EconomicL qd Dispqtchof

Thermql G nerqring Units

3.1 INTRODUCTION
Electricalenergyc4nnotbe sto , but is generatedfrom natural sourcesand delivered as demrand
arises.A transmissionsystemi usedfor the delivery of bulk power over considerabledistances,
and a distributionsvstemis for local deliveries.As depictedin Figure 3.1, an intercornerct€d
power systemconsistsof mainl three parts: the generatorswhich produce the electrical energy,
the transmlssionlines which it it to far away places, and the loads which use it. Such a
configuration applies to all i nnectednetworks (regional, national, int€rnational),where the
qumberof elpmentsmey vary. he transmissionnetworks are interconnectedthrough ties so that
utilities q3n exchangepower, s reservesand render assistanceto one anotherin times of need.

Generator
Tiansmissionlines

Bus2
Figure 3.1 A simpl configuration powersystem.
of an interconnected

Since the souucesof en are so diverse (coal, oil or gas, river water, marine tide, a
radioactivematter, sun power), e choice of one or the other is made on economic,technicalor
geographicbasis. As there are few facilities to store electrical energy, the net production of a
utility (generationplus the i cws over its ties) must clearly track its total load. For an
interconnectedsystem,the fu mental problem is one of minimtzing the source expenses.The
economicdispatchingproblem is to define the production level of each plant so that the l;otal
cost of generationand transmi ion is minimum for a prescribedscheduleof loads.
. Forecastingincludes ining the peak rate of supply (power demandand volume), i.e.
energy dernandfor long-terminvestmentdecisionsand short-termoperatingdecisions.
131

_-_-J
t32 PowerSystemO Zation

Operating applicati s includp allocation of output (dispatching)'unit start-up splsctign'

(unit commitment), hydrothermalcoordination,and maintenancescheduling.
The investmentp ning applicationscover the generationand the transmissionsystems.

O RAilNG COST
3.2 GENERATOR
The majority of generators I extant systemsare of three types-nuclear, hydro, and fossil (coal'
oil or gases).Nuclear p tend to be operated'atconstantoutput levels and hydro plants have
essentiallyno variable ting costs. Therefore, the componentsof cost that fall under th-e
category of dispatching dures are the costs of the fuel burnt in the fossil plants. The total
cost of operationincludest fuel cost,costsof laboua supplies'andmaintenance. Generally,costs
of labour, supplies and ma ntenanceare fixed percentagesof incoming fuel costs. Figure 3.2
shows a simple model of a fossil plant. The power output of the fossil plant is inpreased
sequentiallyby openinga of valves at the inlet to its steamturbine. The throttling lossesin a
valve are large when it is j st openedand small when it is fully opened.

iler Turbine Generator'

Fuel input

gure 3.2 Simple model of a fossil plant.

As a result, the operating of the plant has the form shown in Figure 3.3. For dispatching
purposes,this cost is usuall approximatedby one or more quadratic segments.So, the fuel cost
curve is modelled as a q ic in the active power generation

F(Ps)= qP?,+ biPr, + ci Rs/h (3.1)

Fi(Ps)

u)
&
ct)
o
O
bo
tr
cE
k
C)
a
o

'p-in
8i OutPutPower -pmu
(MW1 8;

Figure3.3 Operati
^ llll0
costs of a fossil-fired generator, f 8i
r
anct Pil'* arethe lower
and the er limits on its outPut.
 Econonic Inad Dispatch of rhernml Generatilry IJnits r33

The PTin
6i
is themini um loadinglimit belowwhichit is uneconomical (or may be techni-
cally infeasible)
to operate e unit and P;:* is the maximum output limit.
The fuel cost curve y have a number of discontinuities.The discontinuitieseccur when
the output power has to extended by using additional boilers, steam condensers,or other
equipment.Discontinuities a lso appear if the cost representsthe operation of an entire power
station,so that cost has di tinuities on parallelingof generators.Within the continuity range
the incrementalfuel cost ay be expressedby a number of short line segmentsor piece-wise
linearizations.

3.3 ECONOMTC
DIS ATCH PROBLEMON A BUS BAR
Let us assumethat it is k n a pnon which generatorsare to be run to meet a particular load
demand on the station. S there is a station with NG generatorscommitted and the active
power load Pp is given, the I power generationPr, for eachgeneratorhas to be allocatedso as
to minimize the total cost. optimization problem can thereforebe statedas

NG
Minimize F(Pr) = I F;(Pgr) (3.2a)
i=l

subject to (i) the energy anceequation

NG
T Pr, = P, (3.2b)
f=l

(ii) and the ineq ality constraints

pT s psts PJ* (i= 1,2,...,NG) (3.2c)

Pr, is the deci variable, i.e. real power generation

PD is the real demand
NG of generationplants
Pil" is the lower issible limit of real power generation
Pffx is the upper issible limit of real .power generation
Fi(P s,) is the operati fuel cost of the ith plant and is given by the quadraticequation

F i(P,) = of?, * brPst * ci Rs/h (3.2d)

The above constrainedop ization problem is converted into an unconstrainedoptimization
problem. Lagrange multipli method is used in which a function is minimized (or maximized)
with side conditions in t form of equality constraints.Using this method, an augmented
function is defined as

( N G
L(Pei,1,)= F(Ps)+ hl Po - I (3.3)
\ i=l

where ,?.is the Lagrangian ultiplier.

134 Power System

A necessary conditionfor a function F(Ps,),subjectto energybalanceconstraintto havea

relativeminimumat point Pi, that the partial derivativeof the Lagrangefunction definedby
L = L( Pr,, h) with respectto h of its argumentsmustbe zero.So, the necessary
conditionsfor
the optimizatioRproblemaro
aL(P8i'
I = E{(Pr,)
- )"=o ( i = I , 2 , . . . ,N G ) (3.4)
EPs, dPe,
and
aL(k' L) -
'I' Pp-YPr,=o (3.s)
U" i=l
From Eq. (3.4),
F (P8t)
(i = 1,2, ...,NG) (3.6)
dP,,

whereAFe)la Ps is the incre fuel cost of the ith generator(Rs/TvIWh).

Optimal loading of gene tors correspondsto the equal incrementalcost point of all the
generators.Equation (3.6), cal the coordinationequationsnumberingNG are solved simulta-
neously with the load demand to yield a solution for Lagrangemultiplier /, and the optimal
generationof NG generators.C idering the cost function given by Eq. (3.2d), the incremental
cost can be defined as

{9 =zarPs,
+ bi (3.7)
dP,,

Substitutingthe incremental into Eq. (3.6), this equationbecomes

2 iPr, + bi = )" (i = 1,2, ...,NG) (3.8)

RearrangingEq. (3.8) to get Pr,

8i
-- L-u, ( i = I , 2 , . . . ,N G ) (3.e)
2o,

Substitutingthe value of P* in Eq. (3.5), we get

y+3 -Pp
i=l LAi

or

)"=
NG
(3.l0)
I
i=l' 6
Thus, )" can be calculatedusin Eq. (3.10)and Pr, can be calculatedusing Eq. (3.9),Now
consider the effect of the ge tor limits given by the inequality constraintof Eq. (3.2c), If a
 Economic Load Dispatch of Thermal Generating units 135

particulargeneratorloading Pr, reachesthe limit P#" or Pffx, its loading is held fixed at this
valueand threbalanceload ir sftaredbetweenthe remaining generatorson an equal incremental
costbasis.

3.3.1 LimitConstrai Fixing

To fix up the limits the fo ing strategycan be applied [Famideh-Vojdani,I e801.
Let
Rr R2
M=Zrf*
i=l i=l

where
hf* =,Psi - Pf* ( l = l, 2, ..., R1 upper bound violations)
hY" =P#n - Pr, (j= l, 2, ..., Rz lower bound violations)

(i) It M > 0, fix all 1 upperboundviolationsto the upperlimits, i.e. Pt *

(ii) If Lh < 0, fix all 2 lower bound violationsto the lower limits, i.e. P#t"
(iii) On the otherside,if Lh = 0, fix both R1 upperand R2 lower bound violationsto their
respectiveupper maxand lower Pmrn limits.
8i 8i

Determine the new which is original Pp minus the sum of fixed generationlevels,

*'ft"r,
Pff* - Pp-
i=l

The new demandis alloca to various committed generatorson an equal incrementalcost basis.

EXAMPLE 3.L Two units of the systemhave the following cost curves:

F(Ps)= o.o5Pg1
+ 22Pn+ l2o Rs/tr
F(Pe)= o.o6P?,
+ l6Pr, + l2o Rs/tr
whereP, is in MW. Both units operateat all times and the maximum and minimum loads on
each unit afe 100 MW and MW respectively.Determine the economic operating scheduleof
',
the plants for lsads of 80 120MW, and 180MW neglectingthe transmission line losses.
Solution Using Eq. ( .10)to calculateL
22 16
Pn* -.------- + -
--1 2 x 0 . 0 5 2 x 0 . 0 6Rs/IVIWh

(i)
f
,136 Power Sys'ternOptimizat

When PD - 80 MW
Substitutingthe value of Pp in . (i) and solving for h, we get
)" = 23.6368 Rs/]vfWh
Using Eq. (3.9) to calculateg lons
I

n = -23.6369 - 22 23.6368- 16
_- 6.36MW; tDg z - = 63.64MW
51 1lx 0.05 2 x 0.06

But Ps = 16.36IvtW < 20 MW fix Pst at the lower limit 20 Mw and the rest of the demancl
will be met by the secondg
So Pr, and Pr, =80-20=60MW

When Po= 120 M'W

Substitutethe valuerof Pp in Eq. ( ) aqd solve for ,1,,i.e.

). = 25.81B18 Rs/IvIWh
From Eq. (3.9),
-22 2 5 . 8 1 8 1-8 1 6
tDr s _
r =2 5 . g l g l g _ .' , 1818MW
= ; Psz= = 8 1 . 8 1 8M2W
2*o-05 2 x 0.06
When Po i 180 M'W
Substitutethe value of Pp in Eq. ( ) and solve for ),, i.e.

), = 29.0914Rs/lvIWh
From Eq. (3.9),

- 22
29.10914 29.0914- 16
P
or
= -- -1f\
9142MW; 'D9 2 - = 109.095
MW
2l x 0.05 2 x 0.06
But Pez= 109.095MW > 100 , so fix Pr, at the upper limit 100 MW and the resr of rhe
demandwill be met by rhe first g

So Pr, = 100 IvfW and Pr, = 180 - 100 = 80 MW

EXAMPLE 3.2 Inrcrementalfuel ts in rupeesper MWh for a plant consistingof two units are

_1F'= o.2o + 40 and = 0.40Prr+ 3O

dP^ , *dPr,

md the generatorlimits are

30 I\dw < Pr, < I MW and 20 MW < Pr, < lZ5 MW

isumethat both units are operatir at all times. How will the load
be sharedbetweenthe two
its as the system loaclvaries over the full range of the load values?
What are the corresponding
of the plant incremental c s ?
 Economic l-oad Dispatch of Thermat Generating Units t37
Solution FromEq. (3.1

40 30
P P+
),= 0 2 * o A Rs/IvIWh
1 1
0.2 0.4

-- Po +275 Rs/rvIWh
7.5

PD- Q.sL- 27s)MW (i)

The values of )" at mini um and maximum operatinglimits can be obtainedas follows:
When unit 1 is operatin at minimum limit, P#tn = 30 MW then

nun = 0.2 x 30 + 40 -- 46 Rs/lvIWh

When unit 2 is operatingat minimumlimit, p#n = 20 MW rhen

Pin = 0.4 x 20 + 30 = 38 Rs/IvIWh

When unit I is operatingat maximumlimit, PrTo = 175MW then

= 0.2 x 175+ 40 = 75 Rs/lvIWh

When unit 2 is operatingat maximumlimit, Pil* = 125MW then

= 0.4 x 125+ 30 = 80 Rs/IvIWh

So, there are three ng conditions for L, which are obtained from minimum and
maximum operatingvalues, i.e
(r) 38 < X"s 46
(ii)M < 2vs75
(iii) 75 < r, s 80
The range of load can be obtainedas given below:
(i) For 38 < )" S 46, the lower power limit of unit I is violated because ), = 46 Rs/IvIWh
when A*in = 30 MW.
The mininnumdemand the system that can be met will be, Pp - P,'"tn+ Pfin , i.e. Po =
3 0 + 2 0 = 5 0 M W w h e2n" = 3 8Rs/lvtWh.For 1,= 46, the value of Pp can be determinedfrom Eq.
(i), i.e.

Pn = 7.5 x 46 - 275 = 70 MW
So, the range of demand 50< PD<70.
 beyond50
fu Pp increases the load increments unit 1 is
are placedon unit 2 because
fixed at 30 MW (minimum ng limit). So, the incrementalcost equation of unit 2 can be
written in terms of Pp as
h=0.4x(Po-30)+30

On simplification
L = (0.4 o + 18) Rs/TvIWhfor 50 < PD < 70
i'
(ii) For 46 < )" 3 75, the ation of generatorswill be within the operatinglimits' When
- 75, the value of Pp can be ined from Eq. (i), i.e.
= 7.5 x 75 - 275= 287.5MW

So, the range of demandis S Po < 287.5, and

Pp
Rs/IdWhfor 70 3 Po S 287.5
- 75 Rs/lvIWh
(iii) For 75 upper timit of unit 1 is violated because )"
when 1. = 80 Rs/
when Prnax= 175 MW. The max mum demandof the systemthat can be met
MWh, iS
- 175 + 125 = 300 MW
P n = P m a x * P2'"*, i.e Pp

So, the range of demandis 8 7 . 53 P o < 3 0 0 .

As Pp increasesbeYond28 5 MW the load increments are placedon unit 2, becauseunit
1 is fixed at I75 MW. So, the i ntal cost equationof unit 2 canbe written in termsof Pp as

A"=0.4 x (Pp - I75) 30 = 0.4PD- 40 Rs/IvIWhfot 287.5S Po < 300

The generationscheduleis ivenin Table3.1.

9.4 OPTIMALGENERATI SCHEDULING

the plant can be
From the unit commitment tabl of a given plant, the fuel cost curve of
least squaresfit. If the
determinedin the form of a PolY mial of suitable degreeby the method of
divided among the
transmissionlosses are neglect' , the total system load can be optimally
(3.6). It is, however,
variousgeneratingPlantsusing e equal incrementalcost criterion of Eq.
of power is
unrealisticto neglecttransmissi lossls particularly when long distancetransmission
low load density' The
involved. A modern electric util ty serves over a vast area of relatively
. transmissionlossesmay vary m 5 to 15 per cent of total load. Therefore, it is essentialto
policy.
account for transmissionlosses hile developingan economic load dispatch
-
The econonnicdispatch lem is defined as that which minimizes the total operating cost
generatorlimits.
of a power system while meetin the total i-oadplus transmissionlosseswithin
Mathematically, the problem is

Minimize Pr,)=Y (r,r],,*b,Pr,*6;) Rs/h (3.r2a)

 Economic Inad Dispatch of Thermal Generating Units

Thble3.1. Generationschedule(Example3.2)

h t l, Pr, Unit 2, Pr, Po=Psr*Ps,

(Rs/IvIWh) o{w) (Mw) o4w)
38 30 20 50
40 30 25 55
42 30 30 60
44 30 35 6s
46 30 40 70
48 40 45 85
50 50 50 100
60 100 75 175
75 r75 112.5 287.s
78 175 r20 295
80 175 125 300

subject to (i) the energy balan equation

NG
2Pn,-Pp+Pt (3.r2b)
i=l

(ii) and rhe inequalit constrarnts

P's', - ^9, - t gi (3.r2c)

where
ai, bi, and c; are the cost oefficients
PD is the load demand
Pr, is the reatrpower ge ration and will act as decisionvariable
NG is the nunnberof ge ration buses
PL is the transmission loss.
One of the most importan simple but approximatemethodof expressingtransmissionloss
as a function of generatorpo is through B-coefficients.This method uses the fact that under
normal operatingconditions, transmissionloss is quadraticin the injected bus real powers.
The generalform of the loss fr ula (derivedlater in this section)using B-coefficientsis

NG NG
Pt=7 ZPs,BijPsi MW (3.13)
i=l j=l

P* and Pr, *" the real po injections at the ith and 7th buses,respectively
Bii are the loss coefficien which are constantunder certain assumedconditions
NG is numberof generati buses.
The transmissionloss f, la of Eq. (3.13)is knownas the George'sformula.
 140 Power System Optimi rcn

Another more accuratef of transmissionloss expressior,&equently known as the Kron's

loss formula is
NG NG NG
Boo+ Z, B*Pr, + ) I Pr,BiPri lvrw (3.14)
i=l j=l

where
Pr- and P, ue the real injections at ith and 7th buses,respectively.
Boo,Bio,and B,; are the coefficientswhich are constantunder certain assumedconditions
NG is numberof generati buses.
The aboveconsffainedop mization problem is convertedinto an unconstrainedoptimization
problem.Lagrangemultiplier thod is used in which the function is minimized (or maximized)
with side conditionsin the form of equality constraints.Using Lagrangemultipliers, an augmented
function is defined as

l r 'L ) = F(Pe,'t
* t(r, +Pr .r,) (3.1s)
i:
where ,?,is the plier.
Necessary le opti mization problem are

t(Pr,
)
. ^(W , )= o ( i =1 , 2 , . . , N G )
)P,,

Rearrangingthe above equation

aF(P8
( i = 1, 2 , . , . ,N G ) (3.16)
dPs,

where
aFest)
is the incremen cost of the ith generator(Rs/IvIWh)
at;
dP,
= u is the irrcremental missionlosses.
dP,
Equatipn(3.16) is known the exact coordinationequation,and

L(Pe,,L) NG
-=' P
I )n +' 'P
Lt P r ,= Q
ah ?2 r ' 8 i (3.17)

Equation (3.16), the so-called ination equation,nurnberingNG is solved simultaneously

with Eq. (3.17) to yield a soluti n for Lagrange multiplier )" and the optimal generationof NG
generators.By differentiating transmissionloss equation,Eq. (3.14) with respectto Pr_,the
incremental transmission loss c be obtainedas
NG
B io + >,.28UPBj ( i = I , 2 , . . . ,N G ) (3.18)
j=L
 Economic Inad Dispatch of Thermal Generating Units l4l

and by differentiatingcost on Eq. (3.12a),with respectto Pr-, the incrementalcost can be

obtainedas

)
=Za;Pr, + 6i ( f = 1, 2 , . . . ,N G ) (3.le)

Equation(3.16) can be rewri

=h

(3.20)

the penalty factor of the ith plant.

To obtainthe solution, ' )and

r (3.19)into Eq. (3. 6)

NG
ZaiPsi + bi t n,lijPe)
,>, 22 ( i = 1 , 2 , - .N
.{G
, )
,
Ji r= l )

Rearrangingthe above equati te

ave
wer h OVt

NG
t
)
2(ai + LBii)Pe, Bi,
iPt
iPci l-bi ( r = 1 2,
, ,, NG)
j=l
j+i"2r )

The value of Pr, can be o

'['

P$ ( l = 1, 2 , . . . ,N G ) (3.2r)
Z(ai + )"9 ii)

If the initial values of b,(i = 1,2,..., NG) and 2uare known, the above equationcan
be solved iteratively until . (3.17) is satisfied by modifying )". This technique is known
as successiveapproximation. The stepwise procedure is explained below. For simplicity it is
considered that the solution ins within limits.
142 Power System

Algorithm 3.1: Economic (ClassicalMethod)

l. Readdata, namely cost Bii, Bio,Boo(i = 1,2,..., NG;
fficients,ai, bi, ci; B-coefficients,
j = 1 , 2 , . . . ,N G ) ; tolerance,q step size a; and maximum iterationsallowed,
ITMAX, etc.
Compute the initial val of Psi(i = l, 2, NG) and L by assuming that the
transmissionlossesare , i.e. Pt= 0. Then the problemcan be statedby Eqs. (3.2a)and
(3.2b) and the solution be obtaineddirectly using Eqs. (3.10) and (3.9).
3 . Set iteration counter,IT
4. ComputePs,Q= 1,..., G) usingEq. (3.21).
5. ComputetransmissionI usingEq. (3.14).
NG
6. Compute L,P = Pp + Py ZPs,
i=l

7. Check I AP | < €, if 'yes' then GOTO Step 10.

CheckIT > ITMAX, if yes' then GOTO Step 10. (It meansprogramterminatedwithout
obtainingthe required vergence.)
8 . Updateffn = ),+ al l, where a is the step-sizeusedto increaseor decreasethe value
of )" in order to meet Step 6.
9. IT = IT + l, ),= ff"* GOTO Step 4 and repeat.
1 0 . Compute optimal total loss from Eq. (3.1a).
from Eq. (3.12a)andtransmission
l l . Stop.
Consider now the effect f the generatorlimits given by the inequality constraint of
Eq. (3.12c).If a particulargenera r loading p* reachesthe lower limit, P#i, or the upper limit,
Pf"* , its loading from then on i held fixed at this value and the balance load is shared between
the remaining generators.A step ise procedure to obtarn the optimal generation schedule when
the operating generationlimits imposed is given below.

Algorithm 3.2t EconomicDi tch ConsideringLimits (ClassicalMethod)

l. Readdata,namely cost Bii, Bn, Boo(i = 1,2,..., NG;
cients,ai, bi, ci; B-coefficients,
j = l, 2, ...,NG); con q
ence tolerance, step size u, and maximum allowed iterations,
ITMAX, etc.
Computethe initial val of PBi(i= 1,2,..., NG) and )" by assumingthat Pt= 0. Then
the problem can be sta by Eqs. (3.2a) and (3.2b) and the values of ,1.and P s i ( =
i I,2,
..., NG) can be obtaineddirectly usingEqs. (3.10)and (3.9),respectively.
3 . Assume no generator been fixed at either lower limit or at upper limit.
4. Set iterationcountel IT 1 .
5 . Compute Prn(i= l, 2, .. R) of generatorswhich are not fixed at either upper or lower
limits, using Eq. (3.21),
ere R is the numberof participatinggenerators.
6. Computethe transmissio loss using Eq. (3.14).
NG
7. Compute AP = Pp + P2 L Ps,
i=l
 Ecorbmic l-oad Dispatch of Thermal Generating Units 143

8. ChecklAF I S e, if ' ' then

GOTO Step 11.
'yes', thenGC)TOStep11. (It meansthe progra,m
Checkm>ITIYIAXi movesfonvard
without obtaining convergence.)
9. Modify ,un* = 2v+ , where a is the step-size.used to increaseor decreasethe value of
l. in order to meet Step7.
10. IT = IT + l, 2v=2y GOTO Step 5 and repeat.
1 1 .Check the hmits of ators, if no more violations then GOTO Step 13, else fix as
following.

If Ps,(P#" then Pr, P#"

If Ps,) P#* then Pr, P#"'
12. G0TO Step4.
13. Compute'theoptimal cost from Eq. (3.12a) and fansmission loss from Eq. (3.1a).
14. Stop.
The above sfrategy is demons in Example 3.4.

EXAMPLE 3.3 The tuel i per hour of two plants are given as

n(Pe) = (0.00889P:t + 10.333Pr, * 200) Rs/tr

Fz(P' ) = (0.0074r
P:r+ 10.833
Pn +240) Rs/h

Determine the economic to meet the demandof 150 MW and the correspondingcost of
generation.The transmission are given by

+ o.wzr!, - 2 x 0.w2PstP82
0.001Pr2

Solution Follow the s se procedurementionedin Algorithm 3.1.

.Stepl: Recognizn,the and assumea = 0.05, t = 0.0001,and ITMAX = 15
Step 2: Compute l. Eq. (3.10),i.e.

10.333 10.833
1 5 0 +- - L -

0 . 0 0 8 8 9 ' 2 x 0 . 0 0 7 4=1 11.81812

Rs/]vIWh

2 x .00889 2 x 0.00741
Compute P^ and Psz usingEq. (3.9)

11.81812- 10333 =
i P8 83.5276MW
2 x 0.00889
r 1.818t2 - 10.833=
Pe 66.47239MW
2x O.O074l
144 Power System Optimi

Step 3: Set IT = I
Step 4: Compute Pr, Prr, usingEq. (3.21)
1 r . 8 1 8U
12 - 10.333
0 x (-0.0002) x 66.472391
Pgt = = 43.44557MW
2t0. + (11.818r2
x 0.001)l
"11.81812U x (-0.0002)x $.5n61 - 10.833
tDg z - - 22.22453MW
210. 4r+(1r.fr8r2x0.002)l
Step 5: Computetransmi ion loss,
P t = ( .001X43.44s5T2+ (0.002)(22.224fi)z
_ 2.489154MW
2(O.OO02)(43.44ss7)(22.224s3)
Step 6: Compute the balancerequirementto be met
2
A , P = P p + p r - Z Pr, = 150.0+ 2.489154- 65.6701=+ 86.81905MW
i=l

Step 7: I f t A P t < 0 . 0 0 0 the

1 criterion is not met then GOTO Step 8.
Step 8: Modifyh - 11.8812 + 0.05(+86.819C5) = 16.15907Rs/IvIWh
IT= IT+landGOTO
Step 4 and repeat.

The iteration-wiseobtainedresultsof Prr, Pr, , L, LE and Py are given in Table 3.2.

Table 3 2 Generationschedule(Example3.3)

IT Pe, L LP PL
(Mw) (Rs/IvIWh) (lvIW') MW)
I 43.4456 2.2245 1r.8t8t2 + 86.81905 2.489rs
2 I 19.1606 0.5657 t6.1,5907 - 18.93145 20.79482
3 110.l3t6 7.4586 15.21250 -g .3316A4 19,,25956
4 r0r.7692 1.8196 r4.74592 r.894872 t 5.49376
5 102.7077 2.1681 14.84067 + 0.848730 15.72457
6 r03.4826 2.6953 1 4 . 8 8 31r -0.203037 15.97491
7 r03.3739 2.6494 14.87295 -0.077766 15.94553
8 rc33Ar2 2.5993 14.86907 + 0.021338 15.92185
9 103.3133 2.6048 14.870r3 + 0.007057 15.92521
l0 103.320r 2.6096 14.87049 -0.002229 15.92743
11 103.3188 2.6089 14.87037 - 0.000624 15.92705
t2 103.3181 2.6085 t4.87034 + 0.000239 15.92684
l3 103.3183 .6086 14.8703s + 0.000041 15.92689

The final optimal schedule

Pst = 103.3183MW, pr, = 62.6a86MW ), = 14.87035
Rs/Ivrwh,and cosr= 2309.77Rs/h
 Economic l-oad Dispatch of Thcrmal Generating IJnits l4s

EXAMPLE 3.4 For a enerator system, the fuel cost coefficients and the operating
generatorIimits are givenin le 3.3(a).The B-coefficients
for transmission
loss aregivenin
Table 3.3(b). Determine the ic schedulefor loads 160 MW and 210 MW.
Solution Algorithm 3 is followed to get the optimal generation schedule. The achieved
generationscheduleis given Table3.4. The numberof iterationstakenby the algorithmare
given in Thble 3.5. The val of steplength a chosen,and the achievedconvergence are given
in Thble 3.5.
The method is very sen ve to the value of a,, i.e. the step size. The number of iterations
dependupon the assumedval of u. An incorrect valueof a,, sets the solution procedurein the
oscillations.To avoid this pro lem the Newton-Raphsonmethod can be implementedto get the
solution.

Table 3.3(a) Fuel coefficientsand operatinggeneratorlimits (Example 3.4)

Generator a; Ci
r (R$/]WW2h) (Rs/h)
I 0.00608s 10.04025 t36.9r2s 5.0 150.0
2 0.00s915 9.t60576 s9.1550 15.0 100.0
3 0.005250 8.662500 328.t250 s0.0 250.0

.3(b) B-coefficientsMW-r (Example 3.4)

0.0001 0.0000175 0.0001839

0.000075 0.0001545 0.0002828
0.000139 0.0002828 0.0016
r47

Thble 3.4 Optimal generationschedule(Example 3.4)

PD Pe F PL
Mw) O4w) (Rs/h) (Mw)
No generationlimits imposed
160.0 57.5577 7 .5238 37.9172 11.09701 2176.023 s.998648
210.0 83.4010 9 .6169 39.4862 1r.52315 2741.473 8.503935
Generationlimits imposed
160.0 53.3906 .6094 50.0000 I 1.08013 2179.r59 7.999945
210.0 79.9043 .5531 50.0000 11.51164 2743.905 10.457380

3.5 ECONOMICDISPACH USINGNEWTON-RAPHSON

METHOD
The economicdispatchproble is expressedby Eqs. (3.L2a),(3.Lzb), and (3.12c) and is converted
into an unconstrainedoptimi ation problem as in Eq. (3.15). Necessaryconditions for the
optimization problems [Eq. ( .15)l are given by Eqs. (3.16) and Eq. (3.17). The solurion of
nonlinearEq. (3.16)can be ob ined using the Newton-Raphsonmethod in which any changein
 146 Power System rcn

(Exarnple3.4)
3.5 Optimalschedule

PD Iteratiorts a
Mw)
No generation
li its imposed
160.0 20 0.7009506X l0-4 0.00s
210.0 20 0.7820129X 104 0.005
Generation
limits imposed
160.0 l3 0 . 5 5 3 1 3 1X l l0-4 0.00s
210.0 T4 0.2t93451X l0-4 0.005

control variables,about their i itial valuescan be obtainedusing Taylor's expansion.Taylor's

expansionto secondorder of . (3.16)and Eq. (3.17)can be writtenas

a2L
" = L P ,
:rD r L ) " =- a L (3.22)
. O t L dP.*,
8,

#
HffiLPsi-.#L)'= (3.23)

The above equationscan be itten in matrix form as

Der,frVStr )s cal
lval ives ca be
)e(
f;r:Y:lt*1=[-r;1
lows:
(3.24)

AL _ dF, NG
^[ 2a1Pr, + I zB,Ptt
IP,, aP_
aP_ aP- o,
j=l

AL NG
I P*, (3.2sb)
AL i=l

Taking derivativesof Eq. (3 ) with respectto Pr,,

dzt dzF, ^dZP,

-------==-:-r*i =Zdi +2M,, (j= I , 2 , . . .N
, G) (3.26a)
aP: aP; apl

( i = 1 , 2 , . . . , N G ;j = 1 , 2 , . . . , N G ;i * j ) (3.26b)

.--c
 Economic Load Dispatch of Thernnl' Generating Units 147
Takingderivatives
of Eqr. (3. ) and (3.25b),withrespecrto i.,
d2L d2L
aTaPs aPstaA
*,{ 2Brip,
-l
#-r-4,
-d2t
-- =Q (3.26d)
a1'
Equations(3.22)and (3.23)[or . t . (3.24)l are iteratedtill no further improvement is obtained,
or
single derivativeswith respectto control variablesbecome zero. The stepwise procedure is
outlinedhere.

Algorithm 3.3: Economic (Newton-Raphson Method)

l. Read data, namely c;, , ci (cost coefficients);B;i, Bis,86 @-coefficients)(i = 1 , 2 ,. . . ,
NG;/ = 1,2,...N
, G ) c o vergence tolerance,q and ITMAX (maximum allorved iterations),
etc.
2. Compurethe initial val of P.*,(i = l, 2,..., NG) and .1.by presumingthat P1 = 0. The
valuesof )" and Po(r = ,2, ..., NG) canbe,computed directlyusingEqs.(3.10)'and(3.9),
respectively.
i .

3 . Assumethat no genera has beenfixed eitherat lower limit or at upper limit.

4. Set iterationcounter IT l.
5 . ComputeHessianand J bian matrix elementsusing Eqt. (3.25) and (3.26).

,[arnJ
IHll l=-tJl
L^^)
Deactivaterow and col mn of Hessianmatrix and row of Jacobianmatrix representing
the generatorwhose ge tion is fixed either at lower limit or at upper limit. This is
done so that tixed g tors cannotparticipatein allocation.
6. Gausseliminationmeth is employedin which trian_gularization and back-substitution
processesare perf, to find Pr,(i = L,2,..., R and A,l.).Here R is the numberof
_qeneratorswhich can icipate in allocation.

7. Check either (APs,

If convergenceconditio is 'yes' then GOTO Step 10.

CheckIT > ITMAX. if ndition is 'yes', GoTo Step 10. (It meansthe procedureproceeds
without obtaining requi convergence.). I
8. Modify control variabl

= Ps,* LPr,
,P,,ilt* (i = I ,2, ...,R and inew- ), + Ltr)
14E PowerSystem

9. I f = I T + 1 , p r , = p , , L = Xo" andGOTOStep5 andrepeat.

10.If no more violations GO(fO Step 12, else check the limits of generatorsand fix up
as followc

11 Pei . P#tn then Pr, P;un

fr Pr, t P#* thenpr- Pf*

11. GOTOStep4 and t.
12. Compute the oprimal cost and transmissionloss.
13. Stop.

EXAMPLE 3.5 Determine economicscheduleto meet the demand of 150 MW using the
Newton-Raphsonmethod. Use the data of Example3.3.

Solution Giv6n PD = 150 MW. Initial values are presented below as calculated in
Example3.3.

MW;, '&rd i. = l1.8tBI2 Rs/tvIWh

Pr, = 83.5276MW Pn = 66.47239

The HessianMatrix e are obtained and are given below:

2ar + 2)Bn = 0.041416RsfivtW2h

2a2+ zhBzz= 0.062092Rs/IvfW2h

dzt = ZXBz= -0.004727Rs/lvtw2h

= 2Bn Pe,* ZBnPrr=- 1.0= -0.859534

alLaPs,
^7t
a2L =2821Pr,+2822Pr, - 1.0= -0.767522
a).a42 6t

a2L
0
*
The Jacobian matrix ele ts are computedand are shown below:

dt
+ )u(2BrrPr,* 2B21Ps,- 1) = 1.66N47 Rs/IvIWh

+2822Pr,- 1)=2.747459Rs/lvIWh
bz + 1"(zBzLPs,
 Economic Load Dispatch of Thermat Generating
Units 149

P:, + 2 B z P r , P g =
, I 3 . 5 9 3 1M
I W

_=pp+py Pr, * Prr) = 13.59311

MW
AL
The aboveelementsof Hes and Jacobianmatricesare written in matrix form as

Using the gauss-elimination

method,

.767522

.066314

Using back substitution,

LA - 2.977077RsAvI h, LPr, = -5.846895MW, Mr, = 27.035510MW

Pr, = 83.5276+ 21. 5510= 104.563110 MW
P r r = 6 6 - 4 ' 7 2 3 9 - 5 . 95 = 60.6255MW
L - 1 1 . 8 1 8 t 2 + 2 . 077 = 14.7952Rs/lvIWh
Checkthe convergence,

t(M8,) + (LPs)z + (A,X)'l'' = 22.035014

ComputeAP = | Prr+pr, ' L - P p l = 0 . 5 6 0 0 7 1 >0 . 0 0 1
The Iteration-wiseschedule s given in Table 3.6.

