Copyright2010CleanPowerResearch 1

Pv Powei 0utput vaiiability:

Abstract
UtilityplannersandoperatorsareresponsibleforguidingwherePVsystemsarelocatedandare
[Link]
[Link],theyareconcernedaboutPVpoweroutputvariability.
Thispaperintroducesanovelapproachtoestimatethemaximumshorttermoutputvariabilitythata
[Link],
orabsencethereof,existingbetweenindividualinstallationsinthefleetattheconsideredvariability
timescale.
[Link],thepaper
focusesonanalyzingthecorrelationcoefficientofthechangeintheclearnessindexbetweentwo
locationsasafunctionofdistance,timeinterval,[Link]
[Link]
becapableofdescribingsitepaircorrelationacrosstimeintervalsfromsecondstohours.
Themethodisderivedempiricallyandvalidatedusing12yearsofhourlysatellitederiveddatafrom
SolarAnywhereinthreegeographicregionsintheUnitedStates(Southwest,SouthernGreatPlains,and
Hawaii).Resultsattimeintervalslessthanonehourarecorroboratedusingfindingsfromrecent
investigationsthatwerebasedon10secondtooneminutedatasets.
Thestrengthandstructureofthemethodissummarizedbythreefundamentalfindingsthatboth
confirmandextendconclusionsfrompreviousstudies:
1. Correlationcoefficientsdecreasepredictablywithincreasingdistance.
2. Correlationcoefficientsdecreaseatasimilarratewhenevaluatedversusdistancedividedbythe
consideredvariabilitytimeinterval.
3. Theaccuracyofresultsisimprovedbyincludinganimpliedcloudspeedterm.
ThepresentapproachhaspotentialfinancialbenefitstosystemsthatareconcernedaboutPVpower
outputvariability,rangingfromdistributionfeederstobalancingregions.

Copyright2010CleanPowerResearch 2

Introduction
[Link],utilityplannersandoperatorsaregrowingmore
[Link]
controlsystemoperatorsneedtoadapttheirplanning,scheduling,andoperatingstrategiesto
accommodatethisvariabilitywhileatthesametimemaintainingexistingstandardsofreliability.
Itisimpossibletoeffectivelymanagethesesystems,however,withoutaclearunderstandingofPV
[Link]
severalhoursaheadorplanningforreserveresourcesyearsintothefuture,theindustryneedstobe
abletoquantifyexpectedoutputvariabilityforfleetsofuptohundredsofthousandsofPVsystems
[Link]
[Link]
requirementsmayresultinanunnecessaryexpenditureofcapitalandhigheroperatingcosts.
ThepresentobjectiveistodevelopanalyticalmethodsandtoolstoquantifyPVfleetoutputvariability.
Variabilityintimeintervalsrangingfromafewsecondstoafewminutesisofprimaryinterestsince
[Link],regulationreservesmight
[Link]
variabilityinfivesecondintervalscouldbeusedtodeterminetheresourcesnecessarytoprovide
frequencyregulationserviceinresponsetopowerfluctuations.
VariabilityofaPVfleetisthusameasureofthemagnitudeofchangesinitsaggregatepoweroutput
[Link]
thechangeinoutput,ratherthantheoutputitself,[Link],foreachtimeinterval
thechangeinoutputmayvaryinbothmagnitudeandsign(positiveandnegative).Astatisticalmetricis
thereforeemployedinordertoquantifyvariability:thestandarddeviationofthechangeinfleetpower
output.
Itishelpfultographicallyillustratewhatismeantbyoutputvariability.TheleftsideofFigure1presents
10secondirradiancedata(PVpoweroutputisalmostdirectlyproportionaltoirradiance)andtheright
sideofthefigurepresentsthechangeinirradianceusinga10secondtimeintervalforanetworkof25
weathermonitoringstationsina400meterby400metergridlocatedatCordeliaJunction,CAon
November7,2010(HoffandNorris,2010).Thelightgraylinescorrespondtoirradianceandvariability
forasinglelocationandthedarkredlinescorrespondtoaverageirradiancedistributedacross25
locations.Resultssuggestthatspreadingcapacityacross25locationsratherthanconcentratingitata
singlelocationreducesvariabilitybymorethan70percentinthisparticularinstance.

Copyright

Figure1

Afleetco
follows:id
periodof
synchroni
foreachP
obtainthe
calculate
Thisfleet
severalre
geograph
insolation
hourtime
satellitete
withafou
approach
selection.
Finally,ca
Amorevi
variability
theobser
individual
etc.)dete
HoffandP
theoutpu
t2010Clea
.Twentyfive
mete
omputation
dentifythePV
concern(e.g.
izedsolarirra
PVsystemusi
ecombinedf
theresulting
tcomputatio
[Link],
icalresolutio
ndataforthe
[Link]
[Link]
urminutetim
isonlyapplic
Thestudyw
alculationsare
ableapproac
ymethodolog
vationsthats
lPVsystemsa
rmineoveral
Perez(2010)
utvariabilityr
anPowerRes
locationnetw
erx400mete
approachcan
Vsystemstha
.,oneminute
adiancedataf
ngstandardm
leetoutput;c
statisticalou
napproach,
insolationdat
n).Forexamp
continentalU
hasapractica
eetcomputa
[Link]
cabletostudi
ouldhaveto
ehighlycomp
chistostream
[Link]
skyclearness
andthatphys
lfleetvariabi
developeda
resultingfrom
earch
workreduces
rgridatCord
nbetakento
atconstitutet
echangeseva
foreachlocat
modelingtoo
calculatethe
utputvariabili
whiletechni
taisnotavail
ple,SolarAny
[Link]
lrealtimelim
tioncouldno
econd,PVvar
eshavingam
berecommi
putationinten
mlinethecalc
odneedstoq
andsunposi
sicalparamet
ility.
simplifiedmo
manensembl
s10secondv
deliaJunction
ocalculateout
thefleettob
aluatedovera
tionwherea
ols;sumtheo
changeinfle
tyfromthes
callyvalid,is
lableinsuffic
where(2010)
waii,iscurrent
mitofonehal
otbeperform
riabilitydeter
matchingtime
ssionedwhen
nsive,andthu
culationsthro
quantifyshort
itiondrivethe
ters(i.e.,dime
odelasafirst
eofequallys
variabilitybym
,CAonNove
tputvariabilit
estudied;se
aoneyearpe
PVsystemis
outputfrome
etoutputfor
streamofvalu
difficulttoim
cientresolutio
),whichprov
tlybasedona
lfhouranda
medfor,say,s
rminedusing
eintervalofi
neveradditio
usarenotsui
oughtheuseo
ttermfleetp
echangesint
ensions,plan
tsteptoward
spaced,ident
morethan70
ember7,2010
tyforafleet
lectthetime
eriod);obtain
tobesited;s
eachindividua
reachtimein
ues.
mplementinp
on(eithertim
videsoperatio
a10kmx10
fewkmbase
systemsspace
thefleetcom
nterestanda
onalPVsystem
itableforrea
ofageneralp
poweroutput
theshortter
ntspacing,nu
sageneralm
ticalPVsystem
0percentina
0.
ofPVsystem
intervaland
ntime
simulateoutp
alsystemto
nterval;andfi
practicefor
meresolution
onalrealtime
kmgridanda
edoncurrent
ed0.5kmapa
mputation
afixedfleet
mscameonl
ltimeoperat
purposePVo
tvariabilityu
moutputfor
mberofplan
methodtoqua
[Link]
3
a400
sas
time
put
nally
or
e
aone
art
ine.
tions.
utput
sing
ts,
antify
ined

