Adv

ancedDat
aSyst
emsforEnergyConsumpt
ionOpti
mizat
ionandAirQuali
tyCont
roli
n
Smar
tPubl
icBui
ldi
ngsUsingaVer
sati
leOpenSourceApproach.

[Link]
oundoft
heSt
udy

Thiswor kdi scussessmar tbui l

dingappl i
cat i
onsi nv olvingt heI nter netofThi ngs( I
oT)whi ch
arefocusedonener gyconsumpt i
onmoni toringandf orecast i
ngsy st ems, aswel l asindoorai r
quality( I
AQ)cont r
[Link]-costhar dwar ei ntegr atingsensor sandopensour cepl atformsar e
i
mpl ement edf orclouddat at ransmi ssion, datast orageanddat apr ocessi [Link] anceddat a
analyticsisper formedbyt heseasonal aut oregr essiv eint egrat edmov ingav er age( SARI MA)
methodandal ongshor t-t
er m memor y( LSTM)neur al net wor kwi thanaccur atecal culation
perf
or manceaboutener gypr edicti
[Link] oposedr esul t
sar edev elopedwi thint he
fr
amewor koft heR&Dpr ojectDat aSy stem Pl at formf orSmar tCommuni ties( D-Sy SCOM) ,
whichi sor ientedtoasmar tpubl i
cbui ldingappl [Link] ngoal oft hewor kwast odef ine
agui deline-mat chingenergyef f
ici
encywi t
hwel lnessi npubl ici ndoorenv ironment s,bypr ov i
ding
modul arl ow-costsol uti
onswhi char eeasi l
yimpl ement abl eforadv anceddat apr ocessi [Link]
i
mpl ement edt echnologiesar esui tabletodef ineanef fi
ci entor gani zational userpr otocol based
onener gyef fi
ciencyandwor kerwel l
[Link] i
mat edper for manceofmeansquar eer ror
(MSE)of0. 01oft headopt edal gorithmspr ov est heef ficiencyoft hei mplement edbui lding
moni tor i
ngsy stem intermsofener gyconsumpt i
onf orecast [Link] ti
on, thepossi bil
ityof
designi ngandi mplement i
ngamodul arlow- costhar dwar e–sof twar esy stem wasdemonst rated
uti
li
zingopensour cetoolsinawayt hatwasor ientedt osmar tbui ldingsappr oaches.

[Link]
iewofRel
atedLi
ter
atur
e.

Inener gysy stems, takingdecisionsont hebasi sofdat acollect

edbyapower fulenergy
moni t
or i
ngi st heonl ywayt omi ni
mi zeconsumpt i
onandt otrytoavoidhighsimultaneousloads.
Theener gydat apr ocessingisakeyel ementf ort heest imationofKeyPer f
ormanceI ndi
cator
(KPI)insmar tbuildings[ 1]
.Energymoni t
or i
ng, andt her el
atedeffi
ciencymanagementcan
severelyreducecost s(suchasel ectri
cal uti
lit
ycost s)
,thusincreasingthelif
etimeofener gy
storagesy stems[ 2] .Thef i
rststepinmanagi ngel ect r
icloadconsumpt ioni
stoknowt he
electr
icit
yconsumpt i
onpr ofi
l
ev ersust i
[Link] ect r
icalpowerconsumpt i
ongener al
lyvari
esdue
tobot huserbehav i
ors[ 3]andambi entcondi tions[ 4],
whi chleadstot hedemandofcont i
nuous
[Link] dingt odat aanalyti
cs,datami ni
ngt echni quesar eusedf orenergydata
classi
ficati
on, clusteringandpr edicti
on[ 5].Cloudsy stems[ 6]aregoodcandi datestoprocess
datadet ectedbyI nt ernetofThings( IoT)sensor s.

