Академический Документы
Профессиональный Документы
Культура Документы
doc
Loadflow
Formulationoftheloadflowproblem. GaussSeidel,NewtonRaphson,andStott'salgorithm.
Calculationoflineflows,systemlosses,andareainterchange.
1
FormulationoftheProblem
Theloadflowproblemisoneoftheclassicpowersystemengineeringproblems. Duringthe
earlydaysofdigitalcomputers,manyadvancesintechniquesforsolvinglargesetsofequations
werebroughtaboutspecificallytohelpsolvetheloadflowproblem.
Inmostelectricalcircuitanalyses,thenetworkconsistsofknownimpedances,voltagesources,
andcurrentsources.However,intheloadflowproblem,activeandreactivepowers,ratherthan
shuntimpedances,arespecifiedatmostnetworkbusses,becausemostloadsbehave,onaverage,
asconstantpowerloads(activeandreactivepower),aslongastheirappliedvoltageremains
withinreasonableranges. Consider,forexample,theairconditioningloadofabuilding. A
certainamountofenergyisrequiredtomaintain T betweeninsideandoutsidetemperatures.
Eventhoughtheairconditionercyclesonandoff,andthevoltagemaychangeslightly,theair
conditioning load appears, on the average, as a fixed power load, rather than as a fixed
impedanceload.
Powersystemloadsarecloselymonitoredatsubstations,atlargecustomers,andfortotalelectric
utilitycompanies.Loadstendtohavepredictabledaily,weekly,andseasonalpatterns.Annual
peak demands and energies for electric utilities are forecasted for generation and planning
purposes.
Mostloadflowsmodelthreephasebalancedsystems. Positivesequencevalues,usuallyinper
unit,representR,L,C,P,Q,S,,V,andI.
The purpose ofthe loadflow programis to compute bus voltages andline/transformer/cable
powerflowsoncenetworktopology,impedances,loads,andgeneratorshavebeenspecified.
Ideally,thecomputedbusvoltagesforthestudysystemshouldremainwithinacceptableranges,
andline/transformer/cablepowerflowsshouldbebelowtheirratedvalues,forareasonableset
ofoutagecontingencies.
Fromaloadflow perspective,therearefourparameters ateverybusvoltagemagnitude V,
voltageangle ,activepowerP,andreactivepowerQ.Twomaybespecified,andtheother
twocalculated.Formostbusses,PandQarespecified,andVand arecalculated.Obviously,
PandQcannotbespecifiedatallbussesbecausethatwouldimplythatsystemlossesareknown
apriori. Therefore,theloadflowproblemmustincludeone"swingbus"atwhichthe P can
assumeanyvaluesothatit"makesup"systemlosses. Theswingbusisusuallyacentrally
located large generator whose voltage magnitude and phase angle (usually = zero) are
specified.
Page1of15
_18_ELC4340_Spring13_Loadflow.doc
Althoughanytwoofthefourparameterscanbespecified,theusualwayinwhichpowersystem
bussesareclassifiedisgiveninTable1.
Table1.LoadflowClassificationofPowerSystemBusses
Classification
Knowns
Unknowns
PQ(LoadBus)
P,Q
V,
PV(GeneratorBus)
P,V
Q,
V (SwingBus)
V,
P,Q
Theloadflowprogramsolvesforthesetofunknownsthatproducespowerbalanceatallbusses,
orasillustratedforbusiinFigure1,
Pispec jQispec Picalc jQicalc ,
where
Picalc jQicalc Vi I i* .
Inotherwords,thepowerspecifiedateachbusmustequalthepowerflowingintothesystem.
NoteinFigure1thatspecifiedpowerisdrawnaspositivegeneration,tobeconsistentwithKCL
equationYV=I.