Copyright2010CleanPowerResearch 4

outputvariabilitytobethestandarddeviationofthechangeinoutputoversometimeinterval(suchas
oneminute),usingdatatakenfromsometimeperiod(suchasoneyear).
Thesimplifiedmodelcoveredthespecialcasewhenthechangeinoutputbetweenlocationsis
uncorrelated(i.e.,cloudimpactsatonesitearetoodistanttohavepredictableeffectsatanotherforthe
consideredtimescale),fleetcapacityisequallydistributed,andthevarianceateachlocationisthe
[Link],HoffandPerezshowedthatfleetoutputvariabilityequalstheoutput
variabilityatanyonelocationdividedbythesquarerootofthenumberoflocations:
1

o
At
PIcct
=
o
At
1
N

(1)
whereo
At
PIcct
isthestandarddeviationofthechangeinoutputofthefleetusingatimeintervaloft,
o
At
1
isthestandarddeviationofthechangeinoutputofthefleetconcentratedatasinglelocation,andN
isthenumberofuncorrelatedlocations.
MillsandWiser(2010)havederivedasimilarresultthatrelatesvariabilitytothesquarerootofthe
numberofsystemswhenthelocationsareuncorrelated.
MaximumOutputVariability
Equation(1)[Link]
[Link],supposethatthevariability
ofasinglesystemwas10kWperminuteandtherewere100uncorrelatedidenticalsystemsinthefleet.
Totalfleetvariabilityequals0.1MW(
10010kw
100
) [Link]
confidencelevel(e.g.,theymaychoose3standarddeviations)todeterminetherequiredreserve
capacity(e.g.,3x0.1MW=0.3MW).
Thiscalculationisapplicablewhentwofundamentalconditionsaresatisfied:(1)theoutputvariabilityat
asinglelocationcanbequantifiedand(2)thechangeinoutputvariabilitybetweenlocationsis
uncorrelated.
[Link](o
At
1
)istoanalyze
[Link]
aratethataccommodatesthetimeintervalofinterest(perhapsdowntoafewseconds)overa
substantialandrepresentativeperiodoftime(perhapsoverseveralyears).Suchhighspeed,high
resolutiondataisnotgenerallyavailable.
2

[Link]
theoreticallyworstcasevariabilityofasinglePVplantwouldbethatitcyclesalternatelybetween0and
[Link],supposethatthePVplantisratedat1

1
SeeEquation(8)inHoffandPerez(2010).
2
OneofthefewexamplesofthissortofdataisprovidedbyKuszamaul,[Link].(2010).

Copyright2010CleanPowerResearch 5

MWandthetimeintervalofinterestis1minute.AsillustratedinTable1,maximumvariabilityoccurs
whenthePVplantisatfullpowerat12:00,zeropowerat12:01,fullpowerat12:02,[Link]
intherightsideofthetable,[Link]
standarddeviation
3
[Link],a1MWPVplant
thatisexhibitingmaximumvariabilityovera1minutetimeintervalhasa1MWperminutestandard
deviation.Thiswouldimplythat1MWofreservecapacityisrequiredtocompensatefortheoutput
variabilityforasingleplant.
[Link].
Time Power(MW) Change(MW/min)
12:00 1 1
12:01 0 +1
12:02 1 1
12:03 0 +1
12:04 1

SupposethatthePVfleetcapacitywassplitbetweentwolocationsandeachweretoexhibit
[Link],illustratedinTable2,
[Link]
concentratedatasinglelocation,thechangeinoutputfluctuatesbetween1and1MWandthe
standarddeviationforthisscenariois1MWperminute.
Thesecondscenario,illustratedinTable3,assumesthattheplantscycleonandoffalternatelywitha
[Link],thechangeinoutputfromthefirstlocationcancelsthechangein
outputatthesecondlocation.Theresultofthisscenarioisastandarddeviationof0MWperminute.
Itisincorrecttoconclude,however,thattheupperboundofoutputvariabilityfor1MWofPVis1MW
perminutebecausethisisthelargervalueofthetwoscenarios(thefirstequals1MWperminuteand
thesecondequals0MWperminute).Thisisbecauseeachofthetwoscenariosviolatestheassumed
[Link],thechangeinoutputbetweenthetwo
locationshasperfectpositivecorrelationinthefirstscenario(i.e.,correlationcoefficientequals1)and
perfectnegativecorrelationinthesecondscenario(i.e.,correlationcoefficientequals1).

3
ThestandarddeviationofarandomvariableXequalsthesquarerootoftheexpectedvalueofXsquaredminus
thesquareoftheexpectedvalueofX. = E|X
2
] -E|X]
2
.

Copyright2010CleanPowerResearch 6

[Link](scenario1).
Time Power(MW) Change(MW/min)
Plant1 Plant2
Fleet
(1+2)
12:00 0.5 0.5 1 1
12:01 0 0 0 +1
12:02 0.5 0.5 1 1
12:03 0 0 0 +1
12:04 0.5 0.5 1

[Link](scenario2).
Time Power(MW) Change(MW/min)
Plant1 Plant2 Fleet1+2
12:00 0.5 0 0.5 0
12:01 0 0.5 0.5 0
12:02 0.5 0 0.5 0
12:03 0 0.5 0.5 0
12:04 0.5 0 0.5