Table 3.6 Schedu of generationduring iterations(Example 3.5)

IT PIl )" AP PL
(Mw) (Rs/LIWh) (Mw) (Mw)
2 103.2330 2.56359 r4.86603 0.1043829 15.9020r
3 1 0 3 .134 0 2.60577 r4.870r2 0.0578308 15.92552
4 103.3183 2.60843 14.81034 0.0031089 15.92682
s 1 0 3 .r38 3 2.60858 14.87035 0.0001812 rs.92690
= 2309.771Rs/h
 Power System Opti

3.6 ECONOMICDISP TCH USINGTHE APPROXIMATE

NEWTON-
RAPHSONMETH
The economicdispatchpro lem is solvedby solvingEq. (3.22) and Eq. (3.23) iteratively.
Equation(3.22) can be imatedwithout much loss in the accuracyof solution by neglecting

Eq. (3.22) can be rewritten as

AzL
* a z L -'
M=_aL (3.27)
lLPr' arr.M" dPr,

In view of Eq. (3.26d),Eq. ( .23) is rewritten as

\o atr *AF-8i -= -aL (3.28)

fr il.dPr, M.

SubstitutingEqs. (3.26c)in Eq. (3.27),

.[#-')^^=-
[#.^(# '))

or

=[, E:)e+^^)-t#)
^W)*,, (3.2e)

Let
AF, -
# 2atPs,+ bi Rs/TvIWh (3.30)
ut gi

azFi ^ d2P1-=Zar+Z)"Bii
+ IW Rs/lvIWh (3.31)
q

= )" + L)', Ki =
Anew (3.32)
#
FromEq. (3.29),
( 1- K ) L * -Xi
LPr,=T (3.33)

SubstitutingEqs. (3.26c) and (3.25b)in Eq. (3.28),

NG
^ , P s i = - [+* P r
j=l
#-r)

.*d
 Economic l-oad Dtspcttchof Thermal Generating [Jnits 151

Rearrangingthe above get

NG

j=l
r(l- K)Mr, =P; (3.34)

NG
where P; = P p + P y LPs, (3.3s)
i=l

SubstitutingEq. (3.33) in Eq. we get

(1 - Ki)L""*
I (1- Ki)
j=i Yj

Nc( _ K )' NG (1 - K1)X1

I i Lnew I _P;
j=I j=l Yj

NG (1 - Kj)Xj
P; + I
j=L Yj
(3.36)

By iteratingEq. (3"36)and Eq (3.33), the solution can be obtained.The stepwiseprocedureis

givenin Algorithem3.4.

Algorithm 3.4: EconomicDis tch (Approximate Newton-Raphson Method)

1. Readdata,narrrclyai, br, ci (costcoefficients);B;i, B;s,86 (B-coefficients);(i = l , 2 , . . . ,
); e (convergencetolerance);and ITMAX (maximum allowed

2. Computethe initial v uesof PsiQ-- 1,2, ...,NG) and)" by assumingP1= 0. The values
of ,1,and Pr,(i = l, be obtaineddirectly using Eqs. (3.10) and (3.9),
respectively.
3 . Set iteration counter, - l.

4. Compute Lnewusing . (3.36).

5 . F i n d L P r , ( i= 1 , 2 , . . . NG) usingEq. (3.33)and P;"* = P,-.+ APs,

6. Compute P1 using Eq (3.r4)or Eq. (3.I 3).
NG
7 . Compute AP = Po + L - ZPe,.
i=l

8 . ChecklAP | < e, if 'y 'GOTO Step10,

or li,.n"*- A) ( t, if 'y 'GOTO Step10,
or IT > ITMAX , if ' s'GOTO Step10 (premature
endof procedure).
ts2 Power System Opt
I

9. Modify Pr, = t1"* i = 1 , 2 , . . .NG)

, and )' = frnew

IT=IT+l,G Step 4 andrepeat.

10. Computeoptimal tal cost and transmissionloss.
1l. Stop.
The limits of generations be handledby the procedureoutlined in Section3.4.

EXAMPLE 3.6 Determine he economicscheduleto meet the demandof 150 MW using the
approximateNewton-Raphs method.Use the data of Exarnple3.3.

Solution Find the ini values(as calculatedin Example3.3),

2 "= 1 1 . 8 1 8 1 2 ; Psr = 83.5276i|dWi Prr, = 66.47239MW
Calculate incrementalI , using Eq. (3.32),

Kr =2&1Pg,+2BpPr, - 0.1404663

Kz =ZBzrPgt2B22Pg,= 0.23324785

Calculate ffansmission I

_B
Py ,P!, + B22P!,+ZBpPrrPr,= 13.59311
MW

P; Pp +Py - (Ps, +Prr) = 13.5931

1 MW

Find constantsas define in Eqs. (3.30)and (3.31),i.e.

- 2arPr,+ b, - 11.81812
Rs/lvIWh
= 2a2Prr+bz = 11.81812
RsAvIWh
= 2(ar + hBLr)= 0.041416Rs/IvIWh
= Z(az+ )"Bz) = 0.06209Rs/lvIWh

Computethe new value ,1.,usingEq. (3.36),

14.81919Rs/lvlWh
2 | - K;
t '
iir Yj

Find the change in gen tions usingEq. (3.33),

-'Kt)''n"* -Xl (1 - Kz) )"n"* - X2
APr,-(1 MW; LPs,=
22.20076 - -7.151834MW
Yl Y2
 Economic I-oad Dispatch gf fficrmal Generating Uryits 153

Check the convergence

I hnew- )"1 - 11.81812

| 14.8r9r9 | = 3.00106
> 0.001 ,.
The required convergenceis not met, so modify the generationvaluesand updateL

P,, = P g , * = 83.5276+ 22.20076= 105.7284

MW

Pr, = P s r * 'r, = 66.47239

- 7.151834= 59.32056
MW
The above procedureis ated till convergenceis obtained.The generationschedule)".
load mismatchand advanceme in X.during each iteration are given in Table 3.7.

Thble 3.7 Gene on schedule during each iteration (Example 3.6)

IT Pg h AP lLnew- 11
(Mw) (Rs/ivlWh) (Mw) (Rs/It'IWh)

1 I 05.7284 59.32056 14.81919 0.6586790 3.001060

2 103.0107 62.88092 14.87678 0.03664s7 0.057589
3 r03.34s6 62.57994 14.86997 0.0003290 0.0068
i2
4 103.3156 62.61r34 r4.87035 0.000003
8 0.000442
5 103.3186 62.60832 t4.8704r 0.0000172 0.000058
Cost= 2309.771Rs/h, Pr = | .92689MW

3.7 ECONOMICDISPATH USINGEFFICIENT

METHOD
For the economic dispatch blem, the necessaryconditions for optimality are given by
Eq. (3.16) and Eq. (3.17) as

dFi
(3.37)
dPr,
NG
ZPs,-Py+Po (3.38)
i=i

The initial values po and A.oat calculatedby assumingthat the transmissionlossesare absent
(r.e.Py- 0) and usirfbeq. (3.10 and Eq. (3.9).
The nonlinearequationsin P* andX, canbe solvediteratively.L,et P:t(i = 1,2,..., NG) and
Lo be approximatedsolutionsto Eq. (3.37). To find the new approximation,let

Lnew _ 9+A,L
pnew - (i = l, 2, ...,NG)
'gi + LPr,
8i

So, here the aim is to in

( i = 1 , 2 , . . .N, G ) (3.3e)

rr- -2
t54 Power System O

and
NG
t*:'*- P/'* * Po (3.40)
,l
Taylor's expressionto first r of Ple* is

Plt* = Pot^
* H d P | . M*'
np (3'41)
,.?,3,po
where Pf is the initial tra ission losses.
loss from the initial valuecan be
The modified value f the incrementaltransmission
obtainedA S

dpT* azPottp_
-= d P o , *' Y -''sl
dPr, dPr, i"=, aPsiaPsj

To retain the classicaform and at the sametime to improve the convergence,it is possible
to include only the trh it of the summationof the aboveequation,i.e.

apf* _ dPL d' Po,(pnew

- *' '' (3.42)
dP*, dPr, aP: 3i

Similarly, the modified i tal cost from the initial value can be obtainedas

(P;.* - (3.43)

SubstitutingEq. (3.42) and Eq. (3.a3)into Eq. (3.39),

aFP, a'F?(P;"* - P!; - ,.- l-, d Pot

)' ='1 .-n 'L* 1 E
1 P *

Theabo""rr::.: terms of P*lt* and

rewritten in ten

t { p *-w * ( a r ? - a'r?-oo)=
-='v7 ^ " * (Pr -) - (Ao + LA,)
aP? s' [Er*, aPr' r' ) (.- dPr, )

trinally, the following is arranging the above

tained by rearri

aPg)
+ (w
)new - ; - q p :- - ^ o (3.44)
8i
[dPt'
aPz
8i
8, a'.)

^ 1 dt Po'
Here the term un
6rtgt ! is ignored,being small.
 Economic Inad Dispatch of Thermal Generating Units 15s

SubstitutingEq. (3.41)ir

NG

i=l

By separatingthe Prl"*'
Ncr
>l (3.4s)
i=l \

(
tDg+i -- l l rr
\

Substitutingthe value of Pi, i to Eq. (3.45),

NG
I Pi,= P ; (3.46)
i=1

w h e r eP i = P p + P f - Y
a4
i=L aPr
Substitutingthe value of

I 1,,r_ -ll

l + + L o azI
dp
\- si - inew (3.47)
(
1 1 -
\
The aboveEqs. (3.47) ) can be rewritten of equalincrementalcost,
when the transmission of considered,as

(3.48a)

(3.48b)

(3.48c)
 Power System O

(3.48d)

Pi, P;"* (3.48e)

P; - P D + P tf P " :'
iJ dpr, si (3.48f)

In view of the defi problem in Eq. (3.12),the problemin Eq. (3.48) is redefinedas
zai f|, + bi = Lnew (3.49a)
and

NG
P;, = P;
i--L (3.4eb)
where

NG
= 2, ...,NG)
j=l Q.a9c)

* ar + ho Bii
a;= (3.4ed)
(1- KT,

,* bi - 1oBiiP!,
D ; =
(3.49e)
I l - Kr. 0,
p*
'8i - (1 :P)
KI r Til'*
)P (3.4ef)
\IC
NG N
p 0 -
t L \ IP!,
I >, r, 11.
B ..Po
BuFi
I gi
f = l j=
;- 1 Q.aee)
NG
D * -
t D - PD +r P f - IIr(,9
r P:,
i == ll
(3.4eh)
The problemin Eq. (3. 9) can be solvedusingEqs. (3.9) and (3.10)
as explainedin Section
3.3. The stepwiseprocedure implement the efficient method is outlined
below (Algorithm 3.5.).
Algorithm 3.5: Economic ispatch Using Efficient Method
1. Read,data,namelya br, ci (costcoefficients);Bu(B-coefficients)
(i = l, Z, ..., NG; j = l,
2, ..., NG); e (conv )nce tolerance);and ITMAX (maximum
allowableiterations),etc.

*-d
 Economic Load Dispatch of Therrnal Generating units

2. Compurethe inirial ues of Ps,(i- 1, 2, .-., NG) and A, by assumingthat pt= 0. The
valuesof 2" and psie , 1,2,..., NG) can be obtaineddirectlyusing
Eqs.(3.lOiand (3.9),
respectively.
3. Set iteration counter
4. ComputeKl, ai, bi, , ilrd Pj usingEqs.(3.49c),(3.49d\,(3.49e),(3.499),and(3.49h),
respectively.
5. ComputeL** , Pi,, PrI* usingEqs.(3.10),(3.49a),and (3.49f),respectively.
6. Check I Lo* - Lo I s if 'yes' thenGOTOStep8.
7. Modify Pro = Ps:"* (i 1,2,...,NG)and1.0= i.,"*
IT=IT+lGOTOS 4 and repeat.
8. Compute the optimal I cost, and the transmissionloss.
9. Stop.
The limits of generationscan added by the procedureoutlined in Section 3.4.

EXAMPLE 3.7 Determine economic scheduleto meet the demand of 150MW usingthe
Efficient method. Use the data f Example3.3.

Solution Initial values calculatedin Example3.3 are:

ho = 11.81812
Rs/IvI , Prl = 83.5276
MW P:, = 66.47239
MW
Find the incrementaltran ssion loss with initial values of generation,i.e.

? =zft,Prl +2npP!, = e.t404663

8=2BrrPrl*znzzP!,= 0.233247gs
Calculate transmission

P 2 = B n )'+ Brr7:)z +Znr2r!rr!,= 13.5931

Mw
Find the constants in Eqs. (3.49),i.e.

= 1j * 7!!r: =o.o11oze
Rsrrrw2h
(1 - Kl)z

az * LoBn
= 0.232479RsfirdW2h
(1 - KD'
bt - to nnP!,
=9.724716Rs/IUWh
1-r(f
158 Power System

bz - lo nzz$
= 10.02016RsA'IWh
,1 - v t t 2
0

z
- P p + P f t rc?P!,
= 136'40690
IvIw
, i=l

Find the new value of using Eq. (3.10), i.e.

2 u l
P;+L3
i =t l Zai
LUi
=-=l4.8 l9l9Rs/IVIWh
2 I
I, o *
i=l LAi

Calculate new ge NS AS

*
P,, =Ln*
- bi
=90.84408MW ; = hnew- b;
Ptr, = 45.5298IvIW
zai : zai
The required new mod ed generationsare

Pi, Pi, =
P;"* = 105.7283
IvtW; P;:* = I\dW
59.32055
I - ,Kto r-K8
Checklrl'* - Lol= J. 107> 0.001
For iterationIT = 2 ::

P!, - ro5.7z8MW MW 7o - 14.8LgrgRs/lvIWh

P:, = 59.32055
The above procedureis till convergenceis obtained.The generationschedule,1,,{rans-
mission losses,load mis and advancement in 1. during each iteration are given in Table 3.8.

Thble 3.8 ion scheduleduring each iteration (Example 3.7)

ru Psl L P L LP lLnew- Ll
(Mw) (Rs/IaWh) ' (MW) (Mw) (Rs/lvIWh)

2 103.0107 62.8 3 14"87678 15.9:2827 0.0366220 0.0575914

t3
103.3456 62. 9 3 14"86997' I s.92s86 0.0034237 0.0068140
4 1 0 3 .135 7 62.6 134 r4.8704r 15.92699 0.0000038 0.0004434
s 103.3186 62. 8 3 3 14.87035 rs.92699 0.0000095 0.0000582
Cost - 2309.77 Rs/h

3.8 CLASSICAL OD TO CALCULATELOSS COEFFICIENTS

The simple rhethod to dte the penalty factors and total transmissionlossesfor generator
-coefficient'or the 'loss-coefficients'method.This analysiswas

.-J
 Load Dispatch of Thermal Generating Units

initiatedby Kron [95], 1952], popularizedby Kirchamayer tl958l; it hashad numerous

extensions by Happ [963, 1963, 1964, 19641,and has been formulated using efficient sparse
matrix computationtechniqueb Meyer tl97ll. Physically, these constantsrepresentthe
parameters of a fictitiousnetwork hrough which.the.powerflows from the generatingbusesto a
hypotheticalload. Thesecoeffici or constantSare.in fact not constantsbut depend on the
loading conditions as well' as the iguration of the power system.
In classical approach the Bii, B,.eand Bsg coefficients can be calculated directly as discussed
here. The voltage magnitude and phase angles at each bus of the network power system are
obtainedby performing the load w. The variable load or generatorcurrentsare separatedfrom
the fixed network. All shunt elem ts, namely,line chargingcapacitances
from bus to ground and
transformer neutrals, etc. are con ed to dummy bus L as shown in Figure 3.4.

c)
o
9l 1,,
.9
o
N.

Figure 3.4 Powersystem ith separatgdshuntelementsfrom load and generators.

Select the reference bus and unt elementsto bus L to be part of the network. Form fr-bus
using the Z-bus algorithm and zero row and column in order to obtain (NL+NG+I) x
(NL+NG+l) matrix, where NL is number of loads, NG is the number of generatorsand NB is
the number of buses.
Es = \Is (3.s0)
where /s is a vector of load and r currentsand

The voltage vector transform into the new reference Frame I (Figure 3.5) by the conjugate
of Tr, i.e.
frunspose
I 'Eo=(f\r4T)lr=ZJr
I h-(ri (3.51)
I

Iwhere
I (3.s2)
4Tt
0
I
I
pvith
i
II
i
II
L-- _1
 In
+

Transformationof power system to Frame 1.

7
#Lg , B- l )
( k= 1 , 2 ,. . . N (3.53)

| - L; ( k = l , 2 , . . . ,N B - l ) (3.54)
NB -I

\ Loto+t, (3.ss)
k=l

Now,

(3.56)
zu
where /n is the neuffal cu
f1 is complex calcul off nominal turns ratio betweenbus k and the reference bus.
21 is calculated with the fo owing rule [Kirchamayer,1958]:

Zt(n, k) - 4(n, ) - L* Zs(n,L); (n = l, 2, ...,N8; k = 1, 2, ..., NB) (3.56a)

Zr(n,L) = 4(n, ) - A,;Zs(L,L); (n = L, 2,..., NB; L = NB + 1) (3.56b)
Zt(L, n) = Zr(n, ) ; ( n = 1 , 2 , . . .N, B ; Z = N B + l ) (3.s6c)
Eachof the NL load curren is assumedto vary as a constantcomplex fraction of the total load.
I*=l*Ir (k = l, 2, ...,NL) (3.s7)
NL
1 7 -> i r (3.s8)
i=l

where
ft is the complex
/7 is the total load c
 Economic Inad Dispatch of rhermal Generating (Inits

The voltages are accordingto

Ez= ebrE, (3.se)
which maintain the generatorb at €1r- tp€, and defines an equivalentbus

€ 1 -t p , = z t ; @ * - t * e , ) (3.60)
k=l

Transmission
lossis as
Pt+ = (Ihrq - Kl;r;fzrlrrlr)l = (r)r4rz (3.61)

= (T;)rZ{z and Ir = TzIz

=l'o
I
l
LO J
I = llt, 12,...,
/Nilr
irnplement on compute\ 22 be obtainedas (Figure 3.6)

a
4- (3.62)
21

l i r
.+

et- tF,
1,,

Figure3.6 ransformationof power system to Frame 2.

NL NL
w=>zt ztT,ilt
i=l j=L

NL
oi= Z fi zr&,j) l, 2, ...,NB + 1)
/c=l

. . J
 ,.ry
i

162 PowerSystemO

NL
b,= 2 Zre,k)I* (l = NL * 1,..,NB + l)
&=l

4(m, n) = ZrG,I (i = NL * 1 , . . . , N lB=;N L * 1 , . . . . , N+Bl ; m = i - N L + 1 ; n = i - N L + 1 )

The load bus is eliminated ich is .a further dimension reducing transformationto Frame 3
(Figure3.7)
jet = e)'8, = le;rhrq (TrI)) = (I;)rhI3 (3.63)

where
Iz = T{t, Ez = h. Iz

Q.e)
with
/ = [fNl* y 2, ..., fNI-*Nc*t]

. t*io I
( k = 1, 2 , . . . ,N G ) (3.6s)
tk= - , INL+NG+I - - - F
ty ty

NL
tL= \ forr
k=l

To implement on cornputer, can be obtai

2 3 - l-t (3.66)

where
'na^
Zt(n, m) = TrW; - brT; - , ) ( n = 1 , 2 , . . . ,N G + I ; m = 1 , 2 , . . . ,N G + 1 ) ( 3 . 6 7 )
* Z 1 ( nm

INL+l
L+t- ',
,,

En=-2
tt

Figure 7 Transformation'ofpower system to Frame 3.

A
 conomic Inad Dispatch of Thermal Generating Units. 163

It is assumedthat the real and tive powers at each generatorbus are linearly related as shown
in Figure3.8.
*.
€*t (k = NL * 1, ...,NB) (3.68)

rfi
= & m P k +F * (m= k-It[L) (3.6e)
€1a

Qr
Q * = Q o r +s f r
l-{
0)
'
o
P.
(.)
o
Cg
a)
&

Real power

Figure 3.8 Linearlyrelatedreal and reactivepowers.

The real power transmissionloss is expressedas

= Re t(/)r Zt Il (3.70)

where * ftuo]t
= la1P1* Fr,...,aNcPNc

Equatingto the loss formula gi by Eq. (3"r4)

Bii = aiui Re lrz3(t, i)l (3.71)

I
3rj=R" lrwi aj ztL,i) ait 4(i, j)l *o' i = r,2, ...,NG)
(i= 1,2,..1' (3.72)
[ -l
Bro= Reldt"z't(i,NG +' 1) a,ti zt(Nc + l, i)l (3.73)

Boo= l"lj Ro[Z3(NG+ 1, + l)l (3.74)

Algorithm 3.6: Evaluation of -coefficientsUsing Classical Method

1. Perform load flow to find % and Pi + jQi at each bus.

2. Build Zs-busrusingthe Z- algorithm by fixing all charging to n"uou:.common Point, L.
Then add zero row and c umn for the referencebus.
3 . Compute )a, urd f1 using qs. (3.53) and (3.54),respectively.
Compute neutral current and Ip currentusing Eqs. (3.56) and (3.55), respectivelY.
4. Build 21 bus using Eqs. 3.56a)-(3.56c).
5 . Computefu using Eq. (3. 8) and then l* (k = 1,2,..., NL) from Eq. (3.57)).
6. Build Q-bus using Eq. .62).
7 . Computef1 using Eq. (3. 5).
8 . Build Z3-busfollowing . (3.67).

L.*-
 l& Power System )pt

9" Computedilk = l, Z, ... Nc) usingEq. (3.69).

10. Build B-coefficientsfoll ingEqs.(3.7rH3.74).
11. Stop.

EXAMPLE 3.8 Use the clas method to determine the B-coefficients for a 5-bus system
shownin Figure3.9.Bus 5 is as the slack bus. The seriesimpedanceand line charging of
each line is taken liqn Table

0.08+70.20
I .<f
o
O
o
+ +
N
O C\
O
o
0.02+70.06 o

4 3
Flg 3.9 Powersystem(Exampleg.g).
Table 3.9 Power systemnetwork data (Example 3.g)

Impedance Une charging

z afil2
I-2 O.U2+ j0.04 70.010
1-3 0.15+ j0.40 j0.02s
1-5 0.08+ 10.20 j0.020
?-3 0.A2+ yO.06 J0.010
34 0.04+ j0.20 p.020
4-5 0.A2+ j0.M fi.020
Solation The base case wer flow (p"u.100MVA) is achievedfor I:0 x
lOa convergence
in l0 iterationsusing the decou led load flow method. The voltage
and power at each bus are
given in Table 3 . 1 0 .
umber of load buses,NL = l
numberof buses,NB = 5
mber of generatorbuses,NG = 2
Referencebus = 5

Table 3.1.0 S data after load flow calculations (Example 3.g)

Bus
Pi + iQi
1 0.865666- 70.125030 - 0500 - j0.25000
2 0.858383- 70.133531 - 0.400- J0.1s000
3 0.87t962- j0.r25507 - 0.4s0- i0.20000
4 - .10.008090
0.984324
5 0.350+ fl.15000
1.000000
+ 70.000000 1.089+ ;0.55608
 Economic Load Dispatch of Thermal Generating Unrts 165

The total line charging ittance to ground at each bus is given in Thble 3 . 11 .

Table 3.11 Li e charging'admittanceto ground (Example3.8)

Adminance,Yo Impedanct, Z,

I j0.0s5 - jl8.1818
2 j0.020 -jsO.0
3 y0.055 -jl8.1818
4 j0.04 - j25.0
5 jo.M - j2s.0

The bus impedancealg hm is applied to obtain the Zo for the power system shown
in Figure 3.10. All line chargin are fixed to neutral and treated as bw Z (common point). The
bus 5 is a reference bus. The to build the Za-busis given in Table 3.12.

R
0.08+/0.20

+ j0.20 0.02+/0.06 ^$'1,)

oo
@

Figure3.10 Power em configuration (Example

for fr-bus algorithm 3.8).

0.444087 0. 0.029675 0.008838 0.0 0.024534

+j0.123897 +j0.1I 712 + j0.093206 +j0.015219+j0.0 +j0.070400
0.037080 0. 785 0.035772 0.010502 0.0 0.025537
+,r0.Lll7lZ +j0.1 +j0.108 r98 + j0.0r7624 +j0.0 + j0.073160
0.u9575 0.03 0.Mr282 0.012351 0.0 0.024344
+ j0.093206 +j0.1 1 9 8 +j0. 130368 + j0.02r325 +70.0 +j0.072924
4- 0.008838 0.01 02 0.0r235r 0.017940 0.0 0.w9967
+j0.015219+j0.01 624 +j0.021325 + j0.037025 +70.0 +j0.018302
0.0 0.0 0.0 0.0 0.0 0.0
+j0.0 +j0.0 +j0.0 +70.0 +j0.0 +j0.0
0.024534 0. 537 0.0243M 0.w967 0.0 0.017r32
+j0.0702100+j0. 60 +j0.072924 +70.018302+70.0 - j 4.713606
t66 PowerSystemOptimiza

Thble3.1 Data for buildingZs-bus(Example3.8)

"Line Sending Ending 2"., Type of link
bus bus (p.u.) frcm bus to bus
1 R I 0.08+ j0.20 I (new to reference)
2 1 2 0.02+ jO.M 2 (new to old)
3 2 3 0.02+ 70.06 2 (new to old)
4 3 4 0.04+ j0.20 2 (new to old)
5 4 R 0.02+ j0.04 3 (old to reference)
6 3 I 0.15+ j0.40 4 (old to old)
7 1 L -jl 8.1818 2 (new to old)
8 L 2 -js0.o 4 (old to old)
9 L 3 -jl8.1818 4 (old to old)
l 0 L 4 -jzs.0 4 (old to old)
11 L R -jzs.0 3 (old to reference)

From Eqs. (3.53)and (3. , the values of ),1,and tp arraobtainedand are givenin Table3.13.

ThbI 3.13 )" and t values (Example3.8)

)"k tp

I -0.01 16 + .p.005259 1 . 0 1 4 9 1+6 70.005259

2 -0.01 628 + j0.005475 + j0.@5475
1.015628
3 -0.01 2 + 70.005221 1.015452+ j0.005221
4 -0. 875 + j0.002r29 + j0.002129
1.003875
5 0. + "10.000000 1.000000+ 70.000000

From Eq. (3.56) the neuff culrent, In = -0.000771- j0.212149.

\\e 2"1,are used in matrix i to ftansfer the systemto

0.o4t823 0. 849 0.03M22 0.009083 0.0 0.M9579

+j0.r248r8 +j0. 12678 + j0.094166 +j0.01s440 +j0.0 +i 0.000r80
. 0.037849 0. 7588 0.036552 0.010758 0.0 0.051609
+j0.112678 +j0. 35479 +j0.r092M +j0.017855+j0.0 +j0.000188
0.03M22 0. 6552 0.042039 0.012601 0.0 0.M9217
+j0.094166+j0. w2M +j0.131368 +j0.021556+i0.0 +j0.000179
Zt=
0.009083 0. 10758 0.012601 0.018018 0.0 0.020067
+j0.0r5440 +j0. 17855 +j0.021556 + j0.037075 +j0.0 +j0.000073
0.0 0. 0.0 0.0 0.0 0.0
+j0.0 +j0. +j0.0 +j0.0 +j0.0 +j0.0
0.0 0. 0.0 0.0 0.0 0.017132
+j0.0 +j0. +j0.0 +i0.0 +j0.0 - j4.713606

-_-{
 Ecorwmic Load Dispatch of Thermal Generating Llnits 167

The load current is defi AS the sum of bus injection culrents.Using Eq. (3.58)'

Ir -r.426628- i0.903495
I1 = -0.524929- i0.3646r2
I2 -0.4284r - i0.241396
,: = -0.473258 - i0.297487
UsingEq. (3.57),we
f,
- j0'016095
Ir = + = 0.378144
rT

+ j0.014g75
- 1 9 7 = 0.290831
b=

lz=
- 1! 7 = 0.331025+ j0.001116

The ratios I; are used in trans tion T2.Using Eq. (3.62), 4 ls obtainedas

.,7
z4-

UsingEq. (3.65), we calculat

tq= L015304+ 70.005298
/r = 0.988727
+ 70.003063
tz= 0.984900+ 70.005139
fg= 0.984900+ 70.005139
In Step 4, the turn-ratios are u to definetransformation73,to eliminateload. Equation(3.67) is
used to build 4 as
+ i0.092r90-0.00n80+ i0.091959
rr7r} 0.026638
+ i0.109709 -0.012408
0.036885 + i0.r09302
+ i0.1,W709 0.W724 - i4.6M303
0.036885
The currents are calcul as implicit functionsof the real power using Eq. (3.69),
u r = r.0r2279- i0.4437r7
e z - 1.000000- i0.510634

The B-coefficientsare tainedusingEqs.(3.71),(3.72),(3.73)and(3.74)as

B t t = 0 r49MW-r, Bn = 0.03277MW-r
Bzt -- o 3277MW-l, Bzz= 0.04650MW-l
B o t = 0 t 2 MW, Boz= 0.00255MW Boo= 0.00021MW

L
 168 Power System

3.9 LOSS COEFFICI CALCULATION

USING Yaus
Another method to evaluate Ioss coefficientsis presentedhere.The method inverts the remote
equivalentindependent
(RED ix [Meyer, l97t) to obtainthe loss coefficienrs.
Exploiting the load flow ysts, we compute the phase angle and voltage magnitudefor
everybus in the system.The and reactiveloads are convertedinto lumped elementsso that
the systemis linearized.
All utotransfonnersare convertedto fI equivalentssuch that shunt
currentsto groundmay be co bined at adjoining buses.
All the load busesare nected to a common node called Z node. The potential at the L
nodeis arbitrarilyset to zero. network will appearas shown in Figure 3.1I .

':
r1<--

Network

Figure3.11 Lumpedloa connectedto a common node L, where L1 to Lp1 are the

lumped

An artificial node R, is whose injection current and power is equivalentto the NL

load buses,i.e.

NL
0R-
r
Ltt (3.75)
i=l

NL NL

Iq=I
d .tr
- €iIt
vl(
(3.76)
i=l i=l

^SR
.+ (3.77)
Ip

Equivalent admittancefrom R e to 'zero potential' Z node is calculatedas

Ynr
.si
tt (3.78)
€ReR

The network can be redrawn as Figure3,I2.

The system nodal equation are

f -
IBUS - Esus (3.7e)
 Load Dispatch of Thermal Generating Units

+iNL

Network

Figure3.1 Remoteequivalentindependent
of a network.

wherc
tu[-NL+NG+2
/sus = bus injection = [itr.., fp;, tL, jNL*I,...., iNr*Nc, inJr

Esus= bus voltagevec = [0,.., 0, fNl*1,...., INL+Nc,in]?

Eliminating zero injecti currentsfrom Eq, (3.79),

/sus= lYzz- YnYll frzl Egus (3.80)

Y22is partitionedmatrix f s i z e ( N G +I x N G + 1 )
Y21is partitioned matrix f s i z e ( N G + lx N L + 1 )
Yp is partitioned matrix f s i z e ( N L + 1 x N G + 1 )
hr is partitioned matrix size(NL+lxNL+1)
Another admittance Y3 is consideredby eliminating bw R elementsfrom the admitt-
ancematrix of Eq" (3.80). inverseof Y3is the required impedancematrix, i.e.

4= Ytr (3.81)

Transmissionloss can obtainedas

PL- ne t(4 )l'ztttl (3.82)

Re(Ir,) = 2,..,NG)

P; as the power injecti at the generatorbus

I V; I as the voltage mag itude of generatorbus

The above procedure is trated with Example 3.9.

 :il
L70 Power System Opti,

EXAIIPLE 3.9 Usethe Isus to determine for a 5-bussystemshownin

the B-coefficients
Figure 3.9. Bus 5 is taken as slack bus.

,Solution Performing I flow, the voltageand power at all busesare calculatedand given
in Thble 3.14.

Thble 14 Load flow solution (Example 3.9)

P+JQ
I -0.50 - j0.2s - j0.r2s030
0.865666 -0.524929+ j0.364612
2 -0.40- 70.15 - 70.133531
0.858383 -0.42844r+ j0.241396
3 -0.45 - j0.20 - p.L2ss07
0.871962 -0.n3258 + j0.297487
4 - 0 . 1 s- j 0 . 1 0 - p.008090
0.984324 -0.151544+ j0.102838

I-oad buses1, 2, 3, and are convertedinto lumped admittancesas below:

L= -0.653587 + j0.326793
= -0.530f,/;6
+ j0.198767
= -0.579845+ j0.257709
= -0.154805+ /0.103204
IJsingEq. (3.78),
= 1 . 9 1 7 8 8-1 j 0 . 8 9 5 0 1 1
lfhe network will be as shownin Figure 3.13.
ifhe system admittance x is obtainedusing the l/-bus algorithm as
v
IBUS -

2r.93304 - 10.0 -0.82192 0.65359-1,1.76471 0.0

0.0
- j24.75rr7 +j20.0 j2.r9178 - j0.32679 +j0.0
+j2.94rr8 +j0.0
-'.10.0 14.46995 - 5.0 0.53005 0.0
0.0 0.0
+j20.0 - j34.78123 i15.0 - j0.19877 +70.0
+j0.0 +j0.0
--0.82192 --5.0 6.20362 0.57985 -0.961,54
0.0 0.0
+.j2.19178 +j15.0 i2t.68676- j0.25771 +j4.80769
+j0.0 +j0.0
0.65359 0.53005 0.57985 0.0 0.15481 0.0 - 1.91788
- j0.32679 -j0. re877 j0.2s77| -j0.00854 -j0. 10320 +j0.0 +j0.89501
0.0 0.0 -0.96154 0.15481 10.80674 - 10.0 0.0
+70.0 +70.0 j4.80769 - j0.r0320 -j24.6w9 +j20.0 +70.0
-17.7&71 0.0 0.0 0.0 - 10.0 2r.7&71 0.0
+j2.94118 +j0.0 j0.0 +j0.0 +j20.0 -j22.90rr8 +j0.0
0.0 0.0 0.0 - 1.91788 0.0 0.0 I .91788
+j0.0 +j0.0 j0.0 +i0.89s01 +j0.0 +j0.0 +j0.89501

^d
 Economic l-oad Dispatch of Thermal Generating Units 171

Ynr

Figu 3.13 The power system(Example3.9).

