Arent: PNNL CSP Energy Policy Paper

14 Pages·2010·1.28 MB·English

by  Yabei Zhang

EnergyPolicy38(2010)7884–7897
ContentslistsavailableatScienceDirect
Energy Policy
journal homepage: www.elsevier.com/locate/enpol
$
Modeling the potential for thermal concentrating solar power technologies
Yabei Zhanga, Steven J. Smithb,n, G. Page Kyleb, Paul W. Stackhouse Jr.c
aUniversityofMaryland,JointGlobalChangeResearchInstituteandDepartmentofAgriculturalandResourceEconomics,SymonsHall,Room2200,CollegePark,MD20742,USA
bPacificNorthwestNationalLaboratory,JointGlobalChangeResearchInstitute,5825UniversityResearchCourt,Suite3500,CollegePark,MD20740,USA
cNASALangleyResearchCenter,21LangleyBoulevard,MS420,Hampton,VA23681,USA
a r t i c l e i n f o a b s t r a c t
Articlehistory: In this paper we explore the tradeoffs between thermal storage capacity, cost, and other system
Received22December2009 parametersinordertoexaminepossibleevolutionarypathwaysforthermalconcentratingsolarpower
Accepted3September2010 (CSP)technologies.ArepresentationofCSPperformancethatissuitableforincorporationintoeconomic
Availableonline28September2010
modeling tools is developed. We also combined existing data in order to estimate the global solar
Keywords: resource characteristics needed for analysis of CSP technologies. We find that, as the fraction of
Solar electricity supplied by CSP technologies grows, the application of thermal CSP technologies might
CSP progressfromcurrenthybridplants,toplantswithamodestamountofthermalstorage,andpotentially
Thermalstorage eventoplantswithsufficientthermalstoragetoprovidebaseloadgenerationcapacity.Theregional
and global potential of thermal CSP technologies was then examined using the GCAM long-term
integratedassessmentmodel.
&2010ElsevierLtd.Allrightsreserved.
1. Introduction CSP plants either need backup auxiliary generation or storage
capacity to maintain electricity supply when sunlight is low or not
Concentratingthermalsolarpower(hereafterCSP)technology available. All existing commercially operated CSP plants are hybrid
isapotentiallycompetitivepowergenerationoption,particularly plants (Kearney and Price, 2004). They generally either have a
in arid regions where direct sunlight is abundant (Emerging natural-gas-firedboilerthatcangeneratestreamtoruntheturbine,
EnergyResearch(EER),2006).TheroleofCSPwillbedetermined or an auxiliary natural-gas-fired heater for the solar field fluid
by its cost relative to other electric generation technologies (Kearney and Price, 2004; National Renewable Energy Laboratory
(NationalRenewableEnergyLaboratory(NREL),2007a).Although (NREL),2005),althoughotherfuelssuchasbiomasshavealsobeen
previous studies have projected future CSP costs based on used.ThishybridstructureisanattractivefeatureofCSPcomparedto
assumptions for technology advancement and the effect of othersolartechnologiesbecausetheauxiliarybackupcomponenthas
economies of scale and learning curves, few studies have a low capital cost and can mitigate intermittency issues to ensure
consideredthecombinedeffectsofintermittency,solarirradiance systemreliability.Theadditionofthermalstoragewouldallowbetter
changesbyseason,anddiurnalandseasonalsystemloadchanges useofavailablesolarenergyandwouldfurtherreduceintermittency
(Blairetal.,2006).Becausethegenerationofasolarplantvaries issues and potentially lower overall generation costs. However, the
overadayandbyseason,theinteractionsbetweenCSPgenerators cost-effectiveness of adding solar storage depends on the tradeoffs
andothergeneratorsintheelectricsystemcanplayanimportant betweenstoragecapacity,costandotherCSPsystemparameters.
roleindeterminingcosts.Ineffect,CSPelectricitygenerationcost CSPtechnologieshavebeenexaminedpreviously.Perhapsthe
willdependonCSPmarketpenetration. mostdetailedlong-termanalysisofthistechnologyisthatofBlair
etal.(2006)whouseda spatiallyexplicitmodelto examinethe
potential penetration of CSP in the United States. A number of
technologiescompeteonaneconomicbasistosupplyelectricity
$This manuscript has been authored by Battelle Memorial Institute, Pacific demand. Their model uses CSP capacity factors for each solar
Northwest Division, under Contract no. DE-AC05-76RL01830 with the US resourceclassandtimeslicethatareconstantovertime.Hybrid
DepartmentofEnergy.TheUnitedStatesGovernmentretainsandthepublisher,
backupoperationonlowirradiancedaysdoesnotappeartohave
by accepting the article for publication, acknowledges that the United States
Governmentretainsanon-exclusive,paid-up,irrevocable,world-widelicenseto been explicitly considered. Fthenakisa et al. (2009) have con-
publishorreproducethepublishedformofthismanuscript,orallowotherstodo sideredacombinationofsolartechnologiesaspartofananalysis
so,forUnitedStatesGovernmentpurposes. of the feasibility of supplying a large fraction of US energy
nCorrespondingauthorat.Currentaddress:nowattheWorldBank.
demand using solar energy. CSP is included in this analysis,
Tel.:[phone];fax:[phone].
although it appears that fixed capacity factors may have been
E-mailaddresses:[email protected](Y.Zhang),[email protected]
(S.J.Smith),[email protected](P.W.StackhouseJr.). usedtomodelCSPoutput.CSPthermaltechnologiescanalsobe
0301-4215/$-seefrontmatter&2010ElsevierLtd.Allrightsreserved.
