Sei sulla pagina 1di 15

22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA

CT&FCiencia,TecnologayFuturo ServicesonDemand
PrintversionISSN01225383
Article
C.T.FCienc.Tecnol.Futurovol.5no.5BucaramangaJuly/Dec.2014
English(pdf)

Articleinxmlformat
Journalofoil,gasandalternativeenergysources
Articlereferences
ASTATISTICALANALYSISOFWINDSPEED Howtocitethisarticle
DISTRIBUTIONMODELSINTHEABURRVALLEY, Automatictranslation
COLOMBIA
Sendthisarticlebyemail
ANLISISESTADSTICODEMODELOSDE Indicators
DISTRIBUCINDELAVELOCIDADDELVIENTOENEL
VALLEDEABURR,COLOMBIA CitedbySciELO

Accessstatistics
ANLISEESTATSTICADEMODELOSDEDISTRIBUIO
DAVELOCIDADEDOVENTONOVALLEDEABURR, Relatedlinks
COLMBIA Share

PaulaAndreaAmayaMartnez 1,AndrsJulinSaavedraMontes 1andEliana More


IsabelArangoZuluaga 1
More
1 UniversidadNacionaldeColombia,Medelln,Antioquia,Colombia

Permalink
email:ajsaaved@unal.edu.co

* Towhomcorrespondenceshouldbeaddressed

Howtocite:AmayaMartnez,P.A.,SaavedraMontes,A.J.&ArangoZuluaga,E.I.(2014).Astatisticalanalysisof
windspeeddistributionmodelsintheAburrValley,Colombia.CT&FCiencia,TecnologayFuturo,5(5),121136.

(Received:Mar.17,2014Accepted:Dec.17,2014)

ABSTRACT

Theprobabilitydensityfunctions,thatmodelthewindspeedbehaviorinfiveurbanplacesandoneruraloftheAburr
ValleyinAntioquia,Colombia,arepresentedinthispaper.Theprobabilitydensityfunctionsareusedtocalculatethe
windpowerdensityforeachlocation,whichareusedtorecommendseveralapplicationsthatcouldtakeadvantageof
suchpotentialpowers.Windspeedsarerecordedatfivemonitoringstationslocatedintheurbanareaofthevalleyand
atonenearbyruralstation.Fourprobabilitydensityfunctions,namely,Weibull,Rayleigh,Gamma,andLognormal,are
usedtorepresentthewindspeeddatahistogramsofeachlocationandtoselecttheprobabilitydensityfunctionthat
bestfitthevariabilityofthewindspeeddata.Also,fourgoodnessoffittestarecalculated.Thewindpowerdensityis
calculatedwiththeprobabilitydensityfunctionthatbestrepresentsthewindspeeddistributiontoevaluatethewind
poweravailability.Thepowerdensitiesreportedforthefiveurbanstationsrangedfrom1.38to4.54W/m2,andthat
reportedfortheruralstationwas911.1W/m2.Takingintoaccountthepowerdensityofeachstation,several
applicationsaresuggested.

Keywords:Powerdensity,Probabilitydensityfunction,Goodnessoffittest,Windspeed.

RESUMEN

Enesteartculosepresentanlasfuncionesdedensidaddeprobabilidadquemodelanelcomportamientodela
velocidaddelvientoencincolugaresurbanosyunoruraldelValledeAburrenAntioquia,Colombia.Lasfuncionesde
densidaddeprobabilidadsonutilizadasparacalcularladensidaddepotenciaenelvientoparacadalugar,lacuales
utilizadapararecomendarvariasaplicacionesquepodranaprovecharlapotenciadisponible.Lasvelocidadesde
http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 1/15
22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA
vientosonregistradasencincoestacionesdemonitoreoubicadasenelreaurbanadelvalleyunaestacinrural
cercanaalValle.LascuatrofuncionesdedensidaddeprobabilidadWeibull,Rayleigh,GammayLognormalsonusadas
pararepresentarloshistogramasdelosdatosdevelocidaddelvientodecadaubicacinycuatropruebasdebondad
deajustesoncalculadasparaseleccionarlafuncindedensidaddeprobabilidadquemejormodelalavariabilidadde
losdatosdevelocidaddelviento.Adicionalmente,ladensidaddepotenciaenelvientoescalculadaconlafuncinde
densidaddeprobabilidadquemejorrepresentaladistribucindevelocidaddelvientoparaevaluarladisponibilidadde
potenciaenelviento.Lasdensidadesdepotenciareportadasparalascincoestacionesurbanasestnenelrangode
1.38a4.54W/m2yladensidadreportadaparalaestacinrurales911.1W/m2.Variasaplicacionessonsugeridas
teniendoencuentaladensidaddepotenciadecadaestacin.

Palabrasclave:Densidaddepotencia,Funcindedensidaddeprobabilidad,Pruebasdebondaddeajuste,Velocidad
delviento.