Partitioningthe Ysusma x into four matrices,namely,Y'1, Yp, Y21,Yzzgives

2r.93304 10.0 -0.82t92 0.65359
- j24.75r|7 + 20.0 +j2.r9r78 -i0.32679
- 10.0 14.46995 - 5.0 0.s3005
+ j20.0 4.78123 +715.0 -i0.19877
Ytt =
-0.82192 5.0 6.20362 0.s7985
+j2.19178 + il5.0 - jzr.68676 -i0.2s77r
0.65359 0.53005 0.57985 0.0
- j0.32679 j0.19877 - j0.2s771 -i0.008s4 4 x 4
0.0 1 .7&71 0.0
+70.0 +j .94118+70.0
0.0 .0 0.0
+70.0 .0 +70.0
v, 1 2 -
- 0.96154 .0 0.0
+j4.80769 .0 +j0.0
0.15481 .0 -1.91788
-j0.10320 .0 +j0.895014 x 3
0.0 .0 - 0.96154 0.15481
+70.0 .0 +j4.80769 -i0.10320
-17.7&71 .0 0.0 0.0
Yzt =
+j2.94rr8 .0 +70.0 +70.0
0.0 .0 0.0 - 1.91788
+70.0 .0 +70.0 +70.89501 3x4

10.80674 10.0 0.0

- j24.6W9 i20.0 +70.0
- 10.0 2r.tu7r 0.0
Yzz=
+j20.0 i22.90118+70.0
0.0 0.0 1.91788
+70.0 j0.0 +70.89501
3x3
 t
Power System Opti

FromEq. (3.80),the

Eliminating the referencebus elements

-8.422v1+il8.7914801
y a = [ t o ' l o a s r c -.17223
i2
" - i24.49slsol
+ jr .7sL48Ors.7276s}
L- 8.M2224
'The
inverseof Y3 matrix

+ 70.
lo.o+ozl5 + i0.027e74f
66 0.034555
4- + 70.036764)
fo.orotts+70.74 0.M6476
Using Eq. (3.82), B-coeffi ients are real valuesof Z matrix, i.e.
Bn = 0.M1'5240 B 0.035t046
Bzt = 0.035t046 0.0464764

3.10 LOSSCOEFFICIE USINGSENSITIVITYFACTORS

3.10.1 DC Load Flow
The complex power flow from s i to 7 is expressedas

-. l-v, vil--
sr=v,LTl (3.83)

where
'Z^ is the impedanceof branch connecting the ith and 7th buses
% i s the voltage of the i bus
7; is the voltage of 7th
'Ihus,

= * t l v ? l - l v ,l l v ,|/ ( o i - o i ) l

where
llv; I is the magnitudeof Itage of the ith bus
llv; I is the magnitudeof oltage of the 7th bus.
Therelore,

','.1- 14 V;
| ll I {cos(0i - 0) + 7 sin(0i - gj)}] (3.84)
 Economic Inad Dispatch of Thermal Generating Units

The reciprocal of conjugate i ance ls

R^+iX^
= { i Ix * =Rm
*+iX^
W

R. is the resistanceof mth hne

X, is the reactanceof mth hne.
the above equati
Subst.ituting into Eq. (3.84),

s,Su
..= = l y r i l v , t ( c o (s0 i - 0 , ) + 7 s i n ( 0 i - 0 i ) ) ] ( & , + i x ^ ) (3.85)
_Tllv,'l
The real part of Eq. (3.85) is activepowerflow from the sendingbus i to thereceivingbus7,
i.e.

P^=Pi t v i z t - t v , n v j l ( R - c o s( 0 i - 0 ) - X - s i n( 0 i - e j ) ) ] (3.86)

Assume
(i)lV;l=lV1l=1
(ii) X*D R*
(iii) (0i - 0) is small such cos (9, - 0) = 1 and sin (0i - 0i) = ei - ej

Thus,,Eq. (3.86) can be rewri n, in view of the aboveassumptions,

as

1 - -
P^- Pri= A LRrn
tLml

o,-oi (3.87)
x^
Generalizing,we have

) (3.88)
J

where k is the number of rlv connectedbranchesto i.

In the matrix form
O = [ i lp
a,0- tn LP (3.8e)

3.10t.2PowerLoss in Line
The active power flow from e sendingbus 7 to the receivingbus i can be expressedin the
similar way as Eq. (3.8O,i.e.

P^ = Pii t v j l z - t v i n v j l ( R . c o s( 0 i - 0 , ) + X " , s i n( 0 i - e j ) ) ] (3.90)

174 Power System Opti rcn

So, power loss of line can be ined as

PL*- Pi1+ \i

= f f |i | (,l, , Y
,,
t +l t v j l 2 ) R ^- 2 l v i l l v j l R , c o s( 0 i- e ) l (3.e1)

Equation(3.91)is rewrittenas

f z n ^ - 2 R ^ c o s ( g- 0; ) l ('.' It v i I = l V i l =11)
lz^l

2R
Rk+ (
c Q = =')z, s
1 -' ccos ,in2
$ r)
-E
4R^
l x* R^ 2',
on 3.10.t
l _ , Sectior

_4R^
E t'.' (0i - e)n is small,Section
3.10.2f

_ R .(
I\
The above equation can be tten as a function of real power generation by substituting
Eq. (!i.87) into it. Thus,

PL,n = R^P} (3.e2)

3.10.3 GenerationShift Distribution(GSD)Factors

Using a DC load flow model, t GSD factor is expressedas

A(m, i) = =*t#
+=#(+) #)
From Eq. (3.89),it is concl that and
W=xii W=xr,r.rhus,

A(m,rr=+ (3.e3)

where,
P, is the real power on transmissionline m from sendingbus 7 to receivingbus k.
X7i and Xpi are the el of the X matrix.
X. is the reactanceof li m.
 Economic Inad Dispatch of Thermal Generating Units 175
In Eq. (3.93), since all changesare compensatedby the reference bus, the total
systemgeneration
is assumed
t be unchanged.
NG NLOAD

I Pr,= Zr, = consrant

i=l i=l

where
litlG is number of genera
I\ILOAD is numberof I
Using GSD factors, the n line flows, after rescheduling generation, can be expressedas

NG
+) A(m,i) Ms, ( m = 1 , 2 , . . ,N L ) (3.e4)
i=l

where
P: is the basecase line fl
I,IL is the ntrmberof lines in the system.
Ilv this incrennentalsu tion, any changein line flow due to changein generationcan
easily be calculated.

3.10.11GeneralizedGen ation Shift Distribution(GGSD)Factor

Since all the generationchan are absorbedby the referencebus, the total syster4generation
remains unchanged.If not, the initial line flows are known quantities,a new load flow
must be executed to re-establish initial values.That is, the solution accuracyof GSD factors is
guaranteed only when the total tem generationremains unchanged.A new sensitivity factor,
the generalizedgenerationshift ibution factor (GGDF) was developedby Ng tl98ll to update
the GSD factor. Instead of an i tal form, an integral generation form has been defined as

NG
P* D (m, i) Ps, ( m= 1 , 2 , . .N
, L) (3.es)
i=l

where
Lt(m,f) is the GGDF
IttG is number of generati buses
IttL is number of lines.

3.10.5i Derivationof GG F
From GGDF, if a particular generatork is increasedin generationby
some ne ru will be

NG
P'rn=I O i) Ps,+ D(m, k) Msr (m = 1,2,..,NL) (3.e6)
f=l

I
L
t76 Power SystemOpt nn

The referencegenerator(R * k) will adjustthe changein Mro by decreasing

its generation
by
LPro. The real power flow orr line m afterthis generation
shift will be
NC
F ^ =| n ( m , i )'r,* D(m,k)Mro - D(*, R)APs& ( m = 1 , 2 , . . .N, L ) (3.e7)
i=l

SubtractingEq. (3.95)from . (3.e7),

F^ P* - [D(m, k) - D(m, R)] Mrr (m = l, 2,...,NL)
or
P* m - D(m,
k) - D(m, R) (m = l, 2,...,NL) (3.e8)
M

The changein line flow with pect to the generationchangeis termed the GSD factor.So, the
above equationcan be rewri n a s

A( k) = D(m, k) - D(m, R) (3.ee)

p
- m P* = A(m,k)Mso (3.100)
Ther,eis a need to calculate R). By shifting all the generationsfrom all the generators
to the
referencegeneratorR, i.e. by aking
NG
-\f tDg i - ^AgpR
.L
i=l
i*R

Equation(3.100) can be rewri AS

NC
p P^- - L ot*,i) Ps, ( 3 .1 01 )
d=l
i*R

FromiEq. (3.95),
NG

Y o<*,i)Fr,+ D(m,n)Pr^ (3.r02)

i=l
i*R

when:
Pr, is final generationfi m generatori, which is now reducedto zero

Pr* is final generationfrom the reference generator which contains the total system
generation.
Equation (3.102) is red

NG
'*= D(m. R)Psn= D(m, R) Pr, (3.103)
i=l

----da
 Economic Load Dispatch of Thermal Generating Units

SubstitutingEq. (3.103)into Eq ( 3 . 1 0 1 ) ,
NG NG
D(m,R) = pm- I A(m,i)Ps,
I d=l
i=l
i*R

or
NG
P*-) a(m,i)Pr,
i=l
. d*R
D(m, (3.104)

where
P. is basecaseflow of m
Pr, is generationof the i generator.

3.10.6 Evaluationof cients

The power loss expressedby usi B-coefficientsand bus generationin a quadraticform was first
developedby Kusic [1996] as
NG NC
Py- I L Bi Pr,P, (3.105)
i=l j=l

The incrementalloss can be o ed by differentiatingwith respectto Pr,, i.e.

{+ = y 2B,iP,, (3.106a)
dPr, fr
Taking the derivativesof Eq. (3

(3.106b)

We know from Eq. (3.92),

PL*=rt^P3
The systempower loss is ex by the summationover all transmissionlines as
NL
pr;=Z n^Pl (3.107)
m=l

SubstitutingEq. (3.95) into Eq. (3.107),

NL
Pt= (3.108)
m=l
 178 Power SystemOptimi.

Takinlgthe derivativesof Eq. ( .108)with respectto Pr,,

NL NG
EP
AP
= I ^ '\ tzr(m,i)D(m,j) Pr,l (3.10e)
m=l j=l

and

NL
a2p
'r,aPr, 2 R^ D(*, i) D(m, j) (3.I 10)
m=l

ComparingEq. (3.105)with (3.110),

NL
s,j= I *,, ,i) D(*, j) (3.11)
m=l

where
,R, is the,resistanceof li
.D(m,i) are Generalized ion Shift DistributionFactors.

Algorithm 3.7: B-coefficientsUsing SensitivityFactors

The stepwisealgorithm to cal te B-coefficients is outlined below:
l. Form the reactancema using Zsvs algorithm.
2. ComputeA(m, i) using. (3.e3).
3. ComputeD(m,R) using . (3.104).
4. ComputeD(m, i) using . (3.ee).
D(m, i) = A(m, i + D ( m ,R ) ( m= 7 , 2 , . . . .N, L ; i = 1 , 2 ,. . . ,N G )
5. ComputeBy using Eq. ( . 1 1 1 ) .

EXAIIPLE 3.10 Using the DF method, determinethe B-coefficients for a 5-bus system
shownin Figure 3.14. Bus 2 is taken as the slack bts. The line data is given in Table 3.15.

2 5 4
Figu 3.14 PowerSystem (Example 3.10).

L---,,- -,. -- . 2

_,.4
 Economic Inad Dispatch of Thermal Generating Units t7g

3.15 Line data (Example3.10)

Sending Ending Z(series)
bus bus
I I 2 0.060+ j0.206
2 2 5 0.160+ j0.524
3 5 4 0.066+ j0.236
4 2 4 0.210+ j0.694
5 4 3 0.212+ 70.806

llolution From the base ase load flow, the line flows are obtainedas given in Thble 3.16.

Thble 3.1.6 ine flows from basecasedata (Example3.10)

P^ '^ + jG^
PL* + jQL* tv^l fl
I 58.00 + j39.023 86.0 + j2O.0 1.015 rl.l139
2 12.78 10. + jr3.549 46.8 + /5.8 0.980 7.9999
3 -04.77 34. + jl0.528 0.0 + J0.0 0.953 15.6829
4 8.00 0. + 710.884 36.2 + /3.0 0.981 5.8130
s -33.07 0. + j25.343 17.4+ j12.0 0.972 6.3267

lb obtain the Z-bus,the Z- us algorithmis used.The data for Z-busis describedin Thble3.17.

Thble 3.1 Data for Z-busbuilding (Example3.10)

Z(series) Typeof link

1 n-R 0.060+ j0.206 I (new bus to slack bus)
z R-5 0.160+ j0.524 1 (new bus to slackbus)
3 R-4 0.210+ j0.694 I (new bus to slack bus)
4 +3 0.212+ 70.806 3 (new bus to old bus)
5 54 0.066+ j0.236 4 (old bus to old bus)
Note: SB-Sending bus node, ing bus node

T'he reactance matrix is partof Z-bus,i.e.

imaginary
.206 0 0 0 0
0 0 0 0
X= 0 jr.r692 j0.3632j0.2s00
0 j0.3632j0.3632 j0.2s00
0 j0.2500 j0.2500 70.335
The GSD factors are calcu ted using Eq. (3.93) for generatorl, l e

xrr xzr = 1.0

A(1, ( m = 1 , i = 1 , j = 1 a n dk = 2 )

For generatorsl, 2 and 3, the G factors are calculatedand are tabulatedin Thble 3.18.
 Power System Opti

Table3.18 GSDfactors

A(m A(m 2) A(m 3)

1 1. 0.0 0.0
2 0. 0.0 -o.477099
3 0. 0.0 -0.47966r
4 0. 0.0 -0.523343
5 0. 0.0 - 1.0

Using Eq. (3.104),the val of D(m, R) are calculatedas

8.0- (1.0x 144.0+ 0.0x 34.60) -0.454382
D(r, =
144.0+ 10.668+34.60
D(2, R) = 0"154742; R) = 0.062484;D(4, R) = 0.137940;D(5, R) = 0.008084
Similarly,ttr" GGDF are culatedusing Eq. (3.99) and are tabulatedin Table 3.19.

3.19 GGDF factors(Example3.10)

D(m 1) D(m 2) D(m 3) Rn,

I 0.545618 -0.454382 - 0.454382 0.060

2 0.1547
42 0.154742 -0.322358 0.160
a
J 4.062484 0.062484 -0.411770 0.066
4 0.137940 0.137940 - 0.385403 0.210
5 0.008084 0.008084 -0.991916 0.212

UsingEq. (3.111), the B- icientsare calculatecias

Brr = [0"06x (0.54 18 x 0.545618)
+ 0.16x (0.154742x 0.154742)
+ 0.066x (0. + 0.21 x (0.13794x 0.13794)
x 0.062484)
+ 0.212x (0. = 0.02596
x 0.008084)l
Bn = 0'06 x (0.54 x -0.454382)
+ 0.16 x (0.154742x 0.154742)
+ 0.066x (0. x 0.06248$+ A.2r x (0.13794x 0.13794)
+ 0.212x (0. = -0.006777MW-r
x 0.008084)
Siimilarly,
Bn = -0.037441M I t

Bzt= -0'006777M I , Bzz = 0.020486MW-l , Bzz= -0.010178MW-l

Btr = -0.03744LM I , Btz = -0.010178MW-r, 833= 0.280279 MW-t

3 . 1 1 T R A N S M I S S I O N L SS COEFFICIENTS
An exact transmissionloss fo ula has beenderivedby Dopazo et al. fl9671,usingbus powers
and sy'stemparameters.The I systemloss is the sum of the bus powers,i.e.
N B N B
Pr+jQr= Ii=lt ' v,Ii (3.r12)
i=l
 Economic Load Dispatch of Thennal Generating units

INB is the number of

,Pais real powerlossof powersystem
8t is reactive power loss f the power systern
]l/; is the voltageof the i
.i; is the injected power the ith bus
llor the network shown in Figure3.15,
NB
v,=Z zuri (3.113)
j=l

3.15 Power system network.

where ,Zii are the clements of i matrix,i.e.

Zi = Rij jXA (l = 1, 2,..,NB; j = 1,2,...,NB)

Eq. (3.113) into Eq

Substitruting (3.rr2),
NBNB
Pt+ier= > I zuIj ri (3.114)
j=l i=l

Equation(3.114) can be rewri in rectangularform as

NB NB
Pt+iQr=I I ( t r , t c o s g ;+ i l l j l s i n g ; ) ( tl r t c o s0 i - j t l t t s i n g , ) (3.il5)
j=I i=l
where
0 i = , 6 i- 0 i
 *r,t

t82 Power System O

4 is voltage angle at bus.

,o,
0,= tan-'T
Equation(3.115) is exl nded as
NB
Pr+ iQr= Z U II i l l I j l ( . o t 0 , c o s0 i + s i n9 , s i ng r ) (3.116)
j=l

NB
+j 2 tr l /, ll 1, | (cosg; sinei - coso, sin0,)
I .l=l

Equating the following

) NB /Ns )
z i j t l j sl i n g r l =) r / ; l c o s r r l l z i ; t t i sl i n o i l ( 3 . 1 1 6 a )
) j=t \i=t )

and because Zsg5 is symmetri matrixn so Z;i = Zii.

SubstitutingEq. (3.11 into Eq. (3.116) and simplifYing

NB NB
Pr+ jQt= (3.117)
i=l ,l=l

Substitutingthe following in . (3.117) and separatingthe real and imaginary parts

Zij = Rij jxu
cos (Q - 0) = cos cos 0; + sin 0i sin 0j
NB NB
pr= R i | / , l l / j I c o s ( 0 -r 0 i ) (3.118)
I
i = 1 i=l
NB
-oi) (3.1le)
h=Z Xij I Ii ll /j lcos (9t
i=1 =1
We know o
.s.
S;=lVilllil or lfl= (3'l2o)
W
From Figure 3.16,
Pr= S; cos Qi (3.r2r)
Qi = S; sin Qi (3.t22)
From Eqs.(3.120)and (3.121

-P.
f
l f - l =
' -
(3.123)
l V ,l c o s @ ,
Figure 3.16 Powertriangletor ith bus.

" --d
 onomrc Inad Dispatch of Therrnal Generating Units lg3

FromEqs.(3.120)and(3.tZZ),

r{l=eil (3.r24)

Substiturting
Eq. (3.123)into Eq. ( 3 . 11 8 ) ,

S$ D 4Pi cos(o,-oi)
^'i (3.125)
* fr wltvi I t* 4'r* oI
Equation(3.125)can be rewrittenin terms of B-coefficientsas

NB NB
PL- I I Pi Bij Pj (3.126)
i=l j=l

c o s ( 0 ;- 0 i )
B,
lvi ll V, I cosQ, cosQ;

P = Ps,- Pai (i = I ,2,.., NB)

Similarly, substitutingEq. (3.124) into Eq. (3.119)

c o s ( 0 ;- e i )
er= sinp, sin@r.
(3.r27)

Equation(3.127) can be rewritten

Q r -I I Q,cuQi (3.128)
f=l j=I

c o s ( 0 ,- 0 i )
gii
sin @,sin p,

Qi Qs,- Qa, (i = 1, 2,..,NB)

Let NG be the number of erating busesand for rest of buses, Pr,= 0.0.Equation(3.126)
be rewritten as

NB NB
P L -I > (ps,-po)B,i
i=l .l=l

NB NB NC INc
=I L po,Bih,*Z
i=l j=l i=l
IT
L,t=t
Pr,BiPr,
 184 Power System O

NB NB Nc lrvs I NG NG
= L.tT LtY Po, B,j -I
i=l j=l l I @ u + B i Pr,
i=l Lr=t
) h*I, l
j=l
I
j=l
pr,Bipr,
J
The above equation be written in the form of Kron'slossformula
NG NC NG
,-= Boo+
I B,oPr,
+f Z ,r,Bipr, (3.tzg)
i=l r'=l ,l=l
where
NB NB
Boo=I I Po,B,iPo,
i=l ./=l

NB
Bio=-I @u+ 81) pa,
j=l

Similarly from Eq. (3 128), we can obtain

et- coo
+ X o, er +f X e* c,ier, (3.130)
j=l i=l j=l
where
NBNB
Coo =
i=l j=l

NB
cio= -I (cu+ ci) ea,
j=l

The system power I are based on the assumptionthat (i) the generatorbus-voltage
magnitudesand angles are c stantand (ii) the power
factor of each sourceis constant.However,
the: use of loss coefficien can accountfor any change
in load demand at the buses while
schredulingthe generations the system.

3.112TRANSMISSION
LOSSFORMULA:F UNCTION
OF GENERATIO AND LOADS
. (3.r2s),
(di-d;)-@i-Q)l
cos@,cos@,

- 6 ; ) s i n ( Q ;-
Qi)
cos@;cospy (3.131)
 Economic l-oad Dispatch of Thermal Generating Units 185

R,',
cos (6; - 6); 6ri=ffi sin(6;-6,
can be rewritte , consideringai1 and b;;, as

cosQi C O S ( , * s i n Q i s i n | , sin@;cos@,- cos@;

sin@,
* bii
cos ; c o s @ y cos@;cos@;

( l + tan Q i Q)+bilGanQi-tanQ)f

Substitutingtan Q, = * in the
ri

PL ,,,,1",['.
)iI w + bij

On simplification,
NB NB
= -
t"i1(4Pi+ QiQ) + i1(Q;P1
)i \ 48j)l (3.r32)
i==l j=7

Similarl'yin Eq. (3.127),subsri nE 0i = 4 - Qi and

Xii X,,
sii= -d;), 4ii= -6 j), weget
lfficos(d, ffisin(6,
NB NB
Qt= )f > QiQl [c,7(cot Qi cot j + l ) + d i j @ o t Q , - c o t @ ; ) l (3.133)
i='l j=l

Substitutingtan Q, = * and cot = + in Eq.(3.133),

we get
r; tanp,
l.IlB NB
Qr= ), I ['ii (4Pt+ ete) + (QiPi - PiQi)l (3.r34)
i=l j=l

e abo've method requires the luation of bus impedance matrix.

.13 ECONOMICDISPATH USINGEXACTLOSS FORMULA

economic dispatch problem defined as to minimize the total operatingcost of a power
ystem 'while meeting the total load plus transmissionlosses
within the generator limits.
athemartically,
the problem is ined as
NG NG
Mirnimize F(Ps)= I i=I @ , P i , + b i P s i + rRs/h
,) (3.l35a)
i=l i=l
 Power System

to (i) tLreenergy equation

NG

Ii=lt , = Po,+ P, (3.r35b)

(ii) the inequali constraints

P#'n s P, < P#* (i = 1,2,...,NG) (3.I 35c)

u'here
ai, bi, C; &ta the cost
Po, is the load dema d at the ith bus
Pr, is the real power generation(decision variable)
NG is the number of generationbuses
P1 is the transmissio power loss.
This method uses fact that under normal operatingconditions, the fransmissionloss is
quadraticin the injected real powers. The generalform of the loss formula using B-coefficients
is [Eq. (3.126)]
NB NB
P L - I > PiBijPj MW (3.136)
i=l j=l

where

sQi
Ai= 6i- Qi ( i = I 2, ...,NB)
Pr=Pr,-Po, . (t 1 , 2 , . . .N
, B)

Qr= tan-'(Q/Pt)
4 and P1 are the real power injections at ith and 7th buses,respectively
Qi ffid Q are the ive powerinjectionsat ittr and7th buses,respectively
NB is the number of ses in the network
Zi = Rq + jXii (e of impedancematrix).
The above constrain optimization problem is converted into an unconstrainedone.
Lagrange multiplier meth is used in which a function is minimized (or maximized) with
conditions in the form of ity constraints.Using Lagrangemultipliers, an augmentedfunction
is defined as

, 4)= F(Ps)
+4[X (3.r37)
"-
where 4 it the Lagrangian multiplier.
 conomic l-oad Dispatch of Thermal Generating Units 187

The ndcessary conditions fo optimization Problem, Eq. (3.137), btatethat the derivatives,
the
- l, 2,...,N G) and Lo, are equal to zero, i.e.
with reslpect to control/decision v iables Psr(i

aL(Pvi' ),py a + ^o
2 ( u r ,- r')) = o ( f = I , 2 , . . . ,N G ) (3.l38a)
EP,, t , [a"o
NG
h ,)
aL(Ps,, (3.138b)
D
+Pr
dho i=l
where

(Bii + Bi)P1 ( l = I , 2 , . . . ,N, G ) (3.138c)

'o,*bi (i= 1,2,"', NG) (3.138d)

The solution of nonlinear s. (3.138a)and (3.138b)can be obtainedusing the Newton-

Raphsornmethod in which chan e in Pr,(i = 1, 2, NG) and )"0 is obtained by expanding

Eqs. (3.138a)and (3.138b)a the initial value using Taylor'sexpansion,i.e.

AL (3.139a)
"^ *-^i" *
t = L P8i" . *t y aPs,aPsi i
"6p2 dPr,

NG
dzr (3.l3eb)
j=l
M" odrr

In the rnatrix form the above eq tions can be rewritten as

Y -l o",-l
Io",. ,,^olI o"- --
_-_ f (3.140)
l'T,^ vz",oAolLo^"J
Lor"J
Differentiating transmissionloss (3.138c)with resPectto Pr,,

d, P, - ( i = l , 2 , . . . ,N G ;j = 1 , 2 , - - .N
, G) (3.141)
Bii + B1i
aPsiaPs
and available
All the,derivatives for exPressi s given in Eqs. (3.139a)and (3.139b) are known
(3 141). Using Gausseliminationmethod, Eqs. (3-139a)and
from tiqs. (3.26a-3.26d)and
The
(3.139b)are solved to find chan e in controlvariables,namely PsiQ= 1,2,..., NG) and,ln-
is
controll variables are updated. ne above procedure is repeatedtill no further improvement
achieved. The detailed stePwise procedureis outlined in Algorithm 3.8-

Algorithm 3.8: Economic tch Using Exact Loss Formula

l. Read data: NG is the n ber of buseshaving generators,NB is the number of buses,NV

--
is the numberof PV bu Vr, 6, for slackbus, P7., Qai(i 1,2, . . . ,N B ) . Y i t( t = 2 , 3 , . . . ,N V )

L
 Power System

for PV buses,Vr^tn,Vi (f = NV + 1,NIV* 2,...,NB) for PQ buses.

ey", OiY (i = 2,3,..,
M) for PV buses,4;, ;, c{cost coefficients);i = 1,2, ..., NG. Rl, R2 (maximumnumberof
iterations), 11, E (tole rn convergence), etc.
z. Obtain )reusand by i ing it obtain Zsv5.
3 . Compurethe initial that pL - 0. So, the
of PBi(l= 1,2,..,NG) and), by assuming
initial valuescan be ined directly using Eqs. (3.9) and (3.10).
4 For PQ buses,the t reactivePower demandis distributedto various generatorsin such
a manner that the po factor at all the generator buses remains the same.

Qs'=

:t. Computecost on initi schedule and consider as previous cost (F pt"u).

6. Set iteration counter II - l .
',1
. Compute Pi= Ps,- Pi (i=1,2,...,NB)

Qi-.Qr,-l i Q = l, 2,..., NEI for PQ-buses

only)
Thke Ps. = g, Qr, = O non-generatingbuses.
Perform load flow to btain real and reactivepower, p,,
e; and voltage magnitudeand
anglesI Vil, 4 at each as explainedin Section2.I1.
Check at slack bus that I Pr, - Pd,- Pr l< er, if 'yes' GOTO Step 23.

Compute Pu, = P, + Pa for slack bus and Qr, = Qi +

Qa, for slack and pv-buses.
Compute loss coefficie 84, usingEq. (3.136).
Assume/setthat no ge ion has been fixed either at lower or at upper limits.
Set iteration counter,I - t .
EvaluateHessianand J matrix elementsusingEq. (3.138)and Eq. (3.141).Deactivate
the row and column o Hessianmatrix and the row of Jacobianmatrix
representingthe
generatorwhose on is fixed either at lower limit or at upper limit. This is done so
that fixed generatorsd not participate in generationallocation.
15,, Gauss elimination is used in which triangularization and back substitutibn
processesare perfo to find MsiQ - 1, 2, ..., R) and A/,'. Here R is the number of
generatorswhich can icipate in the allocation.

R
16. Check +( A,r)z s e,
Ii=l ro+,)z
if 'yes' then GOTO 20.
17. Modify P;"* = Ps,*
, ( i= 1 , 2 , . . .R, )^ d A * = 4 + L 4 .
18. CheckIII > Rr,
if 'yes' then GOTO S 20 (without convergence),
elseIII = III + l, ps.= t s : : "(*t = 1 , 2 , . . . ,R ) , =
4 Ltr*
,
GOTO Step t4 and
 Economic Load Dispatch of Thermal Generating tJnits 189

If, no limit is violatedfurther,then GCIIO Step20, elsefix

19. Checkthe limits of gonerators.
the limits as following:

ff Pr,. p#tn then pr, =p#t

tr P, t P#"* then Pe,= P#"*

GOIO Step13.
',20.Computethe optimal loss P2, atc.
total cost 4 transmission
'.21.CheckcostlFPrcv- 'yes'GOTO Step23.
Fl S E, if
122.CheckII > R2
if 'yes' then GOTOStep23 (withoutconvergence)
elseII = II + l, F Prev= 4 GOTOStep7 andrepeat.
:23.Stop.

EXAMPLE 3.11 A 5-bus$ystemis shownin Figure3.17.The seriesimpedance andthe shunt

of eachline are givenin Table3.20. The systemhas three generatorsThe operating
adnnittance
of three generatorsare given below.Find the economicgenerationschedule.
cost characteristics

fi(Pr,)= 60P:,+ 200Pr, + 140 Rs/h

4(Prr) = 75P;,+ 150Pr, + 120 Rs/h

4(Prr)= 704, + 180Pr, + 80 Rs/h

0.02+j0.06
N

o
+
t+
O
0.01+70.03 0.08+.10.24 o

Figure3.17 Powersystemnetwork(Example3.11).

Table3.20 Line data(p.u.)(Example3.11)

Sending end bus Ending end bus Ysn 4,ER
1 1 2 70.06 0.02+ j0.06
,z 1 3 7O.05 0.08+ j0.24
3 2 3 j0.04 0.06+ j0.18
,4 2 4 ,o.04 0.06+ j0.18
5 2 5 0.04+ i0.12
"p.03
6 3 4 i0.02 0.01+ .p.03
'7 5 j0.0s 0.08+ i0.24
4
1e0 Power System Optimization

Solation Ynusis obtainedand its elementsare given below:

6.2s - 5.0 -r.25 0.0 0.0

+ jr8.64 +715.0 +j3.75 +j0.0 +70.0
- 5.0 10.83333 -r.66667 -r.66667 * 2 . 5
+715.0 - j32.33 +75.0 +75.0 +j7,5
-1.25 -r.66667 12.91667- 10.0 0.0
I'Bus=
+j3.75 +j5.0 -i38.640 +730.0 +y0.0
0.0 -r.66667 - 10.0 1291667 -t.25
+70.0 +75.0 -j30.0 -i38.640 +j3.75
0.0 -2.5 0.0 -r.25 3.7 s
+70.0 +j7.5 +70,0 +j3.75 - jll,t70

By taking the inverseof Ysu5,Zsus is obtainedand its elementsare given below:

0:0126019 0.000M94 -0.0042793 -0.0052528 - 0.0051958

- jr .686547 - jr .722815 - j1,736837 - jr .739733 - jr.739s77
0.0004494 0.0052821 - 0.0037327 *0.0038368 -0.0014543
- jr .722Srs - jr ,708342 -- jir.735222 -il.735530 - jr.1284&
-0.0042793'-0.0037327' 0.0089317 0.0045316-0.0047161
Zgus=
- jt.736837 - jr .73s222 - jr .697483 - jr.t 10662 - jr.738149
- 0.0052528 - 0.003 8368 0.00453 16 0.0090162 -0.0032723
- jr .739733 - jr .735s30 - jr .710662 - jL.697234 - jr.733854
- 0.0051958-0.0014543 -0.0047161 -0,0032723 0.0211987
-jr.x39577 -jr.7284& -'jr.738149 - jr.7338s4 - jr.660936

5 5
Total real demand, Pa.= 1.65 p.u. and total reactivedemand, Z Qo,= o'40P'u'
i=l l'=l
The initial I, is obtained as

4= = 253.0147

The initial PEiQ= L, 2,3) and Qr,{t = l, 2, 3) are calculatedfollowing Step2 andStep 3 of
Algo;rithm 3.8 and are tabulatedin Table 3.21 along with loads at each bus. The load flow
solutironis obtainedusing the decoupledload flow method (see Section2.11). The convergence
0.00001 is achievedin six iterations.The voltage magnitudes,voltage angles,real and reactive
pow€rs injected at each bus are obtained and are given in Table 3.22. Loss coefficients are
calculateduging Eq. (3.126). The valuesof angles 4 0, and 0 are tabulatedin Table 3.23. For
exam-ple,Qr = tan-L(QrlPrl= -0.8224847, 0r = 6, - Qr = 0.8224847;and so on for Qz, Qt, Q+
and Qs.

.ra
I Dispatch of Thermal Generaing Units 191

gd to start procedure(Example 3.11)

Pa Qa Type
(p-u.) (p.u.)
0.00 0.00 Slack
0.20 0.lo PQ
0.45 0.15 PQ
0.40 0.05 P8
0.60 0.10 PQ

after six iterations(Example3.11)

6 P 0
ad) (p.u.) (P.u.)
0.4600578 -0.4955128i
08256 0.48674M 0.0664834
03014 0.0715068 -0.023574r
95604 -0.4000043 -0.M99996
28022 -0.6000013 -o.099999E

r 0 and 0 (Example3.11)
g(rad)
Otr"Ol
-0.8224847 0.8224847
0.1357489 -0.1565744
-0.3184559 0.2781545
0.1243527 -0.1739131
0.1651480 -0.237954t

low:

rphsonmethod, elaboratedin Steps9, 10, 11, and

are obtainedin five iterations.The implementation
 r
i tgz Power System Optimization

llteration= l: The elementsof Hessianare computedas

below:
r32.2s9200 0.165799_2.s3r326-0.97r%z Lp s,
-7.326804
0.165783152.374900_ r .591480- 0.gg1264 LP -2.210387
s.
-2.531279 -L591449 144.389900 - 1.004788L p s t t.2tr333
-0.97rMT *0.991264 _ I.004788 - 0.019300
0.0 LLo
llriangularizingthe above equations,we get

t32.259200 0.165799 -2.531326 -0.971u2 Lp s,

0.0 152.374900- 1.588307- 0.990046 Lp s,
0.0 0.0 144.3249 *1,033692 A,p
s,
0.0 0.0 0.0 -0.020966
LLo
Clhange
in generationPr andincremental
cost 4 *" ob-taineS usingback substitution,
'APg,= 4-027130p.u., LPs, = 0.010359 p.u,, Lpe, = 0.034212p.u., N1'-= 3.762gRVp.u.h
The differentconvergence
criteriacan be appliedas
3 5

I"r,-IPo,-P, = 0.45626248-03
> 0.0001
i=l d=l
or
lt7;lt
l> | #l
/.,r,
*l'\ar,)
# | =-'.'-
7.7a826000>
o:oool
llfi[aP,,
)
or
-
,l-.........--...-
,/I f**l' +(M,")2= 3.76299591
> o.ooor
I i=r
No convergencecriteria is satisfied,so to go for next iteration,
i.e. 2, the valuesare updated8s
Pr,= pr, + dPr,. (i = l, 2, ...,NG) and 4= 4* &
Po = 0.441756_ O.V27|3O
= 0.414626p.tr.
Pc,= 0.6g6Vit + 0.010359= 0,697097p.u.
Pg:= 0.521505+ 0.034212= 0.555717p.u.
1, = 253.0107+ 3.7628= 256.7735Rs/p.u.h
' The aboveprocedureis repeatedtill any one convergence
criterionis satisfied.After the
fourth iteration, one of the criteria is satisfied.