doi:10.1016/j.enpol.2010.09.008
Y.Zhangetal./EnergyPolicy38(2010)7884–7897 7885
usedforprocessheat,butthispotentialisnotconsideredfurther three technologies. The parabolic trough is currently the most
inthispaper(TriebandMu¨ller-Steinhagen,2008). matureandcommerciallyavailableCSPtechnology(EERE,1997;
The interaction between CSP generation and the rest of the SargentandLundy,2003).Thistechnologyusesparabolictrough
electric system along with changes in seasonal and daily shapedmirrorreflectorstofocusthesun’sdirectbeamradiation
irradiance make analysis of this technology challenging. This on a linear receiver located at the focus of the parabola. A heat
paper examines these relationships and develops a new metho- transfer fluid circulates through the receiver and returns to a
dologyforrepresentingthermalCSPtechnologiesthatcanbeused series of heat exchangers in the power block where the fluid is
in aggregate economic models. Most current analyses of this used to generate high-pressure superheated steam. The super-
technology are static, in that economic, and often technology, heatedsteamisthenfedtoaconventionalreheatsteamturbine/
assumptions are constant over time and penetration level, generator to produce electricity (EERE, 1997). We focus on this
althoughseasonaleffectsareoftenconsidered.Therepresentation technologyanditscharacteristicsinthispaper,althoughmostof
developedhereallowsadynamicrepresentationofthistechnol- ourinsightswouldalsoapplytopowertowertechnologies.
ogy to be incorporated into long-term economic analysis. Three ForCSPplantswithoutthermalstorage,aminimumirradiance
applications of CSP technologies are examined: (1) CSP as level is required for operation. We have assumed a minimum
intermediate and peak (I&P) load power plants without thermal irradiance level of 300W/m2 (Kearney and Price, 2004). For CSP
storage, (2) CSP as I&P load power plants with thermal storage, plant system efficiencies, we assume constant optical efficiency
and(3)CSPasbaseloadplantswiththermalstorage. (e ),heatcollectorelementthermalloss(Loss )(W/m2),solar
OPT HCE
We begin in Section 2 with a description of our analytic field pipingheat losses (Loss ) (W/m2), turbine gross efficiency
SFP
approach, including the temporal analysis periods and seasonal (e ),andelectricparasiticloss(Loss ),whichmeansthat
turbine parasitic
solar irradiance data used in the analysis. Section 3 provides these parameters do not vary with irradiance level. These
resultsforthethreeCSPapplicationsoutlinedabove.InSection4 parameters are used to formulate the relationships among solar
these results are embedded within a long-term, integrated output,solarfieldsize,anddirectirradiancelevel,asexpressedin
energy-economicmodel,includinganewdatasetofglobal,solar the following equation,1 where Asf is the collecting area of the
resourcedataapplicabletoCSPanalysis,toexaminethepotential solarfield:
roleofCSPtechnologiesintheglobalenergysystem.Weconclude
Output ¼Output ð1(cid:2)Loss Þ
withadiscussion. net gross parasitic
¼e AsfðDNIe (cid:2)Loss (cid:2)Loss Þð1(cid:2)Loss Þ ð1Þ
turbine opt HCE SFP parasitic
2. Methodologyandassumptions
Inthefollowingsections,wewillfirstdetailthemethodologies
2.1. Overview andassumptionsusedforthecasewhereCSPservesasI&Pplants
withoutthermalstorageandthendiscussthecaseswiththermal
storage.
Any long-term analysis of energy systems must strike a
balancebetweentechnologydetailandtractability.Manymodels
thatexaminetimespans25–100yearsintothefutureoperateon 2.3. Costcomponents
anannualaveragebasisoftenin5–10yeartimesteps.Inorderto
facilitatetheinclusionofCSPtechnologieswithinsuchlong-term To calculate the electric generation cost for a CSP plant, we
analysis, the methodology used here first examines CSP plant needtoconsidercapitalcosts,variablecostsofrunningthesolar
operationanditsinteractionwiththeelectricgridusingseasonal component,andvariablecostsofrunningthebackupcomponent.
average irradiance and electric load distributions, split into ten Eachofthesecomponentswillbediscussedbelow.Fundamental
timeslicesdesignedtocapturekeytemporalfeatures.Theresults tothisanalysisaretheloadcurveforelectricitydemandand its
are then aggregated to the annual scale appropriate for use in relationshiptosolarirradiance,whicharedescribednext.
long-term analysis. CSP technologies are considered in two
categories: plants supplying intermediate and peak power, and
2.4. Systemloadcurveandclassificationoftimeslices
thosesupplyingbaseloadpower.
This approach provides a level of detail appropriate for long-
Electric system load, usually measured in megawatts (MW),
term modeling while incorporating key considerations such as
referstotheamountofelectricpowerdeliveredorrequiredatany
changing daily irradiance and load patterns by season. We
specificpoint orpoints on a system. A system load curve shows
considerthesesimplificationsreasonableasappliedtolong-term
theloadforeachtimeperiodconsidered.
analysis exercises. More detailed analysis, such as those per-
Electric system load can be classified as base load and
formedwithhightemporalresolutiondispatchmodels(Lewetal.,
Intermediate and Peak (I&P) load. While these divisions are
2009), would be warranted for shorter-term planning. Such
somewhatarbitrary,thesecategoriesareoftenusedinmodeling
analysis would also, for example, be able to examine in greater
and serve to capture the difference between plants that operate
detail the hourly and daily correlations between load and CSP
aroundtheclockandthosethatoperateonlyduringsomeorallof
plant output, as well as correlations between CSP output and
the day and evening. In principle, base load is the minimum
other renewable resources. These considerations are, however,
amount of power that a utility must make available to its
somewhat less important for CSP plants as compared to, for
customers and I&P load as the demand that exceeds base load.
example PV or wind plants, since the CSP hybrid mode can
Thus base load power plants do not follow the load curve and
providefirmpowerregardlessofsolarirradiancevariations.
generally run at all times except for repairs or scheduled
maintenance. I&P generation must, in aggregate, follow the load
2.2. CSPtechnologies
1ThisfunctionalformisfromtheSolarAdvisorModel(SAM)developedby
Three different types of CSP technologies have been devel-
NREL,inconjunctionwithSandiaNationalLaboratoryandinpartnershipwiththe
oped: (1) parabolic trough, (2) power tower, and (3) parabolic
USDepartmentofEnergy(reference:personalcommunicationwithSAMsupport
dish.Thereexistsignificantdesignandcostvariationsamongthe staff).