RESUMO

Nesteartigosoapresentadasasfunesdedensidadedeprobabilidadequemodelamocomportamentoda
velocidadedoventoemcincolugaresurbanoseumruraldoValedeAburr,Antioquia,Colmbia.Asfunesde
densidadedeprobabilidadesoutilizadasparacalcularadensidadedepotncianoventoparacadalugar,aqual
utilizadapararecomendarvriasaplicaesquepoderiamaproveitarapotnciadisponvel.Asvelocidadesdevento
soregistradasemcincoestaesdemonitoramentolocalizadasnareaurbanadovaleeumaestaoruralprxima
aoVale.AsquatrofunesdedensidadedeprobabilidadeWeibull,Rayleigh,GammaeLognormalsousadaspara
representaroshistogramasdosdadosdevelocidadedoventodecadalocalizaoequatroprovasdebondadede
ajustesocalculadasparaselecionarafunodedensidadedeprobabilidadequemelhormodelaavariabilidadedos
dadosdevelocidadedovento.Adicionalmente,adensidadedepotncianoventocalculadacomafunode
densidadedeprobabilidadequemelhorrepresentaadistribuiodevelocidadedoventoparaavaliaradisponibilidade
depotncianovento.Asdensidadesdepotnciaregistradasparaascincoestaesurbanasestonafaixade1.38a
4.54W/m2eadensidaderegistradaparaaestaoruralde911,1W/m2.Vriasaplicaessosugeridas
considerandoadensidadedepotnciadecadaestao.

Palavraschave:Densidadedepotncia,Funodedensidadedeprobabilidade,Provasdebondadedeajuste,
Velocidadedovento.

1.INTRODUCTION
Theusageandapplicationofrenewableenergieshaveincreasedsignificantlyinrecentyears(Dincer,2011Fyrippis,
Axaopoulos&Panayiotou,2010Hernndez,Espinosa,Saldaa&Rivera,2012).Windenergyisanimportanttypeof
renewableenergythatprovidesacleanandinexhaustibleenergysource(Li&Li,2005aSalameh&Nandu,2010).
Thepresenceandbehaviorofwindareknownthroughlargescalemapsofwindathighaltitudesaroundtheworld
however,lessinformationisavailableregardingthepresenceandspeedofwindinurbanareasatlowaltitudes
(Hernndezetal.,2012Li&Li,2005bSeguro&Lambert,2000).

Localanemometriccharacteristicsareusedinwindstatisticalanalysis.Amongtheavailableanalyticaltools,the
ProbabilityDensityFunction(PDF)standsout,whichisusedtoassesstheavailablewindenergyatagivenlocation
anddescribesthevariationofwindspeedtooptimizethedesignofwindenergyconversionsystems(LoBrano,Orioli,
Ciulla,&Culotta,2011Chang,2011).Thisapproachreducesthelongtimeandresourcesthatareassociatedwith
theprocessingofhourlywindspeedinformationrecordedduringmanyyears(Li&Li,2005b).Themostwidelyused
PDFscorrespondtotheWeibullandRayleighdistributionsbecausetheyrepresent,withhighprecision,windspeed
datafromvariousgeographicallocationsaroundtheWorld(Acosta&Djokic,2010Akdag&Gler,2010Celik,2004
LoBranoetal.,2011Safari,2011Safari&Gasore,2010Zhou,Erdem,Li&Shi,2010).OtherPDFs,suchasthe
Gamma,Lognormal,InverseGaussianandMaximumEntropyPrinciple,havebeenusedwithpositiveresultsinsome
studies(Zhouetal.,2010Li&Li,2005a).InthePalermourbanareainItaly,theWeibull,Rayleigh,Lognormal,
Gamma,InverseGaussian,PearsontypeVandBurrPDFs,weretheonesthatbestrepresentedwindspeeddata(Lo
Branoetal.,2011).Basedonthereferencesdiscussedabove,itisclearthatinurbanareasthereiswindpotential
whichdependsontheenvironmentandbuildingconfigurations.AlthoughtheAburrValleyisatropicalanddeepvalley
wherethewindspeediscomparativesmallerthanothercities,itisnecessarytocharacterizethewindspeedusing
commonPDFstoknowthewindpowerandtotakeadvantageofurbanapplicationssuchasbatterychargersand
lighting.

BecausethereareseveralPDFsthatcouldmodelthewindbehaviorinaspecificplace,itisnecessarytoselectthe
PDFthatbestrepresentsthewindspeed.Zhouetal.(2010)usedthechisquare(2),coefficientofdetermination
(R2),RootMeanSquareError(RMSE),loglikelihood(LogLH),KolmogorovSmirnov(KS),andIndexofAgreement
(Index_A)goodnessoffitteststodeterminewhichPDFsshowedthebestadjustmenttogivenlocationsavoiding
biasedresults.MorethanonegoodnessoffittestiscommonlyusedtoguaranteetheselectionofthePDFthatbest
modelsthewindspeedinaspecificplace.

Theassessmentofavailablewindresourcesinagivenlocationandtimeperiodisbasedonthecalculatedwindpower
density,whichcanbeobtainedbyseveralmethods.InChang(2011)thewindpowerdensityiscalculatedfromwind
speeddataasatimefunction.Toobtainconfidentwindpowerdensityresults,thismethodrequireswindspeeddata
measurements,whichnotalwaysareavailable.AnotherisbasedonthewindPDF,whichisusedtocalculatetheWind
http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 2/15
22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA

PowerDensityDistribution(WPDD).AcomparisonoftherelativeerrorbetweentheWPDDobtainedfromthePDFand
theWPDDobtainedfromwinddataisconsideredwhenmakingadecisionifawindPDFisagoodrepresentationofthe
windbehavior(Li&Li,2005a).Therelativeerrorbetweenthepowerdensitiescalculatedbythesetwomethodstakes
thepowerdensitycalculatedfromthewinddataasthereferencevalue.OnceaPDFischosentomodelthewind
behaviorofaspecificplace,itisnecessarytocalculatethewindpowerdensityandvalidatetheresultcomparingit
withtheWPDDobtainedfromwinddata.