.,"d
 Economic Inad Dispatch of Thermal Generating Units 193

The real generationsare given in Thble 3.24 which are used as initial values for the next
iteration. The cost and transmissionloss after the first overall iteration is
Cost = 696.1357 Ryh, PL - 0.01829998p.u. and Lp = 256.9161Rs/p.u.h

Table 3,24 Initial valuesfor next iterations(Example3.11)

'lype
P8 Q8 Pd Qa
(p.u.) (p-u.) (p.u.) (p-u.)

0.412303 0.107092 0.00 0.00 Slack

0.697811 0.166482 0.20 0 .1 0 PQ
0.558186 0.126426 0.45 0.15 PQ
0.0 0.0 0.40 0.05 PQ
0.0 0.0 0.60 0.10 PQ

The above procedure is repeatedfollowing Algorithm 3.8 till the slack bus balance is
achieved.After three iterations, the convergenceis obtained and the final solution is given in
Tables3,,25and 3.26. The slack bus mismatchis

- Pa,-4 = 3's077338-05
< o'cool
lt, |
4 = 256.9665 Cost= 695.8972Rs/tr,Py - 0.0L739662
Rs/p.u.h, p.u.

Thble3.25 Final [email protected])

Ps Q8 Pd Qa P o
(p.u,) (p.u.) (p.u.) (p.u.) (p,u.) (p.u.)
0.4127
r7 0.107092 0.00 0.00 0.412752 -0.499097
o.698472 o.166482 o.20 0.10 0.498471 0.066481
0.556208 0.t26426 0.45 0.15 0.106206 -0.02357
4
0.000000 0.000000 0.40 0.05 -0.399999 -0.049999
0.000000 0.000000 0.60 0.10 -0.600000 - 0.100001

Table 3.26 Voltagemagnitudeand angle (Example 3.11)

Bus V d
(p.u.) (rad)

1.060000 0.0000000
1.071436 -0.0191460
r.069649 -0.o37t636
1.066864 -0.0466896
1.052804 -o.w06622
 t94 Power System Optimization

3.14 ECONOMIC DI$PATCH WHICH

USINGLOSSFORMULA
IS FUNCTION
OF REALANDREACTIVE
POWER
The economic dispatch problem is defined as to minimize the total operating,cost of a power
system while meeting'total real load plus real transmissionlosseswithin the generatorlimits of
real power. Mathematically,the problem is definedin Eqs. (3.135a)-(3.135c).
This method uses the fact that under normal operating conditions, the transmission
loss is quadratic in the injected bus real powers. The general form of the loss formula using
Eq. (3.132)is
NB NB
' PL= - 4e))
eiPi+ eie) +bu(e,r, (3.r42)
ZZU.
j=l
i=l
where
Rii Rii
ai= (6;- 6);6,i=
cos ,
sin(d;-d,
ffi ffi
P i +i Q i = ( P s , - P D +j ( g r , - Q ; ) (i = 1,2,...,N8)
P; and P; are real power injectionsat ith and 7th buses,respectively
: Qi Nd Q1 are reaqtivepower injectionsat ith and 7th buses,respectively
' 'P*,and
Qa Ne real and the reactivepower load demandsat the ith bus, respectively
Po and Qr, are real and the reactive power generationsat the ith bus, respectively
NB is the number of busesin the network
Zij = Ry + jXii ,(elementsof impedancematrix).

, , Using the Lagrange multiplier method, the constrained,optimization problem given ,in
Eq. (3.135) is convertedinto unconstrainedone and is given in,Eq. (3,L37).
The necessaryconditions for the optimization problem,:given by Eq. (3.137), state that the
variables( Pr,,Qr, (; = lrZ,..,NG) and 4), *"equal to
O"riuitiueswith r"rp"r, to controVdecision
: -
zero,1.9.

=Q (f=1,2,...,NG) (3.L43a)
#=#+ho[#-')
(i = l, 2, ...,NG) (3.143b)
6Qgt

NB NG

q = I P a . + P L - I P r ,= 0
AL
(3.1a3c)
i=l i=l

where incrementaltransmissionlossesare expressedas

NB
AP
= hii4. i,; r, + (bii- bi) Qil
[@;1'+ (i = 1,2, -..,NG) (31M)
d E
j*i

L .. .-.'.---_. .--l
 Economic Load Dispatch of Thermal Generati4g Units 195

NB
=; hi,Q,+
I l@u+ ai) Qi + @u- bi) Pil (3.145)
'l=l
j*i

The solution of nonlinearEqs. (3.143a-3.143c)can be obtainedusing the Newton-Raphson

method in which change in Pr,,Qr, (t = l, 2,.., NG) and 1, is obtained by expanding
Eqr. '(3.143a-3.143c) about the initial valuesusing Taylor's expansion.

AL
(i = 1,2, ...,NG) (3.146a)
dP,,

AL ( i = 1 , 2 , . . . ,N G ) ( 3 . 1 4 6 b )
dQr,

= _ - AL Q.ra6c)
dLo

Second-orderdifferential expressionsare presentedbetow which are requiredfor expressionsgiven

by Eqs. (3.146a-3.146c).In additionto F;q.(3.26a4.26d), the following expressionsare obtained

,
L, d'P,
(i = 1,2,...,NG;i = 1,2,...,NG;' * i) (3.r47
a)
Affi
From Eq. (3.143b),the following can be obtained:

dzr ^ dzP, ( i = 1 , 2 , . . . ,N G ; j = 1 , 2 , . . . ,N G ) (3.r47b)

W=/LPffi
dzt dzr dPr ( i = I , 2 , . . . ,N G ) (3.147c)
W=aer,W=q
The second-orderdifferential expressions,requiredfor Eqs. (3.26a-3.26d)and trqs. (3.147a-
3.147c) are presentedbelow. Theseare obtainedfrom Eqs. (3.14) and (3.145), respectively.

d, p, d, p, - nii + aji (3.r47d)

Q - l, 2, ..., NG; i = t, 2, ..., NG)
AP^ AP
lPrdPr, dQedQsi

d, p, d, p, (3.r47
(r = 1,2, ..,NG, j = L, 2, ..,NG) e)
dPrdQr, aQs,aPt
81

The rnonlinearequations(3.146a) to (3.l46c) are solved using the Newton-Raphsonmethod. To

solve this problem, Algorithm 3.8 can be implemented.

l* --;,
 196 Power System Optimization

EXAMPLE 3.12 Considerthe 5-bus systemof Example 3.1I and obtain the optimum schedule.

Solution Ysuselementsand ZilJ.eelementsare same as those given in Example 3.11. The

initial values and load flow solutions are also same as those of Example 3.tl and are given in
Table 3.21 and Table 3.22, respectively.The units are in p.u. system.
The loss coefficients are\tabulatedin Tables 3.27 and 3.28.

Thble 3.27 a-coeffrcientsfor real transmissionloss P2 (Example 3.12)

4lt

I 0.0112156 0.0003959 -0.0037756 -o.ffi46442 -0.w6475

2 0.0003959 0.0046074 -0.N32624 -0.0033610 -0.0012894
3 -0.0037755 -0.0032623 0.00782/+9 0.0039799 -0.w41942
4 -0.w46442 -0.0033610 0.0039799 0.0079387 -0.0029183
5 -0.0n/i6474 -0.0012894 -0.0041942 -0.0029183 0.019r583

Thble 3.28 b-coeffrcientsfor real transmissionloss P; (Example 3.12)

bt brz bR bio bis

I 0.0000000 0.0000082 -0.0n0r522 -0.00023M -0.0003389
2 - 0.0000082 0.0000000 -0.0000635 -0.0000966 -0.0000671
3 0.0001522 0.0000635 0.0000000 0.0000369 -0.0001364
4 0.0002304 0.0000966 -0.0000369 0.0000000 -0.0000678
5 0.0003389 0.0000671 0.0001364 0.0000678 0.0000000

Using the Newton-Raphsonrnethod elaboratedin Algorithm 3.8, the real and reactive power
generationsare obtained in six iterations
The Hessianmatrix elementsare obtainedas given below:
l,2,
Ht" = += ?-ar+ Z)ra;; (i = 1, 2, 3)
dPr,

,1ii=&=A'o(aii+ai) ( r = 1 , 2 , 3 j; = 1 , 2 , 3 i a n d i * j )

rr dzl
trr;1s- = lvp(bii-bi) (f = 1, 2, 3; j = l, 2, 3; k - - 4 , 5 , 6 )
ffi

i --------._* ..-l
-d
 Economic Load Dispatch of Therrnal Gencrating Units 797

lrvg
r r * = f faiz=t L } ( { o , * a i ) P i + ( b i ,i ._^ b
H , , i. ll n a
) ). - , , l . 0 ( f = 1 , 2 , 3 ; k = 7 )

rr
H * i = azt A n ( b ibi -1 ) ( r ' =1 , 2 , 3 ;j = 1 , 2 , 3 i k = 4 , 5 , 6 )
ffi=
t,
H u = f fd zi t_ \ ( a i 1 + a 1 i ) ( l = 1 , 2 , 3 ; j = I , 2 , 3 i k = 4 , 5 , 6 ; I = 4 , 5 , 6 )

NB
rr azr
H *t=ffi=Zk,*ai)Q1+@u-bi)Pi)(i=l,2,3;k:4,5,6;t=7)

, p 1 -dzr
H W = f f i azt (f=1,2,3
k=; 7)

rr p - dzr
H azt (i=1,2,3
k=; 1 tl = 4 , 5 , . . . , 7 )
m= ffi
tr azt =0.0
Hn= ^
EL",

The Jacobian matrix elements are given below:

rr =uro +b;+x,[(* (to,+ai)pi+(bi:-aul

or)
.) I
ro e=r,2,3)
fr III L\r=t
l- I ) J

l'xs . )
#-=ti = 4lI +ai)Q;+@1;-b1ltP)
(<o, | ti= r, z,r)
\.Fr )

rr NB fNs Nr \ Nc
= >Pr, *l I \ aa(4pi - 4e) I
+ee)+ b;;(QtP1
#- " p l-
i=l \i=r;=r ) i = r "r,

Seven simultaneousequationsare solved using Gausselimination method. The modified

generationswhich are uscd as initial valuesfor the next iteration are given below:

Pq = o.412342p.v., Psz=0.697811p.u., Psr = 0.558187p.u.

Qr, = 0.U17638p.u., Qn= 0.177555P.u. Qca= 0.163107p.ir;

198 Power SystemOptimizetion

The convergenceobtained in six iterations is given below:

3 5
I ar,- I Pd,:P, = 0.00000002
< 0.00001
i=l i=l

The valupsqf Ap,Py, and cost at presentobtainedscheduleare givenas

trv = 255.8830Rs/p.u.h;Cest= 695,35550Rs/h, PL.= 0.01558135

p.u.
The initial valuesfor the next iterationare givenin Table3.29.

Thble 3.29 Initial valuesfor next iterations (Example 3.12)

P8 Qs Pd Qa
(p.u.) (p.u.) (p.u.) (p.u.)
I o.426539 -0.151419 0.00 0.00 Slack
2 0.691533 -0.116962 0.20 0.10 PQ
3 0.547509 -0.062548 0.45 0.15 PQ
4 0,0 0.0 0.40 0.05 PQ
5 0.0 0.0 0.60 0.10 P8

After four _iterationsthe convergenceis obtainedto get the final solution as

- Par-4 6'437302E-06
< o'ooool
It, l=
The final solution is given in Thble 3.30. The convergenceis obtainedin each iteration and
is tabulatedin Table 3.31.

Thble 3.30 Final solution@xample 3.12)

P8 Q8 Pa Qa P o tvl 6
(p,u.) (p.u.) (p.u.) (p.u.) (p.u.) (p.u.) (p.u.) rad
'1 0.426513 - 0.150459 0.00 0.00 0.426959 -0.013799 1.06 0.00
2 0.691320 -0.tt4766 0.20 0.10 0.491319 -0.214768 1.0503470 -0.0134539
3 0.548050 -0.056980 0.45 0.15 0.098050 -0.206977 r.M23970 -0.0305688
4 0.0 0.0 0.40 0.05 -0.400000 -0.050002 r 1.04Q5910-0.0408579
5 0.0 0.0 0.60 0.10 -0.599999 -0.099999 r.0292560 -6.[667rU

Thble 3,31 Convergence

during iterations(Example3.I2)

Pr,- Po, -Pr,-4

Pl
I tn, |
4.4t75628-01 4.600578E-01 -1.830158E-02
4.2653938-0r 4.n3$6E-01 -8.M2753E-04
4.265130E-01 4.26ds87E.-0r -4.4569378-M

*-*
 Ecorromic Load Dispatch of Thermal Generating Units

lossaregivenbelow:
Theoptimumvaluesof 4, cost,andtransmission

4 = 255.g3g0Rs/p.u.h,Cost= 695.43360 Rs/h,Pt = 0.01612151

p.u.

3..I5 ECONOMIC FORACTIVEAND REACTIVE

DISPATCH
POWERBALANCE
The objective of economic dispatch problem is to minimize the total operating cost of a power
system while meeting the total real load plus real transmissionlosseswithin the generatorlimits
of real power as well as reactiveload plus reactivetransmissionlosseswithin the generatorlimits
of reactive power. Mathematically,the problem is defined as
NG :.
Minimize F ( P s )= I @ , P ? , + b i P s i + c ; )R s f t r (3.1a8a)
i=l
N G N B
subject to Z P,, (3.148b)
i=l i=l

NG NB

Zer, =I ea,+Q, i=l

(3.la8c)
i=l

P#" s Ps,< P#* (i = 1,2,...,

NG) (3.148d)

Q['" s Qs,3 QI"* (i = 1,2,...,NG) (3.1a8e)

where
ar, bi, and c; are cost coefficients
Po, is the real power load demand at the fth bus
Pr, is the real power generation(decisionvariable)
Qa, is the reactive power load demand at the ith bus
Qr, is the reactive power generation(decisionvariable)
Ps is transmissionreal power loss
Qr is transmissionreactiygpower loss
NG is the number of generationbuses
NB is the number'of busesin the network. : .

This methrod uses the fact that under normal operating conditions, the transmission loss
is quadratic in the injected bus real powers. The general form of the loss formula obtained in
Eqs. (3.132)and (3.134)is
NBNB
p2- -
I It, i1(4P,+ QiQi\+ ba(QtPi 4Q)l (3.149a)
i=l j=l

NBNB
- (3.14eb)
Qr- I It' u(4r, + QiQil+ d;i(QiPj 48)l
i=l j=l
 #
200 Power SysternOptimization

where

R,,
sii= cos(d;- dr)
ffi
R,,
bii=ffisin(d;-,t,
X
=N,ffi cos(4-6,
X' i i
d
) ri=ffi
s- : i n. / c( d ; - 6 ,
Pi + iQi= (Ps,- Pa) * j(Qs, - Qa) (i = I ,2, ..., NB)
P; and Pi are ttre real power injections at the ith and 7th buses,respectively
Q; and Q1 are the reactive power injections at the ith and 7th buses,respectively
Zi = Rij + jXil @lementsof impedancematrix).
Using the Lagrange multiplier method, the constrainedoptimization problem given by
Eq. (3.148) is convertedinto an unconstrainedone.

Lr) = F(ps,)
L(ps,,er,, +A, p4, L $ p ] *,ta
+ ' p-- -i
[i t'X ea+er \,'=, A'r')' \i=r i=l
n*I (3.1s0)
)
where 4 it the Lagrangianmultiplier.
Necessaryconditions for optimization problem statedby Eq. (3.150) are
_
a L = a F *, ,Lo ' ) * , d Q r-_ n
, ( r r r _-r)*
q Aoq 0 (i = 1'2'"'' NG) (3.151a)
aP& ttr*

aL = n, (' El dP)r*'^,l .u, t( a g , . ) A

q t4-')=o ( r =r ' 2 ' " ' ' N G ) (3.1s
lb)

Nts NG
AL
= Pa+PL-I =o (3.151c)
4 I "''
-\r NB NG

q =E Qa+Qr-Znr,--o (3.ls ld)

wherethe incrementaltransrnission
loss expressioniare expressed
hereas
NG
dP,
=Za;iP;+ t i = 1 ,2 , . . . , N G ) (3.r52a)
dr. Zlf"u*aii)Pi+(bi,-bilQi]
j*i

NG

q
,dP, = 2aiiQ;
+ * ai)Qi+@u-bi)Pj] t; = t, z, ...,NG) (3.tszb)
Zrr,
j+i

-
 EconomicLoad Dispatchof ThennalGeneratiryS
_Utt!!!

NG
=2ciiPi* - (3.152c)
l@u + c i) Pi + (d ii di)Q il
W .i=l
j*i

NG
dQr
dQr,
- 2ci1Qi+
I
j--l
-
[(",i + ci)Qi + @u d i)P;]
(i = 1,2,...N
, G) (3.152d)

j*i

The solutionof nonlinearEqs.(3.15la) to (3.15ld) can be obtainedusing the Newton-Raphson

methodin which changein variables,P--,Qr,(i = 1,2,.., NG), A, and ho areobtainedby expand-
ing Eqs. (3.15la) to (3.15ld) aboutthe initial valuesof the variablesusing Taylor'sexpansion.In
the matrix form the above equationscan be rewritten as

Y,,,, Y ,re, Y ,r^, Y ,r^o LPr

v nrr, Ynrn, Ynr^o v nr^o tes
(3.1s3)
v\^o v\r^, v^o^, Y^r^o LL,
vT,^'nv6r^o Y^o^, Y^o^o Lhn

Elementsof Hessianmatrix derivedfrom Eqs.(3.151a)to (3.l5ld) are as discussedin previous

section.Equation (3.153) can be solved on the basisof the detailedAlgorithm 3.8.

EXAMPLB 3.13 Considera 5-bus systemof Example 3.ll and obtain the optimum schedule.

Solution Ireuselements and Z,u'5elementsare sameas those of Example 3.11. The initial
'values
and load flow solutions will also come out same as in Example 3.11 and are given
in Table 3.2I and,Table 3.22, respectively.The loss coefficients a;i and b;i are tabulatedin
Tables 3.27 and 3.28. (see Exarnple 3.I2). The loss coefficientscy and dii are tabulatedbelow in
Tables3.32 and 3.33

gri= (d;- 6,
ffilcos
X,,
dii= sin(d,- 6r) (i = 1,2,...,5;i = 1,2,...,5)
ffi
Table 3.32 c-coefficientsfor reactivetransmissionloss Q1 @xample 3.13)

Ct Cp C6 C6

I - 1.501021 - 1.517614 -r.532397 -l.53818r - 1.555997

2 - 1.517614 -t.490116 - 1.516583 -r.520337 -r.532566
3 -r.532397 - 1.516583 -1.487124 -r.502384 -t.545794
4 - r.53818r -r.520337 -r.502384 -t.49M12 - r.546260
5 -r.555997 - r.532566 -t.545794 -r.546260 - r.50r063

I
L
tr
'202 Power System Optimization

loss Q1 (Example3.13)
Table 3.33 d-coefficientsfor reactivetransmission

dil diz dts d;a d6

I 0.0000000 -0.03rffi97 -0.0617913 -o.u|62953 -0.1134805

2 0.0316098 0.0000000 -0.0295405 -0.@36986 -0.o797294
3 0.0617913 0.0295405 0.0000000 -0.0139109 -0.0502571
4 0.0762953 0.0436986 0.0139109 0.0000000 -0.0359M3
5 0.r r34805 0.0797294 0.0502571 0.0359M3 0.0000000

The Hessianmatrix elementsare computed using equationsas given below:

Hrr= - 2ai + ZAna;i

+ 2).oci; (i=1,2,3)
" * aP;,

gri = dzt = (i=1,2,3; j=L,2,3; and i+j)

Lo@i1+ ai) + )'o@i1+ cii)
aPsiaPs
j

t, =
H* azr - du)
= )r(b1'- b') + Lo(d1i ( f= 1 , 2 , 3 ; j = I , 2 , 3 ; k=4,5,6)
ffi

Hn=
azr -1.0; ( i = 1 ,2 , , 3 )
)Pi+
aPsiaLP

dzr Pi+( ( i = 1 ,2 , 3 )
.Hn=
aPsia)"q

, azt = 1 /, - b1)
,- + ).n(ds- d1i)
H *3i = Tm Lo(b6 (i- 1,2,3; j = I,2,3)

Hi*3,j*3 =
azr = 4@U + a1) + I'oGi1+c;) ( i = 1 , 2 , 3 ; j = I,2, 3)
dQrdQs

azr = NB - (f= 1,2,3)

Hi*3,7 =
p Z
aQs.u" @u* air)Q1+ @u u1)ri)
j=l

NB
a2r = ( i = I,2,3)
Hi*3,8 = ( t " u* c i ) Q i + @ u - d j ) P j )
aesM,q Ij=l

Hlt =
dzr a2t 3)
aPsiahp

H7,i+3 =
d2t ( l = 1 ,2 , 3 )
aQs.M"
e

.*rd
 Economic Inail Dispatch of Thermnl Generating Units 203

onofi=ffi (i=r'2'3)
Hsi= -Plt dzt

-H
- o'ttJ
E , i * 3#=-
azL dzr (i= 1,2,3)
u, qaQs aQ\.a;,q

Hn= *= F/s=
--oo s *- = H 'u
tB=-# = Ht.=
dto dL'n dlv pdLq

The computedJacobianmatrix elementsare given below:

aL -- ^?n;pr, l-frr-B
lfltr ) I
+ b,+^"l[](tou + a1) p1+ (bii- bil O)l- t.o
q I
) J
I
+ LofX(tr,* c1)p1+ (d1i - dij)o)
| ,, = r,2,3)
L.Ft J
:
I
dt = LLE({"u*ai)Q1
^[mn
+@u-bi,'t,).1 r . f, , \r
ft
[(rn .'\ I
) i + @ ,-i d i , pl ; ) - t o
+ L o l| ( t ' u * c 1 e = r , 2 ,3 )
I | |,t
L\r=r ) J

-x
ioi)l
A1
u "
P
z-r
i=l \;=t j=l ) ",,
i=l

1r NB / r'rg NB ) NG
ctL -- s ea +l c;1e;pi+erei+dii(eipi-pieill-ZO*
'
dhn > > u\ri'.r wiYi
# \i=r j=t ) i=l

The modifiedgenerations
which are usedas initial valuesfor the next iteration'are given below:
Pr,= 0.4265079p.u., Pr, = 0.6916468 p.u.
P.u., Prr= 0.5479347
Qr, = -0.005769
p.u., Qrr= 0.22L1785p.u., Qrr= 0.2322112
p.u.
The convergence in six interations,
obtainedduringNewton-Raphson is givenbelow:
3 5 -

L,,,
i=I i=l
= -{.00000019< 0.00001

The values of Lp, Lq, Py and Qp for the obtained schedule are given as

L, = 255.9044,L, = 0.34058850
Pr= 0.01608950p.u., Qr = 0.04761850
p.u.
 Power System Optimization

In two iterations, the ovbrall convergencevalue obtained is 4.0252518-03. The final solution
achievedis given in Tables3.34 and 3.35.

Table3.34 Final solution(Example3.13)

Bus P8
(p.u.)
Qt
(p.u.)
Pd
(p.u.)
Qa
(p.u.)
P o
(p.u.) (p.u.)
0.426485 -0.006340 0.00 0.00 0.430511 -0.66V293
0.691690 0.220934 0.20 0.10 0.491695 0.120932
0.547717 0.232224 0.45 0.15 0.097717 0.082224
0.0 0.0 0.40 0.05 - 0.400000 -0.049996
0.0 0.0 0.60 0.10 -0.600002 - 0.I 00001

Thble 3.35 Voltagemagnitudeand angle (Example 3.13)

Bus I V I (p.u.) 6 (rad)

I 1.060000 0.0000000
2 r.077662 4.0216892
3 1.080308 -0.0411964
4 1.076726 4.05m282
5 t.060467 4.0729917

b PL - 0.01589258
= 255.8771,Lq = 0.3357378, p.u., er - 0.046g0209
p.u.,
Cost = 695.4353 Rs/h

3.16 EVALUATION
OF INCREMENTAL
TRANSMISSION
LOSS
The transmission loss can be expressedin terms of B-coefficients. Hence penalty factors or
incrementallossescan be evaluatedin terrnsof B-coefficients.The transmissionlossescan also be
expressedin terms of power flow equations.The transmissionloss in terms of power injection at
various buses is
NB NG NB
P r =I e (3.154)
i=l i=l i=l
where
Pr, is the power generatedat the ith unit
Po, is the bus power demandat the ith bus
& is the bus power at the ith bus
NB is the total number of busesin the power system network
NG is the number of generatingbusessuppryingreal power.
The incrementaltransmissionloss for the ith generatingunit is

( i = 1 , 2 , . . . ,N G )

-.{
 Economic Load Dispatch of Thermal Generating Units

The real and reactive power injection p; and ei tre

NB
Pi = I t v,llvi llcu cos(d,- di) + Busin(d,- 6j)l (3.15s)
j=l

NB
Qi= I I V,llVillcusin(d;- 6j) - Bucos
(d,- dj)l (3.1s6)
j=l

where
Y,j = Gu + jB,i are the elementsof the bus admittancematrix
I Vi I is the voltage magnitude at ttre rth bus.
4 is the voltage angle at the ith bus.
For slack or referencebus, angle is zero or fixed one in some cases.
SubstitutingEq. (3.155) into Eq. (3.154),
NB NB
PL- I I lq ll vj llcucos(6,- di) + Bijsin(6,- 6j)l
i=l j=L

On rearrangement,
NB NB NB
Pr= I I v , l 'G , , +) I t V l l v i l f c u c o(sd ,- 6 i )+ B , i s i n (-d 6; j ) ] (3.157)
i=r i=l j=l

Equations(3.155) and (3.L57) show that the distribution of 4 and P, dependon rhe bus voltage
magnitude and angle.

=$ laP,,.dp,. aei, Dp,I

dPr=
til hltf "fr*;5;"6) (31s8)
dPr _ $ | ar, ,.dp, , dQi _dprl
tiv;= hlm"fr*q t4l"q) (3'15e)
Equations(3.158) and (3.159)can be written in matrix form as

(3.160)

or

(3.161)
 -
^ l
!

206 Power System Optimization i

(3'157):
Expressionsfor the elemenrsof the precedingmatrix can be obtainedfrom Eq-

NB
dPt (3.162a)
E6t
.l=l
j+i

NB

ly, l Gii+ )
-6r)]
[ Z lV j I G Uc o s ( d ; ( i = M B + 1 ,M B * 2 , - - . , N B ) ( 3 . 1 6 2 b )
j=l
j*i

where MB is the voltage.controlledbus.

From Eq. (3.155),we can obtain

= cos(4 - 6j)l
(d,- dr)+ -Bu (3.r62c)
d d , $ v,llvj ll-Gusin
33
#j+i '
- l v i ll v i l [ c u s i(nd -; 6 j) - Bu cos(6, - tj)]
# (3.r62d)
, B ; i = 2 , 3 , " ' ,N B ; i + i )
( i - 2 , 3 , . . .N
NB
- z l v ; l G i iI+ t vi llcti cos(d,- 6i)- Busin(6,- 6i )]
# j:l
j*i
(i = MB + 1; MB + 2, ...,NB) (3.r62e)

#- r - lvil lcu cos(6, - 6j) + Biisin(6;- 6j)]

dt tf

r'
( i = M B + 1 , . . - , N B/;= M B + 1 , . . ' , N B ; i * i ) (3.r62t)

From Eq. (3.155),we can obtain

= y. v, vj (6;- 6j)l
(d;- 6;) * Biisin (i = 2, 3, ',NB) (3.162e)
d 6 ' A ' I ll llcucos
P
j*i

= lvillviII-G,icos
(6;- 6 j ) - B u s i n ( 6-;d j ) ]
#
(i -- 2, 3,-.-,NB; i = 2, 3,-..,NB; i * i) (3.r62h)
NB
- 2l v , l n , I, +tvj llcu sin(6,- 6i)- Bucos(6,- dr)l;
#J- j:l
j+i
(i = MB + l, MB * 2,...,NB) (3.1621)

| 4 I [cu sin(6i - d;) - Bucos(di - 6i)l;

(f = MB * 1,..,NB; i*j) (3.r62i)

.^.{
 Economic Inad Dispatch of Thermal Generating Units

The (2NB + MB - 1) simultaneous equationsrepresented by Eq.(3.160) or Eq. (3.161)can be

solved using the Gausselimination method or GaussJordon method or matrix inversemethod.
An intrinsic characteristicof any practicalelecfric power system operatingin steaCystateis
strong'inter-reliancebetweenreal power and bus voltage anglesand betweenreactive powers and
voltage magnitudes.Equation (3.158) can be simplified if it is consideredthat for the given
voltage,real power, Pi dependson the bus voltage angles.'Then the new expressionappearsas

(3.163)

Equation(3.163) can be written in matrix form as

=[#]
t#lt#l (3.164)

The elementsfor the abovematrix are alreadyexpressedin Eqs. (3.162a),(3.I62c), and (3.162d).
The (NB - 1) simultaneousequationscan be solved by any suitable numerical technique,viz.
Gausselimination,GaussJordon,etc.

3.16.1 AlternativeMethodto EvaluateIncrementalLoss

Further, from the Newton-Raphsonpower flow algorithm the real and reactive bus power
mismatchescan be written as

(3.165)

The slack bus (referencebus) real power mismatchis

' " -1= l #

-gill fnat (3.166)
L EiiTlLo'u'l
where
d=[4, 4, ...,4n]'
IV| - [ Vr'rn*r[, I Vlas+2t....t
IyNB |]r

FromEqs.(3.165)and(3.166),

(3.167)

(3.168a)

IL
 Power System Optimization

The elementsof Jacobianmatricesare givenby Eqs. te.r6zq-(3. r6zi)1.

Equation(3.168a)can be rewrittenas

- l:]' f*l
APr (3.l68b)
LpI Lnol
where

lal
I l =
Lp)
and
A = La2,d3, ..., Gr.rg]r
F = [F*s+b Fus*Z,..., fin"]t
Eq. (3.168b)can be rewrittenas
NB NB
APr- (3.16e)
\a,Lp,+ \F,tO,
i=2 i=MB+l
From Eq. (3. 154),

L,P2=AP1
+ X oO (3.170)
i=2
To get the changein total transmission
loss,substitureEq. (3.169)into Eq. (3.170)
NB NB
Lpy-Ift*d)A4+ (3.171)
i=2
\F,tO,
i=MB+l
The incrementaltransmissionloss is definedas
the changein transmissionloss due to a changein
generationi, keeping all other generatorsconstant,
i.e.
dP,
= dP, = 1 + d i (i -2,'.., NG)
;4 (3.r72a)
{
For slackbus
dP, dP,
=o'o (3.r72b)
tro=;4

3.17 ECONOIVIICDISPATCHBASED ON PENALTY

FAGTORS
The problem of optimal allocationof generation
to various plants utilizes a set of penalized
incrementalcost functionsfor generatingunits
as describedby Eq.(3.16), i.e.

(3.r73)

. --. . ...-..c
 Economic Load Dispatch of Thermal Generating units 209

Penalty.factor is rermedby Eq. (3.20) as

( 3 . 147)

From this equation,

(3.r74a)
or
dPt_ _ r _ l
q-t-T Q.r74b)
Incrementaloperatingcost is defined by Eq. (3.19),

AF = Z a ; P r , +
q 'bi

Substitutingthe above equationinto Eq. (3.173),

( f,,,\
z a i P r ,b+i - L l t - # | f3 . r 7 s )
( dPr,
)
From Eqs. (3.174a)and (3.175),

?n,;Po,
6i
* bi= +
Li
or
'\
h ( l' ,
' s ' = [ e - o ' ) -rlo- , (r= I ,2, ...,NG) (3.176a)
or

P-si
-= -- -L (i = 1,2, "', NG) (3.176b)
?-,L,- ?4
The total transmissionloss is expressedas a function of initial power loss Pf and,the changein
total transmissionloss A,Pp,i.e.

Pr= Pf + A'P-
Since power changesin load busesare zero

Pr=Pf+ y, + te - rr?) (3.177)

dPr,\-8i
fr
SubstitutingEqs. (3.174b)and (3.176b)in Eq. (3.177),

P7--Pf y
L/+ rt- +) l^+ - L- P-el
^
\. L,)Lh, L, 2o, t'J
fr

L - j
210 Power SysqemOptimization

pL-pr*}^W (3
17E)
+G+.";)]
#) }l(++p,e)
The power balance equation is

NG NB

I "r, = Pr*Z ,0, (3.r7e)

i=l i=l

NG NB
When Pt= 0, then Z,r, = Z Po, (3.t79a)
j=l i=l

SubstitutingEq. (3.178)and Eq. (3.176a)into Eq, (3.179),

yl:r +l=xpa+pf
*},ffi hl }l(+*p:)
fr LZa,L," +(+.4)]
On simplification and cancelling the equal terms,

NG ^ l-NB NG NG - / t \
P a i - r "+' ?P : . I+ l * * r ! , )
Zm=L:
ConsideringEq. (3.L79a),

p!*;+w.",:)
)"- (3.180)

To obtain the final schedulethe detailed algorithm is outlined below:

Algorithm 3.9: Economic Dispatch Based on Penalty Factors

l. Readdata ei, bi, and c; (i = 1, 2,...,NG), load on eachbus, line data for the power system
network.
2. Obtain Ysus using the l-bus algorithm.
3. Calculateinitial valuesof PsiQ= L, 2,...,NG) and h by assumingthat Pr = 0. Then the
problem can be statedby Eqs. (3.2a) and (3.2b) and the solution can be obtaineddirectly
using the equationsalreadystatedas Eq. (3.10) and Eq. (3.9).
4. The total reactive power demandis distributedto various generatorsin such a mannerthat
the power factor at all the generatorbusesremainsthe same.
MB+l /NB
Qr,= Pr,x Z eo,12 ,0, (f = 1, 2, ..., NB (in caseof pe buses))
i=l I tt
 Economic Load Dispatch of Thermal Generating Units 2ll

5. Calculate Pi= P*- Po, (l = I ,2, ...,NB)

Qi = Qs, - Qa, (i = 1, 2, ..., NB)

taking Pr, = 0, Qr, = 0 for non-generatingbuses.

6. Perform load flow to obtain the real and reactivepowers, P;, Q; and voltage magnitudeand
anglesI Vil, 4 eachbus.
7, Calculatea{i ="t 2,.., NG) from Eq. (3.168)and then calculateLiQ = 1,2,..., NG) from
Eq. (3.r74a).
8. CalculateX,from Eq. (3.180)and Pr,(i= 1,2,..., NG) from Eq. (3.176a).
9. Calculatethe total cost.
10. Stop.
Generatorlimits can be implementedby fixing the generationto the maximum or minimum limit
as per requirement and then such generatoisare not allowed to participate in generation
scheduling.

EXAMPLE 3.14 Considerthe 4-bus systemof Figure3.18. The seriesimpedanceof eachline is

given in Thble 3.36. The system has three generatorsThe operatingcost characteristicsof three
generatorsare given below:

Ft= 50P?t+ 35| Pr,+ 44.4 Rsfti

F2- 50PA+ 389Pr, * 40.6 Rs/h

F r - 60PA+ 340Pr, * 40.0 Rs/h

Given the numberof busesNB as 4, the number of lines NL as 4, and the number of voltage-
conffolledbusesMB as 2, find the economicgenerationschedule.