FeasibleMaximumOutputVariability
Thesescenariosdemonstratethatitisimpossiblefortwosystemstoexhibitthebehaviorofworstcase
varianceindividually(bycyclingonandoffateachinterval)withouthavingeitherperfectpositiveor
[Link],foreachsystemtoexhibititsmaximumvariance,itsoutput
changesmustbeexactlyintempowiththetimeinterval,looselyanalogoustoeachmemberofan
orchestrafollowingintimetoitsconductor,inwhichcasethesystemswouldbydefinitionhaveperfect
correlation(whetherpositiveornegative).Bythisreasoning,themaximumoutputvariabilityscenario
describedabove(1MWofvariabilityforeach1MWoffleetcapacity)[Link]
havelessthanperfectcorrelation,asmustbethecaseforanyrealworldfleet,thevariabilityofthe
combinedfleetmustbelessthanthetotalfleetcapacity.
Tocorrecttheworstcasescenario,retaintheassumptionthateachpowerchangeiseitheratransition
[Link]
conservativesincetheimpactsofcloudtransientsonPVsystemswillalmostneverproducechanges
withmagnitudesashighas100percentofratedoutputandwillgenerallyproducechangesmuchless
[Link],ratherthanbeingsynchronized,eachsystemisassumedtocycleonand
offinarandomfashion,representingfleetsofPVsystemswithoutputsthatareuncorrelated.
RandomtimingofpoweroutputchangesisillustratedforasinglelocationinTable4fora1MWPV
system.Supposethatitis12:00andthetimeintervalis1minute.Thereisa50percentchancethatthe

Copyright

plantison
a50perce
at12:00,t
12:[Link]
timeinter
chanceof
Notethat
wouldact
wouldnee
systemw
needtoo
demonstr
Withthes
standard
point,and
equals:
Hoximum

Themaxim
numerica
intoEqua
squarero
Hoximum

4
SeeFigur
5
o = |(u
t2010Clea
nanda50pe
entchanceit
thenthereis
erightcolum
rval,thereisa
fnochangein
twhilethisis
[Link]
edtobesod
ouldhaveto
perateasa
ratedthat,in
Table
secaveats,th
deviationof

dthefleetha
m o
At
1
=
C
PIc
2
mumoutputv
ldefinitionof
tion(1).The
otof2times
m o
At
PIcct
=
C

re13inKuszam
u.2S)(-1)
2
+
anPowerRes
ercentchance
willturnoffa
a50percent
mninTable4p
a25percent
noutput,and
themaximum
irst,weather
arkthatthere
turnonando
pointsource
fact,asmoot
[Link]
eabovedistr
1
2
times1MW
dacapacityo
cct
2

variabilityfor
fthemaximu
resultisthat
thenumber
C
PIcct
2N

[Link].(20
([Link])(u)
2
+
earch
ethattheplan
anda50perc
chanceitwil
presentsthep
chanceofa1
da25percent
mpossiblech
conditionsw
ewouldbeno
off,rathertha
ofoutput;K
thingeffecto
mchangeinp
ributionistak
W.
5
Iftheent
ofC
Fleet
,then
rafleetofun
umoutputvar
tthatmaximu
ofuncorrelat
10)andFigure
(u.2S)(1)
2
] -
ntisoffat12
centchanceit
llstayoffand
probabilitydi
1MWpermin
tchanceofa
ange,itisext
wouldhaveto
ooutputwhe
anasubseto
Kuszamaulet.
occursassyste
poweroutput
kenforthecu
tirefleetofPV
themaximum
ncorrelatedlo
riabilityfora
umoutputva
tedlocations.
7inMillset.a
-|(u.2S)(-1)
:[Link]
twillremain
da50percent
istributionof
nutedecrease
1MWperm
tremelyunlik
beexception
encoveringa
ofthearrays.
.al.(2010)an
emsizeincre
tassumingra
urrentpurpos
Vsystemswe
mstandardde
ocationscanb
singlesystem
riabilityequa
.
al.(2009).
+ ([Link])(u) +
antisonat12
onat12:01.
tchanceitw
thechangei
einpower,a
inuteincreas
kelythatsuch
nallyerratic.S
PVsystem.T
Fourth,each
[Link].
ases.
4

ndomoutput
[Link]
ereconcentra
eviationofch
becalculated
mbysubstitut
alsfleetcapac
+(u.2S)(1)]
2
2:00,thenthe
Iftheplantis
illturnonat
[Link]
50percent
seinpower.
adistributio
Second,cloud
Third,theent
PVsystemw
.(2009)have
t.

ibutionhasa
atedatasingl
hangeinoutp
usingthis
tingEquation
citydividedb
=
1
2

7
ereis
soff
each
n
ds
ire
would

le
put
(2)
(2)
ythe
(3)

Copyright

Equation
thechang
practicalu
Perezeta
Exampl
Supposet
outputva
MWbase
composed
uncorrela
installeda
MWpert
5,000loca
or1perce
Thisexam
spreading
illustrates
reservesw
resources
battery,o
fleetsmad
amounts.
Figu

6
Appendix
t2010Clea
(3)placesan
geinoutputb
upperbound
al.,2010a).
le
thatautilitys
ariabilitycalcu
dontwoflee
dofuncorrela
ted100MW
at50location
timeinterval,
ations(N=5,0
entoffleetca
mpleillustrate
gthemapartg
stheunneces
withoutadeq
sthatwouldb
orotherdispa
deupofunco

[Link]

xAillustratesh
anPowerRes
nupperboun
betweenlocat
onsinglepoi
systemplans
ulatedusingE
etcompositio
ated1MWsy
[Link]
ns(N=50)with
or10percen
000)withunco
apacity.
6

esthepotenti
geographical
ssarypotentia
uatetoolsfo
beprocuredw
atchableresou
orrelatedsyst
umvariability

howtoverifyth
earch
ndonthemax
tionsisuncor
ntoutputis
toincorporat
Equation(3)
nstrategies.
[Link]
illustratedin
huncorrelate
ntoffleetcap
orrelatedcha
albenefitof
lysothatout
alcostthatco
rquantifying
wheneachM
[Link]
temswould
for1MWan
heseresultsus
ximumoutpu
[Link]
substantiated
te5,000MW
forPVfleets
Theblueline
redlineisthe
thefigureat
edchangesin
[Link]
angesinoutpu
dividingtheP
tputchanges
ouldbeincurr
PVvariability
WofPVwas
N=5,000exam
dresultincap
nd100MWsy
singanExcelsp
utvariabilityf
ualresultsare
dbyawealth
[Link]
withcapaciti
eisthevariab
variabilityw
the5,000MW
output,max
ver,if1MWP
ut,maximum
PVcapacityin
areuncorrela
redifsystem
[Link]
fullybacked
mple,suchap
pitalexpendit
ystemsizesw
preadsheet.
foranytimein
elikelytobe
hofempirical
e2presentst
iesrangingfr
bilitywhenthe
henthefleet
Wlevel,if100
imumoutput
PVsystemsar
moutputvaria
ntosmallsyst
[Link]
plannerswe
dlinereprese
dupwithaM
planningprac
tures99time
withuncorrela

ntervalaslon
[Link]
evidence(se
themaximum
om0to5,000
efleetis
tiscomposed
0MWsystem
tvariabilityis
reinstalledat
abilityis50M
tems,and
mportantly,it
retoprocure
entsthereser
MWoffossil,
cticeatlea
estherequire
atedchanges.
8
ngas
ee
m
0
dof
msare
s500
t
MW,
talso
e
rve
stfor
ed
.