ThemainobjectiveofthisworkistopresentthePDFsandtheirparameters,whichmodelthewindspeedbehaviorin
fiveurbanplacesandoneruraloftheAburrValleyinAntioquia,Colombia.Fourgoodnessoffittestareusedtoselect
thePDFthatbestrepresentthewindspeedbehaviorineachassessedlocation.Additionally,thewindpowerdensity
foreachlocationiscalculatedwiththeselectedPDFanddifferentapplicationsarerecommendedtotakeadvantageof
suchpotentialpowers.

Therestofthepaperisorganizedasfollows:InSection2themethodologyusedtoassessthewindresourcesinsix
locationsoftheAburrValleyisdescribed.TheresultsandtheirdiscussionarepresentedinSection3.Finally,the
conclusionsarepresentedinSection4.

2.METHODOLOGY
ThemethodologyusedtoassessthewindresourcesinthefiveurbanlocationsintheAburrValleyandtheone
nearbyrurallocationispresentedinthissection.First,thestudyareaandthemonitoringstationdescriptionare
presented.Then,thefourPDFsoftheWeibull,Rayleigh,LognormalandGammadistributionusedtofittheobserved
windspeeddataaredescribed.ThefourteststoevaluatethegoodnessoffitofeachPDFarepresentedtakinginto
accountthattheyaretheonesmostwidelyusedinpreviousworks(Akpinar&Akpinar,2007Chang,2011).Laterin
thissection,thepowerdensitycalculationusingthePDFthatbestfitsthehistograms,andtheaveragepowerdensity
calculationusingtheobservedwinddataaredescribed.Thelatermethodisusedasareferencetovalidateifthe
selectedPDFrepresentsthewindbehaviorinthespecificlocation.

StudyAreaandMonitoringStations

ThestudywasconductedinfiveurbanlocationswithintheAburrValleyandinonerurallocationoutsidebutcloseto
thevalley,asshowninFigure1.TheAburrValley,anaturalbasinoftheMedellinRiver,islocatedinthecentral
southernpartoftheAntioquiaprovinceinColombia.Thevalleyhasanirregularandsteeptopographywithaltitudes
rangingfrom1300mto2800mabovesealevel.Thelower,flatpartofthisnarrowvalleyhasamaximumwidthof10
kmfromeasttowestandalengthof70kmfromsouthtonorth.

Thenamesofthelocationsofthefiveurbanmonitoringstationsare:FacultaddeMinas(FM),GirardotaColegio
Colombia(GCC),PolitcnicoJaimeIsazaCadavid(PJIC),UniversidadCES(UC),andDitaires(DI).Theruralstationis
calledSantaElena(SE).TheanemometeratFMmonitoringstationislocated4mabovetheground.Thestationis
influencedbyhighvehiculartraffic.Thedistancefromthestationtotheborderroadis3mandtheroadwidthis12m.
Thelocalenvironmentofthestationisresidential,universityandmultipleactivities.ThePJICstationislocatedina
residentialandmultipleactivitiesenvironmentandisinfluencedbyhighvehiculartraffic.Thedistancefromthestation
totheborderroadis5mandtheroadwidthis7m.Theanemometerislocated6mabovetheground.

http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 3/15
22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA

TheGCCmonitoringstationisfixedandislocatedinasuburbanareai.e.biggreenareasaremixedwithbuiltareas.
Thestationisindirectlyinfluencedbythetraffic.Thedistancefromthestationtotheborderroadis4m,theroad
widthis5m,andtheanemometeris10mabovetheground.TheDImonitoringstationisfixedandislocatedinan
industrialareaalsotheareaischaracterizedasmandatorysocialuse.Thereisnotsignificantinfluenceofvehicular
trafficortallbuildings.Theanemometeris4mabovetheground.SEmonitoringstationislocatedinaruralareawitha
verysmallresidentialindexandbasicallythereisnobuildingorroadsaffectingthestation.

Inallstations,thesamplingtimeusedwas60minutesthissamplingtimehasbeencommonlyusedtorecordurban
winddatainpreviousstudies(Acosta&Djokic,2010Georgakis&Santamouris,2008Li,Wang,&Yuan,2010Lo
Branoetal.,2011Zhouetal.,2010).ThewinddatafromurbanareaswereprovidedbytheAirQualityLaboratory
(CALAIRE)atUniversidadNacionaldeColombia,andcorrespondtooneyear.Thenumberofmeasurementsandthe
minimum,maximumandaveragewindspeedsforeachstationaresummarizedinTable1.Also,thelatitude,longitude
andheightofeachstationareprovided.Incertaincases,thenumberofmeasurementswaslowerthanitshouldhave
beenasaresultofdatalossduringcertaintimeperiods.

ProbabilityDensityFunctions(PDFs)

PDFsaremathematicalfunctionsthatcharacterizetheprobablebehaviorofadataset.Thisdatasetcancorrespond
tothebehaviorofarandomvariablethatiscontinuousintime.Sincewindspeedisrandomvariable,itisappropriate
todescribethoseusingPDFs.TheWeibull,Rayleigh,LognormalandGammaPDFshavebeenusedinmanystudiesto
evaluatetheinstallationofwindturbinesinagivenlocationortocalculatetheavailablewindresources(Acosta&
Djokic,2010Fyrippisetal.,2010Li&Li,2005aLoBranoetal.,2011Safari,2011Zhouetal.,2010).