0.08+70.20

c!
o
+
.+
a
o
0.M+70.14

Figure 3.18 Powernetworksystem.

212 Power System Optimization

Thble3.36 Line data,(Example

3.14)
Line no. From bus To bus Zsen(p.u)
1 I 2 0.08+ j0.20
2 1 4 0.05+ j0.10
3 2 3 0.Ot+ 10.12
4 3 4 0.04+ j0.14

Solution l/su5 is given below

5.724138 -1.724138 0.0 - 4.0

- jr2.3r034 +j4.310345 +i0.0 +j8.0
-1.724138 4.224138 -2.5 0.0
+j 4.310345- jrr.810340 +j7.5 +j0.0
Ygus=
0.0 -2.5 4.386792 -r.886792
+j0.0 +j7.5 - j14.r037;r0 +j6.603774
- 4.0 0.0 -r.886792 5.886792
+j8.0 +j0.0 +j6.603774 - jr4.603770

Initial values are obtainedand tabulatedin Thble 3.37.

L- 608.2354
[*Pa+**,)/*
P r ,= ( L - b i l ( 2 x a ) (l= I,2,3)

p.u.; Pr, -- 0.235295p.u.

Pr, = 2'572354 p.u.; Prr= 2"192353

Thble 3.37 Initial values(Example 3.14)

Bus P8 Pa Qa Typt
(p.u.) (p.u.) (p.u.)
1 2.572354 1.85 Slack
2 2.192353 1.45 PV
3 2.235295 2.10 PV
4 0.0 1.60 0.80 PQ

Load flow is performedusing the decoupledload flow method.The convergenceis obtained

in four iterations(see Tables 3.38 and 3.39).Bus 1 is taken as the slack bus, buses2 and 3 are
taken as PV busesand bus 4 as PQ bus

Pt = 1.079061E-01
p.u.
 Economic Load Dispatch of Thermal Generating Units 213

Tbble 3.38 Load flow solution (Example 3.14)

Ps
(p.u.)
Pd
(p.u.)
Qa
(p.u.)
P o
(p.u.) (p.u.)
I 2.572354 1.85 0.8302581 0.2982M9
2 2.192353 r.45 0.7423536 -0.3050831
3 2.235295 2.r0 0.1352946 1.0886370
4 0.0 1.60 0.80 -1.60 -0.7999994

Table 3.39 Voltagemagnitudeand angle (Example 3.14)

lvl d
(p.u.) (raO
I t.02 0.0
2 1.04 0.03898002
3 1.06 -0.03029456
4 0.94239460 -0.w235717

The expressionsof Eq. (3.161) are representedin matrix form as

-8.0574 0.0
14.75836 - 6.467n62
-6.7ffi96r 13.76972
- 1.56199 3.849879 I

whereat = dPtldPi (i = 2, 3, 4), Fo= aprlae4

Using the Gausselirninationmethod,the abovematrix,is triangularizedas

150 -8.0574 0.0

9.559563 -6.467162
0.0 9.236439
0.0 0.0

Back substitution gives

dP, = . ap, =
o'o (slack,bus),
OP,
0.089306, = 0.079922
E E ?Pg
FromEq.(3.174a),
we get

L,=rrl@ -,0,k-"lF h-'rl@ H) =1086865

#) W)=,0e8064,

L--,_:.
214 . Power System Qptimization

FromEq. (3.180),new incremental as

costis'cOmputed

)"= = 646.711900
+ l
k h,4
From Eq. (3.176a),generationsare computedas

Pr, = 2.957119
p.u.

-P9 2 = 1.999563
p.u.

'
-P8 3 = ( n - k, ) r Lrzszro
p.u.
tu )tr=
The overall computed cost is
F - 3571.481'
Rs/h

3.18 OPTIMAL POWER FLOW SASED ON NEWTONMETHOD

The rptimal power flow is a power flow problpm in whicH cgrtain controllable variablesare
'active power generationor the
adjustecito minimize'an objective function such as t[re cost/of
losses,while' satisfying physical.and operatinglimits on various controls, dependentvariablesand
function of variables. The types of controls that an optimal power flow must be able to
accommodateare active and reactive power injections, generatorvoltages, transformertap ratios
and phase-shiftangles.In other words, the optimal power problem seeksto find an optimal profile
of active and reactive power,generations;"^along with voltage magnitudesin such a manner as to
minimize the total operating costs of a thermal electric powet' system, while satisfying network
security constraints.For example:
Minimize,,operatingcost of thermal stations

F'= > 4 = I + bi Psi+c;) Rsftr

@,P?, (3.181a)
i=l i=l

subject to (a) active power balance in the network

Pi(V,A- Pr,*P4,=0 ( f = 1, 2 , . . . ,N B ) (3.181b)

(b) reactive power balancein ttre network

Q i ( V , A - Q e , + Q d= 0 ( i = N V + 1 ' N I V* 2 , ' , ' N B ) (3.181c)

(c) Security-relatedconstraintscalled soft constraints.
: limits on real power genenations

P#" 3 P, < PrT"* Q= L,2,...,NG) (3.181d)

 Econonic l-oad Dispatch of Tlrcrnul Generating Urtits 215

- limits on voltage magnitudes

YminsU <ymax (f = NV + NV + 2, ...,NB) (3.l8 1e)

- limits on voltage angles

a 6, . 6;nu*
d,lntn (i = 2,3,..., NB) (3.l8lf)
(d) Functional constraint which is a function of control variables.
- limits on reactive power

Ot sQs,sOY (r= l, Z,...,NG) ( 3 .1 8 1 e )

- limits on active power flow of line and reactive power flow of line can be
applied.
Real power flow equationsare

4(vd) = t,i l l j ( G i i c o s ( 4- 6 j ) + B i i s i n ( d , - d , ) ) (3.l8lh)

i:l

Reactive power flow equations are

NB
Qi(v,d) = v,Z vj(Gijsin(d,- Aj)- Br cos(4 - dj)) (3.l8li)
j=l
I
hvhere
i NG is the numberof generatorbuses.
NB is the number of buses
NV is the number of voltage controlled buses
Pi is the active power injection into bus i
Qt is the reactive power injection into bus i
Pa,is the active load on bus f
Qa,is the reactive load on bus I
Pr, is the active generation on bus i
Or, is the reactive generation on bus i
Vi is the magnitude of voltage at bus i
4 is the voltage phase angle at bus i
Yij = G,i + jBii @re the elementsof admittancematrix).
The constraint minimization probleq can be transformed into an unconstrainedone by
lugmenting the load flow constraints into the objective function. The additional variables
me known as Lagrange multiplier functions or incremental cost functions in power system
lerminology.The cost function becomes:

; N B N B
L(Ps,v,A = F(Ps)+I )'r, lPi(y,d)- Pr,+ Po)+ Z ^0,[Qi(vd)- er +ed,l (3.182)
i=l i=NV+l
r
F 216 Power System Optimization

The optimizationproblem is solved, if .the following equationsof optimality are satisfied.

(i = I ,2, i..,NG) (3.183a)

# =Xl^,,#]*,=H.,I^,,H]
(i=2,3,
NB) (3.183b)

# =X[^,,#]*,=H.,#l
L^,,
(r=NV
+,,NV
i 2,,NB)(3.183c,[
- Pt(%O - pr, po, (r= 1,2,...,N8)
! (3.183d)
#
=Qi(V,D-Qs,+Qa ( f= N V + l , N V * 2 , * . , N 8 )
T .li
(3.183e)
Any small variation in control variablesabout their initial values is
obtainedby formingthe
total

#ft;av,+Idha,pi
,=H.,
+ y, ;a,*- s'vq'
LL-=-- L (l=1,2,...,NG) (3.184a)
drrM"n dPr,
7=fi*r

X#b '.r,.X#,#46,
+,=H.,
#; ^vi+I ^hpi
"h
*,=$.,#;qLho, (3.184b)

$ dzt NB
) s dzt
aviaPsj LL,
fr j=2 aviaLej

+ $ --a'r
LLo' =-- dt (i = NV + 1, NV * 2,..,NB)
.,=fi*rav1i- dvt (3.184c)

$ a ' r &w.Z
n * u
fi dtr,a\
NB ^"
-r=fr*,
+ ! d'L ^^ dr (i = l, 2, ...,NB) (3.184d)
Walni= dL,,
 Economic Load Dispatch of rhermal Generating units 217

NG

T
j=l

*r=H., =-# (,=NV+r,NV* 2,.-.,rvB) (3.184e)

diqMr,
Menow differentiate Eqs.(3.183a-3.183e) with respectto confrolvariables,i.e. Pr,,6,, V,, Lr,
urd hr.. Second-orderpartialderivativesrequiredfor Eq. (3.183a)are obtainedby differentiating
iq. (3.183a)with respect-tocontrolvariables.

a z t - Zna t
(f = 1,2,..,NG) (3.185a)
q

azt = tnt ( r = 1 , 2 ,. . . , N G ;i = 1 , 2 ,. . . , N G ;i * i ) (3.18sb)

Ep;
dzr = azr A
=o (i=l'2'"" NG;i=2' "''NB) (3.185c)
@Fq ffi
dzt = 4a =4' !;==o u ( i = r , 2 , . . . , N G/ =; N V + 1,NV * 2; ...,NB) (3.185d)
w
dzt dzr 1 (i=1'2'"''NG) (3.185e)
ffi=W=-l
azr azr -o
,rdL = aL (r= 1'2'""'NG;
i = L'2'"" NB;i"i) (3.18s0
ffi o, ffiP.aPl.
dzt dzr ( i = 1 ' 2NG;
' " ",t = NV + 1,NV * 2, ...,NB) (3.1859)
' W=ffi=o
iecond-order partial derivatives required for Eq. (3.183b) are obtainedby differentiating
iq. (3.183b) with respectto control variables.

(3.l86a)

dzr
ad,ayk
(3.186b)
azr dPi
(3.186c)
e5;
zlE OPtimization
Power SYstem
t

(3.186d)
...,NB; j = NV + 1, NV + 2, .-.,NB)

obtained by differentiating
Second-order partial derivatives required iot Eq- (3-183c) are
Eq.(3.183c) with respectto control variables,i'e'

azr =$f, 1 NBl-

dzp,
i:%l
il^,'Eifl*,R., l^"Dv,aanJ
aqade
(3.187a)
( f = N V + 1 ,N V + 2 , " ' , N B ; k = 2 ' 3 ' " " N B )
1 NBl-
dzp,
azr =$f, *,R., :gl
aviavk
fil^,,;itrhj l^" Ftr[] (3.r 87b)
(i=NV+1,NV*2,.",NB;1s=NV+1'NV+Z""'NB)
t
a'2 -= DPi =
^ ' \ v+ 1,NV + z, .-.,N8; ,l l, 2, .",N8)
\1;&=-NV
(3.187c)
W ay,

&=v(l=NV+1,NV*2,...,NB;/=NV+1,NV*2,...,NB)(3.18
by differentiating;
Second-order partial derivatives required for Eq. (3.183d) are obtained
Eq. (3.1S3d) with respectto control variables,i'e'

( i = 1 ,2 , . . . ,N B ; i = 2 , 3 , . ' . ,N B ) (3.188a,)

( i = 1 , 2 ,- . . , N B ; / = N V + l , N V + 2 ' " ' ' N B ) (3.188b)

azn = 0 ( i = 1 ,2 , . . . ,N B ; j = t , 2 , . . . ,N B ) (3.188c)
7Lo,r^o,

* 2, "', NB) (3.188d)

Second-order partial derivatives reqr-niredfor Eq' (3.183e)are obtainedbY differentiating

Eq. (3.183e) with respectto control variables,i'e'

(f = NV + 1,NV * 2'--,NB; i = 2,3, .'.,NB) (3.189a,)

, B ; / = N V + 1 ,N V * 2 ' " ' ' N B )

( i = N V + 1 ,N V * 2 , - -N (3.18etD

a2L =0 (i = NV + I, NV * 2, ...,NB; ,J = 1, 2, .-.,NB) (3.189c)

a/-q.ah
P.
 units 219
Ecorwmic Load. Dispatch of,Thermal Generating

. , : l

azt I N 1, Nn/ + 2,,...,N8) (3.189d)

- n (/; l-=rrr\/
NV -r-+ 1 ,V+ .2-...
NV * ZIr' " ' , N B ;
,i=,NV* : r. '.,,r '
ffi=o : ' :
derivativetermshaving
second-partial
(3.tg4a)to (3.1g4e)canbe rewrittenconsidering
Equations
zerovaluesas definedby Eqs. (3'185-3'189)'
(3.190a)

NB
+' L\ it
AL
d6t
,t=NV+l
(3.1e0b)
NB
AL
+ t
'
'Lrl dvi
./=NV+l
(3.190c)

AL ( r = 1 , 2 , . . . ,N B ) ( 3 . 1 9 0 d )
dL,,

z#A6,+,=H-,ffi^vj=#;(i=NV+1,NV*2,...,NB)(3
3.18.1 Limits on Variables
Kuhn-Tuckerconditions
inequality constraints are assumedto be
The Kuhn-Tucker approach can be adopted and thus'
t: its lower or upper limit
inactive during the,initial solution.process.Any variable gun^9r,1red.
-trre can be deactlvatedso
values. Then, partial derivative terms with Tqpect to^the fixld variables
of rest of variables' A variable should be
that these terms do not participate in the evaluution
correctionof all other variables'
moved to its relevant limit in coordinationwith the simultaneous'

t,9td'+ At; )'x,Fu '

* Lxi <x;t* (3.191a)

",9t0

necessaryconditions for minimization of L

In accordancewith the Kuhn-Tucker theorem, the
under constraints are
=0 if t,lt"n .1 xi {tt*"*

if x; = x,ltin
(3.191b)
<0

>0 if xi=xl*
 Power System Optimization

Penaltyfunction method
Penalty functions ideally fulfit the requirementsfor inequality constraintsenforcement.
The
penalty function is describedbelow which is quadratic.

f r R r t
1-t \-r J.
-t) (3.r92a)
where
* " ' =I | &
yi is the target value
y; is the current value
S; is the weighting facror.
The weighting factor is automaticallycontrolled to give the appropriateamount of hardness
of enforcement.The first and secondderivativesof a; ari:

s '( y i - y i ) (3.rezb)
#=
(3.r92c)
The proper value of S, is automaticallycontrolled.For the larger values
of ^t, the target limit acts
as the rigid4rard limit and for small value of .Si,the target limit acts as
the soft limit.
The detailed algorithm for optimal load flow is outlined below.
,
Algorithm 3.10: optimal power FIow Based on Newton Method
1' Read dataai, bi, andc; (i = 1,2,..", NG), Ioad on eachbus,
line data for the power system
network.
2. Obtain Irsususing the I-bus algorithm.
3' Calculatethe initial valuesof Pr,(i = 1,2,...;NG) and2,by assuming
-(3.2a) that pt=g. Then the
problem can be statedby Eqs. and,(3.2b) and the solution can be obtaineddirectly
usingEg.(3.10)andEq. (3.9).Initializeall Io,= A, = 1,2,...,
1i NB), 2q- = 0 (j = NV + l,
N V + 2 , . . . ,N B ) , V r = | p . u . ( t= 2 , 3 , . . . , N B ) a n d d ; = 0 ( j = 2 , 3 , . . . ,
NB).
4' Calculatethe Jacobianand Hessianmatrix elements
from Eqs.(3.1g3)to (3.1g9)

AL
LPg
a"'
dt
Ad e-
AL
LHI aAe
AV
q
AL
Llq fr
dt
dLu
The Gausseliminationmethodis usedto find LP*
A6,A4, AV, andLLn.
 Economic Load Dispatch of rhermal Generating (Jnits ?:21
5. Chpck convergence

fg NB NB NB r*
lI (^4,)2 +l{uo1:t+ I ( A v , ) 2\+t t t , r f l ( s
+f rm,)z FI

l_ i=l i=2 j=l i=lW+l i=NV+l J

and optimality conditions.If conditionis not satisfiedthen GOTO Step6 else GCIIO Step g.
6. Modify the variables,
Pr,= Pr,* Mr, (r = I , 2, ...,NG)
4= 4+ A6; ( t = 2 , 3 , . .N
. .B
, )
L o , = L o ,* L L , , , ( i = 1 , 2 , . . . ,N B )
Vi= Vt+ LV; (i = NV + 1, NV * 2, ...,NB)
L n , = L n ,* M r , (i = NV + l, NV *2,..., NB)

7. Check the limits, if any limit of a variableis violated,then imposeor removepower florv
equationor a penalty for inequality.Add or remove derivativesfor penalty or equation
changeand GOTO Step 4 to updatethe solutlon.
8. Calculatethe total cost.
9. Stop.

EXAMPLE 3.15 Considerthe 3-bussystemof Figure 3.19. The seriesimpedanceof eachline is

givenin Table 3.40. The systemhas threegenerators. Find the economicgenerationschedule.The
operatingcost characteristicof two generatorsare given below:

Ft = 50P:, + 351Pr,+ 44.4 Rs/h

Fz - 50 P;, + 389Pr, * 40.6 Rs/tr

0:02+70.08

o.o2+jo91

Figure 3.19 Power networksystem.

Thble 3.40 Line data (Example3.15)

Line From To 7+pn Ysn
no. bus bus (p.u.) (p.u.)
1 1 2 0.02+ 70.08 j0.02
2 1 3 0.02+ j0.08 j0.02
3 2 3 0.02+ 7O.08 j0.02

L Z
222 Power SystemOptimization

Solution Number of generators,NG = 2; Number of buses,NB = 3; Numb€f of lines, NL =

3; Numberof PV buses,NV = l.
Y-busis given below:

Initial valuesare obtainedand tabulatedin Thble 3.41.

)'=[*Pa,*E*)/Er')
= 291.1111

-p8"d, - * (i=1,2)
Zxa,

Pr, = 0.759259; Pr, = 0.940741

Thble 3.41 Initial values(Example3 .l 5 )

P8 Pd Qa Type v 6 hP )rq
(p.u.) (p.u.) (p.u.) (p.u.) (rad)
I 0.7592s9 2.O 1.0 Slack 1.04 0 291,.nn 0.0
2 0.94074r 0.0 0.5 PV 1.04 0 291.1111 0.0
3 0.0 1.5 0.6 PQ 1.00 0 0.0 0.0

Ps, Qs,P, and O are tabulatedin Table 3.42 which are obtained with the initial values
tabulatedin Thble\3.41.

Thble3.42 Othervaluescalculated
with initial values(Example3.15)
P, (n.u.) 0r (n.u.) Pa (p.u.) 0a Q.u.) P (p.u.) 2 (p.u.)
I 0.7592590 0.546t460 o.20 0.10 0.1223528 0.M6r460
2 0.9407406 0.4461460 0.00 0.00 0.1223528 0.4461460
3 0.0 0.0 1.50 0.06 -0.2352939 -0.981176/.
Cost = 653.9258Rs/h
a
With the above-mentionedvalues,the following equation is solved to initiate the iterative
processto implement the Newton-Raphsonmethod.

Y ,rr, o vrr^o o o LPB _y pc

0 Voo Vu^o vN Yu^, Ad -v6
Y ^or, V^ou o Y^0, o Lhp -l
Le (i)
0 vva Yr^o Vw Yr^, AV -vv
0 Y^ro 0 Y^r, 0 L)' q Y^,
 Econontic Load Dispatch of Thennal Generating Units 223

To solve the precedingequation,there is a need to calculatethe Jacobianand Hessiair

matrix elements.Theseelementsare obtainedas describedbelow:
Partial derivativeswith respectto P*:
a r . = Z'oarPs,
* br - lPt = -o'o@ool4
E (tl

ar.= + bz - 1 p2= -o'0000038

Z'oazPsz
i, 82

Partial derivatives with respect to 62:

aL_- ^ dP,*,
= + ^o';y.* 7"'dQt= -3561
'82e
fr h At;
wherethe partial derivativeof P1 with respectto d2is

- vtVz (Gnsin (p1- 6) - Bncos(d1 6)) = -12.72471

+w v 2 :
Partial derivative of P2 with respect to d2:
an 1

#doz = VzL
' - 6 + Bzicos(62
v;(-Gzisin(d2 - dj)) = 24.960
)
F,
j*2

Partialderivativeof P3 with respectto d2:

{{t sin (6r - 6z}-832 cos (dr - D) = -Lz.ziszg

= VtVz(Gcz
vv2

Partialderivativeof Q3 with respectto 62:

{€t = VtVz?G32cos (6r - 6) -832 sin (dr - D) = 3.0588240

vv2

Partial derivative of L with respect to d3:

?P3
+A*aQ3
= Z ho,F,f aq
#
where the partial derivative of Pt with respect to 63 is
fp
=VrVE(Gn sin (d1- ds) - Bn cos (d1 - dJ) =-12.23529
a6,-
Partial derivative of P2 with respect to Q:

+P = VzVt(Grsin (62- 6r) - 823c& (62- 4)) = --r2-2352g

wLr3

Partial derivative of P3 with respect to 63:

a n= J -
V3Lvi?Gti s i n ( d 3- 6 j ) + h i c o s ( d 3- 6 j ) ) = 2 4 . 4 7 0 5 9
E- -*r j=l
j*3
Partial'derivative
of 0i'with respect
to d3:

P = v, vi(Gtlcos(d3- 6ii) + _rr \_ r - dj)) =

(63
____ -6.1 11647
Edr i hisin
ji.\

Partial derivative of L with respectto V3:

aP'
=:aL )"
aq =
l
-l7go.9l5
M A ^ ' , e * ' - q ,d v ,
where the partial derivative of Pr with respectto V3 is
an =V{Gn
___ cos (dn- 6r) * Bn sin (d1- dg))= -3.058824
*
Partial derivative of P2 with respectto V3:
AP
= Vz(Gx cos (62 - 6s) + Bzl sin (62 - 6r)) = -3.058824
e
Partial derivative of P3 with respectto V3: :

E P . n '
=2VtGt *I V i ( q i c o s ( d 3- 6 i ) + B t i s i n ( d 3 ' - d i ) ) = 5 i / 7 0 5 9
;%
i=\
Partial derivative of p, with respectto V3:
3
dQl
- -2vtBt
.
.I vi(Gzisin(63 6c j)-
r
4i cos(63-6)) = 22.,508240
rr /? c \\ .r/r

M
j*t

Partial derivative,of L with respectto ), ;

or:

dr - Pt - Ps * Par=o'4369062
q

Partialderivativeof L with respectto h or,

AL - '
P2' Ps'* Paz= o'8183877 :':
q

Partialderivativeof L with respectto ).

*:
AL -
P3 - Pr' * Pat= -l '264706
q
Partial derivative of L with respect to hnr:

a L A - - .
= QE-,Qr, * Qo,- 0.9211764
A
q
Hessianmatrix elementsare obtainedfrom Eqs. (3.185-3.189).
 Economic had Dispatch of Therrnal'Generating Units ?;25

Secondpartial derivativesof L with respectto P*:

4 = 2 . o a tr2o.o;
= =oo
AP; #
azt = o.o;+ = Z.oaz=
t5o.o
aPsr
aPsz dPi,
Second partial derivatives of L with respectto )"u:

azr --ro; = 0.0

w #=oo;
azt - o o ; = 0.0
aPszaL
pl #--ro; aPhzaLr,3

Secondpartialderivativesof L with respectto 6.2:

azt
wl
=;^',W*u'# = 2742.609

d, P, - -VtVr(Grrcos(di -
6r)+ Bnsin (dt . 6r)) = 1.181176
a67
3
d, P,
= vrL viFG4 sos(62- di) - Bzisin(62- 6;)) = 6.240000
asi j=l
j*2

d, P, = -vtvz(Gncos (63-
6) + Bn sin (d3 - 6r)) = 3.058824
a63
a'(h (6r - 6J + 832cos (6r - 6r )) - 12-23529
as3.
Second partial with respectto 62 and 63:

A,Q,
= -890.4575
ad2a63
where

= 0.0

VzVt(Grrcos(62- dr) + Bzt sinr(d2- dr D = -3.058824

lr1'vz(q2cos(6s - dJ + 832sin (dr - dz)) = -3.058824

ad2a63
 226 Power SystemOptimization

d'g:-
= \v2(G32sin (d'3- 6r) - Bn cos (d_j- d, = -t2.23529
at?adr ))

Second partial derivativesof L with respect to

d3 and rD2:
3 12o
dre, -890.4575
'* adnt =
where
d2P,
ffi=o'o
azn =
\v2(G32cos(6r- dJ + 832sin(dr - 6,)) = -3.058824
ffi
d2o^
- v3vr(Gs,cos(63- 6) + Bnsin(d3-
tffi sr))= -3.058824
dzm-
= VIV2(G32sin (63 - 6r)- Bncos (63 -
t45h A)) = -t2.23SZe
secondpartiarderivativesof L with respectto d3:

d ' t r=- i . , A ' P r .^ d ' e ,

pE I ^ottE- * 70, = 1780.e15
dt
where

-VJ,(GB cos (d'1- dr) + 813 sin (dr - d, )) = 3.05gg24

= -VzVt(Gu cos (d2 - 6r) + 84 sin (d2 -

d3)) = 3.05gg24

= vrf viF;rj \ - 6i) - 83;sin(dr- di)) = s.rr7@7

- -(dr
#d b i "f r \ r r cos

= -
#--r t rn,(-cysin (d3 6i)* Bticos
(d'3- dr)) = 24.4705e
j=\
second partial derivativesof L with respectto
d,2and v3:

-dzr = + n d ' P , ^ d l o " =-3424.836

W h"otfin*Aqrffi
where
 Economic Load Dispatch of Thertnal Generating Units 2n

j+ = V2(Gs sin (62- dr) + B23cos(dz- dr)) = -1 r.7647r

' J

.l- P

(d3- dr)) = 11.7647r

= Vr(G32sin(d3- dJ - Bzzcos
EErt

*3* = vzFG*
' cos(63- 6r) - Bt sin(63- drD = -2.94rr77
a62a% z\
Secondpartial derivativesof L with respectto 4 and V3:

a ' L =+ ^ d'P, A,Q, - :7

t23.66
a 6 " a vA, ^ o t f f i r * L oa\av3
'
vhere

(6r - 6) - 813cos (4 - 6E)) = -12.23529

(62- dr) - 823cos (62 - 6r)) - -12.23529

-* 4- = ,,(t-c' sin(d3- 6i) a Bticos(d3- 6i)) - 24.4705e

a63ay3 f
,;\

j?'-
= -5 -di)) = 6.117647
Btisin(63
ad3ay3 Lr,(o",cos(,83 j)+
i=l
Secondparrial derivativesof L with respectto d and \:

dzr = g:t _ _12.7247r;

- - ' , L + I L =$
' = 33 _ _r2.2352e
W= d6, Wr: ad,

dzr = 9? = 24.e60,'v' ==r1! =- dPz- -12'23529

,.
ffi;' or, aa3aLpz fr
dzr- =9:, --r2.2352e;
L u ' E e r - " =$=9:,- --24.4705e
d6z ad3a[ Eq
W-
Secondpartial derivativesof L with respectto 6 and )'o:

azt dQl ,,ncoo.tA- azt dQt - -6.117

647
a;t=
effi=fr=3'058824; Edr
Second pantial derivatives of L with respect to V3:

# =oo
=p.^,,#*',W
r
i;

i:
x28 Fower System Optimization

where

#
d,e,
= o.o;
W=
o.o;
#
- 2.0G33=
n .7647r

T-- = -2.0h3 = 46.97882

dv:
Secondpartial derivativesof L with respectto V3 and )r:

dzt a4 =-3.05gg24;
: = a2t ,oP', dzt dP,
= 5'64705e
w=ait dw=')fr=_3.0588240; ffi= ft
Secondpartialtderivativesof Z with respectto V3 and Ln:

=22'50824
&=W
Using the Gausselimination method,the changein variablesis obtainedand the updated
values of variablesare shown in Thbles3.43 and 3.M.

Thble3.43 updatedvaluesof variables(Exampre3.15)

Bus P8
(p.u.)
Q8
(p.u.)
Pd
(p.u.)
Qa
(p.u.)
P o
(p.u.) (p.u.)
1 0.7284207 0 . 1 2 9 5t 81 0.20 0.10 0.5485317 0.0295118
2 0.9690878 -0.0730638 0.00 0.00 0.9946736 -0.0730638
3 0.0 0.0 1.50 0.06 -r.5205370 0.0056424 v

Thble 3.44 Updatedvaluesof variables(Example3.15)

Bus v d LP )",r
(p.u.) (rad)
1 1.04 0.0 287.4105
2 1.04 0.01I69348 295.3632
3 1.025293 4.05167275 300.2588 -10.70845
Cost= 653.3180Rs/h

The final solution after four iterationsis given in Thbles 3.45 and 3.46.

Table 3"45 Final solution obtained after four iterations (Example3.15)

Bus P8
(p.u.)
QB
(p.u.)
Pd
(p.u.)
Qa
(p.u.)
P o
(p.u.) (p.u.)
1 0.7767037 0.1521580 o.20 0.10 0.5767038 0.0521580
2 0.9452465 -0.0327489 0.00 0.00 0.9452462 -0.0327489
3 0.0 0.0 1.50 0.06 - 1.5000000 -0.0s99991
 Economic Load Dispatch of Thermal Generating Units

Thble 3.46 Final solution obtainedafter four iterations(Example 3.15)

Bus V d ).e ).q

(p.u.) (rad)
I 1.04 0.0 293.2044
2 1.04 0.00966759 29r.7870
3 t.02297 -0.05t42289 301.0320 0.58963730

Cost = 660.3356Rs/h

3.18.2 DecoupledM for Optimal Power Flow

An important characteristicof ny practical power transmissionsystemoperatingin steadystateis
that the changein real power the specifiedvalue at a bus is more dependenton the changes
in voltage angles at various es than the change in voltage magnitudes,and the change in
reactive power from the spec ied value at a bus is more dependent on the changes in voltage
m4.gnitudesat various buses the changesin voltage angles. This is due to the fact that
transmissionlines are mostl -reactive, the conductance G's are very small compared to the
susceptance, B's. Also, under normal operatingconditionsthe angle (4 - 6) is small (typically

less than 10'). In view of this, theterms, *o O"tngsmallcanbeignored.

# ffi
,Q
rherermr and
## av will also be small and can be ignored. Ignoring these terms
beingsmall,Eqs.(3.186b),(3. 86d),(3.187a),(3.187c),(3.188b),and (3.188a)can alsobe ignored
being small. Equations(3.1 ) to (3.190e)become:

azt dr (f=1,2,...,NG) (3.193a)

'rrdL Mo, dPr,
r,
NB
a2t LLo, = AL
I #a da, a
j=2
a dr/+
- -f" i ' 4 a,dLr, Ed,
( i = 2 , 3 , . . . ,N B ) (3.1e3b)

NB
+f dzr Ad; = AL (i=1,2,...,N8) (3.193c)
j=2
),p.a6j dL o,

dzr LLu, = AL (f = NV + 1, NV * 2, ...,NB) (3.193d)

'dL
n,

dzt AL
( i = N V + 1, NV * 2, ...,NB) (3.193e)
LVi =
L,rtavj dL n,
The above equationscan be upled and can be solved separately.In the matrix notation these
can be representedas

Y ,ru

I
vao (3.r94a)
pPs
Y ^ru
ThesizeoftheHessianmatfixis(NG+2xNB-l)x(NG+2xNB-l)andthesizeofJacobi
is (NG + 2 x NB - l) x l.

lI:. =F;;l
I:;,ltil,l (3.le4b)

n a l t r i ixs [ 2 x ( N B - N V -
T h e s i z e o f t h e H e s s i am l)] x[2 x (NB -NV- l)] andthesizeof
Jacobianis [2 x (NB NV - - I )] x l. The elementsof Hessianand Jacobianmatriceshas already
beendescribedin Eqs. (3.1F3)-(3.189).
The detailed algorithrfr is 6utlined below.

Algorithm 3.11: Decoupl0d Method for optimal power Flow

l. Readdata a;, b;, and c; (i = 1,2, ..., NG), load on eachbus, and the line data for the power
system nretwork.

..., NG) and2vby assumingthat PL- 0. Then the

(3.2b) and the solution can be obtained directly

I n i t i a l i zael l ) v o .= ) v ( i = 1 , 2 , . . . , N B ) ,r , q -= 0 ( i = N V + l , N V * 2 , . . . , N B ) , V i = | p.u.
( i - - 2 , 3 , . . . ,N B ) a n d 4 = 0 ( i = 2 , 3 , . . . , N B ) .
4. Calculate the Jacobipnand Hessianmatrix elementsfrom Eqs. (3.183a) to (3.189) and
solve Eqs. (3.194a)qnd (3.194b).
AL
4
dr and [H2)
a6
AL
il P
Gauss elimination m$thod is used to solve these simultaneous equations separately to find
LPs, L6, L2,,,A% and LLo.
5. Checkconvergence

NB
(s and l - N B (LV,)z+ t
1g
>
ll_i=NV+l i=NV+l I
and optinnalityconditions.If convergencecondition is not satisfiedthen GOTO Step 6 else
GOTO Srep 8. i
6. Modify the variablesl

Pr, -- Pr, n M* (i = 1, 2, ...,NG)

4=4+A4 (i = 2, 3,'...,NB)
Lo, = Ao, * OLlo, (i=1,2,-..,N8)
V i = Vi + LVi (i = NV + l, NV * 2, ...,NB)
Ln, = 10,* O^i, (f = NV +:1,I{V * 2, ...,NB)
 Economic Load Dispatch of Thermal Generating Units 231

7. Check the limits, if limit of a variable is violated then impose or remove power.flow
equationor a penalt for inequality. Add or remove derivatives for penalty or equation
change and GOTO 4 to updatethe solution.
8 . Calculate the total
9. Stop.

EXAMPLE 3.16 Consi the 3-bus systemof Figure 3.19 (see Example 3.15). The series
impedance
of eachline is in Table 3.40. The systemhas two generators.The operatingcost
characteristics of two are given below.

Fr= 50P:,+ 352Pr,+ 44:4

Fz - 50P:r+ 38gPB2+
40.6
Obtain the optimal schddul usingthe decoupledmethod.

Solution Number of NG = 2; Number of buses,NB = 3; Number of lines,NL =

3; Number of PV buses,
Y-bus and calculati for initial values are given in Example 3.15 and tabulated in
Table3.41. Ps,Q,g,P and Q are tabulatedin Table 3.42 which are obtainedwith the initial values.
The updated values after iterationare tabulatedin Tables3.47 and 3.48.

Thble 3.47 Upda valuesof variablesafter the first iteration(Example3.16)

Bus Ps Qs Pa, Qa P o
(p.u.) (p.u.) (p.u.) (p.u.) (p.u.) (p.u.)

