldquoAn integer linear programming based optimization for homedemand-side management in smart gridrdquo in Proceedings of theIEEE PES Innovative Smart Grid Technologies (ISGT 12) pp 1ndash5Washington DC USA January 2012
[15] A-H Mohsenian-Rad and A Leon-Garcia ldquoOptimal residen-tial load control with price prediction in real-time electricitypricing environmentsrdquo IEEE Transactions on Smart Grid vol1 no 2 pp 120ndash133 2010
[16] K M Tsui and S C Chan ldquoDemand response optimizationfor smart home scheduling under real-time pricingrdquo IEEETransactions on Smart Grid vol 3 no 4 pp 1812ndash1821 2012
[17] GWood andMNewborough ldquoDynamic energy-consumptionindicators for domestic appliances environment behaviour anddesignrdquo Energy and Buildings vol 35 no 8 pp 821ndash841 2003
[18] M Avci M Erkoc and S S Asfour ldquoResidential HVAC loadcontrol strategy in real-time electricity pricing environmentrdquo inProceedings of the IEEE Energytech pp 1ndash6 Cleveland OhioUSA May 2012
[19] Z Q LuoW KMa A C So Y Ye and S Zhang ldquoSemidefiniterelaxation of quadratic optimization problemsrdquo IEEE SignalProcessing Magazine vol 27 no 3 pp 20ndash34 2010
[20] K F Fong V I Hanby and T T Chow ldquoHVAC systemoptimization for energymanagement by evolutionary program-mingrdquo Energy and Buildings vol 38 no 3 pp 220ndash231 2006
[21] D L Ha F F de Lamotte and Q-H Huynh ldquoReal-timedynamic multilevel optimization for demand-side load man-agementrdquo in Proceedings of the IEEE International Conferenceon Industrial Engineering and Engineering Management (IEEM07) pp 945ndash949 Singapore December 2007
[22] S Noh J Yun and K Kim ldquoAn efficient building air condi-tioning system control under real-time pricingrdquo in Proceedingsof the International Conference on Advanced Power SystemAutomation and Protection (APAP 11) pp 1283ndash1286 BeijingChina October 2011
[23] G L Nemhauser and L A Wolsey Integer and CombinatorialOptimization Wiley-Interscience New York NY USA 1988
[24] S Kay Fundamentals of Statistical Signal Processing EstimationTheory Prentice-Hall New York NY USA 1993
[25] ldquoContiki the open source OS for the internet of thingsrdquohttpwwwcontiki-osorg
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014