Inthisstudy,themaximumlikelihoodmethodwasusedtocalculatetheWeibull,Rayleigh,LognormalandGammaPDF
parametersthatmaximizetheprobabilityoftherecordedwindspeeddata.Themaximumlikelihoodmethodis
consideredtobeapreciseandrobustmethod(LoBranoetal.,2011Seguro&Lambert,2000),anditisalsothe
mostappropriatecomputationalmethodfortheanalysisofwindenergywhentheWeibulldistributionparametersare
estimatedbasedonthewindspeeddataseries(Seguro&Lambert,2000Zhouetal.,2010).ThePDFparameters
werecalculatedusingthemaximumlikelihoodmethodfromtheStatisticalAnalysisSoftwareSAS .Weibull,Rayleigh,
LognormalandGammaPDFdefinitionsaregivenbyEquation1,2,3and4respectively.

InEquation1,2,3and4,xrepresentstheobserveddata.TheWeibulldistributionhasthreeparameters:theform
parameter,thescaleparameter,andthelocalizationparameter.Thefamilyofcurvesthatcanbegenerated
fromtheWeibullPDFusingdifferentparametersresultsinagoodfittothemeasuredwindspeeddatabecauseofthe
flexibilityoftheparametersandthecapabilityofthefunctiontoadjustbothexponentialandnormalshapes(Chang,
2011Zhouetal.,2010).

TheRayleighdistributionisaspecialcaseoftheWeibulldistributionandisusedfrequentlyintheliteraturetofitwind
PDFs(LoBranoetal.,2011).OneadvantageoftheRayleighdistributionisthatthePDFandthecumulative
http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 4/15
22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA

distributionfunction(F)areobtainedfromtheaveragewindspeed.InthisPDF,theformparameterisequalto2and
theparameterisequalto2timesthebparameterfromtheRayleighdistribution(Acosta&Djokic,2010Zhouet
al.,2010).

TheLognormaldistributionisusedtofitwindspeedvariationsandtogeneratefrequencydistributionsofwindspeed
foragiventimeinterval(Li&Li,2005bLoBranoetal.,2011Zhouetal.,2010).Itisaprobabilitydistributionofa
randomvariablewhoselogarithmhasanormaldistribution.TheparametersofthePDFincludethescaleparameter,
thestandarddeviation,andthelocalizationparameter.

Thefirststatisticalstudiesofwindspeedwereperformedapproximately60yearsago,andthesestudiesconsidered
windspeedtobeadiscreterandomvariable,andmodeleditusingaGammadistribution.ThegeneralizedGamma
distributionisappropriatefordescribingthesurfacewindspeeddistributioninmostofEurope(LoBranoetal.,2011).
Thisdistributionrepresentsthesumofexponentialfunctionsdistributedasrandomvariables,anditsPDFincludesthe
formparameter,thescaleparameterandthelocalizationparameter.

ThecumulativePDFsoftheWeibull,Rayleigh,LognormalandGammadistributionsaregivenbyEquations5,6,7and8
respectively.ThesecumulativePDFsareusedtocalculatethegoodnessoffittestdescribedinnextsubsection.The
observeddataarerepresentedbyx.

GoodnessofFitTests

Toevaluatethegoodnessoffitofthechosendistributionstothewindspeedhistograms,fourtestswereselected
basedonaliteraturereview.ThemostcommonlyusedtestsaretheR2,RMSE,KSandIndex_A.Thedefinitionsof
thefourtestsarepresentedinEquations9,10,11and12respectively(Li&Li,2005aLoBranoetal.,2011Ramrez
&Carta,2005Zhouetal.,2010).R2,RMSEandKStestshavebeenusedextensivelytocomparethestatistical
distributionsobtainedinwindspeedmodelingstudies.TheIndex_Atesthasbeenappliedintheanalysisofstatistical
distributionstopredictaircontaminantsandwinderosion.

CoefficientofDetermination(R2)Test

http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 5/15
22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA

R2measuresthedegreeofthelinearrelationshipbetweentheexpectedandtheobservedwindspeedfrequency
values.Nrepresentsthenumberofbinsfortheintervalsdefinedinthehistogramandpisthenumberofparameters
oftheanalyzeddistribution.ThetotalsumofsquaresSS Tofthedifferencebetweenthefrequencyoftheobserved
windspeeddata(Oi),whichisthesumofthenumberofwindspeeddatathatareincludedintheithhistogrambin,
andtheaverageofthefrequenciesobserved( )isgivenbyEquation13.ThesumofsquarederrorsSS Eofthe
differencebetweentheobservedandtheestimatedwindspeedfrequencyisgivenbyEquation14.Theestimatedor
expectedwindspeedfrequency(Ei)isgivenbyEquation15,whereFrepresentsthecumulativeprobabilitydistribution
functionevaluatedattheupper(Ui)andlower(Li)limitsoftheithhistogrambin.Theseupperandlowerlimits
correspondtothelargestandsmallestvaluesofthewindspeedforthebinanalyzed.nisthetotalnumberofwind
speedmeasurements.WhentheR2valueiscloserto1,thefitofthedatatothedistributionusedisbetter.