Copyright2010CleanPowerResearch 9

CorrelationversusDistance
Background:CriticalFactorsAffectingCorrelation
Thecriticalfactorsthataffectoutputvariabilityaretheclearnessofthesky,sunposition,andPVfleet
orientation(i.e.,dimensions,plantspacing,numberofplants,etc.).Toimproveaccuracy,HoffandPerez
(2010)[Link]
incorporatesthelayoutofafleetofPVsystems,thetimescalesofconcern,andthemotionofcloud
[Link]
fromthedeploymentofmultipleplantsdecreasedquasiexponentiallyasafunctionofthegenerating
[Link](1)decreasesas
thedistancebetweensitesincreases;(2)decreasesmoreslowlyasthetimeintervalincreases;and(3)
decreasesmoreslowlyasthecloudtransitspeedincreases.
MillsandWiser(2010)analyzedmeasuredoneminuteinsolationdataoveranextendedperiodoftime
for23timesynchronizedsitesintheSouthernGreatPlainsnetworkoftheAtmosphericRadiation
Measurement(ARM)[Link]
7
thatthecorrelationofthechangeinthe
globalclearskyindex(1)decreasesasthedistancebetweensitesincreasesand(2)decreasesmore
slowlyasthetimeintervalincreases.
[Link].(2010b)analyzedthecorrelationbetweenthevariabilityobservedattwoneighboringsites
asafunctionoftheirdistanceandoftheconsideredvariabilitytimescale.Theauthorsused20second
tooneminutedatatoconstructvirtualnetworksat24USlocationsfromtheARMprogram(Stokesand
Schwartz,1994)andtheSURFRADNetworkandcloudspeedderivedfromSolarAnywhere(2010)to
calculatethestationpaircorrelationsfordistancesrangingfrom100metersto100kmandfrom
[Link]
correlationofthechangeinglobalclearskyindex(1)decreasesasthedistancebetweensitesincreases
and(2)decreasesmoreslowlyasthetimeintervalincreases.
Theconsistentconclusions
8
ofthesestudiesarethatcorrelation:(1)decreasesasthedistancebetween
sitesincreasesand(2)[Link](2010)add
thatthecorrelationdecreasesmoreslowlyasthespeedofthecloudsincreases.
Objective
UtilityplannersclearlyrequireatoolthatcanreliablyquantifythemaximumoutputvariabilityofPV
[Link](3)wouldpotentiallymeetthis
requirementifthechangeinoutputbetweenlocationswereuncorrelated(i.e.,correlationcoefficientis
zero).Inrealfleets,PVsystemswillgenerallyhavesomedegreeofcorrelation,soanyplanningtoolwill
havetoincorporatecorrelationeffectsincalculatingactualfleetvariability.
Thispapertakesasteptowardsageneralmethodbyanalyzingthecorrelationcoefficientofthechange
inclearnessindexbetweentwolocationsasafunctionofdistance,timeinterval,andotherparameters.

7
SeeFigure5inMillsandWiser(2010).
8
TheresultsapplytoeitherchangesinPVoutputdirectlyorchangesintheclearskyindex

Copyright2010CleanPowerResearch 10

ItuseshourlyglobalhorizontalinsolationdatafromSolarAnywhere(2010)tocalculatecorrelation
coefficientsfor70,000scenariosacrossthreeseparategeographicregionsintheUnitedStates
(Southwest,SouthernGreatPlains,andHawaii).Thecorrelationcoefficientstakenfromthesescenarios
arethencomparedtoamethodthatcouldproveusefulwhenintegratedintoutilityplanningand
[Link](several
secondstoseveralminutes),itsresultsarecomparedtostudiesbasedon10second,20second,and1
minuteinsolationdatasets.
Approach
HoffandPerez(2010)definedPVfleetvariabilityasthestandarddeviationofitspoweroutputchanges
usingaselectedsamplingtimeinterval(e.g.,suchasoneminuteoronehour)andanalysisperiod(such
asoneyear),[Link],theyformulateditinterms
ofthechangeininsolationratherthanthechangeinPVpower.
Asstatedearlier,skyclearnessandsunpositiondrivethechangesinshorttermoutputforindividualPV
[Link](2010)andPerez,[Link](2010)subsequentlyisolatedtherandomcomponentof
outputchangeandexaminedchangesattributableonlytochangesinglobalclearsky(orclearness)
[Link]
skyinsolation.
ThispapercontinuesinthedirectionofMillsandWiser(2010)andPerez,[Link].(2010)andfocuseson
changesintheglobalclearnessindex.
ChangeinGlobalClearnessIndex
TheglobalclearnessindexataspecificpointintimeistypicallyreferredtoasKt*.Itequalsthe
measuredglobalhorizontalinsolation(GHI)[Link]
changeintheindexbetweentwopointsintimeasKt*.Sincethechangeoccursoversomespecified
timeinterval,t,atsomespecificlocationn,thevariableisfullyqualifiedasKt

t,At
n
.Thisonly
[Link].
Thus,AKt

At
n
isthesetofchangesintheclearnessindicesataspecificlocationusingaspecifictime
intervaloveraspecifictimeperiod.
AKt

At
n
= |(t
1
, Kt

t
1
,At
n
), (t
2
, Kt

t
2
,At
n
), , (t
1
, Kt

t
T
,At
n
)| (4)

Copyright2010CleanPowerResearch 11

Table5illustrateshowtocalculatethechangeinclearnessindex(Kt*)duringconditionsofrapidly
[Link],Kt*between12:00and12:01equalsthedifferencebetweenKt*at
12:01andKt*at12:00(0.51.0=0.5).
[Link](Kt*)usingt=1minute.
Time GHI ClearskyGHI Kt* Kt*
12:00 1.0 1.0 1.0 0.5
12:01 0.5 1.0 0.5 0.5
12:02 0.0 1.0 0.0 +0.5
12:03 0.5 1.0 0.5 +0.5
12:04 1.0 1.0 1.0

Correlationanddependenceinstatisticsareanyofabroadclassofstatisticalrelationshipsbetweentwo
ormorerandomvariablesorobserveddatavalues(Wikipedia2010).LetAKt

At
1
andAKt

At
2
represent
twosetsofobserveddatavaluesforthechangeintheclearnessindexthathaveameanof0and
standarddeviations,o
1
ando
2
.
9