TotalCurrentFlowingFromBusiintotheSystemis
I=IB1+IB2+IB3
i
IB1
Busi
Branch
Currents IB2
Into
System
IB3
|Vi | i
spec
spec
*
P+jQ
=VI
i
i
i
i
Figure1.PowerBalanceforBusi
Sincetherearetwounknownsateverybus,thesizeoftheloadflowproblemis2N,whereNis
thenumberofbusses.Obviously,tosolvetheproblem,theremustbetwoequationsforevery
bus.ThesecomefromKCL,whichforanybusihavetheform
Pi
spec
jQispec
Picalc
jQicalc
Vi I i*
j 1
Vi
y i , jV j
Page2of15
_18_ELC4340_Spring13_Loadflow.doc
Separatingintorealandimaginarycomponentsyieldstwoequationsforbusi,
N
Pispec
Vi
Qispec
Vi
where Vi
j 1
N
j 1
yi, j V j cos i j i , j
yi , j V j sin i j i , j
Vi i , V j V j j , y i , j y i , j i , j
Theproblemnowisnowtofindthesetofbusvoltagesthatsatisfiestheabove2Nequations.
2.
GaussSeidelMethod
GaussSeidelisanearlyformulationoftheloadflowproblemthatrequireslittlememoryanditis
easilyprogrammed.However,itisusuallyslowerthanothermethods.Itisbasedupontheidea
ofexpandingthecomplexformofthepowerbalanceequationasfollows:
Pi
spec
jQispec
Vi I i*
j 1
Vi
yi, jV j
Vi yi , iVi yi , jV j
j 1, j i
or
Pispec jQispec Vi* yi , iVi Vi*
yi , jV j
j 1, j i
sothat
Vi
spec
N
jQispec
1 Pi
yi , jV j .
y i, i
Vi*
j 1, j i
Thesolutionprocedureisto:
1.
Initialize the bus voltages. For load busses, use V = 1 + j0. For generator busses
(includingtheswingbus),use V Vspec j 0 .
2.
Onebyone,updatetheindividualbusvoltagesusing
Vi
spec
N
jQispec
1 Pi
yi , jV j .
y i, i
Vi*
j 1, j i
ForPVbusses,updatethevoltageangle,whileholdingthevoltagemagnitudeconstantat
thespecifiedvalue.Donotupdatetheswingbus.
Page3of15
_18_ELC4340_Spring13_Loadflow.doc
3.
CheckthemismatchPandQateachbus.Ifallarewithintolerance(typicaltoleranceis
0.00001pu),asolutionhasbeenfound.Otherwise,returntoStep2.
Convergenceisusuallyfasterifanaccelerationfactorisused. Forexample,assumethatthe
voltageatbusiatiterationmis Vim ,andthattheupdatingequationinStep2computes V jnew .
Insteadofusing V jnew directly,acceleratetheupdatewith
whereaccelerationfactor isintherangeof1.2to1.6.
3
NewtonRaphsonMethod
TheNewtonRaphsonmethodisaverypowerfulloadflowsolutiontechniquethatincorporates
firstderivativeinformationwhencomputingvoltageupdates. Normally,only3to5iterations
arerequiredtosolvetheloadflowproblem,regardlessofsystemsize.NewtonRaphsonisthe
mostcommonlyusedloadflowsolutiontechnique.
AneasywaytoillustratetheNewtonRaphsontechniqueistosolveasimpleequationwhose
answersarealreadyknown.Forexample,consider
x 1 x 99 0 ,
whichwhenexpandedbecomes
x 2 100 x 99 0 .
Theobjectiveistofindxsothat
f ( x) x 2 100 x 99 0 .
Ofcourse,inthiscase,thetwosolutionsareknownaprioriasx=1,andx=99.
The NewtonRaphson procedure is based on Taylor's expansion, truncated past the first
derivative,whichgives
f ( x x ) f ( x )
f ( x)
x
x .
Clearly,theaboveequationgivesastraightlineapproximationfor f ( x x ) .