Theenergyconsumpt iondatacanbecollectedandpr ocessedbythedatacenterfacil

i
ties
thr
oughawi rel
esscommuni cati
onnetwor kusingt hemessagequeuingtel
emet r
yt r
ansport
(MQTT)protocol[7](ali
ght
weightprot
ocol speci
ficall
ydesignedf
ormachine-t
o-
machine—M2M—communi cati
on).I
nrecenty ear
s, el
ectr
icall
oadf
orecasti
ngisget t
ingmor e
andmor ei
mpor tantduetoelectr
ici
tymarketderegulati
onandt ot
heintegr
ati
onofr enewable
energypr oduct i
on, operat
edinl argeplantsaswel l
assmal lonesonbui l
dings[ 8].Toov ercome
theincomi ngchal lengesandensur eaccur atepowerpr ediction,pr edictiv
eal gor i
thmscanbe
adopt edsuchasseasonal ARI MA( SARI MA)[ 9],andl ongshor t-t
er m memor y( LSTM)[ 10].
Specificall
y ,SARI MAmodel isv erysimilartot heaut oregr essivei ntegratedmov i
ngav erage
(ARIMA)model [11],exceptforanaddi tional setofaut or egressiv eandmov i
ngav er
age
component s;LSTM i sakindofr ecurr
entneur alnetwor k( RNN)[ 12].Bot har eusedi ngeneralas
forecastingmet [Link] er-i
ncreasingener gyconsumpt ionandut i
li
tycost sofbui lding
operat i
onsr equir
est heexplorationofnewst rategi
est oopt i
mi zeusageper f
or mance, toreduce
energywast e,andt omi ni
mizeenv i
ronment al impact s[13] .Asuccessf ulappr oacht oreacha
highener gyef f
iciencyinbuil
di ngspacesi st headopt i
onofadv ancedener gymanagement
systems( EMSs)[ 14],fori
nstancet hosebasedont hecombi nationofwebi nter f
acesmoni t
ori
ng
energyconsumpt ionandswi tchingelectricpower[ 15],indoorenv i
ronment alqual i
ty(I
EQ)f or
effi
cientindoorcomf ort(t
hermo- hy
gromet ri
c, li
ghting,ai rqualityandacoust ics)[16],anduser
activi
tychecks[ 17] .

Theov erallst ateoft hear tsuggest stheimpor tanceofsmar tbuildingappl i

cat i
onst ocombi ne
well
nessf unct i
onswi thener gyconsumpt i
onaspect [Link] ngoal oft hiswor kist opr ovi
de
modul arandeasi lyappl icabl edesigncriteriaforener gycont rolsystemsbyf ocusingat t
ent i
on
ontheener gysav ingsi npubl icbuildi
ngsgener atedbymeansofl ow- costel ectronic
component [Link] nterf aceswer esodesi gnedandi mplement edf orabasi cpr ototype
demonst rator :onecol lect ingdat atocalculateener gyconsumpt i
onoft hedi f
ferentloadsoft he
buil
ding,andonededi cat edt oindoorairqual it
y( I
AQ)moni toring,capabl eofdet ectinglight
i
ntensity,temper at ur
eandhumi dit
[Link]- costAr duino-
basedi nterfacewast hendev el
opedf or
sensorsdat acol lecti
onandwebpubl i
cationaswel lastodr iveelect roniccomponent sact uating
energy-savingst [Link] oud- basedwebi nterfacepr ovi
desf orthef ormul ati
on
oftheor gani zational pr otocol basedont heaut omat edintervent i
onswhenei theranomal ous
energyconsumpt ionoccur s,highv al
uesofener gyconsumpt ionar epr edi ctedorbadai rqual i
ty
conditi
onsar ef ound( seeschemeofFi gur e1).Thef lowchar tsummar izest heor ganizational
protocoladopt edi nthi sst udyforthepubl icbuildingenergycont [Link] otocol i
smadeof
dif
ferentlev els.

[Link]
eti
calFr
amewor
k.

Ener
gyMoni
tor
ingModul
arAr
chi
tect
ure(
EPI
nter
face)

Thear
chi
tect
ureofthedevel
opedIoT-
basedpowerconsumpti
onmoni
tor
ingsy
stem hast
he
fol
l
owi
ngfour-
lay
ers:per
cept
ion,
network,appl
i
cat
ionandcogni
ti
onl
ayer
s.

Att heper cept

ionl
ayer,
sensordev i
cesareprovi
dedwhichmeasur eandacquirepowerprocess
[Link]
worklayer,
informati
onissenttoacloud-basedserver
.Attheappli
cati
on
l
ay er ,
thereal-
ti
meinfor
mationisstoredanddispl
[Link]
onlayer
,dataarepost-
processedwhi l
eusi
ngmachi nelearni
ngfordataanal
ysi
sandf orecast
ingforassi
sti
nginthe
deci si
onmaki ng.
Per
cept
ionLay
er(
EPI
nter
face)

Thepercept i
onlayerisdirect l
yl i
nkedt othecentralpowersuppl ynodef orpowermeasur ement.
Thesystem measur esther oot-mean- square(RMS)i nstantaneouspower , bycal culati
ngthe
RMScur rentandRMSv oltagef rom theelectri
calnetwor k(theIt
aliannet wor kpr ovides220Vas
anACv oltagelevel)
.Thev oltageismeasur edont heelect r
icl
inespassi ngt hrought hestep-
downt r
ansf or
merandav oltager egulatoradjust
ingtheDCout puti nt
oar angeaccept ablefor
theMicroCont r
oll
erUnit(MCU) .Theadopt edmi crocontroll
eristheAr duinoNanoboar d,
based
ontheATmega328P.I tisanAl fandVegar d’sRISC( AVR)basedmi cr
ocont r
ollerwit h5Vsupply,
8bitand16MHzcl ockfrequency .Thesof twarecodei st heArduino’sWi repr ogr ammi ng
l
anguage, executabl
eont heAr duinoI ntegr
atedDev elopmentEnv i
ronment( IDE) .