7886 Y.Zhangetal./EnergyPolicy38(2010)7884–7897
curve.Forpurposesofthecalculationdescribedherebaseloadis to operate the CSP plant, which we take to be days with irra-
consideredtobeaconstantdemandduringnight-timehoursnot diancebelow3000Wh/m2/day(oranaverageof300W/m2overa
otherwisedetailedbelow(Table1). 10-hour solar day, Kearney and Price, 2004). We then use this
Becausethe system load curve and CSP electricity outputare information to adjust NREL’s annual direct normal irradiance
correlated and both are sensitive to time of day and season, we (DNI)hourlymeandatatoobtainanestimateofhourlymeanDNI
divide the year into specific time slices and then use these to value for each month for days during which the CSP plant is
depictthesystemloadcurve.Wefirstdefinethreeseasonsbased operational (Zhang and Smith, 2008). Fig. 1 shows the hourly
onirradiancelevelsasfollows:(1)summer,thethreecontiguous mean DNI for CSP operational days in Daggett Barstow by
months with the highest irradiance level; (2) winter, the three season. The annual average daily DNI for operational days is
contiguous months with the lowest irradiance level; and (3) 7.75kWh/m2/day.
spring/fall.TheclassificationoftimeslicesispresentedinTable1. Because the CSP solar output profile closely follows solar
The definition of the time slices are chosen for computational irradiance, we approximateCSP outputas solar irradiancetimes
convenience to resolve both the load curve and solar irradiance an average net conversion efficiency as described in Eq. (1). For
curve,althoughtheexactdefinitionsarenotcriticalotherthana simplicity, we idealize daily solar irradiance as an isosceles
requirement that the summer peak should be identifiable trapezoidsymmetricaboutsolarnoon,asillustratedinFig.2.The
as this is a key time period. The assumed system load curve, average daily irradiance, in kWh/m2/day, is the area of ABFE in
denoted as AveSysLoadi, is estimated using data from California Fig.2.
(CaliforniaEnergyCommission(CEC),2003)andisshownlaterin Givenaveragedailyirradiance,noonhours,anddaylighthours,
Table3. we can obtain the maximum irradiance level. We use solar
irradiancedatadescribedaboveandobtaintheleast-squaresfitto
the hourly mean DNI data by varying noon hours. Once these
2.5. SolarirradiancedataandCSPsolaroutput
parameters are determined, we can calculate the daily CSP
operationaltimedenotedasHour ,thelineCDinFig.2,which
CSP solar output depends directly on solar irradiance levels CSP
isdefinedbytheminimumoperatingirradiancefortheCSPplant
(duration and intensity),solar field size, and system efficiencies.
(MinIrradiance),assumedtobe300W/m2.
Solar irradiance data is from the National Solar Radiation Data
Base (NSRDB; National Renewable Energy Laboratory (NREL),
2007b) and we use Daggett Barstow, California, as an example
locationtoillustrateouranalysis.WeusetheNRELsolardatato Irradiance (kw/m2)
calculatelowDNIDays, whichrepresentsthenumberof days ina
season during which there is not sufficient direct sunlight
Table1 Hournoon
Classificationoftimeslices. A B
MaxIrradiance
Slicei Classificationoftimeslices
1 Summermorning(5:00–5:30)
2 Summerdaytime1(5:30–9:00) D Hour C
3 Summerdaytime2(9:00–14:00) MinIrradiance CSP
K
4 Summerpeak(14:00–17:00)
5 Summerevening(17:00–24:00) E Hourdaylight F Time of the day
6 Wintermorning(6:00–10:00)
7 Winterdaytime(10:00–17:30) Solar noon
8 Winterevening(17:30–23:00)
9 Spring/falldaytime(5:00–19:30) Fig.2. SimplifiedanalyticCSPsolaroutputprofilebytimeoftheday(nottoscale).
10 Spring/fallevening(19:30–22:00) Thetrapezoid is defined by the maximum irradiance during the day (MaxIrra-
diance),daylighthours(Hourdaylight)andtheparameternoonhours(Hournoon).
Fig.1. DaggettBarstow(lat(N)34.87,long(W)116.78)hourlymeanDNIforCSPoperationaldaysbyseason,2005.
Y.Zhangetal./EnergyPolicy38(2010)7884–7897 7887
Table2 Table3
Example calculation of key solar geometry parameters by season for Daggett CSPsolaroutputcapabilityandoperationalhoursbytimeslice:DaggettBarstow
Barstow. withasolarmultipleof1.07.
Averagebyseasons Summer Winter Spring/fall Timeslices Ri(%) Houri AveSysLoadi (%)
CSP CSP
Hournoon(hour) 9.6 6.6 7.4 Summermorning(5:00–5:30) 0 0.0 45.5
Hourdaylight(hour) 14.0 10.6 12.3 Summerdaytime1(5:30–9:00) 93 3.2 59.2
DailyIrradiance(kWh/m2/day) 9.2 5.9 8.0 Summerdaytime2(9:00–14:00) 107 5.0 85.7
MinIrradiance(kW/m2) 0.3 0.3 0.3 Summerpeak(14:00–17:00) 106 3.0 96.8
MaxIrradiance(kW/m2) 0.8 0.7 0.8 Summerevening(17:00–24:00) 94 1.2 99.4
HourCSP(hour) 12.3 8.8 10.5 Wintermorning(6:00–10:00) 78 1.9 68.8
lowDNIDays 2 21 15 Winterdaytime(10:00–17:30) 89 6.9 63.5
Winterevening(17:30–23:00) 0 0.0 67.7
Spring/falldaytime(5:00–19:30) 100 10.5 64.3
Table2providesanexamplecalculationofkeysolargeometry Spring/fallevening(19:30–22:00) 0 0.0 59.4
parameters by season for Daggett Barstow. We have implicitly
assumed that the variance in solar radiation over time can be
Wecan definea convenientindicator of CSP solar generation
adequately described by the combination of seasonal irradiance
capability by time slice as Ri, the ratio of average hourly CSP
curves for CSP operational days and the lowDNIDays parameter.
output for time slice i over the CSP capacity (denoted as
We do not consider shorter-term variations in output, thereby
Capacity ) as shown in the following equation. Capacity is
implicitly assuming that these can either be absorbed as part of CSP CSP
themaximumoutputfromtheCSPgenerator
grid operation or offset with operation of the hybrid backup
capacity of the CSP plant. Further, when backup operation is HourlyCSPOutputi
Ri¼ solar 8i: ð4Þ
estimated we assume non-operational days with low irradiance Capacity
CSP
(lowDNIDays) are equally distributed across the seasons (see
Section 2.7). This assumption could easily be relaxed with
Table3presentstheratioRiandCSPoperationalhoursforeach
improveddataonsolarirradiance(seeSection4.2).