Theobjectiveistofind x sothat f ( x x ) isthedesiredvalue(whichinthisexampleis
zero).Solvingfor x yields
Page4of15
_18_ELC4340_Spring13_Loadflow.doc
x
f ( x x) f ( x)
f ( x )
,
x x
whichforthisexampleis
x
0 f ( x)
f ( x)
f ( x)
f ( x ) .
x x
x x
Theupdateequationforiteration(m+1)isthen
x ( m 1) x ( m) x x ( m)
f ( x (m) )
,
f ( x)
x x ( m)
whereinthisexample
f ( x)
2 x 100 .
x
Ifastartingpointofx=2ischosen,thenthesolutionproceedsasfollows:
Iterationm
f(x)
97
f ( x)
x
96
0.9896
1.0193
98.02
0.9999
0.0098
98.00
Additionaliterationscanbeperformediftightersolutiontoleranceisneeded. Notethatifa
startingpointofx=50hadbeenchosen,thepartialderivativewouldhavebeenzero,andthe
methodwouldhavefailed.
Ifx=80isthestartingpoint,thentheprocessyieldsthefollowing:
Iterationm
f(x)
80
1501
f ( x)
x
60
105.02
626.2
110.04
99.33
32.45
98.66
99.00
98.00
Page5of15
_18_ELC4340_Spring13_Loadflow.doc
Therefore,thestartingpointgreatlyaffectstheabilityofaNewtonRaphsonmethodtoconverge,
andtheanswertowhichitconverges.Fortunately,intheloadflowproblem,mostvoltagesare
near1.0puinmagnitudeand0.0degrees,sothatweareabletoaccuratelyestimatestarting
values.
For the loadflow problem, the NewtonRaphson method is expanded in matrix form. For
example,considerasetofNnonlinearequationsandNunknowns,
f1 x1 , x 2 , , x N
f 2 x1 , x 2 , , x N
f N x1 , x 2 , , x N
y1
y2
.
yN
(0)
( 0)
yi f i x1(0) x1 , x 2(0) x 2 ,, x N
x N f i x1(0) , x 2( 0) , , x N
f i
+ x1 x
1 x ( 0)
x2
f i
x2
x N
x (0)
f i
x N
x ( 0)
y1
y
2
yN
f1
x
1
f 2
x1
f N
x1
f1
f1
x2
x N x1 f x (0) , x (0) ,, x (0)
N
1 1 2
f 2
f 2
(
0
)
(
0
)
x2 f 2 x , x ,, x (0)
N
1
2
x2
x N
,
x
( 0) ( 0)
( 0)
f N
f N N f N x1 , x2 ,, x N
x2
x N
or
Page6of15
_18_ELC4340_Spring13_Loadflow.doc
x
J x ,
y f x ( 0 ) , x ( 0) , , x ( 0)
N
2
1 1 1
( 0) ( 0)
y 2 f 2 x1 , x 2 ,, x (N0)
y N f N x ( 0) , x ( 0) , , x ( 0 )
N
1
2
whereJisanNxNmatrixofpartialderivatives,knownastheJacobianmatrix.Therefore,inan
iterativeprocedure,theaboveequationisusedtoupdatetheXvectoraccordingto
X ( m 1) X ( m) X ( m) X ( m) J ( m)
Y
1
spec
Y ( m)
where
Yspec
y1
y
2
, Y ( m)
f x
f x ( m) , x ( m) , , x ( m)
N
2
1 1
( m) ( m)
f 2 x1 , x2 ,, x N( m)
( m) ( m)
( m)
N 1 , x2 ,, x N
yN
In the loadflow problem, the matrix update equation is symbolically written in mixed
rectangularpolarformas
m)
P
P1(calc
1
spec
m)
P2 spec P2(calc
PNspec P ( m) J1 P J 2 P
Ncalc
J3
J4
Q
Q ( m)
V
1spec 1calc
(
m
)
Q2 spec Q
2calc
( m)
Q
Nspec Q Ncalc
m 1 m
1
1 m 1
2 m
2
N m 1 N m
V1 m 1 V1 m
V2
m 1 V m
m 1
m
VN
V N
or,inabbreviatedform,
P J1
Q J
J2
J 4
Page7of15
_18_ELC4340_Spring13_Loadflow.doc
Thedimensionoftheaboveproblemisactually 2 N Number of PV busses 2 since V
updatesatPVbussesarenotrequired,andsince V and updatesattheswingbusarenot
required.