RootMeanSquareError(RMSE)Test

ThistestcalculatestheRMSEbetweentheobservedwindspeedfrequencyOiandtheestimatedwindspeedfrequency
Ei.Theresultofthetestmaydifferbyseveralordersofmagnitude,dependingonwhethertheRMSEiscalculated
usingthefrequencies,theprobabilitydensitiesortheestimatedwindspeedprobabilities.TheRMSEvalueislarger
whenitiscalculatedwiththefrequenciesthanwhenitiscalculatedwiththeprobabilitydensitiesortheprobabilities.
WhentheRMSEofaPDFisclosertozero,thefitbetweentheobservedvaluesandthevaluesexpectedfromthePDF
isbetter.WhentheRMSEvaluesforseveralPDFsarecompared,itisnecessarythatthesevaluesarecalculatedusing
thesamevariable,i.e.frequency,probabilitydensityorprobabilitytomakeafaircomparison.

KolmogorovSmirnov(KS)Test

Thistestisbasedonthemaximumdifferencebetweenthehypotheticalandtheempiricalcumulativedistributions(FE).
TheempiricalcumulativeprobabilitydistributionfunctionisdefinedbyEquation16:

WhereE(i)isthenumberofpointssmallerthanxi,withxivaluessortedfromsmallesttolargest.FEisastepfunction
thatincreasesby1/natthevalueofeachsorteddatapoint.ThenullhypothesisintheKStestisthatthedatafollows
aspecificdistribution.Iftheresultofthetestislowerthanacriticalvalue,thefittothedistributionisconsideredtobe
good.TheKStestwasoriginallypresentedasagoodnessoffittestbyMasseyJr(1951)togetherwithacriticalvalue
table.

IndexofAgreement(Index_A)Test

Index_A,asshownbythedefinitioninEquation12,isthedifferencebetweenthenumber1andthequotientofthe
sumofsquaresofthedifferencesbetweentheobservedandtheexpectedfrequencies,andthesumofsquaresofthe
absolutevaluesofthedifferencesbetweentheobservedfrequencyvaluesandtheiraverageasgiveninEquation17,
andtheabsolutevaluesofthedifferencesbetweentheexpectedfrequencyvaluesandtheiraverageasgivenin
Equation18.ThepossiblevaluesofIndex_Arangefrom0to1.Avaluethatiscloserto1indicatesabetterfitofthe
PDFtotheobserveddatahistogram.

WindPowerDensity

Windpowerdensityiscalculatedtoassesstheavailabilityofwindresourcesinagivenlocation(Li&Li,2005b).Wind
powerdensityisproportionaltothecubeofwindspeedandforaselectedPDF(v)canbecalculatedasshownin
Equation19(Akpinar&Akpinar,2007Li&Li,2005aChang,2011):

http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 6/15
22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA

Wherevminandvmaxcorrespondtotheminimumandmaximumwindspeeddata,respectively,whichwereusedto
obtaintheparametersofthePDF.representstheairdensityinkg.m3andvrepresentsthewindspeedinm.s 1.

Whenatimeseriesofwindspeeddataisavailable,anaveragewindpowerdensity( )canbecalculatedbyEquation
20,whichusestheaverageairdensityvalueandthecubeoftheaveragewindspeed(Carta&Ramrez,2007
Johnson,1985Chang,2011):

AirdensitycanbecalculatedfromtheequationCIPM2007approvedbytheInternationalCommitteeforWeightsand
Measures(Picard,Davis,Glser&Fujii,2008).

3.RESULTSANDDISCUSSION
ComparisonsBetweentheDistributionFunctionsandtheObservedData

Thevaluesoftheform,scaleandlocalizationparametersforeachdistributionandeachstationarepresentedin
Table2.ThethreeparametersofWeibullandRayleighdistributionswereestimatedforthefivestationsintheAburr
Valleyhowever,thewindspeedbehaviorforFMandSEstationswasnotdescribedbythePDFsofLognormaland
Gammadistributions.Forthelattertwodistributions,noreasonablePDFparameterscouldbefound.Thecomparisons
betweenthewindspeedhistogramsandthePDFsofthedistributionsusedtofitthemarepresentedinFigure2.

http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 7/15
22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA

EachofthesixwinddistributionwasrepresentedbytwoormorePDFs.Goodnessoffittests,asdefinedinEquations
912,wereperformedtoselectthePDFofthedistributionthatbestrepresentedthebehaviorofthewindspeeddata
foreachlocation.ThefourgoodnessoffittestconductedtocompareeachPDFwiththecorrespondinghistogramare
presentedinTable3.Foreachstation,thedistributionsweresortedaccordingtothecoefficientofdeterminationR2.A
valuecloserto1forthistestindicatesadistributionwithaPDFthatbetterfitsthewindspeeddata.Thecalculated
valueofRMSEcoincideswiththevalueofR2becauseforeachstationthesmallestRMSEvaluecorrespondstotheR2
valueclosestto1.TheRMSEswerecalculatedusingvaluesoftheobservedandestimatedfrequenciesasdefinedin
Equation10.TheRMSEvaluesobtainedarelargerthanthoseobtainedinsimilarstudies(Akpinar&Akpinar,2007Li&
Li,2005bSafari,2011Chang,2011)becausefrequencies,ratherthanprobabilitiesorprobabilitydensities,were
usedfortheRMSEcalculation.