time slice using Daggett Barstow data, together with the assumed
averagesystemloadasapercentageofthemaximumsystemload
2.6. ElectricityoutputfromtheCSPsolarcomponent for each CSP operational time slice (denoted as AveSysLoadi ) for
CSP
comparison. Although the solar output mostly overlaps with the
ElectricityoutputfromtheCSPsolarcomponentisdetermined systemdemand,thecorrelationisnotperfect.The highestthreeR
not only by the irradiance level, but also by CSP market ratiosoccurduringsummerdaytime2,summerpeak,andspring/fall
penetration. We define CSP market penetration as the ratio of daytime while the three highest system load periods are summer
totalCSPoutputwithinagivenregiontothetotaloutputfromall evening, summer peak, and summer daytime 2. While there is a
I&P load plants. CSP output includes output from both the solar substantialoverlapbetweenpeakdemandandpeakCSPoutput,the
component (denoted as CSPOutputsolar) and the backup system mismatchthatdoesexistimpactsCSPcostsasshownbelow.
(denotedasCSPOutputbackup),or As part of this calculation we determine the optimal solar
multiple by minimizing the levelized cost of generation (see
CSPOutput¼CSPOutput þCSPOutput ð2Þ
backup solar
below).Thesolarmultipleistheratioofthesolarenergycollected
at the design point to the amount of solar energy required to
To calculate the actual CSP solar output (CSPOutputsolar), we operate the turbine at its rated gross power (Kearney and Price,
first determine the potential CSP solar output (denoted as 2004).Ahighersolarmultipleincreasestheplantcapacityfactor
PotCSPOutputsolar). Potential CSP solar output depends only on butalsoincreasescapitalcost.
the solar resource and system efficiency, while the actual CSP Notethattheoptimalsolarmultipleinthiscaseis1.07,thusthe
solar output is also determined by system demand. Because the Rratiocanbegreaterthan100%forcertaintimeslices.Theactual
potentialsolaroutputcanexceedtheI&Ploaddemandforcertain outputcannotbegreaterthantheratedcapacity,whichmeansthe
periods, without thermal storage this excess solar output is excesssolaroutputgreaterthantheratedcapacitywillbewastedif
assumed to be dumped due to the inflexibility of conventional thereisnothermalstorage.Theoptimalsolarmultiplefoundhere
base load generators. A similar issue with respect to the large- issmallerthanthatinsomeotherstudies(KearneyandPrice,2004;
scale deployment of PV is discussed in detail in Denholm and Price, 2003) due to our assumption of optimal system operation
Margolis(2007),whonotethattheabilitytopartiallyde-ratebase insteadofafixedseasonalCSPoutputprofile.
load generation could allow a larger contribution from inter- TocalculateactualCSPsolaroutput,we need toconsiderthe
mittentrenewables. interaction between the CSP plant and the rest of the system.
PotentialCSPoutputistheareaofABCDinFig.2timesthenet SinceCSPplantsinthiscasearedefinedasI&Ploadpowerplants,
system conversion efficiency. In order to calculate the potential wedeterminetheI&Ploaddemandforeachtimeslice(denotedas
CSP output for each time slice (indicated by superscript i), we EDemandi )astheaverageI&Ploadtimesthenumberofhoursin
I&P
calculate the average hourly CSP output (denoted as agiventimeslice,asdescribedinthefollowing:
HourlyCSPOutputi ) and the CSP operational time (denoted as
Houri )foreachsotlaimr eslice,asshowninthefollowingequation: EDemandiI&P¼ðAveSysLoadi(cid:2)GenerationbaseÞHouri 8i: ð5Þ
CSP
PotCSPOuputi ¼HourlyCSPOutputi Houri 8i, ð3Þ
solar solar CSP Similarly, we calculate the I&P load demand for each CSP
where the hourly CSP output is solar irradiance for that hour operational time slice (denoted EDemandCSPi ¼EDemandi
I&P I&P
times the CSP system electric generation efficiency and i ðHouri =HouriÞ),whichisthedemandduringthefractionofeach
CSP
represents the time slice (Table 1). We have defined the time timeslicethattheCSPplantisoperational.Becausesupplymust
slices in such a way that for certain time slices CSP will not be always equal demand, EDemandCSPi is also the maximum
I&P
operational for this location. These times include summer output that CSP can produce for each CSP operational time slice
morning,winterevening,andspringandfallevening. (denoted as MaxCSPOutputi). Any additional output that CSP
7888 Y.Zhangetal./EnergyPolicy38(2010)7884–7897
produces will be wasted. Therefore, the actual CSP output from Table4
the solar component for each time slice can be calculated as BaselineassumptionsforcalculatingCSPLEC(in2004$).
follows.Thiscaseassumesnothermalstorage,anassumptionthat
Variables Value
isrelaxedinthefollowingsection:
(PotCSPOutputi ifPotCSPOutputirEDemandCSPi Capitalcostperunitofinstalledcapacityassuming1.07 2801
CSPOutputi ¼ solar I&P solarmultiple(c)($/kW)
solar EDemandCSPIi&P otherwise FixedO&Mcost(OMfixed)($/kW-year) 47.87
ð6Þ VariableO&Mcost(OMvariable)($/mWh) 2.72
Priceoffossilfuelnaturalgas(Pricegas)($/MMBtu)(HHV) 4.65
2.7. ElectricityoutputfromtheCSPbackupcomponent Gas-to-electricityconversionefficiency(Efficiencygas–electricity) 0.32
Lifetimeoftheplant(n) 30
Capitalchargerate(CCR)(%) 9.38
Because the net conversion efficiency of gas-to-electricity in a
hybrid CSP plant is lower than the efficiency of stand-alone gas
turbines due to parasitic loads such as heaters and heat transfer consideration of environmental externalities or subsidies. LEC is
fluidpumps(NationalRenewableEnergyLaboratory(NREL),2005; an appropriate general comparison metric with respect to other
LeitnerandOwens,2003),underoptimaloperationoftheelectric generators within the same market segment (e.g. intermediate-
system, CSP backup generation would be used only after stand- peakorbaseload),becausehybridCSPplantsprovidefirmelectric
alonegasturbinesorotheravailablecapacityhasbeendispatched. powerasdoconventionalfossilplants.Thecostoffueltosupply
Here we assume that all other I&P load capacity not otherwise auxiliarypowerisincludedinourdefinitionofLEC.Theeffectof
meeting load demand is available for backup and will be fully tax and other financial incentive policies can be considered by
dispatchedbeforetheCSPbackupmodeisdispatched.CSPbackup choosinganappropriatediscountrate(ZhangandSmith,2008).