Net
wor
kLay
er(
EPI
nter
face)

Thisl
ayermainlyt
ransmit
sthedetect
eddataf r
om thesmartenergymonit
ortothecloudsever
.
Datapre-
processi
ngmethods,suchasfi
lter
ingandoutli
err
emov alpr
ocesses,
areexecutedby
thecontr
oll
er,r
educi
ngthequanti
tyoftr
ansmitteddataandthecomputati
onalcost(
are
pr
ocessedonlyusef
ul data)
.Afterpre-
processi
ng,dat aaretr
ansfer
redtotheWi-Fi
module
i
nter
facedwiththemicrocontr
oller
,orstoredi
nt hetransmitqueue,dependi
ngonthedat
a
ef
fect
ivenessdatal
ength.TheESP8266i salow-costWi -
Fichi
poper at
ingwit
htheTCP/I
P
pr
otocolandsupport
ingIEEE802. 11b/g/nWi -
Fistandards.

Appl
i
cat
ionLay
er(
EPI
nter
face)

Dataarepubl i
shedbyt heremoteMQTTbr okerdir
ectl
ythrought
heESP8266gat eway
(publi
sh/subscri
besy st
em) .MQTTr equi
reslowbandwi dt
handhasadatapacketsi zewithlow
overheadmi nimum (>2by t
es)sothatithassmallersuppl
[Link]
is
adat a-
agnosticprot
ocol t
hatcantransmitdatainvari
ousf or
mssuchasbinarydata,text
,XML,
[Link] hewebser verispubl
ished,t
hedashboar d-
basedwebappli
cati
onisaccessible
anywhereandany time,thusenabli
ngr emoteenergymanagement.

Cogni
ti
onLay
er(
EPI
nter
face)

Thi
slay
erdef
inesthepr
ocessf
oll
owi
ngt
hecogni
ti
vecomput
ingal
gor
it
[Link]
nfunct
ions
oft
hecogni
ti
velay
erare

 dat
apr
e-pr
ocessi
ng

 t
hedat
ami
ningpr
ocessi
ng.

ThislayerisabletoexecuteSARIMAandLSTM al gori
[Link] acter
izedbyfiv
ephases
(thefi
rststepisthepre-pr
ocessi
ngphase, andtheot
herf ourstepsar erel
atedthedata
processing).Thedataseti
sspli
tint
otrai
ning(75%)andt est(25%)dat asetstobeprocessedby
themachi nelearni
ngalgori
thms(bestchoiceconcer
ningalgorithm opti
mi zat
ion)
.

[Link]
gni
fi
canceoft
heSt
udy
.

Har dwareandsof t
warel ow-
costsoluti
onswhichar eusefulf
orsmartbuildi
ngenergyandI AQ
moni tor
[Link]-costmodulesallowforthei
nst al
lat
ionofthecomponent si
neachr oom of
thepubl i
cbuil
[Link] ofthepaperwastopr oposeamodul arapproachtoel
ectroni
c
cont r
[Link] edtodefineopti
mizati
onstr
ategi
esf or
electri
calenergyconsumpt i
onreducti
onandwor kerwellnessincr
[Link]
dedsolutions
ar
ecompat i
bleandint
egrabl
ewit
hother“
open”modul escontr
olli
ngot
herpar
ameters,
and
whicharedevotedt
ofurt
herimpr
oveener
gysav i
ngsandwel l
nesscondi
ti
ons.I
nter
venti
onscan
beplannedwhichareal
sobasedonpredi
ctedenergyconsumption.

Compar edtoexi
sti
ngcommerci
alsol
uti
ons,
thet
echnologyproposedhereusesapredict
ive
LSTM algori
thm,whi
chisdemonst
rat
edtobemoreeff
icientt
hanaSARI MAone,forthespecif
ic
datasetty
pology
.

Theformul
ati
onofmorecomplexkeyper
formancebui
l
dingi
ndi
cat
ors,
taki
ngi
ntoaccount
pri
ori
ti
esofel
ect
ri
cal
loads,
isunderi
nvest
igat
ion.

Theadopt edopensour cetoolsar

efull
yintegr
atedintoani nf
ormat i
onsystem managingfr
ont-
endinterfaceswi t
hdashboar dsandback-enddatasy stemscollecti
[Link]
des
alow-costsol ut
iont obuil
dupanadv ancedplatf
orm, suit
abletocontrolwholebui
ldi
ngswitha
modul arimplement ati
onofthehardwaresoluti
onsandcompat iblewit
hdiffer
entsensor
technologiesanddat atransmissi
onstandardprotocols.