http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 8/15
22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA

NoneoftheresultsfromtheKStestwaslessthanthecorrespondingcriticalvaluefromtheKolmogorovtable.
Therefore,thehypothesisthatthedistributionwasagoodfittotheobserveddatawasnotacceptedhowever,the
trendoftheKStestresultswasconsistentwiththoseoftheRMSEandR2tests.TheIndex_Atestresultequals1
whenthePDFofagivendistributionrepresentstheobserveddataexactly.TheresultsfromtheIndex_Atestwerein
thesameorderasthosefromtheR2test.Foreachstation,therewasatleastonedistributionthathadanIndex_A
valuegreaterthan0.99,whichindicatesagoodfitofthedatawiththeanalyzeddistribution.TheresultsinFigure2
areanalyzedbelowforeachlocation.

FacultaddeMinas(FM)

ThestationatFMsampleddataeveryhour.Thisstationwascharacterizedbylowaveragespeed1.1034m/s.The
comparisonoftheobserveddatawiththePDFusedispresentedinFigure2a.TheWeibulldistributionbestfitthe
behaviorofthewindspeedsforthisstation.However,thePDFhasanexponentialshapebecausetheformparameter
wasequalto1.ItshouldbetakenintoaccountthatWeibullfunctiondoesnotrepresentallwindnatural
distributions,e.g.winddistributionswithsignificancelowwinds,whicharerepresentedascerowindspeedinthe
Weibullfunction.BecauseinFMlowwindspeedsarepredominant,Weibullfunctiondoesnotrepresentthewind
distributioninthatplace.

GirardotaColegioColombia(GCC)

Thisstationhadanaveragewindspeedof1.2854m/s.TheLognormaldistributionfitthewindspeedbehaviorforthis
stationbest.Thehistogramoftheobserveddatapresentedasinglepeaktherefore,thegoodnessoffittest

http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 9/15
22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA

calculatedfortheLognormaldistributionwasbetterthanthetestcalculatedfortheWeibulldistribution.Figure2b
showsthecomparisonofthefourdistributionsandthehistogramoftheobserveddataforGCC.

PolitcnicoJaimeIsazaCadavid(PJIC)

TheresultsofthegoodnessoffittestsinTable3showthatthedistributionthatmostcloselyfitthewindspeeddata
recordedatthisstationwastheGammadistributionfollowedbytheLognormalandWeibulldistributions.The
comparisonsbetweenthePDFsandthehistogramoftheobserveddataforPJICarepresentedinFigure2c.

UniversidadCES(UC)andDitaires(DI)

Theobserveddataforwindspeeddistributionsatthesetwostationsweresimilar,asshowninFigure2dandFigure2e.
Forbothstations,theLognormaldistributionbestfittheobserveddatafollowedbytheGamma,WeibullandRayleigh
distributions.

SantaElena(SE)

ThisstationwaslocatedinaruralsectorclosetotheAburrValleyandhadthehighestelevationofthesixmonitoring
stations.ThisdifferenceisreflectedinTable1bythefactthattheaveragewindspeedrecordedatSantaElenastation
wasatleast6.4timesgreaterthanthatrecordedattheDitairesurbanstationwiththehighestaveragewindspeed.
ThewindspeeddistributionatSEwasfitonlybytheWeibullandRayleighdistributions.Weibullwasthedistribution
thatbestfitthedataaccordingtotheresultsshowninTable3.ThecomparisonsbetweentheobserveddataforSE
andthedistributionsthatbestfitthemareshowninFigure2f.

InTable4theselectedPDFs,andtheirparameters,fortheassessedplacesarepresented.Inordertoknowthe
specificplacethatisrepresentedbytheselecteddistribution,thelatitudeandlongitudeisprovided.Oncea
distributionisselected,anditsparametersestimatedforaspecificplace,thewindpowerdensitycanbecalculated.
However,torecommendawindturbineinstallationplaceandtakingintoaccountsomegoodnessoffittestwereno
closetoacceptablevalues(e.g.theR2forFM),itisnecessarytoconfirmthePDFrepresentationthroughthe
comparisonbetweenthewindpowerdensitycalculatedwithPDFandwithatimeseriesofwindspeeddata.

AssessmentoftheWindPowerattheSixMonitoringStations

Thewindpoweratthesixmonitoringstationswasassessedthroughthecalculationoftheirpowerdensitybytwo
methodsateachlocation.InthefirstmethodthepowerdensitywascalculatedusingEquation20andthewindspeed
datasetrecordedduringaperiodofapproximatelyoneyear.Figure3illustratesthewindbehaviorintheAburrValley
basedonaseriesofwindspeeddatarecordedatthePJICstationduringfourweeks.Table5presentsthepower

densityvalues [Wm2]calculatedbyEquation20forallstations.Thesevaluesareusedasreferencesvaluesto
validatethewindpowerdensitycalculatedfromthePDFsselectedinTable4.

http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 10/15
22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA

Inthesecondmethod,thepowerdensityP PDF[Wm2]iscalculatedusingEquation19andthemaximumandminimum
windspeedvaluesattheevaluatedstation.ThewindpowerdensitiesaregiveninTable5.Therelativepercentage
errorbetweenthepowerdensitiesinagivenlocationwascalculatedusingEquation21.Therelativeerrorisusefulin
ordertoknowiftheselectedPDFisenoughrepresentationofthewindbehaviorinthespecificplace.Forexample,for
GCClocation:theR2(0.9628)andtheIndex_A(0.9924)reportvaluesneartoone,whichcreatethefalseideathat
Lognormaldistributionisenoughrepresentationfortheassessedplace.However,therelativeerror18.68%indicates
thatotherdistributionshouldbeevaluatedtoobtainabetterrepresentationofthewindbehaviorinGCClocation.