generation would, therefore, likely be needed not only when the The cost for CSP generation varies as a function of market
electricityoutputfromtheCSPsolarcomponentislowduetolow penetration. To calculate CSP LEC we calculate backup fuel use
irradiance, but electric demands remain relatively high and non- and total CSP output at each CSP market penetration level and
CSPcapacitycannotmeetdemand,suchassummerevenings. thenusethefollowingformula(EEL,1999)tocalculateCSPLEC.
UnderthisassumptionthecalculationofCSPoutputfromthe The baseline economic assumptions used are shown in Table 4
backupcomponentisstraightforward.Foreachtimeslice,wefirst and are taken from Kearney and Price (2004) and National
findthecorrespondingI&Pload,andthencomparethiswiththe RenewableEnergyLaboratory(NREL)(2005).TheLECis
aggregatedoutputlevelfromtheCSPsolarcomponentandnon-
CCRIþOMþF
CSP plants. If there is a deficit, the CSP backup mode will be LEC¼ ð8Þ
CSPOutput
dispatched to make up thedeficit. The output usingCSP backup
modeistherefore whereCCRisthecapitalchargerate;Iisthecapitalcost;OMisthe
8EDemandi (cid:2)CSPOutputi (cid:2)Capacity Houri, annualoperationandmaintenancecosts,whichcanbecalculated
>>>><ifEDemanI&dPi (cid:2)CSPOutpsuotlair (cid:2) nonCSP as OM¼OMfixedCapactiyCSPþOMvariableCSPOutput; and F is the
CSPOutputbiackup¼>>>>:C0apoatchiteyrnwoniCsIS&ePPHouriZ0 solar faðConsSnsPuiOlaulftupeexulptbenancaksutepu=srEafflofircgifaeusne.cly,Tgwhase-heliecficthxricecitdyaÞn,cwhbaehrecgraeelcPruralitaceetegadcshiaosssteFhne¼pfPorrriicceethgoaesf
ð7Þ calculationsinthissectionareequivalenttoasimplerealinterest
rateof8.6%withacapitallifetimeof30years,avaluesimilarto
thatusedinotherstudiesofrenewableenergy.
In addition, when CSP plants are not operational due to low
Fig. 3shows howCSPLEC changeswithCSP marketpenetra-
DNIdays(lowerthan300W/m2)orhighwinddays(greaterthan
tion for our example location in Barstow. In addition to the
35mph) (Kearney and Price, 2004), the CSP backup mode is
baselinescenario,scenarioswith80%ofbaselinecapitalcostand
neededtoprovideoutputfortheCSPplant.Weassumethatthe
50% of baseline capital cost are presented. Assuming idealized
backupamountforeachtimesliceistheaverageCSPoutputfor
systemdispatch,costsareconstantatlowpenetrationlevels.This
thatseason.ThetotalannualoutputfromtheCSPbackupmodeis
impliesthatsubstantialCSPcapacitycouldbeconstructedbefore
thesumofbackup outputateachtimeslicedue tosolar supply
significant cost increases due to intermittency are encountered.
deficitandthebackupoutputduetolowDNIdays.
Notethat,whileafixednaturalgaspriceisusedinthisillustrative
High wind days are relatively rare in most locations so we
calculation, the cost of auxiliary fuel can be endogenously
ignorethiseffectandonlyconsiderlowDNIeffects.Forexample,
calculatedifthisrepresentationisembeddedinadynamicmodel
for a windy site with an average wind speed of 7.75m/s at a
asdemonstratedbelow.
heightof50meters(windresourceclass5),windspeedsgreater
As CSP generators supply a larger portion of I&P power
than 15.65m/s (35mph) would occur only 4% of the time
demand CSP LEC increases due to two factors: increased wasted
assumingaRayleighdistributionforwindspeed.Theoccurrence
solar output and the costs of purchasing natural gas to fuel
ofhighwindwouldlikelybelowerthanthisatgroundlevel.For
increased backup operation as shown in Fig. 4. The addition of
manylandlocationstheoccurrenceofhighwinddayswouldbe
thermalstoragetothissystemwouldmitigatethislossofoutput
evenlessthanthisvaluesincelandwindspeedstendtopeakat
andalsoallowareductioninoperationofthebackupsystem.This
nightwhentheCSPsolarfieldisnotactive.
will potentially decrease CSP LEC, depending on the cost of the
thermalstoragesystem.ThefollowingtwocaseswillexamineCSP
applicationswiththermalstorage.
3. CSPcostandperformance
3.1. CSPasI&Pplantswithoutthermalstorage 3.2. CSPasI&Pplantswiththermalstorage
ThecentraldeterminantofCSPtechnologydeploymentinthe The addition of thermal storage provides a buffer to smooth
long-term will be electric generation cost. We consider here variable solar output to better match the I&P load curve. While
the levelized energy cost (LEC) of CSP generation without thermalstoragesystemsarenotinoperationaluse,suchsystems
Y.Zhangetal./EnergyPolicy38(2010)7884–7897 7889
Fig.3. CSPLECasafunctionofCSPmarketpenetration—CSPasI&Ploadpowerplantswithoutthermalstorage.
0.15
0-Hour
0.14
1-Hour
h) 0.13 2-Hour
W 3-Hour
k 0.12 4-Hour
04$/ 0.11 56--HHoouurr
0 0.10
2
C ( 0.09
LE 0.08
P 0.07
S
C 0.06
0.05
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%100%
CSP Market Penetration-Fraction of I&P Generation
Fig.5. CSPLECasafunctionofCSPmarketpenetration—CSPasI&Ploadpower
plantswiththermalstorage(correspondingoptimalsolarmultipleisinTable5).