[Link]
ini
ti
onofTer
ms.

Theimplementationoft womainf
uncti
ons,suchasenergyandindoorairquali
tymonitori
ng,
provest
hatmanagi ngsensor
sdatadi
splacedincomplexnetworksispossi
[Link]
tion,
Pythonwiththerelatedopensour
celi
brari
esiscompli
antwiththeimplementati
onofdi ff
erent
machinelearni
ngal gori
thms,
prov
idedthatther
eisadatasettobeprocessedandper f
or mance
tobeachieved.

Theadopt i
onofmor ealgori
thmsandsensor sisf undament altoesti
mat einnovati
v eKPIsas
output
sofcompl exmoni tor
ingsystemssuchasener gyr outi
[Link]
ewor ks,impl ement
ing
synchr
onizedel ect
ronicboardswillbethegoal,bothf orcontrolandactuationfunctionsfor
cl
oudcomput ingplatfor
msorf oredgecomput i
ngsy stemsinterfaci
ngmi crocontr
ollerswit
h
[Link]
ementi spossibleast hemodul ari
tyofthesolutionsdepicted
hereal
lowst hedesi gnandtoimpl ementati
onofhy bridnetwor ksconsti
tutedbycloudandedge
dataprocessingsyst ems.

Ref
erences

[Link]
o,A.;
Starace,[Link] ancedandComplexEnergySyst
emsMoni tor
ingandContr
ol:
A
Revi
ewonAv ai
lableTechnologiesandThei
rAppli
cat
ionCrit
eri
[Link]
s2022,22,4929.
[
GoogleSchol
ar][CrossRef
]

[Link]
d,C.;
Iakovl
eva,E.
[Link]
edI
mprovementforSmall
AutonomousEnergySyst
em
Reli
abi
li
tybyReduci
ngHeatandExcessChar
[Link]
ter
ies2019,5,
29.[Googl
eSchol
ar]
[
CrossRef]
[Gr
eenVersion]
[Link]
,C.
-W.;
Chiang,K.
-C.
;Hsieh,H.
-Y.
;Yang,C.-
W.;
Lin,
J.;Chang,Y.-
[Link]
eExtract
ionof
AnomalyEl
ectr
ici
tyUsageBehaviori
nResidenceUsi
[Link]
ectr
onics2022,11,1450.
[
GoogleSchol
ar][
CrossRef
]

[Link]
;Permana,I.
;Chi
ang,W.;Wang,F.
;Liou,[Link]
ionAnaly
sisfor
Coupli
ngAirCondi
tioner
sandColdStorageShowcaseEqui pmentinaConveni
enceStore.
Energi
es2022,15,4857.[
GoogleSchol
ar][
CrossRef]

[Link]
,U.;Buccel
l
a,C.;Cecati
,[Link]
ectri
cit
yConsumpti
onAnalysi
swithDataMi ni
ng
Techniques.I
nProceedingsoft heIECON2016–42ndAnnualConfer
enceoftheIEEEIndustr
ial
El
ectronicsSoci
ety
,Florence,
Italy,
23–26October2016;
IEEE:NewYork,NY,USA,2016;pp.
3966–3971.[GoogleScholar]

[Link]
s,A.;
Sudolska,
A.;Pi
etr
yka,I
.;
Kozaki
ewicz,
[Link]
ingandEnergyEf
fi
ciency:
MappingtheThematicSt
ructur
eofResear
[Link]
gies2020,
13,
4117.[Googl
eSchol
ar ]
[
CrossRef]

[Link],A.I
nter
netofThingsSol
uti
onsinIndust
ry.I
nElectr
oni
csinAdv ancedResear
ch
I
ndustr
ies;Wil
ey:Hoboken,
NJ, USA,
2021;pp.155–202.[Googl
eScholar
][CrossRef
]

[Link],A.;Doost
izadeh,
M.;
Ami nifar
,[Link] Fl
exi
bil
it
y:AnOver
viewofEmer
gence
toEvolut
[Link].CleanEnergy2019,7,987–1007.[
Googl
eScholar
]
[
CrossRef][
GreenVer si
on]

[Link]
an,R.;
Shaaf
iey
oun,A.;Julka,[Link]-
AheadForecasti
ngoft
hePer
cent
ageof
Renewabl
esBasedonTi me-Ser
iesStatist
ical
[Link]
es2021,
14,
7443.[
Googl
e
Schol
ar][
CrossRef]

[Link],S.
;Chri
fi
-Al
aoui
,L.
;Marhi
c,B.
;Delahoche,[Link]
cti
ngEner
gyConsumptionUsi
ng
LSTM,Mul t
i-
LayerGRUandDrop-GRUNeural
Net [Link]
s2022,22,4062.[
Google
Schol
ar][Cr
ossRef][
PubMed]