Table5showsthatwindpowerdensitiescalculatedfromthePDFforthePJIC,UC,DIandSEstationshaverelative
errorslessthan9%.TheLognormaldistributionbestfitthedatafortwoofthosestations.ThepowerdensitiesforFM
andGCCstationshave19.1%and18.68%relativeerrorsrespectively.Thelattertwoerrorvaluessuggestthat
differentPDFshouldbeadjustedforFMandGCC,inordertoavoidoverorunderestimationofthewindenergyin
thoseplaces.

ThefiveurbanstationsintheAburrValleyreportedaveragepowerdensitiesbetween1.38and4.54W/m2,andthe
oneruralstationreportedavalueof911.1W/m2.Inagreementwiththecommerciallyinternationalsystemof
classificationbyElliottandSchwartz(1993),theurbanlocationscorrespondtothewindpowerclass1,sincethe
densityvalueislowerthan100W/m2.Therefore,thefiveurbanlocationspresentwindcharacteristicsthatarenot
suitabletogridconnectedbigscaleapplications.Thepowerdensitylevelsreportedintheurbanlocationsare
appropriatetoelectricalandmechanicalapplicationsthatarenotconnectedtothepowergrid,e.g.batterycharging
andwaterpumping.

SantaElenalocationcorrespondstothewindpowerclass7.Itisgenerallyacceptedthatwindpowerclass4and
highertypesaresuitabletobigscaleelectricitygenerationusingwindturbinemoderntechnology,sincethehighpower
windgeneratorsincombinationwithahighwindresourcearecompetitiveincostincontrastwithotherelectricpower
sources.

4.CONCLUSIONS
InordertoevaluatetheavailablewindpowerinfiveurbanplacesandoneruraloftheAburrValleyin
Antioquia,Colombia,thispaperhaspresentedthePDFsandtheirparameters,whichtogether,modelthe
windspeedbehaviorinthoselocations.ThePDFswereusedtocalculatethewindpowerdensityforeach
location,whichareusedtorecommendseveralapplicationsthatcouldtakeadvantageofsuchpotential
powers.TheobservedwindbehaviorisrepresentedbytheWeibullPDFattwoofthelocations,bythe
LognormalPDFatthreelocationsandbytheGammaPDFatonelocation.Thepowerdensitiescalculated
withthefunctionsthatbestfitthewindhistogramshavethelowestrelativeerrorwithrespecttothe
powerdensitiescalculatedfromtheobserveddata.Fromtheresultsofthegoodnessoffittestsandthe
comparisonofthewindpowerdensities,itisconcludedthatthewindbehaviorsofPJIC,UC,DIandSEare
representedbythedistributionsselectedinthiswork.BecausecalmwindispredominantinFM,noneof
theproposedPDFisacloserepresentation.Therefore,differentPDFfunctionsareneededtorepresentthe
windspeedbehaviorinFM.

Theaveragepowerdensitiesobtainedarebetween1.38and4.54W/m2forurbanlocationsand911.1
W/m2fortheruralone.Theurbanlocationsaresuitableforelectricandmechanicalapplicationssuchas
batterychargingandwaterpumping.ForSantaElena,windturbinescouldbeinstalledtoconvertthe
availablewindpower.Also,thoseturbinescanbeconnectedtothepowergrid.Finally,thispaperopens
severalworklines.Oneofthemisthenecessitytomonitoringthewindspeedinseveralcitiesof
Colombiainordertoknowitswindpotentialandtakeadvantageofit.

ACKNOWLEDGEMENTS

TheauthorsacknowledgetheAirQualityLaboratoryatUniversidadNacionaldeColombiaforprovidingthewindspeed
data,andtheDepartamentoAdministrativodeCienciaTecnologaeInnovacinCOLCIENCIASforthedoctoral

http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 11/15
22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA
scholarship0952005.ThisstudywassupportedbytheresearchprojectMICRORED18687,whichwasfinancedby
UniversidadNacionaldeColombia.

REFERENCES
Acosta,J.L.&Djokic,S.Z.(2010).AssessmentofrenewablewindresourcesinUKurbanareas.15thIEEE
MediterraneanElectrotechnicalConference,Malta.[Links]

Akdag,S.A.&Gler,.(2010).Evaluationofwindenergyinvestmentinterestandelectricitygenerationcostanalysis
forTurkey.Appl.Energy,87(8),25742580.[Links]

Akpinar,S.A.&Akpinar,E.K.(2007).WindenergyanalysisbasedonMaximumEntropyPrinciple(MEP)type
distributionfunction.EnergyConvers.Manag.,48(4),11401149.[Links]

Carta,J.A.&Ramrez,P.(2007).Useoffinitemixturedistributionmodelsintheanalysisofwindenergyinthe
CanarianArchipelago.EnergyConvers.Manag.,48(1),281291.[Links]

Celik,A.N.(2004).AstatisticalanalysisofwindpowerdensitybasedontheWeibullandRayleighmodelsatthe
southernregionofTurkey.Renew.Energy,29(4),593604.[Links]