Pearsonsproductmomentcorrelationcoefficient(typicallyreferredtosimplyasthecorrelation
coefficient)equalstheexpectedvalueofAKt

At
1
timesAKt

At
2
dividedbythecorrespondingstandard
deviations.
p
1,2
=
E|AKt

At
1
AKt

At
2
]
o
1
o
2

(5)

Theanalysisisperformedasfollows:
1. Selectageographicregionforanalysis
2. Selectalocationforthefirstpartofthepair
3. Selectalocationforthesecondpartofthepair
4. Selectatimeintervalfortheanalysis
5. Selectaclearskyirradiancelevelbin
6. Obtaindetailedinsolationdata
7. Calculatethecorrelationcoefficient
10

8. Repeatthecalculationforallsetsoflocationpairs,timeintervals,andclearskyirradiancebins.

9
TheexpectedvalueofKt*[Link]
satisfiedwhenthetimeperiodoftheanalysisisperformedoveronedaybecausethestartingandendingGHIboth
[Link](aswouldbethecase,
forexample,ofananalysisofonehourofdatausingaoneminutetimeinterval).
10
AppendixBillustrateshowtocalculateKt*correlationcoefficients.

Copyright

Thefocus
correlatio
paramete
locations.
betweent
data,inp
Results
Scenari
Assumma
analysis:S
yellowsq
Southern
figures)w
coordinat
wereigno
January1
performe
irradiance
Region

Location#
Location#
Time
Intervals
ClearSky
Irradiance
t2010Clea
softhispaper
oncoefficients
erofthecorre
Asdiscussed
thetwolocat
articular,clou
s
ioSpecific
arizedinTabl
Southwest,So
uaresinthef
GreatPlains,
wasselectedb
teswereavail
ored).Hourly
,1998throug
dasdescribe
[Link]
#1 Latitude:
Longitud
GridSize
#2 0.1,0.3
1,2,3,a
e
10irradi
of0.1kW
anPowerRes
risontryingt
[Link]
elationcoeffic
dbyHoffand
tions,timeint
udspeed.
cation
e6,threesep
outhernGrea
figures),wass
andHawaii,
between0.1
lablebutthe
insolationda
ghSeptembe
edabovefort
nalysisresulte
Southwest
:32to42
de:125to1
e:2.0
,,1.9from
nd4hours
ancebinsini
W/m
2

earch
todetermine
theobjective,
cientofthec
Perez(2010)
terval,andlo
parategeogra
atPlains,and
selectedusin
correspondin
and2.9(abo
illustratedpo
atawasobtain
r30,2010fro
timeintervals
edinmoreth
[Link]
t

109
m#1
ntervals
eifpatternse
,amethodis
hangeinthe
,theinputsin
ocationspecif
aphicregions
[Link]
gagridsizeo
[Link]
out10to300
ointsprovided
nedforeach
omSolarAnyw
sof1,2,3,an
an70,000co
mmaryofinp
Southern
Latitude:3
Longitude:9
GridSize:1
0.1,0.3,,
1,2,3,and4
10irradiance
incrementso
existingthath
testedthatp
clearnessind
ntothismeth
ficparameter
sintheUnited
firstlocation
of2.0,1.0,o
ondlocation(
0km)fromth
dsufficientd
ofthetwolo
where(2010)
nd4hoursan
orrelationcoe
utdata.
GreatPlains
35to38
99to96
1.0
2.9from#1
4hours
ebinsin
of0.1kW/m
2
helptobetter
producesthe
dexbetween
hodincludeth
rsbasedone
dStateswere
(denotedby
or0.5forthe
(denotedbya
efirstlocatio
ataandease
cationscover
.Theanalysis
dfor10sepa
efficients.
s
Latitud
Longitu
GridSiz
1 0.1,0.
1,2,3,
10irrad
increme
rquantify
desiredoutp
twoseparate
hedistance
mpiricalweat
eselectedfor
eachofsever
eSouthwest,
aredcirclein
on(othermap
ofanalysis,s
ringtheperio
swasthen
arateclearsky
Hawaii
e:19to
ude:156to
ze:0.5
2,,1.0fro
and4hours
diancebinsin
entsof0.1kW
12
ut
e
ther
r
ral
nthe
p
so
od
y
20
155
om#1
n
W/m
2

Copyright

Correla
Figure3p
foreacht
alternativ
Thesecon
Resultsin
Resultsin
Resultsin
multiplied
(2010).Th
inthispap
usingpara
Figure4a
presented
compress
thesame

11
Thedata
readable.
12
Relative
averageim
fortheben
t2010Clea
ationCoeff
presentsthec
timeintervala
vecandidates
nd,third,and
thetoprow
themiddler
thebottomr
dbyrelatives
hedashedline
perforusein
ametersobta
andFigure5p
dinthefigure
sestheresults
figure.
Figure3

ainallofthefi
speedequalst
mpliedspeedac
nefitoftherea
anPowerRes
ficients
correlationco
andtherows
setsofvariab
fourthcolum
presentcorre
owpresentc
rowpresentc
speed;
12
thist
einthebotto
futuretools,
ainedfromSo
presentcomp
esaresimilar
sforeachloc
[Link]

guresrepresen
theimpliedsp
crosstheentir
adersothatthe
earch
oefficientsfor
presentthe
[Link]
mnsplotthes
elationcoeffi
correlationco
correlationco
termisrelate
omfiguresre
,thatwillbev
olarAnywhere
parativeresult
acrossalltim
ationandpre
ncoefficients
ntrandomlyse
eedderivedfo
egeographicr
escaleofthex
rtheSouthwe
measuredco
columnsumm
sameresultsu
cientsversus
oefficientsver
oefficientsve
edtotheDisp
presentsthe
validatedint
e.
tsfortheGre
meintervalsin
esentsresults
presentedby
electedsample
orthespecificl
[Link]
xaxisremains
est.
11
Thecol
rrelationcoe
marizesresult
usingtimeint
sthedistance
rsusdistance
ersusdistance
persionFactor
resultsofag
hepresentan
eatPlainsand
nthethreege
swherealltim
ytimeinterva
esofpointsino
locationfromS
espeedisonly
constant.
lumnssumm
efficientsvers
tsforatimei
tervalsof2,3
ebetweenthe
dividedbyti
edividedbyt
rintroduced
generalizedm
[Link]
[Link]
eographicloc
meintervalsa
alforSouthw
ordertomake
SolarAnywhere
yusedforpres
arizetheresu
usseveral
intervalof1h
3,and4hours
etwolocation
meinterval.
imeinterval
byHoffandP
method,propo
ltsarecalcula
patterns
[Link]
arecombined
est.
theresultsmo
edatabythe
entationpurpo
13
ults
hour.
s.
ns.
Perez
osed
ated
e6
dinto

ore
oses

Copyright

t2010Clea
Figure4
Figure
anPowerRes
.Correlation
[Link]
earch
coefficientsp
oncoefficient
presentedby
tspresented
timeinterva
bytimeinter
lforGreatPla
rvalforHawa
ains.
ii.
14