Fig.4. PercentageofCSPoutputlossoverthetotalCSPoutputandfractionofplant Table5
electricityproducedbythebackupsystemvs.CSPmarketpenetration. Optimalsolarmultiplebythermalstoragehour.
Hoursofstorage 0 1 2 3 4 5 6
are being developed and could be commercially viable in the
future. In terms of methodology, the key addition to the Solarmultiple 1.07 1.15 1.15 1.25 1.35 1.55 1.65
calculations presented above is that for each time slice, thermal
storage can potentially be used to store the solar output that:
exceeds generation capacity due to an oversized solar field,
exceeds the I&P demand at a specific time, or is below CSP addedtotheplantasthenumberofhoursofpowergenerationat
operationalrequirements(irradiancelevellowerthan300W/m2). rated(turbine)capacitythatcanbedrivenbythestoragesystem.
Storedsolaroutputthencanbeusedatanylatertimeslicewhen Fig. 5 shows the results of CSP LEC as a function of CSP market
irradiance is not available or low, particularly in evenings. We penetration by the number of hours of thermal storage. The
assume the order of dispatched capacity is: CSP solar output correspondingoptimalsolarmultipleispresentedinTable5.The
directlyfromsolarfield,heatenergydrawnfromthermalstorage, shape of CSP LEC versus CSP market penetration is primarily
otherI&Pcapacity,andoutputfromtheCSPbackupcomponent. driven by increased proportions of wasted solar output, as
Wefurtherassumethatanyleftoverthermalenergyinstorageat illustrated in Fig. 6. The share of backup operation is relatively
the end of the day will be wasted. While there may be stable in this case due to availability of stored solar output. At
circumstances where some energy that remained stored could penetration levels above 70–80%, depending on the amount of
beusedthefollowingday(ifthestoragetechnologyusedallowed storage,costsbegintoincreaseassomesolaroutputiswastedin
this),thisassumptionhaslittleimpactontheresultssincewasted seasonswheretotalCSPcapacityexceedsI&Pdemand.
thermalenergyissmallforsystemswiththermalstorage. From Fig. 5, we can see that thermal storage extends the
Forthermalstoragecostsweassumeafixedcostof$140/kW market penetration threshold where CSP LEC starts to increase.
andavariablecostof$23/kW-hourstorage,derivedfromNational This is perhaps the largest benefit from the addition of thermal
Renewable Energy Laboratory (NREL) (2005). In addition, we storagesystems.TheadditionofthermalstoragealsoallowsCSP
assume that fixed O&M costs increase to $58/kW-year(National plantstooperateintotheeveningwithoutusingbackupfuel.The
Renewable Energy Laboratory (NREL), 2005) for units with valueofthiscapabilitywilldependontherelativecostofthermal
thermal storage. We denominatethe amountof thermal storage storageandnaturalgas(orbiomass)fuelatanygivenlocation.In
7890 Y.Zhangetal./EnergyPolicy38(2010)7884–7897
Fig.7. CSPLECbythermalstoragehourwithoptimalsolarmultipleand50%of
baselinecapitalcost—CSPasbaseloadpowerplantswiththermalstorage.
numberofstoragehoursorthepriceratio.We,therefore,findthat
Fig.6. PercentageofCSPoutputlossoverthetotalCSPoutputandfractionofplant for base load CSP operation the primary factor controlling
electricityproducedbythebackupsystemvs.CSPmarketpenetration—CSPasI&P competitivenessisCSPcapitalcost.ThenumberoflowDNIdays
loadpowerplantswith6-hourthermalstorage.
is also a factor, particularly if natural gas (or biomass) prices
increaseinthefuture.
addition,thermalstoragecanreduceCSPLEC(Fig.5),becausethe
addition of thermal storage acts to leverage fixed costs (e.g.
4. THEpotentialglobalroleofCSP
generatorandturbine)byincreasingthesolarmultiple.However,
thesebenefitssaturateataboutfivehoursofstorage.
4.1. Long-termmodeling
3.3. CSPasbaseloadplantswiththermalstorage ThepotentialcontributionofCSPelectricgenerationtechnol-
ogies will depend first on the cost of CSP electric generation,
Inthissection,weexaminethecasewhereCSPplantsoperate which has a strong dependence on the quality of the solar
asbaseloadgenerators.WeassumethattheCSPoutputinwinter resource.TheroleofCSPwillalsodependonelectricitydemand,
is maintained year-round as base load generation. Since DNI is the cost of other electric generation technologies, and environ-
higher and daylight hours are longer in summer and spring/fall mentalincentivessuchasapriceoncarbon.Inordertoexamine
than in winter, CSP generates additional I&P electricity in these thesequestions,therepresentationofCSPtechnologiesdescribed
seasonsinadditiontobaseloadelectricity.Therefore,CSPplantsin above was implemented within the GCAM (formerly MiniCAM)
thiscasereceiverevenuesfromservingbothbaseandI&Ploads. integratedassessmentmodel.
Intermsofmethodology,thekeyfeatureofthiscaseisthatin While the results above were derived for one specific location,
winter,thedailysolaroutputissmoothedthroughthermalstorage we can generalize these findings by noting that the cost and
to provide steady base load electricity, while in summer and performance of CSP technologies depends largely on just two
spring/fall,inadditiontoprovidingbaseloadelectricity,theextra location-specific parameters, the number of low DNI days and the
energysuppliedbytheadditionalsolarirradianceavailableinthese averageirradianceonoperationaldays.TheinflectionpointsforCSP
seasonscanbeusedtomeetsomeoftheI&Ploaddemand.Dueto I&Ptechnologieswhereincreasedcostsareincurred(Figs.3and5)
the relatively small share of CSP I&P load supplied from these dodependonthedetailedloadandirradiancecurves.Thishaslittle
plants, no solar output will bewasted due to exceeding I&P load impactonoursimulationresults,however.Thefinalstepbeforewe
demandandtheCSPgas-hybridbackupgenerationisonlyneeded canimplementaglobalanalysis,therefore,istoestimateglobalsolar
duringlowDNIdays.Thus,theimpactofCSPmarketpenetration resourceparametersasdescribedinthenextsection.
onLECisnegligibleforaCSPsystemoperatingasabaseloadplant.