Inhighlyinductivepowersystems,Piscloselyrelatedtovoltageangles,andQiscloselyrelated
tovoltagemagnitudes.Therefore,intheabovemixedrectangularpolarformulation,thetermsin
J 1 and J 4 tendtohavelargermagnitudesthanthosein J 2 and J 3 . Thisfeaturemakesthe
Jacobian matrix more diagonally dominant, which improves robustness when Gaussian
eliminatingorLUdecomposingJ.
TheaboveformulationoftheJacobianmatrixisoftenmodifiedtotakeadvantageofsymmetry
inthepartialderivatives.Thismodificationis
1
V
P
Q
1
P
V
Q
V
V
.
V
Thepartialderivativesarederivedfrom
N
Picalc
Vi
Qicalc
Vi
j 1
yi , j V j cos i j i , j
j 1
yi, j V j sin i j i , j
andhavethefollowingform
For J1 ,
N
Pi
Vi yi , j V j sin i j i , j
i
j 1, j i
Pi
Vi yi , k Vk sin i k i, k , k i .
k
For J 2 ,
Pi
Vi
yi, j
j 1, j i
V j cos i j i , j 2 Vi yi , i cos i, i
Pi
Vi yi , k cos i k i, k , k i .
Vk
For J 3 ,
Page8of15
_18_ELC4340_Spring13_Loadflow.doc
Qi
Vi
j 1, j i
yi , j V j cos i j i , j
Qi
Vi yi , k Vk cos i k i, k , k i .
k
For J 4 ,
Qi
Vi
yi, j V j
j 1, j i
Qi
Vi y i, k sin i k i, k , k i .
Vk
NotethesymmetryintheJterms.If V
theexpressionsfor J1 are
isusedasanupdatingparameterratherthan ,then
N
Pi
yi , j V j sin i j i , j
Vi i
j 1, j i
Pi
Vi yi, k sin i k i , k , k i .
Vk k
andfor J 3 ,
Qi
Vi i
yi , j V j
j 1, j i
cos i j i , j
Qi
yi , k Vk cos i k i , k , k i .
Vi k
andthereisevenmoresymmetryinJ.
Thediagonaldominanceof J1 and J 4 canbeobservedbytheexaminingthepartialderivatives
asfollows:thedifferencesbetweenvoltageanglesatadjacentbussesisusuallysmall,sothatthe
i j termsaresmall.Theanglesfoundintheadmittancematrixareusuallylarge(i.e.near
90 o )becausemostpowersystemsarereactive.Therefore,thesinetermsinthematrixupdate
equationtendtobenearunity,whilethecosinetermstendtobenearzero.Decoupledloadflow
programsuseonly J1 and J 4 ,treatingthePandQproblemsseparately.
Sothatthebenefits ofoptimalbusorderingcanbefullyexploited, nondecoupled loadflow
Jacobianmatricesareusuallyformulatedinthefollowingalternatingrowform,ratherthanthat
describedsymbolicallyabove:
Page9of15
_18_ELC4340_Spring13_Loadflow.doc
P1
Q
1
P2
Q2
PN
Q
N
V1 1
V1
V2 2
J V2 .
V N N
V
ThesolutionprocedurefortheNewtonRaphsonloadflowproceedswith:
1.
Initialize the bus voltages. For load busses, use V = 1 + j0. For generator busses
(includingtheswingbus),use V Vspec j 0 .
2.
Form the Jacobian matrix, and update all bus voltage magnitudes and phase angles,
exceptforthoseattheswingbus,andexceptforthevoltagemagnitudesatPVbusses.
3.
CheckthemismatchPandQateachbus.Ifallarewithintolerance(typicaltoleranceis
0.00001pu),asolutionhasbeenfound.Otherwise,returntoStep2.