Copyright

Discuss
Theanaly
coefficien
previouss
alternativ
plottedve
furtherde
(bottomr
rowofFig
inputpara
Results
Anencou
directlyw
moreenc
inputdata
methodis
fourhour
methodb
intervalss
t2010Clea
Figu
sion
ysisprovidess
ntsdecreasew
studies,thisd
vewayofview
ersusdistance
ecreasedwhe
rowofFigure
gure6,fitsthe
ameters.
sProjectto
ragingresult
withseveralem
ouragingisth
atothemeth
sshowntopr
rtimeinterva
basedonpara
shorterthano
anPowerRes
[Link]
severalkeyfin
withincreasin
decreaseoccu
wingthisresu
edividedbyt
enarelatives
6).Finally,th
eempiricalda
oShorterT
oftheforego
mpiricaldata
hatthemetho
hodwastaken
roduceaccura
[Link]
ametersderiv
onehour.
earch
tioncoefficie
[Link],
ngdistance(t
ursmoreslow
ultisthatcorr
timeinterval
speed
12
isintr
hegeneralize
ataquitewel
TimeInte
oinganalysisi
sets,topred
odisshownt
nfromtheSo
atecorrelatio
ngpromptedt
vedfromtheS
ntsforallloc
consistentw
oprowofFig
wlywithlonge
relationcoeff
(middlerow
roducedforth
edmethod,sh
lwhencalibr
rvals
istheabilityo
ictcorrelatio
tobevalidreg
olarAnywhere
oncoefficients
theauthorst
SolarAnywhe
cationsandti
ithpreviouss
gure6).Secon
ertimeinterv
ficientsdecre
ofFigure6).T
hefirstlocati
hownbythed
ratedusingth
ofthepropos
ncoefficients
gardlessofth
edatasetwit
sforonehou
toevaluateth
eredatasetto
meintervals.
studies,thec
nd,alsoconsi
vals(toprow
aseatasimil
Third,thesca
ioninthepai
dashedblack
helocationsp
sedgeneralm
swithsuchac
heselectedtim
haonehour
ur,twohour,
hepotentialo
oprojectresu

correlation
stentwith
ofFigure6).
arratewhen
atterinresult
roflocations
lineinthebo
pecificderived
method,valida
[Link]
meinterval.W
rtimeinterva
threehour,a
ofusingthe
ultstotime
15
An

tsis
s
ottom
d
ated
n
While
l,the
and

Copyright

Itwouldb
intervalss
couldbeu
resolution
Whilethe
coefficien
istoshow
PVsystem
Thecircle
ofFigure
Plainsino
timeinter
thesame
Thesolid
therelatio
resultsca
Evaluat
Theabove
downtot
theutilitie
calculated
used10s
Geograp
Millsand
Southern
t2010Clea
beahighlyva
shorterthant
usedtodeter
ndatathatis
edesiredobje
nts(andthere
wthat,forag
ms.
sinFigure7c
[Link]
ordertoachie
rvalto180m
25percentc
linesconnect
onshipislinea
nbeprojecte
[Link]
tionofTim
elinearrelati
theveryshort
[Link]
dfromtheon
econd,20se
hicDiversit
Wiser(2010)
GreatPlains
anPowerRes
luablefinding
theavailable
rminevariabil
currentlyuna
ectiveistode
eforevariabili
ivencorrelati
correspondto
le,Figure6im
evea25perce
inutesandpl
orrelationco
tingthefourt
arlyrelatedto
edintheregio
ultsscaleline
meIndepe
onshipsugge
ttimeinterva
onprovidesa
nehourSolarA
cond,andon
tyStudy
)usedmeasu
networkofth
earch
gifthemetho
[Link]
lityforfleets
available.
emonstrateth
ty)asafunct
ioncoefficien
othemethod
mpliesthatPV
entcorrelatio
antsneedto
efficient.
timeinterval
othetimeint
onwithshort
earlywiththe
endenceCl
eststhatthem
als(severalse
ninitialvalid
Anywhereda
eminutedat
redoneminu
heAtmosphe
odcouldbes
wouldmean
acrosstheU.
hatthemetho
tionofPVspa
nt,itispossib
dresultstake
Vsystemsne
oncoefficient
bespacedtri
observations
[Link]
tertimeinterv
etimeinterva
laim
methodisind
econdstosev
ationoftime
atasetagainst
tasets.
uteinsolation
ericRadiation
showntoprod
thattheexis
.S.,reducingt
odaccurately
acing,amathe
bletoaccurate
nfromthedo
edtobespac
tusinga60m
iplethedista
sforeachloca
gurebegsthe
vals(i.e.,the
alforafixedc
dependentof
veralminutes)
independen
tresultsfrom
ndatafor23t
Measuremen
duceaccurate
tingSolarAny
theneedfor
ydetermines
ematicallyeq
elydetermine
ottedcurvein
ced40kmap
minutetimein
nce(120km
ationinFigur
equestionas
graysections
correlationco
selectedtim
)thatareofp
cebycompa
mindependen
timesynchro
nt(ARM)pro
eresultsfort
ywheredatas
highspeed,h
correlation
quivalentobje
espacingbet
nthebottom
artintheGre
[Link]
apart)toach
re7illustrate
towhethert
softhefigure
oefficient.

einterval,ev
primaryintere
ringresults
ntstudiestha
onizedsitesin
ogramto
16
time
set
high
ective
tween
row
eat
ethe
ieve
that
he
es).
ven
estto
t
nthe

Copyright

character
thecorrel
andWiser
generalm
generalm
intervalsi
Wiserstu
VirtualN
Perezet.a
including
onedime
intospace
clearness
thatstudy
projection
data). Res
timeinter
coefficien
thatarev

13
Figure5
14
Themin
studies.
15
SeeHoff
t2010Clea
izethevariab
lationofchan
rprovidedan
methodpropo
method,corre
intheMillsan
dyevendow
F
NetworkStu
al.(2010)obt
17stationsin
ensionalvirtua
emeasureme
indexforvar
y.Figure10re
nsfromthec
sultscompare
rvalsbetween
ntsbelow40p
veryclosetog

inMillsandW
ordifferences
fandPerez(20
anPowerRes
bilityofPVwi
ngesinglobal
nelectronicve
[Link]
lationcoeffic
ndWiserstud
ntoonemin
[Link]
udy
tained20sec
ntheARMne
alnetworks
15
[Link]
rioustimeinte
eplotstheda
urrentstudy
ewelltovirtu
n20seconds
percentmay
ether.