Because base load CSP plants sell electricity to both I&P and
4.2. Globalsolarresourceestimate
baseloadmarkets,insteadofanLECwecalculateandequivalent
quantity: the breakeven base load electricity price at which
Operational day DNI and the average annual number of low
capitalanoperatingcostsoftheCSPplantaremet.Weassumethe
DNI days were estimated using data at a one-degree spatial
ratioofI&Ptobaseloadelectricpricesis2,basedonCAwholesale
resolution from the National Aeronautics and Space Administra-
market data from Energy Information Administration (Energy
tion(NASA)SurfacemeteorologyandSolarEnergydatasetrelease
Information Administration (EIA), 2008). Using this assumption,
6.02 (e.g. Chandler et al., 2004). Neither of these quantities are
the breakeven base load electricity price for CSP is around
available directly from this dataset, so an estimation procedure
$0.065/kWh,which is notcompetitive withcurrent prices. If we
wasusedwherebysolarpropertiesonnon-operationaldayswere
assumecapitalcostscanbereducedto50%ofthebaselinevalues
estimated using correlations between the ratio of clear-sky to
(Table 4), the breakeven base load electricity price is in about
total ground radiation and the necessary parameter in the US
$0.04/kWh, which is competitive. Fig. 7 presents the breakeven
National Solar Radiation Database (NSRDB). The methodology
baseloadpriceforCSPasafunctionofthermalstoragehour.The
usedforthiscalculationisgiveninAppendixA.
sensitivity to the ratio of I&P load over base load price is also
shown. The lowest CSP cost occurs at 10–11 hours of thermal
storage, although CSP LEC is not very sensitive to either the 2Seehttp://eosweb.larc.nasa.gov/sse/.
Y.Zhangetal./EnergyPolicy38(2010)7884–7897 7891
Fig.8. EstimateofthenumberofdayswhereDNIfallsbelow3000W/m2.
Table6
NumberofdayswithDNIo3000W/m2foreachsolarclassbyregion.
Region NumberoflowDNIdays
Class1 Class2 Class3 Class4 Class5 Class6
(o4.5kWh/m2/day) (4.5–5kWh/m2/day) (5–6kWh/m2/day) (6–7kWh/m2/day) (7–7.5kWh/m2/day) (47.5kWh/m2/day)
USA – 187 126 101 71 58
Canada – – 181 – – –
WesternEurope – – 117 103 68 61
Japan – 190 171 124 58 –
AustraliaandNZ 188 181 124 54 40 34
FormerSovietUnion – 186 98 75 151 128
China 174 156 162 89 85 82
MiddleEast – 186 43 31 22 13
Africa 175 107 62 30 18 14
LatinAmerica 154 136 88 45 46 43
SoutheastAsia 166 132 98 48 22 60
EasternEurope – 173 155 107 78 –
Korea – – 138 – – –
India – – 16 53 153 111
TheestimatednumberoflowDNIdaysfortheglobeisshownin the global map. Since regions poleward of 451 are generally not
Fig.8.Thelightshadedregionsinthefigureindicateareaswherethe well suited for CSP power, these areas are excluded from this
numberoflowDNIdaysislessthan100daysperyear,whichisa studyandhavelittleimpactontheresults.
roughindicatorofareaswellsuitedforCSPtechnology.Thenumber Akeyfeatureofthesolarresourceinmanyworldregionsisits
oflowDNIdaysisanimportantindicatorofthequalityofthedirect geographic concentration. In the United States, for example, the
solar resource. The Middle East and North Africa have the highest highest quality solar resource is found in the southwestern
quality resource, where we estimate that there are very few days portionofthecountry(e.g.Blairetal.,2006).Forthecalculations
where a CSP plant cannot operate. Note that these estimates are presented here, we make a conservative assumption where we
averages over relatively large regions. Specific locations in these treat each solar resource class as a separate sub-region for
areascanhavepropertiesthatdifferfromlarge-areaaveragesdueto purposes of estimating CSP technology characteristics and para-
localmeteorologicaleffectsthatimpactcloudcover.Foruseinthe meters such as the fraction of system load supplied by CSP. We
model calculation we aggregated the resource estimates into six make the assumption that each solar resource class is spatially
categoriesbasedontheaverageDNIlevel(Table6).Onlyareaswith continuousenoughthatwecantreatallloadswithinthatregion
lowDNIdays o200wereincluded,sinceareaswherethemajority together. This is generally true for the higher quality solar
of days are cloudy would not be suitable for CSP technology resourceswhereCSPhasthemostpotential.
deployment.Forestedandcroplandareasareexcluded.
Eachofthe14GCAMregionsisdividedtosixgeographicsub-
regions, with the resource classifications and the area of land in 4.3. ThepotentialroleofCSP
each resource class shown in Table 7. The current SSE uses
different irradiance to DNI parameterizations poleward of 451 We now combine the representation of CSP technologies
than between 451N and 451S which produces a discontinuity on developedinthefirstportionofthispaperwiththesolarresource
7892 Y.Zhangetal./EnergyPolicy38(2010)7884–7897
Table7
Areaforeachsolarclassbyregion.
Region Area(km2)
Class1 Class2 Class3 Class4 Class5 Class6
(o4.5kWh/m2/day) (4.5–5kWh/m2/day) (5–6kWh/m2/day) (6–7kWh/m2/day) (7–7.5kWh/m2/day) (47.5kWh/m2/day)
USA – 264 2773 113,541 227,257 168,091
Canada – – 380 – – –
WesternEurope – – 8569 12,655 12,207 10,648
Japan – 1824 850 145 0 –
AustraliaandNZ 369 14 6608 289,338 850,751 3,843,693
FormerSovietUnion – 115 197,911 267,215 17,612 9805
China 943 1014 11,829 726,966 479,398 439,120
MiddleEast – 277 5212 304,777 970,911 1,498,909
Africa 49 8121 362,691 1,455,713 2,080,459 5,225,091
LatinAmerica 956 3431 39,001 77,048 134,637 83,467
SoutheastAsia 15,459 36,239 65,507 160,462 162,515 11,717
EasternEurope – 156 450 93 56 –
Korea – – 1339 – – –
India – – 35,142 51,337 20,002 18,071
Fig.9. FractionofUSelectricloadsuppliedbyCSPtechnologies.