1 0.7689440 0.20 0.10 0.4721372 -0.1497818

2 0.9404150 0.00 0.00 0.8499779 -0.2373549
3 0.0 1.50 0.06 - 1.3046550 0.3271568

Final solution is in 11 iterationsfor convergence0.0001 and is tabulatedin Tables

3.49 and 3.50

Thble 3.48 Upda values of variablesafter the first itera-tion(Example 3.16)

Bus V d hP hq
(p.u.) Gad) @Vp.u.h) (RVp.u.h)
1 1.04 0.0 292.2733
2 1.04 0.00985892 291.0713
3 1.O+0361 -0.04675326 295.4336 -6.128246

Cost = 656.6735 Rs/h

Thhle 3.49 F al solution obtainedafter 11 iterations(Example3.16)

Bus Pt Qt Pd - Q'a P o
(P.u-) (p.u.) (ptu.) (p.u,) (p.u.) (p.u.)
I 0.7766963 0714597 0.2 0.10 0.5766963 0.0521595
2 0.945?5.35 0885403 0.0 0.00 0.9452534 -0.0327503
3 0.0 0 1.5 0.06 - 1.5000000 -0.0599991

-.2
 Power System O,

Table 3.50 Final lution obtainedafter ll [email protected])

Bus V 6 IP I,r
(p.u.) (rad) (RVp.u.h) (RYp.'r.h)
l 1.04 0.0 293.2036
2 1.04 0.00966797 291.7880
3 r.02297 - 0.05142269 300.8488 0.00000542
Cost = 660.3355 Rs/h

3.19 OPTIMAL POWER OW BASEDON GRADIENTMETHOD

The objective function to be mi imized is the operating cost
NG
F= I Fi(pst) (3.195a)
i=l
where

4(Ps)= aiP?,+b,Pr,+c,
If the systemreal power loss is be minimized,the objectivefunctionis
F = P(V, 6)
In this casethe net injected real wers are fixed, the minimization of the real injected power P1
at the slack bus is equivalent minimization of total system loss. This is known as optimal
reactive power flow problem, to the load flow equations:
(a) Real power balancein the for each PV bus

4(v, ) - Ps,* P6,= 0 (i = 2,3, ..., I\n/) (3.l esb)

(b) Acqive and reactive power ance in the network for each PQ bus

n(v,t)- P,* P 4 ,- 0 (f=NV+ l,NV *2,...,NB) (3.195c)

Qi(V,6)- Q * Q o , = 0 ( i = N V + l , N I V* 2 , . . . , N 8 ) (3.l esd)
with
NB
Pi(V, d) = vi
Z r,(co .or (4 - dr)+ ^Bu
sin(4 - dr)) (3.195e)
j:l
and
NB

Qi(V, d) = vi Z r,(cu rin(4 - d;)-.Bucos(4 - dr)) (3.1esO

j:l
where
NG is the number of gener tor buses
NB' is+he number of buses
ltV is the number of volta e controlled buses
Pi is the-aetive power inj tion into bus i
Qi is the reactive power i iection into bus i
Po, is the active load on b I

,^8
 Economic Load Dispatch of Thermal Generating Units 233

ion on bus i
ion on bus i

% is the magnitude voltage at brs i

4 is the voltage phaseangle at bus i
Yi = 6ii + jB;i we the lements of the admittancematrix.
Equations(3.195b),(3 95c) and (3.195d)can be expressedin vector form as

(3.1e6)

where the vecton of t variablesis

(3.r97a)

and the vector of i variables is

{} } ror slackbus
LorJ
l'p.l
-'
y = { I for PV bus
l.Y'J
=[l (3.1e7b)

fp.I
PQbus
E,lfor
In the above formulation, I objective function must include the slack bus power. The vector of
independent variablesy can be partitioned into two parts-a vector u of control variables which
are to be varied to achie optimum value of objective function and a vector p of fixed or
disturbance or uncontrol partrmeters.
Control parameters be voltage magnitudeson PV buses,or Pr, at buses with control-
lable power. Slack bus vol and'regulating'transformertap-settingrnay be employed as additional
control variables. Qr, may be used as control variables on buses with reactive power control
[Singh, 1993].
The optinrLization em can be restated as
min F(x, u) (3.198a)

subject to equality constrai

B(x,u, P) = o (3.1e8b)
To solve the optimi ion problem, we define Lagrangefunctions as
L(x, u, p) = F(x, u) + f g(x, u, p) (3.198c)

where ,1.is the vector of e multipliersof the samedimensionas 8(x, u, p).

 Power System O,

The .necessaryconditions minimize the unconstrainedLagrangian function are

a L=
_ E F . [ & l t1 .- =''',
E; E'LEJ 'v o (3.resd)

#=#.[#]']'=o (3.198e)

y = g(x,u,p)= o (3.1e80
Equations (3.198d),(3.198e)a (3.198f) are nonlinear algebraic equationsand can only be
solvediteratively.A simpleyet efficient iteration scheme,that can be employed, is the steepest
descentmethod (also called gradient method). The basic technique is to adjust the control
vector u, so as to move from feasible solution point (a set of values of x which satisfies
Eq. (3.198f) for given u and p; it indeed is the load flow solution) in the direction of steepest
descent (negative gradient) to a new feasible solution point with lower value of objective
function.
The computational for the gradient method with relevant details is given below:
1. Make an initial guess z, the control variables.
2. Find a feasible load fl solution from Eq. (z.ltla and b) by the Newton-Raphson
iterative rnethod. The improves the solution .r as follows:
*r+l={+L,x
where Ax is obtained by solving the set of linear equationsgiven below:

Lx= -s(t'v)
[#,"'r,]
Lx=-f+ (",
' y)l
- s(x',y)
Ldx J
The end result of Step 2 is a feasiblesolution of x and the Jacobianmatrix.
3. Solve Eq. (3.198d)for )"

(3.199a)

4. Insert.1,frorn Eq. (3.199) into Eq. (3.198e)and compute the gradient

YL= (3.1eeb)

5. If Vt equals zero within the prescribedtolerance,the minimum has been reached.Other-

wise:
6. Find a new set of contro variables
uinew uold+ Lu where Lu = -aVL (3.199c)
Here Au is a step in ne ve direction of the gradient.The step size is adjustedby the
positive scalar u.

o-od
 Economic Load Dispatch of rhermal Generating Units z3S

In the algorithm, the choice of is very critical. Too small a value of a guarantees the convergence,
but slows down the rate of ; too high a value causesoscillationsaroundthe minimum.
Severalmethodsareavailable optimumchoiceof stepsize.

3.19.1 lnequalityCons aints on Control Variables

Though in the earlier discu oo, the control variablesare assumedto be unconstrained,the
i.e.
permissible values are, in fact always constrained,
dn.,s
t It I lt^u*

Prftn
sPr,<P#"*
These inequality constraintso control variables can be easily handled. If the correction L,u; in
Eq. (3.199c)causesui to ex one of the limits, ui is set equal to the corresponding limit, i.e.

(3.200a)

After a control variablesreac any of the limits, its componentin the gradientshould continue
to be computedin later i s, as the variablemay come within limits at some later stage.
In accordancewith the hn-Tucker theorem, the necessaryconditions for minimization of L
under constraint are:

=o if Jzlnin
<ui< uf*
ft
if ui= uln (3.200b)
#=o
i f u i= u l *
#.0
Therefore, now, in Step 5 of computationalalgorithm, the gradient vector has to satisfy the
optimality condition (3.200b).

3.19.2 Inequality Co ints on DependentV ariables

Often, the upper and lower li ts on dependentvariablesare specified as
,/-n<.r(.tr**
e.g. Ymin
' t -<V,
' t <
- ' Ymax
l on a PQ bus

Such inequality constraints be convenientlyhandled by the penalty function method. The

objectivefunction is augm by penaltiesfor inequalityconstraintsviolations.This forces the
solution to lie sufficiently cl se to the constraint limits, when these limits are violated. The
penalty function methodis val d in this case,becausetheseconstraintsare seldomrigid limits in
the strict sense,burtare in soft limits (e.g. V < L0 on a PQ bus really means V should not
-
exceed1.0 too much and V 1.01may still be permissible).
 236 Power System

The penalty method ls for augmentationof the objectivefunction so that the new
objective function becomes

F =F(x,u)*2wi (3.201)
j
where the penalry W1is ucedfor eachviolatedinequalityconstraint.
A suitablepenalty ion is defined as

(3.202)

where y is a real positive n ber which controlsthe degreeof penalty and is called
the penalty
factor.
The necessaryconditi ns (3.198d)and (3.198e)would now be modified as given
below,
while the condition(3.198 i.e. load flow equation,remainsunchanged.

(3.203a)

(3.203b)
The vector dW/dx obtained om Eq. (3.202)would contain only one non-zeroterm
corresponding
to the dependent variable xii while d$du - 0, as the penalty functions
on dependentvariables
are independentof the I variables.
By choosing a higher alue of
[, the penalty function can be made steeper so that the
solution lies closer to the ri ;id limits; the convergence,however, will become poorer.
A good
schemeis to start with a low value of and to increase it during the optimi zation process,if the
A
solution exceedsa certain to limit.

EXAMPLE 3,17 Considert he 5-bus system of Figure 3.17 (see Example 3.11).The
series
impedanceof each line is ei in Thble 3.20. The system has three generators.Find the economic
generationschedule.The ing cost characteristicsof the three generatorsare given below:

I = 60P:, + z}pPtt + 140.0 Rs/h

z = 75Ps2,
+ 150Pr,+ 120.0 Rsftr

IOP:, + 180Pr, + 80.0 Rsftr

Solution Considering 1 as slack bus, buses2 and 3 as pv buses and4 and 5 busesare
consideredas Pp buses.For samplesystem,the optimization problem is statedas
3
Minimize P- \ f",r!, + b,Pr,+c;) Rsftr (3.204a)
i=l
subject to the load flow equ ons

i(v, d) - Pr,+ p a , = o (3.204b)

 Economic Load Dispatch of Thermal Generating Units 237

Qi( ,6)-Qr.+Qa,=0 (i=4,5) (3.204c)

where
5
- -
Pi(V,d) = Z r,(ou cos(d, 6j) + Busin(d, d;)) (3.204d)
d=l

5
Q{V,o - - - cos(6i - dr))
Z ,,(cu rin(d; 6r) ^Bu (3.204e)
i=l

For this problem, the d variables are

x = f62,4, 6+,6s,V+,Vslr
The independent variables
y=[up]r
The control variables are
a = lPn, PrrJ'
The fixed variables are
p = l t, 4, P2, V2, P3,V3,P+, Q+, Ps,Qslr
Ttp Lagrangian function be written as
4

J 5 5
L o , ( 4 ( v-,P
d )r ,* Pa). > l"u,(ei(y,
d)- er,* eo,)
S't | .t
L(x, u, p) = L l",Pi, + b,Pr,+
i=1
).> i=4
i=2
(3.205)
The necessary
conditions minimization are

= Z a ; P r , + b , + ) " 0( i .= 2 , 3 ) (3.206a)
#

# = (hfs,*a,)#
I W
( i=2 , 3 , 4
s ), (3.206b)

aL _ 14_ f, , , ,, d4
(bfs,+b)#, (i=4,5) (3.206c)
fr= Y
p i ( v , 6p) -r . * ' d ,- o (i--2,3,4,5) (3.206d)
#=
AL =
=_
dL
Qi(V,d)_ Q, + =Q (i=4,5) (3.206e)
q,

Equations(3.206d)and (3.206e) n be solved using the Newton-Raphsonmethod.To implement

the Newton-Raphson method,th following equationis solved.
P
5 (3.207
a)
a
d
l"
?38 Power System

The size of the Jaco matrix is (2 x (5 - 1) 4) = 6. 4 and Vi are calculatedusing the

Gausselimination method are updated till no further improvement is achieved.

dfnt= 6i + A'6, (i = 2,3, 4, 5)

V,'*t= V{ + LV, (i = 4, 5)
Initial data to impl ent the Newton-Raphsonmethod is given in Table 3.51.

1 Initial data for N-R load flow (Example 3.17)

Bus Typt P8 Pa Qa V 6 L P Aq
(p.u (p.u.) (p.u.) (p.u.) (rad) (Rs/p.u.h) (RVp.u.h)

I Slack 0.45 8 0.00 0.00 1.0600 0.0

2 P V 0. 4 0.20 0.10 1.0593 0.0 253.0107
3 P V 0.5211 0.45 0.15 t.0525 0.0 253.0107
4 P Q 0. 0.40 0.05 1.0 0.0 0.0 0.0
5 P Q 0. 0.60 0.10 1.0 0.0 0.0 0.0

After four iterations,the solution obtainedls grven in Thble 3.52.

Thble 3.52 Soluti of N-R load flow methodafter four iterations(Example3.17)

Bus p
" 8 Qs Pa Q a P o
(p.u.) (p-u.) (p.u.) (p.u.) (p.u.) (p.u.)

1 0.4585800 0.1633390 0.00 0.00 0.4585800. -0.1633390

2 0.6867383 0.1656168 0.20 0.10 0.4867383 0.0656168
3 0.52t5049 0.1267469 0.45 0.15 0.0715049 -0.0232531
4 0.0 0.0 0.40 0.05 -0.4000005 -0.0500000
5 0.0 0.0 0.60 0.10 -0.6000000 -0.0999998

The voltage and the Itage angle at each bus are given below:
p.u.
V t = 1.0600 4 = 0'0 rad
v 2 - 1.0593p.u. 6z= -0.01739534
rad

l=-l.ffiirlT',:l
v 3 - 1.0525p.u.
p.
V + = 1.048837
v s - 1.034887
p.
Equations(3.206b) d (3.206c) can be rewritten in matrix form as

I
 Economic Inad Dispatch of Thermal Generating Units

The Jacobian matrix i

36.307790 - 5
- 5.608016 4
t: - 5.604438 -3
-8.35267 0 - 4.1017l0 12.10581 1.258813-4.616217
-r.617749 - 1 13.166r -1.386275 40.53706 -3.848699
Lun 4319.686
hh 1079.045
LroI 0.0
tJl Lo, 0.0
hno 0.0
ho, 0.0

From the above eq the values of 4 ^d A.oare obtainedby implementing the Gauss
elimination method.
L r, = 258'oo3, L o, = 261.2L59, L oo= 262'8730, L o, = 267'5160
Lno = 0j1273 , Lo, - 1.167928

From Eq. (3.206a), gradient can be obtainedas

* bz* L or= -4'993037

= 2azPs,
:!;
- ZasPrr*bg*Lp,= -8:205190
#
If the norm' of t is more than the required tolerance, then update the values.

llvPsll - 9.604976
> 0.r

The total operating c at this scheduleis 704.0132Rs/h.

After updating the the Newton-Raphson method is implemented and the solutionis
given in Table 3.53 and v tage magnitudeand angles at each bus are given below:
Vt = 1.060P.u. dr = 0.0 rad
Vz= 1.0593 p.u. 6z = -0.01598643
rad
Vt = 1"0525p.u. 6r = -0.03360153
rad
Vr - 1.048837P.u 6+= -0.04303627rud
-
V5 1.034892 p. 65- -0.06749434rad
240 Power System Opti

Thble3.53 Solutiono N-R load flow method

after rwo iterations
(Example3.17)
Bus 'P8
Q, Pd Qa P
(p.u.) u.) (p.u.) (p.u.)
0
(p.u.) (p.u.)
I 0.4249496 _0.1 0.00 0.00 0.4249496 _0.rs29420
2 0.6992280 0.1 0.20 0.10 0.4992280 0.0606308
3 0.5420235 0.1 0.45 0.15 0.0920235 _0.0305305
4 0.0 0.0 0.40 0.05 -0.3999993 _0,0499971
5 0.0 0.0 0.60 0.10 -0.6000002 _0.0999995

The final solution as ined after eight iterationsis given in Thble 3.54.

Thble 3.54 Solution of R load flow methodafter two iterations(Exampr

e 3.r7)
Bus 'P8
Pd
(p.u.)
I
u.) (p.u.)
Qa
(p.u.)
P
(p.u.)
a
(p.u.)
I 0.4258551 -0.1 93 0.00 0.00 0.4258551 _ 0.t532493
2 0.6924789 0 . 1 6 552 0.20 0.10 0.4924789 0.06275s2
3 0.5478095 0 . 1I 482r 0.45 0.15 0.0978096 _0.0325t79
4 0.0000000 0 0.40 0.05 -0.3999997 _0.0500030
5 0.0000000 0. 0.60 0.10 -0.s999999 _0.099999s
For the above solution, voltage and voltage magnitude at each bus are
given below:
Vr = 1.06p.u.
4 = 0.0 rad
Vz- 1.0593p.u. 5z= -0.01610616rad
V3- 1.0525p.u. ds = -0.03333623 rad
V4- 1.048837p.u. 5q = -0.04284792 rad,
V5- 1.034894p.u. ds = -0.06751189rad
The obtainedvaluesof )"n are given below:
u
Lr, = 253'8727, = 256.7579, L.oo= 258.4248, Lo, -
263.1094
Lno - 0.1106465, 'q, = l j46384

The gradient as obtained i given below:

- Za2Prrt bz *
L p, = 5.9l5l l7E_02
dP,,
AL =
2qzPe,* bz * Lp, = -5.g525g0E_02
dP,,
l l v P sI = 8.3211368-02< 0.1
The total opeidting cost at is schedule
is = 695.5009Rs/tr
 Economic Inad Dispatch of Thennal Generating Units ut
REFERENCES
Books
Arrillaga, J. and C.P.Arnold, 'omputerAnalysis of Power Systems,John Wiley & Sons, Singapore,
1990.
Elgerd, O.I., Electric Energy Theory:An Intrcduction,Znded.,ThtaMcGrawHill, 1983.
El-Hawary,
M.E.,andG.S. tensen,Optirnal Economic Operation of Power Systems,Academic
Press,NewYork, L979.
Gross,C.A., PowerSystem , Wiley, NewYork, 1979.
Kirchamayer,L.K., Economi Operation of Power Systems,Wiley EasternLtd., New Delhi, 1958.
Kusic, G.L., ComputerAided Power SystemsAnalysis, Prentice-Hall of India, New Delhi, 1986.
Mahalanabis,A.K,, D.P ari, and SJ. Ahson, Computer Aided Power System Analysis and
Control, Thta McGraw-Hi New Delhi, L99I.
Nagrath, I.J. and D.P. Modern Power SystemAnalysis, Thta McGraw-Hill, New Delhi,
1989.
Nagrath, I.J. and D.P. Power SystemErigineertng,Tata.McGraw-Hill, New Delhi, 1994.
Singh, L.P.,AdvancedPower tem Analysis and Dynamics,Znd ed.,Wiley EasternLimited, New
Delhi, L993.
Stagg,G.W.andA.H. Ei- ComputerMethods in Power Systen;tsAnalysis,McGraw-Hill, New
Delhi, 1968.
Stevenson,W.D., Elements Power SystemAnalysis, 4th ed., McGraw-Hill, New York, \982.
Wood,A.J. and ts. lV g, Power Generation, Operation and Control, John V/iley, New
York, 1984,

Papers
Alguacil, N. and A. J., C , Multiperiod optimal power flow using benders decomposition,
IEEE Transactions on r systems,PWRS-15(1),pp. L96-201,February2000.
Alvarado, EL., Penalty fac from Newton'srnethod,IEEE Trans.,YoL PAS-97,No. 6, 2031-
2037, 1979.
Aoki,.K. and T. Satoh,New gorithms for classiceconomicload dispatch, IEEE Trans. on Power
Apparatus and Systems, 103, No. 6, pp. 1423-L431,1984.
Burchett,R.C., H.H. Happ, Verath, and K.A. Wiqgau,Developmentsin optimal power flow,
IEEE Trans.,Vol. PAS-101No. 5, pp. 406414, t982.
Carpentier,J.L., Optimal flow, Electric Power & Energy Systems,Vol. 1, No. 1, pp. 13-15,
1979.
Carpentier, J.L., Optimal er flow: uses, methods and developments,Proceedingsof IFAC
conference,RI Brazil, pp. tr-25, 1985.
Chen,L., S., Matoba,H. In and T., Okabe, Surrogateconstraint method for optimal power flow,
IEEE Transactionson Pow r system,r,PWRS-13(3),pp. 1084-1089,August 1998.
 Power SYstem

B.H. andS.
ChowdhurY, A reviewof recentadvancesin economicdispatch,IEEE Trans.
Vol. PW
on PowerSYstems, , No. 4, pp. L248-1259,1990.
Chun-LungChen and Nanmi Chen, Direct searchmethod for solving economicdispatch
problem considering capacity constraints, IEEE Transactions on Power Systems,
PWRS-I6(4), pp. 76L7 69, 2001.
da Costa,G.R.M., C.E.U., and A.M., de Souza, Comparative studies of optimization
methods for the optimal flow problem, Electric Power SystemsResearch,Vot 56,
pp. 249-254, 2000.
Da Costo,V. M., N. Martins, J.L.R., Pereira,Developmentsin the Newton-Raphsonpower flow
formulation based on injections, IEEE Transactionson Power systems,PWRS'14(4),
pp. 1320-1326, 1999.
Da Costa,V.M., N., Martins, nd J.L.R., Pereira, An augmentedNewton-Raphsonpower flow
formulation basedon culren injections, Int. J. Electrical Power and Energy .Sysrens,VoL 23,
pp.305-312,2001.
Dommel, H.W. and W.F. Ti Optimal Power Flow Soutions,IEEE Trans.,Vol. PAS'87, No.10,
pp. 1866-1876,1968.
Dopazo, J.F., O.A. Klitin, G. Stagg, and M. Watson, An optimization technique for real and
reactive power allocation, P edingsof IEEE, YoL 55, No. 11, pp. 1877-1885,1967.
Early, E.D., RE. Watson, and G.L. Smith, A general transmis-sionloss equation, NEE Trans,
Vol. PAS-74,pp. 510-520, 5.
Ekwue,A.O. and J.F. , Comparisonof load flow solution methods,Electric Power
SystemsResearch,Vol. 22, o. 3, pp.2L3-222,199L.
Famideh-Vojdani, A,R. and F. Galiana, Economic dispatch with generation constraints, IEEE
Traw. on Automatic Vo[ 25, pp.213-217,1980.
Fang,R.S., and A.K., David, imal dispatch under transmissioncontracts,IEEE Transactionson
Power Systems,PWRS-I pp. 732-737,May 1999.
Happ,H.H., Analysisof netw IEEE Trans,Yol. PAS'82,
with complexauto-transformers,
pp. 75-81, t963.
Happ, H.H., Analysis of ne with complex auto-tansformers, II: Relations between all open-
path and open-path-closed- impedance matrices, IEEE Trans, YoL PAS-82, pp. 958-965,
L9;63.
Happ, H.H., Analysis of with complex auto-transformers, III: Invariant reduction for
purposeof loss formula, I Trans,Vol. PAS-83,pp. 707:714,1964.
Happ, H.H., Optimal power IEEE Trans.,Vol. PAS-93,No. 3, pp. 820-830,1974.
Happ, H.H., Optimal power tch-A comprehensivesurvey, IEEE Trans.,Vol. PAS'96, No. 3,
pp. 841-854, 1977.
Happ, H.H., J.F.Hohenstein,L. Kircham ayer,and G.W. Stagg,Direct calculation of transmission
loss formula-Il, IEEE Trans. Vol. PAS-83,pp. 702:707,l9&.
Hill, E.F. and W.D. Stev , A new methodof determiningloss coefficients,IEEE Trans.
Vol. PAS-87,No. 7, pp. I 1553,July,1968.
 Economic IAad Dispatch of Thermal Generating Units 243
'metric
Housoi, E;C. and G.D: Irisarri A sparsevariable optimisation method applied to the
solutionof powersystem IEEE Trans.,Vol. PAS'101, No. l, Pp. 195-202, fan, 1982'
Huneault,M. and F.D. Galiana,A survey of the optimal power flow literature, IEEE Trans. on
PowerSysrens,Vol. PW L99L
No.2, pp.762-770,
Part II
IEEE working grouP, DescriPti and bibliography of major economic security function,
and
and lll-BibliograPhy (195 1972 and lg73-lg7g), IEEE Trans. on Power Apparatus
Vol. 100, No. 1, PP. 2 L 5 - 2 3 5 , 1 9 8 1 .
Sysreru.t,
reference
Kron, G., TensorialanalYsisof integratedtransmissionsystems,Part-I: The six basic
frame, AIEE Trans.,Vol. 70, . 1239-1248,195i.
turn ratios,
Kron, G., TensorialanalYsisof integratedtransmissionsystems,Part-II: Off-nominal
AIEE Trans.Vol. 71, pp. stz, L952.
IEEE
Liang, Z. and J.D. Glover, Im ved cost functions for economic dispatch computations,
Trans on Power System.s, Vol. PWRS-6,No. 2, PP.82I-829, L99L
real-time line flow
Lin, C.8., S.T. Chen, and C.L. Huang, A two-step sensitivity approachfor
calculation, Elecffic Power stemsResearch,Vol. 2L, pp. 6349, 1991'
ordered
Meyer, W.S. and V.D. Al , Improved loss formula corirputation by optimally
eliminationtechniques,IEEE Trans,Vol. PAS'92,PP.6249, l97I-
Mohamed-Nor, K. and A.H.A. Rashid, Efficient economic dispatch algorithm for thermal unit
commitment, IEE Proceedin s C Vol. 138, No, 3, PP.2I3-2L7, l99t'
and P. Ristaanovic,
, M.S. Stott, D. Sun, A. Papalexopoulos,
Momoh J.A., R.J. Koessler,B.
Challenges to Optimal er Flow, IEEE Transactions on Power Systems, PWRS-Ia(L),
pp. 44M47, February1997
Monticelli, A. and W.H.E. Liu, taptivemovementpenalty method for the Newton optimal power
flow, IEEE Trans, on Power stems,Vol. PWRS'7, No. 1, pp.33F342, t992'
dispatch algorithm
Nanda, J., L. Hari, M.L. Koth , and J. Henry Extremely fast economic load
Vol' 139, No' 1, pp' 3946'
through modified coordinati equations, IEE Proceedings,Part C
t992.
evaluations,IEEE
}[g, W.Y., Generalizedgenerati distribution factors for power system security
Trans. on Powe4 APParatus Sysfems,Vol. 100, No. 3, pp. 1001-1005,1981'
Electric Power
Palanichamy,C. and K. Srikris na, A method for short-term generationredispatch,
SystemsResearch,Vol. L7, 129-138,1989.
Palanichamy,C. and K. Srikri Simplealgorithmfor economicpowerdispatch,ElectricPower
SystemsResearch,Vol. 21, 147-153,199r.
S.C. Tripathy,Modificaions to Newton RaPhsonload flow for ill-
Prasad,C. Durga,A.K. Jana, 'nt.
SYsrerns,Vol. 12, No. 3,
conditionedpower sYstems, Joumal of Electrical Power & Energy
pp. 1,92-L96,1990.
Ramaraj, N., R, Rajaram,and A new analytical approachto optimize a generation
Parthasarathy,
schedule,Electric Power tems Research,Vol. 11, pp. L47-I52, 1986'
to power system
Sasson,A.M. and H.M. Merri l, Some applications of optimization techniques
problems,Proceedingsof I , Vol. 62, No. 7, pP.959-972, 1974.
 Power System O,

Shoults,R.R., W.M. G and s. Helmick, An efficient method for computing loss formula
coefficientsbasedupon method of least squares,IEEE Trans.,vol. pAS-9g, pp. 2144 zl5z,
1979.
Sun,D.I., B. r\shley,B. rewer,A. Hughes,and W.F.Tinney,Optimalpowerflow
by Newton
approach,IIiEE Trans., PAS-103,No. 10,pp. 286+2880,1984.
Talukdar,S.N. and F.F. W , computer-aideddispatchfor electric power systems,
Ptoceedings of
IEEE, Yol. 69, No. 10, p . 1 2 1 2 - 1 2 3119, 8 1 .
Tao Guo, Mark I Henw d, and Mariekevan Oaijen, An algorithm for combined heat power
economic dispatch, Transactions on Power Systems,PWRS-11(4), pp. t77}-t794,
November 1996.
Wang, C. and S.M. S hopur, Optimal generationscheduling with ramping costs, IEEE
Transactions on Power stems,PWRS-10(1),pp. 60-67, February1995.
Whei-Min Lin, Fu-Sheng ng and Ming-Tong Tsay, An improved Tabu search for economic
dispatchwith multiple rnrma,IEEE Transactionson power systems,pwRs-l7(l),
pp. l0g-
112, February2002.
Wong, K.P. and K. Doan, recursiveeconomic dispatch algorithm for assessingthe costs of
thermal generatorsched , IEEE Trans.on Power system.l,vol. pwRs-7, No. 2,, pp. s77-5g3,
1992.
Yung-ChungChang,Wei-T Yang, and chun chang Liu, A new method for calculatingloss
coefficients,IEEE ctions on Powersystems,PWRS-9(3),pp. 1665-1671,Augusr lgg4.
 StochqsticMultiobiective
GenerqtionScheduling

6.1 INTRODUCTION
Optimal economicdispatchin electric power systemshas gainedincreasingimportanceas the cost
associatedwith generationand transmissionof electric energy keepson increasing.The p
involves the allocation of total generationrequirementsamong the available generatingunits in
the system in such a manner that the constraints imposed on different system vari les are
adequatelysatisfied and the achieved overall cost associatedwith it is a minimum.
Despite extensive research focussing on thermal power dispatch problem' muc of the
-state
effort todate has involved the developmentof deterministic models applicable to ste
conditions. Most of these attempts assumeth-esystem data to be deterministic. It m s that
all input information is known with complete certainty and the optimal plans of disl tch are
always realized exactly. In practice, there are several inaccuracies and uncertainties in input
information (Figure 6.1), which lead to deviationsfrom optimal operation.
The operating cost functions representingthe perfortnance characteristicsof then plants
are computed by calculating the overall thermodynamicperformanceof a unit consisting f boiler,
turbine, condenser,heat cycle, and associatedptant auxiliaries. Such cost functions are rate
in most cases.The inaccuraciesmay be viewed due to the following reasons[Kirchma ,
19581.
. Inaccuracies in the process of measuring the basic data' used for compu tion of
thermodynamic performance of the unit
Deviations frorn-the computed thermodynamicperformance of the unit becauseof
errors
encounteredin operation due to opdratingat other than standard'pressureand
Effect of time on equipment conditions which influences some of its operating
stics, notably its efficiency
Inaccuraciesresulting from inability to hold generationat exact desired
. Fuel cost variations
. Load forecasting erfors
. Inaccuraciesintroduced by various types of transmission'lossequations.

Furttrer, becauseof great difficulty in determining the dependencYof maintenance

from
very i accurate.
the power output, the additional costs for maintenance,supplies, and water are
387
 System

Unit data

\ _ ' {
I
I
I
I
I
\ - l

|- ------- - - ' : r - - L -
- - - - -; - -: -- -- - &- - - - - : - -- - - - - - 1
1--
! ,Inaccuracy: : I Uncertainty:
I tnfonnation i i tnformation
L------- _-__l t----
Unit commitmentand
economicdispatch
'-------1 procedure
'
i Inaccuracy: 1,,'
i Modelling (
r-____ .._______--J

,i Inaccuracy:
i Control
t-------

Figure 6"1 Optimalpower system inaccuraciesand uncertainties.

If all these factors are taken together.these will causeinaccuraciesof great magnit in the
steady-stateoperation.The effect of inaccuraciesis in an increasein the'overall cost. Vi i and
Heydt [1981] have outlined the computationaldetails of the stochasticoptimal energy ispatch
problem. The stochasticoptirnal energy dispatch algorithm employed the multivaria Gram-
Charlier series to statisticallymodel the probability density function of the control v or. The
applicability of the series has been limited by the high computationalrequirementsof c ulating
high order statisticalmoments.The method obviatessome of the difficulty through a
transformationof the variatesto be modelled in order to enhancenormality. The main ai of the
method was to produce a tool which would be useful from an operationalstandpointbu fails to
considerstochasticcost function.
Yakin [1985] has articulated an approachto the optimal generationschedulingof power
systemby treating the electricity demandat a node as a random variable with a known I
lity distribution. Particularly, a two-stagestochasticprogramming with recoursemodel been
developedfor stochasticeconomic dispatch An equivalentproblem to this two-stage m I has
been defined. The penalties for discrepanciesin the generation have been incl in the
objective function of the equivalent problem. The major difficulty during the implemen tion is
to draw out the exact values of thesepenalties.
El-Hawary and Mbamalu [1988] have investigatedthe perturbationsin the systemthermal
fuel cost and the system equality constraintsas stochasticand normally distributed w th zero
mean and a given variance.In an anotherattempt,El-Hawary and Mbamalu [1989] in uced a
method in which the system power demand was assumedrandom with zero mean unit
variance.In the third attempt,El-Hawary and Mbamalu (1991) consideredthe pertu lons ln
system power demand as random and normally distributedwith zero mean and some v
 stochastic Multiobjective Generation scheduling 389
They observedthat optimality conditionsin terms of the active power generations
were b by
parametersobtalned from the variancesof active power generations.
But these do rpt
provide trade-off betweeneconomy and risk measuresdue to uncertainties
in system tion
cost and randomnessof demand.
Parti [1987] has expounded an gconomic dispatch of thermal generation while
incor ng
the randomness in system production cost and system load through
lenerator outputs, whi were
treatedas random variables.He appendedthe traditional objective firnction
of economic patch
with a penalty term accountingfor the possibledeviationsproportional
to the expectation of the
squareof unsatisfiedload becauseof randomness of generatorpower.This upp.ourhsuppr the
ffue characterof the problem by consideringonly the mon etary aspects
and fails to ex the
non-commensurabilityof the conflicting'objectives.
Besides electric energy, power plants also produce sizeable quantities
of solid wastes,
sludge,and pollutantsthat affect air and water quality.The pollutants
affectingair qualit are of
the greatestinterest.They include particulates,NO' CO' So, and other sundry
oxides of lphur
that can travel over considdrabledistances,and have long-term effects
both in spacear time.
The traditional meansfor controlling emissions,such as precipitatorsand
scrubbers,are h ware
intensive,relatively inflexible and limit ttre ratio of emissionsto energy produced
in eac plant,
but not the total emissionproducedin a region. In contrast,dispatchingr"quires
little I ware
and is flexible and effectiveat the regionallevel. Here, the obje.liu" function
used in dis
can be changedin a few moments.
In general,a large-scalesystemas typified by an electric power system,possesses
r ultiple
objectivesto be achieved,namely economicoperation,reliability, securityand minimal
im on
environment.It may be obvious that tracie-offsamong these objectivesare difficult
of
their different nature.This implies that objectivesare non-commensurable.

6.2 MULTIOBJECTIVE
STOCHASTICOPTIMALTHERMAL
PowER DISPATOH-e,-OONSTRA|NT
METHOD
Extensivestudies,associatedwith the optimal power dispatch,have been centred on
ma ing it
more efficient in algorithm and applicableto online with deterministicdata. In actual
nractice, it
is a misleading assumptionthat data is known with completecertainty. In spite
of this, it is also
,hT the dispatch is to optimize just one specific objective, or single perfo
Iu,9 .ontimal nce
index. Now the trend is to formulate multiobjective optimization pioblem with
due consi on
of uncertaintiesfor a more realistic approach.The multiobjective stochasticoptimization
lem
is describedin the subsequentsections.