Chang,T.P.(2011).Estimationofwindenergypotentialusingdifferentprobabilitydensityfunctions.Appl.Energy,
88(5),18481856.[Links]

Dincer,F.(2011).Theanalysisonwindenergyelectricitygenerationstatus,potentialandpoliciesintheworld.Renew.
Sustain.EnergyRev.,15(9),51355142.[Links]

Elliott,D.L.&Schwartz,M.N.(1993).WindenergypotentialintheUnitedStates.InternationalAcademyofScience
ProjectEnergy,KansasCity,USA.[Links]

Fyrippis,I.,Axaopoulos,P.J.&Panayiotou,G.(2010).WindenergypotentialassessmentinNaxosIsland,Greece.
Appl.Energy,87(2),577586.[Links]

Georgakis,C.&Santamouris,M.(2008).Ontheestimationofwindspeedinurbancanyonsforventilationpurposes
Part1:Couplingbetweentheundisturbedwindspeedandthecanyonwind.Build.Environ.,43(8),14041410.
[Links]

Hernndez,Q.,Espinosa,F.,Saldaa,R.&Rivera,C.(2012).Evaluacindelpotencialelicoparalageneracinde
energaelctricaenelestadodeVeracruz,Mxico.Dyna,171:215221.[Links]

Johnson,G.L.(1985).Windenergysystems.EnglewoodCliffs:PrenticeHall.[Links]

Li,D.,Wang,S.&Yuan,P.(2010).Areviewofmicrowindturbinesinthebuiltenvironment.PowerandEnergy
EngineeringConference(APPEEC),Chengdu,Asia.[Links]

Li,M.&Li,X.(2005a).InvestigationofwindcharacteristicsandassessmentofwindenergypotentialforWaterloo
region,Canada.EnergyConvers.Manag.,46(18),30143033.[Links]

Li,M.&Li,X.(2005b).MEPtypedistributionfunction:abetteralternativetoWeibullfunctionforwindspeed
distributions.Renew.Energy,30(8),12211240.[Links]

LoBrano,V.,Orioli,A.,Ciulla,G.&Culotta,S.(2011).Qualityofwindspeedfittingdistributionsfortheurbanareaof
Palermo,Italy.Renew.Energy,36(3),10261039.[Links]

Massey,Jr,F.J.(1951).TheKolmogorovSmirnovtestforgoodnessoffit.J.Amer.Statis.Assoc.,46(253),6878.
[Links]

Picard,A.,Davis,R.,Glser,M.,&Fujii,K.(2008).Revisedformulaforthedensityofmoistair(CIPM2007).
Metrologia,45(2),149155.[Links]

Ramrez,P.,&Carta,J.A.(2005).Influenceofthedatasamplingintervalintheestimationoftheparametersofthe
Weibullwindspeedprobabilitydensitydistribution:acasestudy.EnergyConvers.Manag.,46(1516),24192438.
[Links]

Safari,B.(2011).ModelingwindspeedandwindpowerdistributionsinRwanda.Renew.Sustain.EnergyRev.,15(2),
925935.[Links]

Safari,B.&Gasore,J.(2010).Astatisticalinvestigationofwindcharacteristicsandwindenergypotentialbasedon
theWeibullandRayleighmodelsinRwanda.Renew.Energy,35(12),28742880.[Links]

Salameh,Z.&Nandu,C.V.(2010).Overviewofbuildingintegratedwindenergyconversionsystems.IEEEPowerand
EnergySocietyGeneralMeeting,Minneapolis,USA.[Links]

http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 12/15
22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA

Seguro,J.&Lambert,T.(2000).ModernestimationoftheparametersoftheWeibullwindspeeddistributionforwind
energyanalysis.J.WindEng.Ind.Aerodyn.,85(1),7584.[Links]

Zhou,J.,Erdem,E.,Li,G.,&Shi,J.(2010).Comprehensiveevaluationofwindspeeddistributionmodels:Acasestudy
forNorthDakotasites.EnergyConvers.Manag.,51(7),14491458.[Links]

AUTHORS

PaulaAndreaAmayaMartnez
Affiliation:UniversidadNacionaldeColombia
IndustrialEngineering,UniversidadNacionaldeColombia
email:paamayam@unal.edu.co

AndrsJulinSaavedraMontes
Affiliation:UniversidadNacionaldeColombia
ElectricalEngineering,UniversidaddelValle
M.Sc.inSystemsPowerGeneration,UniversidaddelValle
Ph.D.inElectricalEngineering,UniversidaddelValle
email:ajsaaved@unal.edu.co

ElianaIsabelArangoZuluaga
Affiliation:UniversidadNacionaldeColombia
ElectronicEngineering,UniversidaddeAntioquia
Ph.D.inElectronicEngineering,UniversitatRoviraIVirgili
email:eiarangoz@unal.edu.co

http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 13/15
22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA

ColombiaSantanderBucaramanga
Km7viaBucaramangaPiedecuesta
BucaramangaColombia
Tel.:05776847350
Fax:05776847444

http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 14/15
22/6/2017 ASTATISTICALANALYSISOFWINDSPEEDDISTRIBUTIONMODELSINTHEABURRVALLEY,COLOMBIA


ctyf@ecopetrol.com.co

http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S012253832014000200007 15/15

Potrebbero piacerti anche