Wiser(2010).
in180minute
010)foradiscu
earch
thdifferentd
clearskyind
ersionofthei
hiletheoneh
cientswereca
[Link]
utetimeinte
mparisonofre
condtoonem
etworkand7
5
usingsatellit
encalculated
ervalsanddis
atafromthev
(basedoninp
ualnetworkst
to15minute
belowerasa
etimeintervals
ussionofvirtua
degreesofge
dexbetweent
rresultsand
hourSolarAny
alculatedthat
ts,presented
rvals.
14

esultstogeog
minuteinsolat
stationsinth
tederivedclo
dcorrelationc
[Link]
virtualnetwo
putparamete
tudydownto
es. Resultsfro
aresultofthe
sareduetome
alnetworkcon
eographicdive
thesegeogra
thesewereu
ywheredata
tcorresponde
inFigure8,a
graphicdivers
tiondatafor
heSURFRADn
oudspeedsto
coefficientsb
re9presents
orkstudyalon
erscalculated
ocorrelation
omthevirtua
enegativecor
ethodologicald
struction.
[Link]
phicallydispe
usedtocomp
setwasused
edtomuchs
arecomparab
sitystudy.
24measuring
[Link]
otranslatetim
betweenthec
ssomeofthe
ngwithcorres
donehourSo
coefficientso
lnetworkstu
rrelationarisi
differencesbe
eportpresent
ersedsites.M
pareagainstth
asinputtoth
hortertime
bletotheMill

gstations,
yconstructed
memeasurem
changein
keyresultsfr
sponding
olarAnywhere
of40percent
udycorrelatio
ingfromlocat
etweenthetwo
17
ted
13

Mills
he
he
lsand
d
ments
rom
e
for
on
tions
o

Copyright

t2010Clea anPowerRes
Figure
Figure10.C
earch
[Link]
Comparisono
sfromvirtual
fresultstovi
networkstu
irtualnetwor
dy.
kstudy.

18

Copyright

HighDen
Athirdda
collection
highdens
Lanai,HI)
thusismo
Thisnetw
squareco
Novembe
thereare
combinat
seconds).
Ascanbe
interestin
andPerez

Figur

Conclu
Theobjec
futureuti
Themeth
PVsites,f
t2010Clea
nsityWeath
atasetwaspr
[Link]
sitynetworks
.Second,itis
obile.
workwasdepl
omposedof10
er7,[Link]
25locations,
ionswaseva
Thus,there
seenbythe
gtonotetha
z(2010)andP
[Link]
Nov
usions
ctiveofthispa
litytoolstoe
odusedsate
fromwhichfl
anPowerRes
erStationN
rovidedbyHo
snetworkisi
providinghig
sdesignedto
oyedatCord
00metersbe
eirradianced
,thereare62
luatedusingn
are2,700un
blacklineint
atthisdatase
Perez,[Link].(
ationcoefficie
vember7,201
aperwastola
enablethecal
llitederivedd
eetvariability
earch
Network
offandNorris
nterestingfro
ghspeeddata
bedeployed
eliaJunction,
etweenstatio
dataassociate
25possibleco
ninedifferent
iquescenario
thefigure,the
etexhibitssom
(2010b)using
entsforhigh
10fortimein
aythefounda
lculationofPV
dataandphy
ycanbederiv
s(2010).This
omseveralpe
a(seeKuszam
tomultiplelo
,CAina400
ns).Figure11
edwiththese
ombinations,
ttimeinterva
os.
eresultsfitst
meoftheneg
gthevirtualn
density,25u
ntervalsfrom
ationforane
Vfleetvariab
sicalfleetpar
ved.
dataisfrom
erspectives.F
maul2010for
ocationsfors
meterby400
1presentsthe
ecoefficients
300ofwhich
als(10,20,30
theempirical
gativecorrela
networkappro
nitnetworka
10secondst
ewmethodth
bilityforplann
rameterstod
anetworkof
First,itisone
ranetworko
shortduration
0meterconfi
ecorrelation
ispresented
areunique.
0,40,50,60,
datafairlyw
ationeffectsi
oach.
atCordeliaJu
to5minutes.

hatcouldbee
ningandope
determineco
f25weather
ofthefewkn
f24sensorsi
nsoftimean
iguration(a
coefficientsf
inFigure1.S
Eachofthese
90,120,and
[Link]
dentifiedbyH
unction,CAon
employedin
rationalpurp
rrelationbetw
19
nown
in
d
for
Since
e
300
Hoff
n
poses.
ween

Copyright2010CleanPowerResearch 20

ThepaperusedhourlyglobalhorizontalinsolationdatafromSolarAnywheretovalidatethemethodby
calculatingcorrelationcoefficientsfor70,000scenariosacrossthreeseparategeographicregionsinthe
UnitedStates(Southwest,SouthernGreatPlains,andHawaii),whilevaryingdistance,timeinterval,
insolationbin,[Link]
[Link],
suchthathourlysatellitedatacouldbeusedtocalculatecorrelationcoefficientsforveryshorttime
intervals(severalsecondstoseveralminutes).Theseextrapolatedresultswerevalidatedusingresults
fromstudiesthatarebasedon20secondtooneminuteinsolationdata.
[Link],correlationcoefficientsdecreasedwithincreasing
[Link],correlationcoefficientsdecreasedatasimilarratewhenplottedversusdistance
[Link],theaccuracyofresultswasfurtherimprovedwhenanimpliedspeed
[Link],theseresultsprovidedthebasisforvalidatingthe
[Link],basedoninputparametersfromhourlySolarAnywheredata,
producedcorrelationcoefficientsforshorttimeintervals(secondstominutes)thatcomparedquitewell
toresultsfromindependentstudiesthatused10second,20second,andoneminutedatasets.
Thepreliminaryconclusionofthisworkisthattheapproachvalidatedinthispapercanbeusedto
[Link]
result,itcanbeusedtosatisfyoneoftheinitialmotivationsofthisstudy:thedesiretoequiputility
plannerswithatoolcapableofplacinganupperboundonthemaximumoutputvariabilityofafleetof
PVsystemsusingamanageableamountofdataandanalysis.
Theresults,however,[Link],theresultsmaybethebasisfor
quantifyingoutputvariabilityevenwhencorrelationexists.
NextSteps
Thisstudydemonstratedtheabilitytopredictcorrelationcoefficientsusingtimeintervalsof1to4hours
[Link]
comparedtohighspeedstudies,againbasedonlongtimeperioddatasets.
Thenextstepswillbetofurthervalidatetheresultsforshorttimeintervalsusingmeasuredhigher
[Link]:1kmx1kmgrid,hourSolarAnywheredatainselectedlocations;
1kmx1kmgrid,oneminuteextrapolatedSolarAnywheredatainselectedlocations;andadditional10
seconddatafromthemobile,highdensitynetworkdescribedearlier(HoffandNorris,2010).A
particularlyimportantfocusofthisworkwillbetoassessthemethodsabilitytopredictcorrelation
betweenlocationsovershorttimeperiodsaswellaslongtimeperiods(severalhoursversusseveral
years).
Acknowledgements
PortionsofthisstudywerefundedunderaCaliforniaSolarInitiative(CSI)GrantAgreementtitled
[Link]
CommissionistheFundingApprover,ItronistheProgramManager,andPG&EistheFunding