Stott'sAlgorithm
Stott'salgorithmtakesadvantageofthestrongdecouplingofPandQthatoccursinmosthigh
voltagepowersystems.HisformulationisbasedupontheassumptionthatPisprimarilyrelated
to ,andthatQisprimarilyrelatedto V .
Hebeginsbywritingtheexpressionforpoweratbusi,whichis
Picalc
jQicalc
j 1
Vi
yi, jV j .
Hethendefines
yi , j Gi, j jBi, j .
Expandingthepowerexpressionyields
N
Picalc jQicalc
Vi i Vi j Gi, j
j 1
or,expandinginrealandimaginaryform
Picalc
jBi , j
Page10of15
_18_ELC4340_Spring13_Loadflow.doc
Qicalc
where i , j
i j .
Fromtheaboveequations,thenecessaryloadflowpartialderivativesare
Pi
j
Vi V j Gi , j sin i , j Bi , j cos i , j
j i
and
Pi
j 1, j i
Vi V j Gi , j sin i , j Bi , j cos i , j
which,bycomparingtotheQequationsimplifiesto
Pi
2
Qi Vi Gi, i sin i, i Bi , j cos i, i
= Qi
Vi
Bi , i .
Similarly,forQ,
Qi
Vj
Vi Gi , j sin i , j Bi , j cos i , j
j i
1 Pi
,
V j j j i
and
Qi
Vi
j 1, j i
Aftersimplifying,theaboveequationbecomes
Qi
Q
i Vi Bi , i .
Vi
Vi
TheformoftheJacobianupdateequationisnow
P H
Q 0
0
L
1
,
V V
where
H i , j Li , j Vi V j Gi , j sin i , j Bi , j cos i , j , i j ,
Page11of15
_18_ELC4340_Spring13_Loadflow.doc
H i , i Qi Vi
Li, i Qi Vi
Bi , i
Bi , i .
Now, since the angular differences are small, then cos i , j sin i , j . In reactive power
systems, Bi, j Gi, j ,sothat Bi , j cos i, j Gi, j sin i , j .Also,inmostcases, Bi , i Qi
,sothat Vi2 Bi, i Qi .SubstitutingtheseapproximationsintotheaboveexpressionsforH
andLyields
H i , j Li , j Vi V j Bi , j
H i , i Li , i Vi
Bi , i .
Theseleadtothefollowingsimpleformoftheupdateequation:
V1 0
0 V
2
P
0 0
B1,1
B
2,1
V N B N ,1
0
0
VB 'V
B1,2
B2,2
B N ,2
B1, N
B 2, N
V1 0
0 V2
B N , N 0 0
0
0
1
2
V N N
V1 0 0 B1,1 B1,2
0 V 0 B
2,1 B2,2
2
0 0 V N B N ,1 B N ,2
B1, N V1 0 0
B2, N 0 V2 0
B N , N 0 0 V N
V1
V
1
V 2
V2
V
N
V N
V
.
V
VB ''V
Stottproposesthefollowingadditionalsimplifications:
1.
2.
Omitfrom B''theangleshiftingeffectsofphaseshiftingtransformers,whichprimarily
affectMWflows.
3.
Sincethevoltagemagnitudesareclosetounity,write
Page12of15
_18_ELC4340_Spring13_Loadflow.doc
B (firstonehalfiteration),
P
'
B V
V
'
and
Q
V
''
''
B V
B V (secondonehalfiteration).
V
V
4.
NeglectseriesresistanceswhencalculatingB'.
Atthispoint,B'andB''remainconstantthroughoutthesolution.Therefore,B'andB''shouldbe
LUdecomposedonce,andreusedforeachhalfiteration.Thesolutionsstepsare
1.
LUdecomposeB'andB''.
2.
Calculate
3.
Update using
4.
Checkconvergence,andcontinueifnotconverged.
5.
Calculate
6.
Update V using
7.
Checkconvergence,andreturntoStep2ifnotconverged.