6.2.1 StochasticProblemFormulation
The objective function fo be minimized is the total operatingcost for thermal generating
Its in
the system. The operatingcost curve is assumedto 6e approximatedby a quadratic fu I on
of
generatoractive power output as

NG
F r =I (o,4' + biPi+ ) (6.1)
i=l
",
i
i where
, N G is the total number of generators

I
I
I
L
 System Optimization

ffil; ,l''ii"ib,,andci arecostcoefficients

i1':.
4 is the active power generationof the ith generator.
A stochasticmodel of function F1, is formulatedby consideringthe otherwise rministic
(p bi, and ci ss random variables.Any possible deviation of operating cost coefficien and load
demandfrom their respectiveexpectedvaluesare manipulatedthroughthe randomness generator
powerPt fParti,et al., 19831.A specificway of reducinga stochasticmodelto its nrstic
equivalentis to take its expectedvalue [Sen Gupta, L972; Fredric Soloman, l9}7l. ine that
the random variablesare nonrrally distributedand statisticallyindependent,the expec value of
operatingcost becomes:
NG

4=| ra,4'*6,4 + c,+ a, var(4ll (6.2)

i=l

where

4 is expectedpower generationof the ith generator

dr, 6,, iltd ci are expectedcost coefficients

The varianceof power Pi is given as

var(P,)= CA4' (6.3)

where Co is the coefficient of variation of random variable p;.

Therefore,the expectedoperatingcost as given by Eq. (6.2) is modified as

NG
s r . - t
Ft= ) tft* c'ilo,1'*6,P,+ (6.4)
f=l
",1
The load demand constraint is

F, +F, (6.s)

P, is the expectedpower demand

f" is the expectedtransmissionloss.
The expectedlimits on the power generationimposed are

P,*" < r, s p,max (i = I ,2, ...,NG) (6.6)

4*n is the expectedlower limit of generatorpower ourput

P-r* is the expectedupper limit of generatorpower output.
 Stoclwstic Multiobiective Generation Scheduling

iltre transmissionline lossesare expressedin terms of B-coefficients as

NG NG NC
+Z ,,u**Boo
Pr=I > 4BuPi
j=li=li=l

whereB;i, Bro,and Bssare B-coefficients'

With 4s as independentrandomvariables,the expectedtransmissionloss can be
as
NG NG NG NG
*Z E,,var(4)
FL= I > F,EuFi Boo
+> E*F,+
j=l
i=l l'=l i=l

NG NG NG NG

FL= > Q,+ * | }F,ErFi*I u"1*E*

czil4,,P,' (6.8b)
i=l i=l i=l t;!

where Eij, B;s,and Bsoare expectedB-coeffrcients.

The varianceof transmission becausein the deterministic
losshasbeenneglected , the
normally not more than 5
B-matrix represents-ody the appreximate transmissionloss which is
generatorsto the using substations[Parti et' al' 1e831.
Orr*;; of tt e total power transfened from deviati ns are
Since generatoroutputs P;s are treated as random variables, the expected
These e
proportional to the expectation of the square of the unsatisfied load demand.
deviations ale given as

Using Eq. (6.5), the above equationcan be rewritten as

This on simplification reduces to

NG
\i var(P,)
'E\-ll
'L
i=l

Equation (6J0) is equatedto F2, the new objective function, i.e.

NG
Fz= I ntt4l
i=l
 v System Optimization

Eq. (6.3) into Eq. 6.11(a),we get

Substituting

NG
Fz= I c'oF,' (6 1lb)
i=l

A multiobjectiveoptimization problem c
operatingcost and (b) the risk associatedwith 1
their expected values while satisfying the el
generationlimits. The multiple objectiveoptimi

Minimize [4, rr]' (6.IZa)

NG
subject
to =
Z P, Fo+F, [6.12b)
i=l

{mins 4 < 4"* (i = I ,2, ...,NG) ko.tzc)

6.2.2 Algorithm
To generatea non-inferior solution to the multi
methodis used. In this method,one specificol
preferablycorrespondingto the most importanto
to be minimized. Tlhe multiple objective optimi
t-constraintapproaclhexpressedas

Minimize fi (6.13a)

subjecrro Fz S ez (O.tgU)
NG
S
\ D - Pr+Pt
Zy,i- (6.13c)
i=l

p,min< Fi < p_l** (i = 1,2, ...,NG) (6.13d)

where s2 is interpretedas the maximum tolerableobjective level. The values of r are chosenfor
which the objectiveconstraintsin Eq. (6.13) are binding at the optimal solution. As a corfstraint
is varied parametrically,a set of non-inferior solutions (with their correspondingtrade-dffs)is
generated.
The well-known method of Lagrangemultipliers is quite popular in the power ystem
planningstudies.The LagrangianL formed for the systemis given by Eq. (6.14)

NG
_)
L- r; + Ltz(Fz + F o I- ,,)
i=1

where 1a2 and IL are Lagrange multipliers.

,
 Stochastic Multiobjective Generation

The necessaryconditions to obtain solution are given

as

dr=;t'.
F, ^"fr.,L'+'-t1
aE ^ AF, faE I
=o (6.15a)

dt = Fzttz=O
dAn (6.15b)

#=Fr+Fo-I"--=o i=l
(6.15c)

The optimal solutionto Eq.(6.14) musr satisfythe Kuhn

Tucker conditions.The main ition is
lrr(4 - ez)- o; h1-> o (6.l sd)
The value of 2'p cortespondingto the binding constraints
indicates the margi I benefit
of the objective function due to an additional unit of e. The
Lagrange rnultipliers rel to the
objectives, as constraints may be zero or nonzero. The
set of nonzero Lagrange ultipliers
setof solutions.
Theserof nonzero
Lagrang"r.ittiplT"rs
:r"::ru{:j:^t!|"":l.int3titr
the set of trade-off ratios betweenthe principal objective
and each of the constraining jectives,
respectively'The systemgiven by Eq. (6.14) is solved using
the Newron-nuphsonmr od for R
valuesof e2' Only those valuesof ),'p > 0.0 which conespond
to active constrain
ts F2, P) = tz',
r = l, 2, "', R are considered,since they belong to the non-inferior
solution [Haimes nd Hall,
re74l.
In the problem, the initial value of e2 is taken such that
s2 > F2" and e2 < F; *. Since
objectivesare of conflicting nature,the value of one objective will
be maximum, when value
of another objective is minimum and vice versa.
To implement the Newton-Raphsonmethod, the following
equation is solved i vely till
no further improvement in decision variablesis achieved.

(6.16)

The Newton-Raphsonmethod shows very effective results when

the initial guessis in I domain
of solution' Utilization of factorized matrix is 'another aspectof
the,aforementionedr m
Algorithm 6.L: Non-Inferior Solution by the e-Constraint Method
l . Read data, namely cost coefficients,emission:coefficients
and B-coefficients,Err conver-
gencetolerance)and ITMAX (maximum allowed iterations),NG (number generr
of ) and
K (minimum number of non-inferiorsolutionsrequiredfor the objective
' as const nt), etc.
2. Fix e2 such that Fft" ( Ez . F{u*.
3. Set iteration'for non-inferiorsolutions,k,=, L , : : :
4. Incrementcountofnon-inferiorso[utions,k=k+|.
5. If (k > ^K) GOTO Step 18.
 Power System OPtimization

6. computethe initialvaluesof 4Q = 1,2, "'' NG) andp'

7. Assumethat no generatorhasbeenfixed eitherat lower limit or at uPperlimit.
8. Set iterationcounter,IT = 1.
9. CompureHessianand Jacobianmatrix elementsusing Eqs.(6.15a)to (6.15c)' v.
Deactivaterow and column of Hessianmatrix and row of Jacobianmatrix rep ine the
generatorwhose generationis fixed eittrer at lower limit or at upPer limit. This done so
that fixed generatorscannot participatein allocation.
10. Gauss elimination method is employed in which triangularizationand back tution
processesare performedto find, LF, (f = 1, 2, ..., R), LL'12,Llt.Here R is the mber of
generatorsthat can participate in allocation.

R R
+ ( tr)z
tol )z+ (LJ"rz)z + (Y)'p)z+ (Yi'
f, tvr-.)z
I i=l
i=l )='"-[
then GOTO Step 14.
12. Modify control variables,

P-new= Fi + Af; (r = 1,2, ',', R)

Lrl"n = Lrz+ Llv12

!F* = p+ Lp
13. Update iteration counter,IT = IT + 1,
Aisign new values to old variablesto continue the process'

Pr, = P;"* (i = I ,2, .-.,R)

Ltz=LrTn and P=Fn'n

GOTO Step 9 and rePeat.

t4. Check the limits of generatot'sand fix up as following:

If F, .F,^'n then Fi = flmin

If |rF,^u then 1 = 4'o

If no more violations of limits are there ttren GOTO Step 16'

r 5 .GOTO Step 8.
1 6 . Check the condition is satisfied
[vtz(Fz(P)- e) = 0.0; )qz > 0'0,

If 'yes'then GOTO SteP17.

else modify e2 and GOTO SteP5.
1 7 . Record it as non-inferior solutions,computevalues of all objectivesand ion loss
and modify t2 for the next non-inferior solution 4qd GOTO Step 4
1 8 . Stop.
 stochastic Multiobjective Generation scheduli

6.2.3 Applicationof the Method

Two sample systemsare taken up to illustrate the method to evaluatethe possible onomic
significancewith respectto risk.
Case lz In this case, a three generatorsample system is selected.The expected erator
characteristicsare given in Thble 6.1. The expectedB-coefficientsof fansmission loss fo ula are
presentedin Thble 6.2. In addition, the following coeffrcientsof variation of random var bles are
assumed.
cn = o'l ( l = 1 ,2 , 3 )

Thble 6.1 Expectedgenerationcharacteristics

4i bi C;
4'* Pi'tn
($/Ivtw2tr1 ($/MWh) ($rn1 (Mw) MW)
I 0.010 2.00 10.0 200.0 10.0
2 0.0r2 1.50 10.0 200.0 10.0
3 0.004 1.80 20.0 200.0 10.0

Thble 6.2 Expected B-coefficients

J J
I I 0.0w2725 I 2 5l
2 2 0.0003090 1 3 679
3 3 0.0032295 2 3 65

Using this data, the resulting expected generation schedules with expected cost and risk
in Thble 6.4 for various values of e. The deterministic results are shown in Table6.3 (C
j = I, 2 , 3 ) .

Thble 6.3 Deterministicresults

Sn PD Fl Pr P2 P3
no. (Mw) ($/tr) (Mw) (MV/) (Mw)
I 140.0 361.263 46.146 54.78r .182
2 180.0 484.010 64.417 68.727 1
220.0 621.695
?'

J 83.441 83.200 .4TI

The percentagedeviation in the cost for different schedulescorrespondingto risk i shown

in Figure 6.2 for different expecteddemands.In representation,€z is interpretedas risk, ich is
proportionalto the expectedpower demand(ez= dPD, where g is a risk factor).The curve ndicates
an increase in the percentage deviation of cost of operation for different levels of ri k. The
operatorhas a'choice to selectthe risk factor from the curve.
Case 2: A large systemconsistingof eight generatorsis selectedfor this case.The d for the
Ioss formula coefficientsas well as the expectedincrementalproduction cost coefficie ts of a
396 Power System Optimization

Thble 6.4 Expected non-inferior generationschedules

Sr n
($rn;
F2 FL
(Mw)
Pr
(Mw) (Mw)
P2 F3
(Mw)
no. (MW2)

Fo = 140MW
I 362.2143 7t.4W 5.7648r 47.982 53.900 43.960
2 362.2M0 71.428 5.79242 47.835 53.975 M.060
3 362.1948 71.456 5.81940 47.692 54.048 M.158
4 362.1868 71.484 5.84578 47.554 54.r18 44.252
5 362.1797 71.512 5.87163 47.419 54.r87 M.345
6 362.1736 71.540 5.89695 47.287 54.253 M.436
7 362.1684 7r . 5 6 8 5.92182 47.t59 54.318 M.524
8 362.t64r 71.596 5.94625 47.034 54.381 M.611
9 362.1604 71.624 5.97028 46.9tr 54.443 44.696
10 362.1575 71.652 5.99390 46.791 54.503 M.780

PD= 180MW
11 485.8152 119.880 8.35676 67.572 67.980 52.921
12 485.7668 119.916 8.4331 67.297 68.054 53.210
l3 485.727r 1r9.952 8.52199 67.05r 68.120 s3.470
l4 485.6942 119.988 8.59468 66.821 68.178 53.709
15 485,6663 120,024 8.66279 66.619 68.232 53932
t6 485.e27 120.060 8.72711 66,425 68.282 54.141
t7 485.6222 120.096 8.78810 66.238 68.335 54.337
18 485.6051 r20.r32 8.84665 66.066 68.377 54.526
t9 485.5905 120.168 8.90289 65.902 68.4r8 54.706
20 485.5782 t20.204 8.95712 65.745 68.456 54.879

PD = 220 MW

2l 624.3509 182.600 11.78805 85.728 83.587 62.642

't82.644 11.90598 85.524 83.556 62.996
22. 624.2985,
23 624.2559 182.688 12.01322 85.342 &3.526 63.316
24 624.2204 182.732 12.1t263 85.175 83.499 63.610
25 624.1908, t8-L776 12,20563 85.021 83.474 63.884
26 624.1658 r82.820 72.29350 84.877 83.450 64.t41
27 624.1448 r82.864 t2.37712 84.741 83.427 64.384
28 624.t272 182.908 12.45700 84.6t7 83.405 64.616
29 624.t125 182.952 12.53376 84.489 83.384 64.837
30 624.0999 r82.996 12.60826 84.368 83:367 65.050

-
 Stochastic Multiobjective Generation

0.45

I
v)
0.4

o
C)
0.35

q) 0.3
\e

0.25

0.2
0.51 0.51060.51120.5il8 0.66640.667 0.68760.83020.8308
0.8314
d +

--*- 1 4 0M W --*-- - - . . * . . . 220MW

1 8 0M W
Figure6.2 Percentage deviation in cost vs. a.

system are given in Tables6.5 and 6.6 respectively.

The valuesof coefficient of ariation are
given below:
Cp = 0.1 ( i = 1, 2 , . . . ,8 ) .

Thble 6.5 Transmissionloss formula coefficients(8,7x 102)

J Bij j Bij j Bii

I I 0.07863 2 3 0.04624 5 6 0.01224
2 2 o.oe6gs 2 4 0.01246 5 7 0.0072r
3 3 0.09163 2 5 -0.01218 5 8 0.00378
4 4 0.02646 2 6 - 0.01810 6 7 0.02166
5 5 0.02311 2 7 -0.01750 6 8 0.01682
6 6 0.03723 2 8 -0.0r754 7 8 0.05768
7 7 0.06285 3 4 0.01242
8 8 0.12010 3 5 -0.01198
I 2 -0.00999 3 6 -0:02204
I 3 -0.01402 3 7 -0.02530
I 4 -0.00695 3 8 -0.02841
1 5 -0.01136 4 5 0.00179
I 6 -0.02076 4 6 -0.00707
I 7 -0.02892 4 7 - 0.00876
I 8 -o.03292 4 8 -0.00992
 Thble 6.6 Expectedgenerationcharacteristics

bi fi.max p.mm
($/Mwh) (Mw) (Mw)
I 0.004100 1.280 200.0 50.0
2 0.002200 0.795 210.0 210.0
3 0.000950 1.809 200.0 10.0
4 0.002145 0.657 400.0 150.0
5 0.001110 0.889 310.0 310.0
6 0.006000 0.300 200.0 100.0
7 0.010400 0.635 100.0 50.0
8 0.006350 0.572 150.0 50.0

The deterministic generationschedulesobtained are given in Table 6.7 for various demands.
Generators2 and 5 remain at their maximum loads of 210 MW and 310 MW respectivel; and are
not included.The expectedcost and risk with expectedtransmissionlossesfor vari expected
demandsare presentedin Table 6.8 which are in the non-inferior set. The correspondin expected
generationschedulesare given in Table 6.9 for various valuesof t.

Thble 6.7 Deterministicresults

F D n F r F 3 P 4 F 6 P 7 P8
(Mw) ($nr) (Mw) (Mw) (Mw) (Mw) (Mw) (Mw)
t240 1731.691 137j99 61.047 272.756 145.826 65.254 01.165
1320 1909.131 153.874 84.645 292.947 157.232 71.308 w.476
1400 2w1.048 170.938 107.708 313.463 169.036 77.534 r7.916

Thble 6.8(a) Expectedcost, risk and transmissionloss when PD = 1240

Sr Fl F2
no. ($n1 (MW2)
I 175r.636 2554.OO .87011
2 1751.4M 2555.4 .89693
3 175r.175 2556.88 .9238r
4 1750.951 2558.32 .95073
5 1750.731 2559.76 .97770
6 1750.51,4 256t.20 .00471
7 1750.302 2562.64 .03177
8 1750.093 2564.08 05887
9 1749.887 2565.52 .08601
t0 1749.685 2566.96 .11319
 Stochastic Multiobjective Generation Sche

Table 6.8(b) Expectedcost, risk and transmissionloss when FD = l32o

Sr Fr F2 FL
no. ($n; (MWz) (Mw)
1 1993.124 2682.000 64J5024
2 1990.328 2683.600 64.30595
3 1987.785 2685.200 64.27084
4 1935/48 2686.800 64.24320
5 hBst.zso 2688.400 .22176
6 t98r.271 2690.000 .20558
7 r975.903 2694.822 .17964
8 r974.342 2696.400 .1803
I

Thble 6.8(c) Expectedcost, risk and transmissionloss when FD = 1400 M

Sr Fl FL
no. ($rn; (Mw)
I 2195.333 2950.80 .79;334
2 2191.831 2952.56 72950
3 2188.678 2954.32 .67876
4 2185.806 2956.08 .63847
5 2183.t64 2957.84 70
6 2180.7t4 2959.60 .58204
7 2r78.429 2961.36 .56348
8 2174.271 296/..88 54130
9 2r72.365 2966.64 36lt6

Thble 6.9(a) Expected generationschedulescorrespondingto the results given in le 6.8(a)

.Sr Pr P3 P4 P6 P7 P8
no. (Mw) (Mw) (Mw) (Mrv) (Mw) (Mw)
1 145.r28 91.628 218.9t2 145.615 74.500 105.194
) 145.035 9r.354 219.479 145.624 74.368 r05.rM
3 1M.943 91.079 220.U4 145.633 74.237 105.094
4 t44.852 90.806 220.604 r45.641 74.t09 105.045
5 144.7
62 90.s34 221.t62 r45.U9 73.981 lM.996
6 rM.672 90.262 221.717 t45.656 73.856 lM.948
7 rM.583 89.991 222.268 r45.664 73.732 1M.900
8 144.494 89.72r 222.8r6 t45.671 73.609 t04.852
9 t44.407 89.452 223.362 145.678 73.489 104.805
t0 tM.320 8 9 . 81 3 223.904 r45.684 73.369 ro+.758
 Power System Optimization

Thble 6.9(b) Expected generationschedulescorrespondingtb the results given in .8(b)

Sr Pl P3 P4 P6 P7
no. (MW) (Mw) (Mw) (Mw) (Mw)
1 r75.r89 132.842 166.741 154.820 114.548 l 2 .527
2 174.948 r32.543 168.513 155:099 113.140 l 2 .378
3 t74.7t3 132.248 170.190 155.348 11r.853 t 2 .232
4 174.483 131.955 t7t.787 t55.572 11,0.667 1 2 .089
5 174.258 131.665 173.318 15s.775 109.566 1 .949
6 t74.037 131.376 174.792 155.961 108.537 1 .8t2
7 173.395 130.503 r78.96r 156.425 105.798 I .40t
8 173.186 r30.234 180.233 r3o.sos r0/.982 I .283

Thble 6.9(c) Expected generationschedulescorrespondingto the results given in Table 8(c)

Sr Pr P3 P4 P6 P7
no. (Mw) (Mw) (Mw) (Mw) (Mw)
194.355 r48.276 r9t.225 169.693 127.r85 1 3 .552
1 9 4 . 1I I r48.103 183.135 169.944 125.534 l 3 .393
193.873 147.927 t84.929 170.t63 124.039 l 3 .237
t93.639 r47.748 186.629 170.357 122.668 l 3 .085
193.410 r47.566 t88.252 t70.529 121.400 I .93s
193.184 147.383 189.809 170.683 t20.2r9 I .787
192.962 147.197 191:310 170.822 119.113 I .642
192.526 146.820 194.167 1,7t.O62 117.088 I .357
192.3t2 146.629 1,95.534 t7t.t67 t l:6.t55 I .2t8

The percenmgedeviation in the cost for different schedules,correspondingto risk, is sho

Figures6.3 and 6.4 for different expecteddemandswhere risk e is proportionalto expected
demand.The curve indicates an increasein the percentagedeviation of cost'of operati
different levels of risk. The operator(DM) has a choice to take risk for minimum ex
and for this minimum risk he has to pay more.

OPTIMALTHERMALPOWER
STOCHASTIC
6.3 MULTIOBJECTIVE
DISPATCH-THESURROGATEWORTHTRADE-OFFMETHOD
In this section,the effect of uncertainsystemparametersis incorporated,explicitly in the
multiobjective power dispatch. Multiobjective problem is stated b$ considering (i) the ex
operatingcost, (ii) the expectedminimum NO, emission,(iii) the expectedtransmissionI , and
(iv) the expected dertiations becauseof the unsatisfied demand. The surrogate worth tr ff
techniqueis discussedto find the compromisedsolution.
 Stochastic Multiob.jective Generation Scheduling

4.5

t 3.5
I
g 3
o

Er.t
' E z
GI

T,
s
1.5

0.5
1.6 t.602 1.604 1.606 1.608 1.61 t.6r2 t.6t4 1.6t6 1.618
d, -+

{- 1240MW - -A- - t320MW

Figure 6.3 Percentagedeviationin cost vs. a.

4.6
A
II

8
o
4.t
.E
tr
o
. 9 4
'oq)
Bq

3.7

r.762 t.764 1366 t,768 1.77 t.772 1.776 t.778

d €

# 1400MW
Figure 6.4 Percentage deviation in cost vs. q,.

6.3.i MultiobjectiveoptimizationProblemFormulation
The multiobjective optimization problem is viewed as a stochasticmultiobjective optimization
problem by consideringthe systempower demand,cost coefficients,NO, emission coefficients
 Power System Optimization

and B-coefficientsas normally distributed,and also as statisticallyindependentrandom variables.

The random generatorpower output P, [Parti et al., 1983] manipulatesany possibledeviationsin
the above-mentionedparametersand in load demand from their expectedvalues. The stochastic
model of operating cost has been defined in Eq. (6.2). The stochasticNq emission model is
describedas under:
The emission curve can be directly related to the cost curve through the emission rate per
Mkcal, which is a constantfactor for a given type of fuel. Therefore,the amountof NO, emission
is given as a function of the generatoroutput P; which is quadratic[Nanda et al., 1988]. i.e.
NG
Fz= ( d , P , tze , P , + f i ) (6.17)
i=l

where dt, er, nd fi are emission coefficients.

As mentionedabove,emissioncoefficientsare realizedas independentrandom variables
becauseof measuringor estimationerror.Moreover,power generationlevel is random,since load
is random.By taking expectationsof emissionEq. (6.17),the expectedNO, emissioncomesout as
NG
F2 = ld,F,'+v,fl + Ii +V,var(4)l (6.18a)
i=l

NG
F2 = r(1+4) a,1'*a,F,+j,1 (6.18b)
i=l

where di, Vi, and fi are expectedemissioncoefficients.

The stochasticmodel of transmissionlossgshas beendefinedin Eq. (6.8a)and is considered
as anotherobjective 4 to be minimized.
To ensurea real power balance,an equality constraintis stated as
NG
P;-Pp=Q (6.1e)
i=l

where PD is the expectedload demand.

The inequality constraintsimposed on generatoroutput are
'
s P-.g P--max
P--min (i = 1,2,..., NG) (6.20)

where 4min and 4t"* are expectedlower and upper limits of generatoroutputs, respectively.
Since generatoroutputs Pi are norrnally distributedindependentrandom variables,so the
expecteddeviations are proportional to the expectationof the square of the unsatisfiedload
demand.These expecteddeviations are consideredas another objective to be minimized and the
objective is given as

F4 (6.2r)
 Stochastic Multiobjective Generation Scheduling

This on simplification reducesto

var(4 )

or
NC
Fq= Z c'oF,' (6.22)
i=l

AggregatingEqs. (6.4), (6.18), (6.8), and (6.22), the deterministicequivalentof stochasticmulti-

objective optimizationprobleln is defined as

Minimize [4,Fr,\,Fflr (6.23a)

NG
st
subject to (6.23b)
LP,-Po=0
f=l

p-. p--max (i = l ,2, ..., NG) (6.23c)

f,mrnS E
where 4 and 4 are expectedcost and transmissionlossesrespectively(see Seetion6.2.1).
F1,F2, F3, and F4 are the expectedvaluesof objectivefunctionsto be minimized over the set of
admissibledecisionvector Pi.

6.3.2 SolutionProcedure
To generatenon-inferior solutions to a multiobjective optimization problem, ffre e-constraint
method is utilized [Haimes, 1977].The t-constraintapproachreplacesthree objective functionsto
consffaintsas given below.

Minimize Fr 6.Zaa)
subject to F,3e, ( / = 2 , 3 ,. . . 4, ) (6.24b)
NG
FT

L P , - P D- 0 (6.24c)
f=l

Emins P-.5 4max (i = I ,2, ...,NG) (6.24d)

where €; is the maximum tolerable objective level for the 7th objective.

Generationof non-inferiorsolutions I

Form the generalizedLagrangian L to the systemrespresentedby

q ( N c
L = 11+ \ I'ri(F,- ej) + trlt; - I (6.2s)
j=z \ i=l

whereLv U = 2,3,4) and p aregeneralized Lagrangian multipliers.The subscript17denotes that

with the jth constraint,wherethe prime objectivefunction
i, is the Lagrangemultiplierassociated
isE
 Power System Optimization

The necessaryconditions to obtain solution are given as

(6.26a)

AL FJ ; - t ;J = 0 (j=2,3,4) (6.26b)
dhtj

N
AL \-r
PD-LP,-o (6.26c)
=--
dl,L
i=l

where

a4 = 2(r+ cf;,)aiFi
+ ui (6.26d)
dP,

oFz +v, (6.26e)

2(t+ CzflV,F,
dP,
NG
= ze+czpE,,P;
+ ) LBUPj * Bio (6.26f)
# j=l
j*i

9 5 =2C2nFt (6.269)
dP,
The Newton-Raphsonmethodcan be appliedto solve the abovenonlinearequations.To implement
the Newton-Raphsonrnethodthe following equationis solved iteratively.

(6.21)

Algorith m 6.22 Non-Inferior Solutionsby the e-Constraint Method

l. Read data, namely cost coefficients,emissioncoeffrcientsand B-coefficients,Err (Conver-
gence tolerance)and ITMAX (maximum allowed iterations), M (number of objectives),
NC (nu*ber of generators) and K (minimum numberof non-inferiorsolutionsrequiredfor
the objective as constraint),etc.
2., Set objectiveindex 7 = 1.
'3. If ( j > IA then GOTO SteP20.
else incrementthe objectiveindex, i = i + 1'
4. Fix e; suchthat F;min< ej. {''* (i = 2, 3, 4).
, i 5. Set iterationfor non-inferiorsolutions,k = O'
6. Incrementthe count of non-inferiorsolutions,k= k + l.
 Stochastic Multiobjective Generation Scheduling 40s
7. If (ft > ,(') GCXI0Step3.
8. Computeinitial valuesof 1 (i = I , ?, ..., NG) andp.
9. Assumethat no generatorhasbeenfixed eitherat lower limit or at upperlimit.
10. Set iteration counter,IT = l.
11. Compute Hessianand Jacobianmatrix elementsusing Eqs. (6.26) and,(6.27), respectively.
Deactivaterow and column of Hessianmatrix and row of Jacobianmatrix representingthe
generatorwhose generationis fixed either at lower limit or at upper limit. This is done so
that fixed generatorscannot participatein allocation.
12. Gauss elimination method is employed in which triangularization and back substitution
processes areperformedto find tF, 1i = 1,2,..., R), 6Lri (/ = 2,3,4), and Ap. HereR is the
numbopof generatorswhich can participatein allocation.

R 4
13. Check either
I ro",lt * I (aLr)z+ (Ltt)z
i=l j=2

if convergencecondition is 'yes' then GOTO Step

14. Modify control variables,

f,new= 4+44 ( i = I , 2 , . . . ,R )
L#'"-)W+L,hry ( / = 2 , 3 , 4 )
[tn* = pt+ L,yt
15. Update the iteration counter,IT = IT + l.
Assign new values to old value variables

Pi = Pin"*(r = I , 2, ..., R)
Lu=Lfr"nand p-Ix**
GOTO Step I I and repeat.
16. check the limits of generatorsand fix up as following

If
4.4minthen 4=4min
If 4 , Pr** ,h"n Pi = P,**

If no more violations then GOTO Step 18.

17. GOTO Step 10.
18. Check the condition is satisfied

Lri [Fi (P) - e;l = 0.0; Alj > 0.0

If 'yes' then GOTO Step 19.
else modify ei and GOTO Step 7.
19. Record it as non-inferior solutions,computevalues of all objectivesand transmissionloss,
and modify €7 for the next nbn-inferior solution and GOTO Step 6.
20. Stop.
 Power System OPtirnizatiott

6.3.3 SurrogateWorth Trade'off Algorithm

are in conflict' The
The SurrogateWorth Trade-off (SWT) analysisis most useful when objectives
A
trade-off analysiscan then be conductedbetween the cost and each of the other objectives'
for the
stepwiseprocedurecorrespondingto an algorithm outlined by Haimes and Hall U9741,
problem is given below.
l. Find the minimum and maximum values for each of the rnultiple-objectivesseparately,
i.e. 4.*,n and {'u* (,1 = l, 2,3, 4).This is carriedout by performingeconomicand
minimum emissiondispatchseparatoly.
Z. Generation of trade-off function: Optimal solutionto problemof Eq. (6.25) must satisfy
Kuhn Tucker conditions.The main condition is

hrilFi- Ejl = 0; )'u >- 0; (i = 2,3, 4) (6.28)

The systemgiven by Eq. (6.25) is solvedfor K valuesof €2,say, e1,,.. ., where elo and
.8f,,
Ff
ef; are held at some level sr9.S"t initial values of e; such that e; > ry" and er'<
= et & - 1,
only those values of 2!rz> 0, which correspondto the active constraint F!
Z, ..., K) are consideredsince they belong to the non-inferior solution. Similarly the trade-
=
off function 213is generated,where Eq. (6.25) is solved for K' different valuesof ef (k
is generated.Regressionanalysisis
1,2, ..., K'), wittr fixed level ef and e!. Simil*!, .1,1a
performedto yield the trade-off functions hnlFzl, hy,IQl and )'slFal.
3. Generation of SWT function: SWT function assignsa scalar value (on an ordinal scale)
to any given non-inferior solution. One way of specifying non-inferior solution is by
trade-off functions.Moreover, there is close.relationshipbetweenthe SWT function 1V1.1 and
the partial derivativesof the utility functions. In multiobjectiveanalysis,it is assumed
implicitly that the DM maximizeshis utility which is a monotonic decreasingfunction
of the objectivefunctions.Given a decisionvector F and the associatedconsequencesFi,
the utility is given by
( J= [ J [ 4 , F r , F, F o ] (6.2e)

By linearizing the utility function for a small changein 4, ,h" following can be obtained
[Haimes,19771.

(6.30)

The SWT function Wry is a monotonic function of Uy, with the property that WU = 0
<-> UU = 0 and is written as
1VU= hi Uti (i = 2, 3, 4) (6.31)

where hi is some monogonicincreasingfunction of its argument,with a range of -10 to

+10 and with the property that hj(O)= 0-
Then
4. Find functionslyri(,1.1)for (j =-2,3,4) by regressionanalysisor by interpolation.
the values .l.y = i; is chosenwhere Wv (h0 = (
0 i = 2, 3, 4).

-'id
 Stochaltic Multiobjective Gencration Scheduling ili l
5. The optimalset of decisionvectoris found by solvingthe following problem.

l--3 I
Minimize 632a)
la.Idr4 l
L i = 2 J
NC
subject to
In-Fp=o
i=l
(6.32b)

sF, s P--max (i = I ,2, ...,NG)

P--min (6.32c)
Form the generalizedLagrangian L to the systemof Eq. (6.32)

L(1,F)- r,+f" ,r,r,* t (rr- X tl (6.33)

j=2 \ i=l )

where p is generalizedLagrangian multipliers.

The first-order derivative necessaryoptimal conditions to obtatn solution are given as

L
5F
=- 4 * $ ^ . & - ( r =l ' 2 " "N' G ) (6.34a)
tr. htrifr-F=Q
a =
=ruL FD-I
; N q =o (6.34b)
d=l

The Newton-Raphsonmethod can be applied to solve the above equaiions.To implernent

the Newton-Raphson method, the following equation is solved iteratively. '

=[-;;]
l1rrI';J[il] (6.35)

Hdssianmatrix elementscan be obtainedfrom aboveequationsby differentiatingwith

respectto to controlvariables,one by one.