Copyright2010CleanPowerResearch 21

[Link](CleanPowerResearch)whodesigned,implemented,andoperated
[Link]
(CleanPowerResearch)[Link].
References
Hoff,T.E.,Perez,[Link].SolarEnergy84(2010)17821793.
Hoff,T.E.,Norris,[Link]:CordeliaJunctionResults.
Kuszamaul,S.,Ellis,A.,Stein,J.,Johnson,[Link]
CharacterizingSolarResourceVariabilityofMWScalePVSystem.35
th
PhotovoltaicSpecialists
Conference,Honolulu,HI.June2025,2010.
Mills,A.,Wiser,[Link]
SolarPower.LawrenceBerkeleyNationalLaboratoryTechnicalReportLBNL3884E.
Mills,A.,Alstrom,M.,Brower,M.,Ellis,A.,George,R.,Hoff,T.,Kroposki,B.,Lenox,C.,Miller,N.,Stein,J.,
Wan,Y.,[Link]
electricpowersystem.LawrenceBerkeleyNationalLaboratoryTechnicalReportLBNL2855E.
Perez,R.,Kivalov,S.,Schlemmer,J.,HemkerJr.,C.,Hoff,[Link]
[Link].
Perez,R.,Kivalov,S.,Schlemmer,J.,HemkerJr.,C.,Hoff,[Link]
[Link].
SolarAnywhere_,[Link],SatelliteDerivedSolarIrradianceData
Forecasted7daysAheadandArchivalDatabacktoJanuary1,[Link].
Stokes,G.M.,Schwartz,S.E.,[Link](ARM)program:
[Link]
MeteorologicalSociety75,12011221.
Wikipedia.2010.[Link]
Woyte,A.,Belmans,R.,Nijs,[Link]:Analysisand
statistics.SolarEnergy81(2),195206.

Copyright

Appen
Table7an
spreadshe
locations
outputva
Tim
12
12
12
12
12

13

t2010Clea
dixA:Ver
ndFigure12i
eetasshown
between12:0
ariabilitybyca
Tab
me
Plant#
:00 1
:01 0
:02 0
:03 1
:04 0

:00 0
Figu
anPowerRes
rification
illustratehow
inTable7by
00and13:00
alculatingthe
[Link]
#1 Plant#2
0
0
1
0
1

1
[Link]

earch
ofWorst
wtoconstruct
yrandomlyge
.Sumtheflee
estandarddev
aseexamplef
Power(
2 Plant#3
0
1
0
0
1

1
atedfleetout
tCaseSce
ttheworstca
eneratingeith
[Link]
viationofthe
for5,000inde
MW)
#

putwithmax
enarioRes
asescenariou
her0or1for
lculatethech
echangeinpo
ependent1M
#5000 Fle
1 2,5
0 2,5
1 2,5
1 2,5
0 2,5

0 2,5
ximumrando
sultsUsin
usingExcel.C
eachminute
hangeinfleet
ower.
MWsystems.
Chang
eet
519
552
558
545
540

525
mvariability.
ngExcel
Constructa
forall5,000
[Link]
ge(MW/min)
33
6
13
5
12

22
culate

Copyright

Appen
Table8an
whichcan
Correlati
Time

12:00
12:01
12:02
12:03
12:04

Correlati
Time

12:00
12:01
12:02
12:03
12:04

Correlati
Time

12:00
12:01
12:02
12:03
12:04
t2010Clea
dixB
ndFigure13p
nthenbeuse
ioncoefficien
GHI(kW/
#1
1.0
0.0
0.0
1.0
1.0
ioncoefficien
GHI(kW/
#1
1.0
0.0
0.0
1.0
1.0
ioncoefficien
GHI(kW/
#1
1.0
0.0
0.0
1.0
1.0
Fig
anPowerRes
presentthree
dtocalculate
Table8.
ntequals1.0
m
2
) Clear
#2 #1
1.0 1.0
0.0 1.0
0.0 1.0
1.0 1.0
1.0 1.0
ntequals0.5
m
2
) Clear
#2 #1
1.0 1.0
1.0 1.0
0.0 1.0
1.0 1.0
1.0 1.0
ntequals0
m
2
) Clear
#2 #1
1.0 1.0
1.0 1.0
0.0 1.0
0.0 1.0
1.0 1.0
[Link]
earch
eexamplesof
ecorrelation
.Datatocalc
SkyI(kW/m
2
1 #2
0 1.0
0 1.0
0 1.0
0 1.0
0 1.0
SkyI(kW/m
2
1 #2
0 1.0
0 1.0
0 1.0
0 1.0
0 1.0
SkyI(kW/m
2
1 #2
0 1.0
0 1.0
0 1.0
0 1.0
0 1.0
geinclearne
fhowtocalcu
coefficients.
ulatecorrelat
) Clearness
#1
1.0
0.0
0.0
1.0
1.0
) Clearness
#1
1.0
0.0
0.0
1.0
1.0
) Clearness
#1
1.0
0.0
0.0
1.0
1.0

ssindexforL
ulatethecha
tioncoefficie
Index Ch
#2
1.0
0.0
0.0
1.0
1.0
Index Ch
#2
1.0
1.0
0.0
1.0
1.0
Index Ch
#2
1.0
1.0
0.0
0.0
1.0
Location2vs.
ngeinthecle
nts.
hangeinClearn
#1
1.0
0.0
+1.0
0.0
hangeinClearn
#1
1.0
0.0
+1.0
0.0
hangeinClearn
#1
1.0
0.0
+1.0
0.0
Location1.
earnessindex
nessIndex
#2
1.0
0.0
+1.0
0.0
nessIndex
#2
0.0
1.0
+1.0
0.0

nessIndex
#2
0.0
1.0
0.0
+1.0

23
x,