5.
OtherConsiderations
P
'
B .
V
Q
''
B V
V
CurrentandPowerFlowinTransmissionLinesandCables
Oncethebusvoltagesthroughoutthesystemhavebeencalculated,thentheloadflowprogram
must calculate power flows through lines/transformers/cables. The standard piequivalent
modelsareusedforthispurposeasfollows:
Ijk
Vj/ j
Q(pu)
2
jX
Icapj
Vk/ k
Icapk
Page13of15
Q(pu)
2
Ikj
_18_ELC4340_Spring13_Loadflow.doc
sofromsidej,
I jk
V j j Vk k
Q
Vj
j 90 0
R jX
2
andfromsidek,
I kj
Vk k V j j
Q
Vk
k 90 0
R jX
2
Notethatthecurrentonthetwooppositeendsofthelinearenotexactlythesameduetothefact
thatthecapacitorcurrentsarenot,ingeneral,equaltoeachother.Thepowerflowsforthetwo
*
endsoftheline,directedinward,are S jk V jk I *jk , S kj Vkj I kj
.
TransmissionlinevaluesforR,X(orL),andlinechargingQ(orC),shownontheprevious
diagram,areenteredinperunitfortheentirelengthoftheline.Therelationshipbetweenthese
parametersandthepermetercalculationsdevelopedinthetransmissionlinechapterfollow.
RandX=Larethepositivesequencepermetervaluesofthetransmissionline(e.g.,
resistanceofaphasebundlepermeter,andLwhereLisapproximately1mHperkm),
multipliedbythelengthofthetransmissionline,andconvertedtoperunitohms.
Cisthepositivesequencecapacitancepermeterofthetransmissionline,multipliedby
thelengthofthetransmissionline.UsingC,thenlinechargingQiscomputedusingbase
2
2
voltage with Q 3CV LN
. Then, Q is converted to per unit, and half is
CV LL
placedoneachendofthetransmissionline.Fora345kVlineat60Hz,andassuming12
pFpermeter,Qisabout0.538MVArperkm.Thecorrespondingcapacitivecurrentto
energizethelineisthen0.90Aperkm.
OneperunitQcorrespondstooneperunitadmittanceY. Itisstandardpracticein
loadflowstorepresentQ/2oneachendofalineasanadmittancejQ/2(inperunit).For
the above example, and using a 100 MVA base, the line charging Q of a 100 km
transmissionwouldbe53.8MVAr,anditwouldbemodeledinternallyintheloadflowas
admittancesj0.269pumhosoneachend.
CurrentandPowerFlowinTransformers
ThetransformerequivalentcircuittakenfromSection3is
Vj/j
Busj
t/ :1
Busk' y
Ijk
Vk/ k
Ikj
Busk
,
soforsidek,
V j j
I kj Vk k
y,
Page14of15
_18_ELC4340_Spring13_Loadflow.doc
andforsidej,
I jk
I kj
t
Thecorrespondingpowerflowsforthetwoendsofthetransformerare S jk
.
*
V jk I *jk , S kj Vkj I kj
AreaInterchange
Largescalepowersystemsusuallyconsistofseveral,orperhapsmany,individualelectricutility
companies. Inthesecases,eachareamayhaveadesirednetinputoroutputpowertosatisfy
salesandpurchaseagreements.
Theareainterchangefeatureinaloadflowprogramsumsthetielineflowsintoeachareafora
netareapowerinput.Ifthesenetsarenotthedesiredvalues,towithinafewmegawatts,thenan
areacontrolgeneratorwithintheareaisadjustedbytheerroramount. Thesumofallinput
powersforallareasmust,ofcourse,bezero.
ConvergenceCriteria
Mostloadflowdatacasesaredevelopedusinga100MVAbase,sothata1%loadcorrespondsto
1MW.AtypicalconvergencecriteriaisthatthehighestindividualbusPandQmismatchesare
within0.0010.01%,or0.000010.0001pu.
Page15of15