Fig.10. FractionofloadsegmentservedintheUSregionwithhighresourcesuitability
estimatesdevelopedaboveinordertoevaluatethepotentialrole (mostofinlandCalifornia,Nevada,Arizona,NewMexico,andsouthwestUtah).
ofCSPtechnologiesintheUSandglobalenergysystem.TheCSP
representation above was implemented in the GCAM integrated
assessmentmodel.AbriefdescriptionoftheGCAMmodelanda Fig.10showsthedynamicsofCSPinthismodelforthoseUS
description of the implementation of CSP technologies in the regionswiththehighestresourcesuitability(ourclasses5and6,
modelareprovidedinAppendixA. whichareinthewestandsouthwestoftheUS).Notethatthese
Fig.9showsGCAMmodelresultsforCSPmarketpenetration results must be interpreted with some caution since neither
in the US under a reference case scenario derived from Clarke electricity supply and demand within each sub-region nor
etal.(2007).CSPplantcapitalcostswereassumedtodecreaseata separate peak, intermediate, and base load markets were
rate of roughly 0.6% per year. This represents a reference explicitly modeled, although some of these simplifying assump-
technology case with continued incremental reductions, but tions will be addressed in future work. However, the impacts of
without the type of focused research effort that would result in increasingCSPpenetrationdoreflectthedynamicsderivedearlier
morerapidcostdecreases,forexampleasassumedinonerecent in this paper. By mid-century CSP is providing the majority of
analysis(NREL,2007a). intermediate and peak demand in these regions. The CSP
The largest role for CSP for the first half of the century is to contribution peaks at about 70% due to the cost of operating
supplyintermediateandpeakloads.Inthenear-term,CSPplants hybridmodeandlostoutputthatwouldbeincurredbeyondthis
without thermal storage are competitive as their penetration is point.ThefractionofbaseloaddemandsuppliedbyCSPsteadily
not sufficient to suffer penalties due to lost solar output or increasesovertimeascostsfall.After2050mostofthegrowthin
increased backup mode operation. As penetration increases and CSPoutputisduetoadditionalbaseloadgeneration.
I&Pplantswiththermalstoragearefullydeveloped,thesebecome Fig. 11 shows CSP generation by world region while Fig. 12
the preferred option as thermal storage allows these plants to showsCSPgenerationasafractionoftotalelectricgeneration.All
serveahigherfractionofI&Ploadwithoutincreasedoperationof regions show some level of CSP generation. At very low
thebackupmode. penetration levels this is a result of the logit sharing algorithm
As CSP technology costs fall, CSP becomes more competitive usedinthismodel(ClarkeandEdmonds,1993),whereevenhigh
forbaseloadgenerationaswell.ThepotentialforCSPbaseloadis costoptionshavesomemarketshare.TheMiddleEast,India,and
ultimatelylargerduetoalargermarketsegmentandtheabsence Africa have the highest penetration rates, with the higher
ofsignificantpenaltiesathigherpenetrationlevels. penetrationinIndiaisduetosomewhathigherelectricityprices
Y.Zhangetal./EnergyPolicy38(2010)7884–7897 7893
Fig.11. TotalCSPgenerationbyregion.
Fig.12. FractionofregionalelectricloadservedbyCSP.
inthisregionmakingCSPcompetitiveeventhoughlowDNIdays oftotalcost,inregionswithexceptionallygoodresourcessuchas
are relatively large (Fig. 8). A second group of regions with the Middle East, and 7–8% in the US Southwest. Later in the
moderatelylargefractionsofCSPincludeLatinAmerica,Australia century backup costs can increase significantly, often reaching
and New Zealand, South and East Asia and the United States. 20%oftotalcostsintheUSSouthwest.Thisincreaseisduetotwo
Alloftheseregionshaveareasthatcontaingoodsolarresources. factors.First,CSPcapitalcostsareassumedtofallovertimewhile
In all the above cases, an additional assumption of enhanced theneedforbackupoperationdoesnotdecrease.Second,backup
transmission infrastructure (not assumed here) would allow a inthesescenarioswasassumedtobesuppliedbyeithernatural
larger expansion of CSP technologies. The remaining regions gasorbiomassandthecostsofbothofthesefuelsincreaseover
(Former Soviet Union, Europe, Canada, Japan, and Korea) have thecenturyinourreferencecase.Ifamoreaggressivedecreasein
relatively poor resources for CSP, although as indicated in these CSPcapitalcostsovertimewereassumed,thefuelcostsforhybrid
results there are some locations where CSP generation may be backup operation would be an even larger fraction of the total
competitive. generationcostforCSPtechnologies.
CSP supplies 4% of US electric generation in 2050 in this TheCSPtechnologyassumptionsusedhereimplicitlyassume
scenario. For the only comparable published study on this time significant water use for the steam turbines, largely for cooling
scale,thecentralcaseofBlairetal.(2006)resultsinaCSPcapacity (SargentandLundy,2003;DepartmentofEnergy,2008).Restric-
55GWin2050,orabout2.75%oftotalUSgenerationcapacity.On tions on water use in the arid regions that often have the most
an aggregate level, the representation developed here produces appropriate solar resources for CSP would change the net plant
resultsbroadlycomparabletothemorespatiallyandtemporally efficiency and costs assumed here if dry-cooling technologies
detailedrepresentationinBlairetal.(2006).Thecontributionof wereimplemented.Power-towerCSPplants,whichhavealower
CSP technologies is much larger here than the aggregate solar efficiency penalty for dry cooling due to their higher operating
technology included in a previous version of the same model temperature, could become the preferred option in these situa-
(Clarke et al., 2007), where only 1% of US electric demand was tions (Department of Energy, 2008). Hybrid cooling systems
suppliedbysolarin2050. wouldallowevenparabolictroughplantstooperatewithonlya
FuelcostsforhybridbackupoperationonlowDNIdayscanbe fewpercentlossinoveralloutput,albeitwithamodestincrease
asignificantportionoftotalgenerationcosts,particularlylaterin in cost (Department of Energy, 2008). Power tower plants with
thecentury.Inearlytimeperiods,backupcostsareverylow,2–4% much higher operating temperatures might some day allow the