=
#=#.L^"# ll ' ' 2
" '
N G) (6.36a)

dzt g
\( ie= -1 , z , : . . , N Gj *; i ; j = t , 2 , . . . , N G ) (6.36b)
W=*n@
azt azt (r=1'2""'NG) (6.36c)
ffi=ffi=-l
,'L^= o (6.36d)
dp'
,fOS- Power SYstem

Utllity functlon
generalutility function for a given DM that can predict
Here, it is assumedthat there exists a very
utility function be defined for each objective function
his behaviour and interest. Let the DM's
function F, to the other objective functions' So,
dependingon the importance of the objective
.rotaror overau utility function is defined as [osyczka and Davies, 1984].
1
rr= t k!F, (6-37)
Maximize L'-
t1-t
' L

t=1

to the
The solution vector P is then found bY maximizing the total utilitY subjected
technology constantsas defined below.
NG
tL P,-Pp=o (6.38)
i=l

P,*" sF,3 4'ou* (i = 1,2, ...,NG) (6.3e)

Further define

(6.a0a)

such that
4

I o; = t; ki>o
(6.40b)
i=l
the aboveoptimiza-
The DM gives the weight w;on the attributebetween(0,99). The solutionof
tion probiem can be obtained as explainedin Section 5'4'4'

6.3.4' SamPleSYstemStudY
its applicability' The'
The method is applied to a six-generatorsample system to demonstrate
expected fuel cost characteristics($ltrl undertakenfor study are as:

Ftt = 0.00542 + 2.0P1 + 100'0

4z = 0.010F22+ 2'0F2 + 200'0

4t = 0.020F: + 2'0F3 + 300'0

Frq= 0.003F42+ 1'95Fq + 80'0

Frs = 0.015Fj2 + t.asP5 + 100'0

\o = 0.010F62+ 0.95P6 + 120.0

 Stochastic., Multiobjective @neration Scheduling

lhe expectedNO, emission (kg/h) characteristicsare:

fr, = 0.000657242 - 0.05497Pl + 4 . 1 1 1

Fzz= 0'00059L6F: - 0.0ss8o 4 + 2.593
Fzt = 0.0004906F: - o.0s0l4P3+ 4.268 '
Fro = 0'0003780F: - 0.03150P4 + 5.526
Fx = 0'0004906F: - 0.05014
4 + 4.268
Fru = 0'00057BF: - o.oss48
4 + 6.132
The expectedB-coefficientsare given in Table 6-10'

Table 6.10 Expected transmissionloss coefficients

0.000200 0.0000r0 0.000015 0.000005 0.000030

0.000010 0.000300 -0.000020 0.000001 0.000010
0.000015 -0.000020 0.000100 0.000010 0.000008
0.000005 0.000001 0.000010 0.000150 0.0000s0
0.000000 0.000012 0.000010 0.000006 0.000020
0.000010 0.000008 0.000050 0.000210
0.000030

at
Table 6.1I shows the conflicting objectives, trade-off functions, utility and SWT function
eabh non-inferior set is shown in
each non-inferior set. The decision vector P; correspondingto

Table 6.ll Expectedcost, emission,risk and transmissionloss along with utility and SWT
function, when demand is 200 MW

Sr. Fl F2 F3 F4 )"t, Wn
no. ($/h; (ke/h) (Mw) (MW2) ($/kg;
74.2663 37r.5998 542.1307 -5
I 1306.917 20.6380 r.9145
1305.927 20.&34 1.9096 73.9515 395.297r 54r.6717 4
2
1304.951 20.6559 1.9071 73.7545 429.8842 541.V+39 4
3
4 1303.951 20.6776 r.9076 73.7r25 no.ss21 540.8398 4
73.9955 500.6307 540.4&5 :7
5 1302.847 20.7r8r 1.9140
74.1388 5W.7724 540.0906 :7
6 1301.824 20.7492 1.9r78
74.0979 500.6307 539.6841 :7
7 1300.818 20.7684 1.9187
1299.788 20.7950 r.9219 74.1986 473.6248 539.2983 4
8
74.1169 415.0005 538.9263 -5
I 1298.892 20.8104 t.9215
20.8202 r.9207 73.935r 328.1359 538.5624 4
10 1298.068
73.8146 22r.8559 538.z;07 -3
ll 1297.315 20.8361 t.922r
73,8709 99.3539 s38.0527 -l
t2 1296.801 20.8&9 1.9263
(Contd.)
410 Power System Optimizntion

Table 6.ll (Contd.)

Sr.
no.
n
($/h1
F2
(ke/h)
F3
(Mw)
n
(MW2)
LB
($nt{w1
wn

I r306.095 20.9202 2.2559 85.5850 ffig.577r 544.1902 -9

2 r305.819 20.9517 2.2899 86.1714 599.0720 544.zrm -'9
3 r305.374 21.0r80 2.3305 87.2824 5r9.2t02 5M.2756 +
4 r3M.7ffi 2t.t077 2.376/, 88.7415 428.t452 5U.3491 -7
5 1303.9U 21.2t35 2.4269 90.4506 325.0275 544.4036 4
6 1303.23r 2r.3202 2.4790 92.r787 207.9039 544.4880 -3
7 r302.859 2t.42t0 2.53t5 93.8470 76.r176 5M.7037 4

F1 Fz F3 F4 )qt Wro
($/h1 Csft) (Mw) (MW2) ($A{w1
I t287.16r 2r.6ffi3 2.&Or 88.7065 27.O5r7 537.4r8r 90
2 1286.748 2t.6ffi 2.4000 88.8003 2r.7904 537.2714 50
3 t286.421 2r.6600 2.4000 88.9000 r8.0675 537.r605 3
4 t286.r56 2r.6ffi0 2.40m, 89.0000 15.2434 537.0745 2
5 1285.938 21.6600 2.40W 89.1000 t2.9794 $7.m71 I
6 t285.757 2r.6ffi0 2.4W 89.2000 r r.0854 536.9547 I
7 r285.607 2r.6ffi 2.4000 89.3000 9.M7r 536.9r47 0
8 r285.485 ?1.6600 2.4p/00 89.4000 7.9907 536.8859 0
9 1285.387 21.6600 2.4000 89.5000 6.6679 536.8669 0

Table 6.12. Iri this'case, the coefficientsof variations of cost, emission and B-coefficientsare
assumedlcfo.The scalar weights are 40.0, 2O.0,20.0 and 20.0 for cost, emission,and power lot;s
and risk objectives in sequence.
By regressionanalysis,the trade-off functions for 20O MW are representedas

trrr(F) = 5717.lgg+ 284.376Fr- 2ffi06F;

hrr(fi = -316.958 - rorz.2nFl

+ 27t9.367Ft
t ro(Fi -- 1550.06
- r7.228F4

with 0.5167,0.9998, and 0.9636 being standarddeviationsrespectively.

The Surrogateworth function as straight line is shown as
V/n()'n)= 0.3583- 0.01427
trrz
wn(\) - 0,1603- o.ors26Lrc
Wu()ru) = -3'8614 + 0'4278)'r4
^ +
with 0.9802,0.9701,and 0.9661 being standarddeviationsrespectively.The valuesof i.1i, Lri,
and ,1.1iare 25.10, 10.50, and 9.026, respectivelyas discussedin Step 3 of the algorithm. Ttre
 Stochastic Multiobjective Generation Scheduling 4tl

Thble 6.12 Generationschedulesfor demand200 MW correspondingto non-inferiorsolutions

Sr Pl Pz P3 P4 Ps P6
no. (Mw) (Mw) (Mw) (Mw) (Mw) (Mw)
I 18.202 28.223 35.16s 33.Mr 29.t02 55.267
J 18.599 28.1I7 35.050 3 5 .I 6 5 28.432 54.637
3 18.684 28.080 34.243 37.t03 27.990 53.900
4 18.458 28.073 33.313 39.318 27.781 53.057
5 17.918 28.O49 32.223 41.989 27.569 52.252
6 17.803 28.291 30.952 44.467 27.387 51.099
7 18.316 28.726 29.443 46.652 27.210 49.654
8 19.r49 29.088 27.750 48.885 26.953 48.174
9 20.870 29.488 25.949 50.623 26.512 46.558
10 23.612 29.855 24.W6 51.751 25.690 45.027
ll 27.294 29.855 21.944 52.352 24.977 43.579
12 3t.205 29.465 19.976 52.523 24.633 42.198

1 4s.895 33.262 26.579 4.368 27.924 61.971

2 47.782 34.575 24.333 3.313 29.4t4 60.583
a
J 49.836 34.931 22.r67 2.569 30.s59 s9.939
4 51.900 34.735 19.927 2.10r 31 . 5 0 3 59.834
5 53.734 34.199 17.548 1.958 32.270 60.292
6 5 5 r. 6 5 33.794 r4.994 2.090 33.023 60.934
7 56.232 33.862 12.276 2.332 33.882 61.416

1 44.092 12.349 6.573 43.918 3 1. 1 8 5 61.884

2 42.698 12.427 6.933 4 5 . 111 30.436 62.394
-
J 41.316 r2.597 7.122 46.274 29.966 62.724
4 39.997 t2.812 7.264 47.337 29.620 62.969
5 38.735 13.0s9 7.388 48.308 29.346 63.163
6 37.516 13.335 7.506 49.202 2 9 . 11 8 63.322
7 36.332 13.638 7.623 50.029 28.924 63.454
8 35.174 13.966 7.744 50.799 28.752 63.566
9 34.O35 r4.320 7.870 5r.5t7 28.596 63.663

optimal decisionvector P is shownin Table 6.14. Correspondingto the non-inferiorset,Tables'

6.13 and 6.14 depict the valuesof variousobjectivefunctionsat the expectedminimum cost,
expectedminimum emissionand maximumexpectedutility schedules.It may be noted that each
step involved in reductionof emissionof NO, becomesincreasinglyexpensive.
In the multiobjectiveframeworkit is realizedthat expectedcost and risk are conflicting
objectivesand are subjectto mutualinterface.The solutionset of the fOrmulatedproblemsis non-
inferior due to contradictionsamongobjectivestakpnand ha$'beehobtainedthroughthe s-constrained
4t2 Power System Optimization

Thble 6.13 Comparison of results

F2 Ft F4
FD FL
(Mw) ($/h) Cs/h) (M!U Ct"tW'l
Minimumcostdispatch
200.0 rzu.365 21.8230 2.4749 93.8997
400.0 1788.286 u.8757 8.7164 355.I 199
600.0 2386.325 37.0950 19j952 813.4999
Minimumemission disPatch
1318.079i 20.2491 t.7949 68.6702
200.0
1878.364 20.6196 6.8843 268.3033
400.0
600.0 2577.228 27.7313 15.3691 605.5116
Ma,rimumutility approach
200.0 r3w.79l 20.3M7 r.7755 6692t5
400.0 1856.690 20.9385 7.2144 268.r749
600.0 2575.991 28.2854 15.3633 600.4974

Table 6.14 Generationschedulescorrespondingto the results given in Table 6.13

PD PL P2 P3 P4 Fs P6

(Mw) (Mw) (Mw) (Mw) (Mw) (Mw) (Mw)

Minimum cost dispatch

14.216 7.108 55.719 27.810 66.716
200.0 28.431
75.490 37.745 18.873 13 4 . 1 5 0 43.497 90.245
400.0
122.549 6t.275 30.63,7 212.582 59.183 rr3.775
600.0
Minirhum emission disPatch
30.409 37.017 35.8r2 35.812 39.125
200.0 ,2r.824
56.052 65.505 70.164 66.409 70.t64 7r.704
400.0
93.992 1M.517 110.994 104.517 r0/..284
600.0 81.696
MaximumUtility
200.0 33.216 32.507 30.186 33.650 fr.zo+ 37.r77
66.707 69.890 56.570 65.79r 68.520 72..522
400.0
98.381 96.977 103.874 104.148 98.156 98.463
600.0

made it possible
technique.The novel formulation as a multiobjective optimization problem has
to quantitatively grasp trade-off relations among conflicting objectives.
The trade-off approachis effective only up to two objectives;as the number of objectives
SWT has
increases,the selectionof the best solution becomescumbersome.An interactivemethod
power dispatch
been applied ro identify the best compromised solution for multiobjective
problem, when conflicting objectives are more than two. The major characteristicshnd advantagen
maker's
of the SWT method are that the surrogate worth functions, which relate the decision
preferencesto the non,inferior solutions through the trade-off functions, are constructed in ttre
iunctional spaceand only then are transformedinto the decision space.
 Stochastic Multiobjective Generation Scheduling 413

OPTIMAL
STOCHASTIC
6.4 MULTIOBJECTIVE POWER
THERMAL
METHOD
DISPATCH_WEIGHTING
The economic dispatch problem was defined so as to determine the allocation of electricity
demand among the committed generatingunits to minimize the operating costs subject to physical
and technologicalconstraints.Most of the existing formulations of the economic dispatch are
solved as static deterministicoptimizationproblems.Actually, there are many inaccuraciesand
uncertaintiesin the input information which lead to deviationsfrom optimal operation and cause
an increasein the cost over the optimal value [Edwin and Machate, 1980]. As a result of the rise
in production costs due to uncertainfactors,the electric energy systemhas been representedas a
network characterizedby random variablesand investigatedby numerousresearchersat various
levels [Dopazo et al., 1975;Parti et al., 1983;and El-Hawaryand Mbamalu, 1991]. Although,
these approacheshave been successfulin applicationsinvolving stochasticeconomic dispatch,
but all the methods do not provide trade-off between economy and risk measuresdue to
uncertaintiesin systemproductioncost and random natureof demand.Typically, such conflicts
exist becauseno such feasiblesolution has been found which would minimize them all-
The pollution minimization problem has attracted a lot of attention due to the public
demandfor clean air. Thermal power stationsare major causesof atmosphericpollution, because
not
of high concentration of pollutants they cause. Since optimum economic dispatch is
environmentallythe best solution, many organrzations in their fight against air pollution have
come up with a new method,the so-calledminimum emission dispatch(MED). MED is used to
minimize the total stack emission(NOr) for the entire system,althoughthis may be controlled
either through post-combustioncleaningsystems(electrostaticprecipitators,stack gas scrubbers)
set
or automatically(controllingunit loading).MED rnay be obtainedby introducinga different
of generatorrepresentations into the economicdispatchproblem. The MED generatorequationfor
($fn)
each unit is a function of stack emission(NOr) versusmegawattoutput, insteadof input
versus megawatt output for the economicdispatch.

6.4.1 StochasticMultiobiectiveOptimizationProblemFormulation
units in
The objective function to be minimized is the total operatingcost for thermal generating
the system and a quadraticoperatingcost curve is assumed'
NG
F r =I ( a i P i 2+ b i P i * c ; ) (6.41)
i=1

where
a;, bi, and c; uto the cost coefficients
NG is the total number of generatingunits'
and load
A stochasticmodel of function F1, is formulated by consideringcost coefficients
be convertedinto
demand as random variables.By taking expectation,the stochasticmodel can
distributed and
its deterministicequivalent.The random variablesare assumedto be normally
function may be
statistically dependenton each other. The expectedvalue of operating cost
1987]' By (aking
obtainedthroughexpandingthe function using Taylor'sseries,aboutmean [Rao,
by
the expectationof the expandedform, the eipected operatingcost function is represented
NG
F =I (a;,
[(o,F,'+6iPi+ c, +d, var(4)+ cov(b,,4) + 2F, cov P;))
(6'42)
i=l
414 Power System OPtimization

where
Ai, bi, and Ei are expectedcost coefficients
P, is the expectedvalue of the ith generatoroutput.
In this study, varianceand covarianceare replacedby coefficientof variation (CV) and
correlationcoefficient(CC), respectively.In general,varianceand covarianceare defined as
var()Q= C'* X' (6.43)
cov(X,If- RxvCxCyXY (6.44)

where
Cy and Cy are the CV of random variablesX and I, respectively.
Rlgyis the CC of random variablesX and Y.
The value of CC is positive or negativedependingupon the sign of the covariance and itri
value lies between-1.0 and 1.0.
Using Eqs. (6.43) and (6.44),Eq. (6.42) can be rewritten in the simplified form as
NG
Fr = fft *C'n*ZRo,r,Co,Cn)o,F,'
+(1 + Ru,r,Cb,Cp)biFi+c;) (6.45)
I
i=l

For the fixed network configuration and random load demand, the equality constraint in the
classical dispatch problem is representedby the expected power balance equation stated as

NG
= Po+Pt (6.46"\
It-
i=l

where P, and FL are the expected load demand and the expected transmission loss, respectively.

Expectedtransmissionloss
The transmissionpower loss expressedthroughthe simplified well known loss formula expressiort
as a quadraticfunction of power generationsis given by [Kusic, 1986]:

NC NG
Pr= I I PiBuPj (6.47"1
i=l j=l

Power generations Pi are dependentrandom variables . Bil are also consideredas inaccuratt:
B-coefficients.The expectedtransmissionlossesusing Taylor'sseries are representedas
NG NG NG-I NG NG
FL= I > FiEuFj+)E,,vu(4)+
I I 28, cov(Pi,P)+ > 2\ cov(Pi,Bii)
i=l i=7 j=i+l i=l

NG NG
+> I 2F,cov(pi,Bu) (6.48)
i=l ,t=l
j#i
 Stochastic Multiobjective Generation Scheduling 415

On simplification, the above equationcan be rewritten as

NG NG NG
Fr= I ft * c7,+ 2Rrr.cncr,,lE,,F,'*
f lft * Rn,p,cncr,
+ 2R4Buc4gn)FiEuFj
6.49)
i=t i=l j;i

8,, *e expected B-coefficients

Rr,t,j are the correlation coefficients of random variables P; and BU
- U are the coefficients of variation of uncertain paramete, Bi
Cp..

Expected deviations
Since generatoroutputs P;s are treated as random variables,the expecteddeviations are propor-
tional tb the expectationof the squareof the unsatisfiedload demand.Theseexpecteddeviations
are consideredas the secondobjective to be minimized. The secondobjective function Fz is
representedas

NG
Fz= +Fr-I (6.s0)
i=l

which on simplificationreducesto
NG NG_I NG
Fz=Iur(4)+I \zcov(P,,Pj) (6.51a)
i=l i=l j=i+L

NG NG NG
F2 = Z r'oF,'
*> I Rnr,cncr,F,Fi (6.s1b)
i=l j=l i=l
j*i

The deterministicequivalentof multiobjectivestochasticoptimizationproblem is formulatedby

taking (a) the expectedoperating cost, and (b) the expectedrisk associatedwith the possible
deviationof the random variablesfrom their expectedvalues.Thesetwo different objectivesare to
be minirnized,while satisfyingthe expectedequality and inequality constraints.Mathematically:

Minimize [Fr,Fzf (6.52a)

NG
subjectto
I "_'= Fo+F, (6.s2b)
i=l

<Fi <fimax
P,min ( f = 1, 2 , . . . ,N G ) (6.52c)

where F, and Fz are the expectedvaluesof objectivefunctionsto be minimizedoverthe setof

admissible
Cecisionvariables,Pt
.
4t6 Power System Optimization

6.4.2 Solution Approach

To generatethe non-inferior solution of multiobjectiveoptimization problem, the weighting
method is used. In this methodthe problem is convertedinto a scalaroptimizationproblem as

Minimize (wfr + w2F2) (6.53a)

NG
subjectto
>i = l 1 = Fo+F, (6.s3b)

4minS 4 <4'* ( l = 1 , 2 , . . . ,N G ) (6.53c)

7
s ,
L*o=l,w?20
it= I

where lr1 &r€ the levels of the weighting coefficients.This approachyields meaningfulresult to
the decision maker when solved many times for different valuesof w1r,k = I,2. The valuesof
weighting coefficientsvary from 0 to l.
To solve the scalaroptimizationproblem (6.53), the Lagrangianfunction is defined as

(_ $q _)
L ( P i , , Ar)r=F r + * 2 F 2 * 2 l F r + P L -pI ,I f6.54)
\ i = l )

where 2 is Lagrangianmultiplier.
The necessaryconditions to minimize the unconstrainedLagrangianfunction are:

aF, ^(?6 \ =o (i=

=*,fr*wzfr-^[fr-tJ
aLaE
r,2,...,NG) (6.55a)
tr
NG
)r
= F ' + F L - I 1= o (6.s5b)
# i=1

The scalar optimization problem is solved using the Newton-Raphsonalgorithm. The size of the
formulated Hessian matrix is the same as that for the deterministicproblem, becausesingle
objective function is solved in both caseswith the same number of constraints.To implement the
Newton-Raphsonmethod,the following equationis solved iterativelytill no further improvement
in decision variablesis achieved.

=[-l;1
l1r^Y,:)[il] (6.s6i)

with respect
Hessianmatrix elementscan be obtainedfrom the above equations by differentiating
to control variables,one by one.

( i = 1 , 2 , . . .N
, G) (6.57ar)
 Stochastic Multiobjective Generation Scheduting 417

a2t dzF; ^ dzF,

anaPj
= r.:
- ::=
L I -:-

aPtaPj anaPj

:
dzt (6.57c)
anu"
azt (6.s7d)
fr
Algorithm 6.3: Non-Inferior Solutions by the Weighting Method
l. Read data, namely cost coefficients,emissioncoefficientsand B-coefficients,demand,Err
(convergencetolerance) and ITMAX (maximum allowed iterations), M (number of
objectives),NG (numberof generators)and K (numberof inferior solutions),etc.
2. Set iteration for non-inferiorsolutions,k = t.
3. Incrementcount of non-inferiorsolutions,k = k + l.
4. If (/<> Kl GOTO Step 17.
5. Feed or generateweights,w; [i = I ,2, ..., M (hereM = 2)]
6. Computethe initial valuesof P; (i = I ,2, ..., NG) and 1, by presumingthat Fr = 0. The
valuesof ,1.and 4 O = 1,2,..., NG) canbe computeddirectlyusingEqs.(3.10)and (3.9),
respectively.
7. Assume that no generatorhas been fixed either at lower limit or at upper limit.
8. Set iteration counter, IT = l.
9. Compute Hessian and Jacobianmatrix elementsusing Eqs.(3.56a) and (3.56e) and
Eqs. (3.57a) to Eq. (3.57d),respectively.Deactivaterow and column of Hessianmatrix
and row of Jacobianmatrix representingthe generatorwhose generationis fixed either at
lower limit or at upper limit. This is done so that fixed generatorscannot participatein
allocation.
10. Gauss elimination method is employed in which triangularizationand backsubstitution
processesare performedto find L4 (i = 1,2,..., R) and A,?".Here R is the number of
generatorsthat can participate in allocation.

I 1. Check either (Ln)' + (Lh)zSeor ffi *t.E (s

J
{}lal
'satisfied'then GOTO Step 14.
if convergenceconditionis
'yes', GOTO Step 14 (It meansthe procedureproceeds
Check IT > ITMAX, if condition is
without obtaining the required convergence).
12. Modify control variables,

4 n ' *= 1 * t E ( i= 1 , 2 , . . . , R )
2,re*=),+ LJ"

13. Update iterationcounter,IT = IT + l,

418 Power System Optimization

Update the old valueswith new values.

Pi = Pnew ( i - L , 2 , ..., R), X,= A,n'*and GOTO Step9 and repeat.

14. Check the limits of generatorsand fix up as following:

If F,.fl^'n then Fi = p--min

F,rF,*^* then pi = flmax

If no more violationsthen GOTO Step 16.

r 5 .GOTO Step8.
r6. Recordas non-inferiorsolution.Compute fog = l, 2, ..., lut) and transmissionloss and
GOTO Step 3 for anothernon-inferiorsolution.

6.4,3 DecisionMaking
Consideringthe imprecisenatureof the DM's judgement,it is naturalto assumethat the DM may
have fuzzy or imprecisegoals for each objectivefunction. The fuzzy sets are defined by equations
called membershipfunctions.Thesefunctionsrepresentthe degreeof membershipin certainfuzzy
setsusing valuesfrom 0 to 1 [Klir and Folger, 1993].The membershipvalue 0, indicatesincom-
patibility with the sets,while 1 denotesfull compatibility.By taking accountof the minimum and
maximum valuesof each objectivefunction togetherwith the rate of increaseof membership
satisfaction,the DM must determinemembershipfunction tt(F) in a subjectivemanner.Here it is
assumedthat p(4) is a strictly monotonicdecreasingand continuousfunction defined as

I ;4<4*"
ri
Fmax - r ir.l
. f , m i n <t lttri ,. < f l m a x
It(Ft) = .-p , . m a-x ; 4ttn (6.s8)
' t '' lm m

0 ;4>Pmax

The value of membershipfunction indicateshow much (in the scale from 0 to 1) a non-inferior
(non-dominated)solution has satisfiedthe Fi objective. The sum of the membershipfunction
values (tt(F) (l = ,1,2, ..., It[) for all the objectivescan be computedin order to measurethe
accomplishmentof each solution in satisfyingthe objectives.The accomplishmentof each non-
dominatedsolutioncan be ratedwith respectto all the K non-dominatedsolutionsby normalizing
its accomplishmentover the sum of the accomplishments of the K non-dominatedsolutionsas
follows [Thpiaand Murtagh, l99l]:

Ii=l
F(F)K
tt|= (6.se)
F(F)K

where M rs the total numberof objectives.

,_..J
 Stochastic Multiobjectivq Generation Scheduting 4le
The function IrB in Eq. (6.59) can be treatedas a membershipfunction for non-dominated
solutions in a fuzzy set and representedas fwzzy cardinal priority ranking of the non-dominated
solutions.The solution that attainsthe maximum membershipttt in the fuzzy set so obtainedcan
be chosenas the best solution or that having the highestcardinal priority ranking.

M a x { p * t k = 1 , 2 , . . . ,K l (6.60)

6.4.4 Resultsand Discussion

The proposedmethod is applied to a three-generator samplesystemin order to demonstrateits
applicability.Expectedgeneratorcharacteristicsare given in Table 6.16..ExpectedB-coefficients
of transmissionloss formula are presentedin Table6.15. In addition,the following valuesof CVs
and CCs of random variablesare assumed.

C o ,= 0 . t , C 6 , = 0 J , C n = 0 . 1 , C B , , =0.1 fl' =.1, 2, 3)

Rqp,= 1.0, R6,4= 1.0, Rn,r,,= ll.0 ( i= i , 2 , 3 )
R n r , = 1 . 0 , R n , B , ,1=. 0 , ( i = 1 , 2 ,3 ; i = 1 , 2 ,3 i i * i )

Thble 6.15 ExpectedB-coefficients

0.0000166 -0.0000102 -0.0000136

- 0.0000102 0.0000219 0.0000025
-0.0000136 0.0000025 0.0000168

Tkftle 6.16 Expected generationcharacteristics

Generator q. bi c; pmax
f,min
no. ($/IvIW2tr; ($/MWh) ($ltr; (Mw) (Mw)
I 0.012 6.51 135.33 105.0 30.0
2 0.005 5.627 261.19 225.0 50.0
3 0.006 5.506 264.63 250.0 70.0

The parameterstatisticsis known from past history. OtherwiseMonte Carlo simulation technique
is a useful tool in simulating the parameterstatistics.Analysis of variation in expectedcost seems
necessarybecauseof the existenceof covarianceof two random variablesin the problem formula-
tion. In the study, covarianceof bivariaterandom variablesis consideredpositive (increasing)or
negative(decreasing)pairwise. Covarianceof one pair of random variablesis consideredat a time,
whereasthe rest of the random variablesare consideredindependent(uncorrelated).All correlation
coefficientsare variedfrom:-1.0 to 1.0 in stepsof 0.5. Assumingboth the weightsw1 and w2 as 0.5,
the percentagerelative changesin Fi from deterministic FL with respectto Rqn, Ru,n,RrB- and
Rnr, (i # i) are shown in Figure 6.5 for the expecteddemandof 350 IvtW. It can be observedthat
(i) there is increasein the percentagerelative cost as Rnp, is changedfrom positive value to
negativevalue, and (ii) there is decreasein the percentagerelative cost as CC of rest of the
random variables are changed accordingly.
420 Power System Optimization

Minimum and maximum values of F1 and Fz

Minimum valuesof the objectives F1and F, *" obtainedby settingweights ln1 to 1.0 and wy to
0.0 and vice versa [Dhillon et al., lg93). Owing to the conflicting natureof the objectives,I will
have maximum value, when 4 ir minimurn. The respectiveminimum and maximum values of
objectivesthus obtained are given below when PD is 350 MW.
max
pimin= 3013.519($/h), F = 3063.917($fn;

4min- n3g.5g3 (Mw2), F{u* = 1267.589

(MW2)

8
(J
L

o
()
a)
O.
x
o
c l.)

'5
't
q.)
\J

6 R 1

-1 -0.5 0 0.5 1
Correlation coefficients
--# C a s e1 --l r-' C aseZ " ' {-" ' C ase3 -' + ' - C ase4

Figure 6.5 Peroentagedeviationin expectedfuel cost with respectto correlationcoefficients.

Over and above, an experienceddecision maker can adopt any other suitablemethod to selectthe
minimum and maximum values of the objectiveswithin which he is expectingthe compromisecl
solution, like minimizing and maximizing each objective function separatelysubject to tht:
required constraints.

of optimalpreferredsolution
Determination
Four different casesare consideredto realize the effect of covarianceof the random variablestcr
eachother (pairwise).
Case 1: In this case all random variablesare considereddependenton each other pairwise.By'
varying weights w1 (from 1.0 to 0.0) and w2 (from 0.0 to 1.0) with a regular decrementandl
incrementof 0.1 respectively,4 and Fz as computedfor non-inferior schedulesare shown inL
 StochasticMultiobiective Generation Scheduling 421

Table6.1:., when F, is 350 MW. Corresponding to the aboveschedules;the expectedgenerations

and expectedtransmissionlossesare shownin Table6.18, Thble 6,17 also showsthe membership
functions p(Fr)k and p(Fz)k of objectives Ft *d Fr, tttpectively, along with the normalized
membershipfunction pt Percentagerelative deviationsof FL and Fz from their respective
deterministicvalues with respectto weightsare shown in Figure 6.6. Best solutionsso obtained
from generatednon-inferiorsolutionsare given in Table 6.19 for expecteddemandsof 350, 400,
450, and 500 MW resPectivelY.

Table G,l7 Expectedcost and risk with their membershipfunctions (expecteddemandis

350 MW)

Sr W1 Fr p(n) It(Fz) PB
no. ($rn1
1 1.0 0.0 3013.519 1267.589 1.00000 0.00000 0.07702
2 0.9 0.1 3013.612 1265.897 0.99815 0.06044 0.08153
3 0.8 0.2 3013.963 1263.94r 0.99119 0.13030 0.08638
4 0.7 0.3 3014.734 t26t.658 0.97589 0.21185 0.09148
5 0.6 0.4 3016.204 1258.96r 0.94672 0.30818 0.09665
6 0.5 0.5 3018.876 1255.734 0.89371 0.42345 0.10145
7 0.4 0.6 3023.724 1251.824 0J975i 0.56312 0.1047e
8 0.3 o.7 3032.785 1247.033 o.61773 0.73425 0.10413
9 0.2 0.8 3050.861 1241.166 0.25906 0.94381 0.09264
10 0.1 0.9 3056.911 1239.891 0.13902 0.98936 0.08691
l1 0.0 1.0 3063.9r7 1,239.593 0.00000 1.00000 0.vlT02

Table 6.18 Expected generationschedulesand transmissionlosses (expecteddemandis

3s0Mw)
Sr W1 Pr F2 F3 FL
no. (Mw) (Mvr) (Mw) (Mw)
1 1.0 0.0 60.869 148.662 146.501 6.0322
2 0.9 0.1 62.952 146.829 t46.0r3 5.7944
3 0.8 0.2 65.424 14./.658 145.438 5.5195
4 0.7 0.3 68.407 142.042 |M.750 5.1982
5 0.6 0.4 72.079 138.828 t43.9ll 4.8r84
6 0.5 0.5 76.714 134.781 t42.868 4.3634
7 0.4 0.6 82.756 t29.523 t4r.532 3.8113
8 0.3 0.7 90.975 t22.4W t39.759 3.1335
9 0.2 0.8 r02.849 112.164 137.289 2.3018
t0 0.1 0.9 105.000 r05.215 141.906 2:I26s
ll 0.0 1.0 105.000 94.313 t52.766 2.0785
 Power SystemOPtimization.

x
'3, z
€
F
C!
v)
o
; r.s
C)
(.)
o
p
xo
. E 1
E
o
.E
€ o.s
bq

0
0.6 0.7 0.8 0.9
Weightageto cost
-f-- Cost --*-- Risk
Figure 6.5 Percentage deviations in expected cost and risk with respect to weight, w1.

Thbie 6.19 Comparisonof results of four cases

FD FL F2 FL PL P2 P3
(Mw) ($n; (MW2) (Mw) (Mw) (Mw) (Mw)
CaseI
350.0 3023.724 1,251.824 3.8113 82.756 t29.523 t4t.532
400.0 3412.855 1641.559 5.1616 93.337 150.823 161.001
450.0 3816.886 2084.501 6.5633 105.000 169.608 181.956
500.0 4232.638 2595.893 9.4991 105.000 187.694 2r6.805
CaseII
350.0 3022.618 125r.642 3.7855 82.r60 r29.996 141.630
400.0 3418.477 1635.635 4.4299 99.79r 145.026 159.6t3
450.0 38r5.815 2083.319 6.4339 105.000 t69.769 181.665
500.0 4230.941 2593.945 9.3079 105.000 187.889 216.419
CaseIII
350.0 2996.655 1251.488 3.7637 82.396 r29.705 r41.663
400.0 3380.846 1641.037 5.097r 92.917 151.043 161.138
450.0 3779.968 2083.282 6.4298 105.000 169:582 181.848
500.0 4189.928 2593.896 9.303r 105.000 r87.639 216.664
CaseIV
350.0 2996.306 436.72r 3.9232 82.219 t35.354 136.351
400.0 3379.355 577.308 5.4529 90.929 t57.r?L 157.403
450.0 3776.r31 739.849 7.3192 99.080 179.538 178.702
500.0 4186.835 929.535 9.7847 105.000 203.656 20t.128

-.-*{
 Stochastic Multiobjective Generation ^Sc

Case2: The resultsfor this caseare computedwith the consideration that

powergenerations areindependent of eachother(Rna, = 0; f =L,2,3, i = 1,2,
randomvariablesare dependenton eachother.With demands350, 400, 450
obtained'best' solutionsare givenin Table6.19.

Case 3: Cost coefficients and power generationsare consideredindependentto each othgr (Ro,h
= 0, Ru,n=0; f = 1,.2,3) along with the consideration that power generationsand B-coefficients
are also independent -
( Rntu 0; l = 1, 2,3, j = 1,2,3), whereasexpectedpower generations EIre
dependenton each other. 'Best' solutions for this case are given in Table 6.L9 for the same
demandsas in the case 2.

Case 4: All random variablesare independentof each other. In this case (R*n - 0, fa,4 = 0,
R n u ,=, 0 ; i = 1 , 2 , 3 , i = I , 2 , 3 ) a n d R n r ,= 0 ; i = 1 , 2 , 3 , i = L , 2 , 3 , i * i ) a r et a k e nT. h e
obtained 'best' solutionsare given along wiih the expectedschedulesand transmissionlossesin
Table 6.19 for various demandsunder consideration.It can be concluded from Figure 616 that a
decreasein the weightageto expectedoperatingcost leads to a decreasein the expectedrisk and
an increasein the expectedoperatingcost. Comparisonof four casesclearly-.showsan increasein
expectedcost as the correlation (dependence)of bivariate random variablescomes into qxistence
(CasesI, 2, and 3) comparedto independentrandom variables(case4). Expectedrisk is always in
conflict with the expectedoperating cost in all the cases.Expected risk is lorver in the case 4
where random variables are independentof each other. Case 2 gives very small decreasein
expectedoperating cost as well as in expectedrisk comparedto the Case 1. This is bepauseof
fixed network configuration. Case 3 gives significant decreasein the expectedcost compared
to the Cases 1 and 2. However,there is insignificant decreasein expectedrisk comparedto the
ones obtained in Cases L and 2, respectively.This is attributed to the fact that the objective
function Fr is a separablefunction and Ro,n, R4Oand Rnr, have no effect on the objective Fz
Considering,the results of all four cases,Table 6.19 shoWsan increasein the expectedoperating
cost as the correlation (dependence)of each random variablesincreases.

LOAD DISPATCH
6.5 STOCHASTICECONOMIC-E]VIISSION
The economic-emissionload dispatch(EELD) problem is a multiple non-commensurable objective
problem that minimizes both cost and emission together.In this section, a stochasticEELD
problem is formulated with the considerationof the uncertaintiesin the system production cost
and random nature of load demand.In addition,risk is consideredas anotherconflicting objective
to be minimized becauseof random load and uncertain system production cost. Validity of the
method has be6n demonstratedby analyzinga sample system consisting of six generators.

ProblemFormulation
6.5.1 StochasticEconomic-Emission
In this section, the multiobjectives with the equality and inequality constraintspertaining to the
power system optimization problem are described.The important non-commensurable objectives
taken into account here are:
. Economic operation
. Minimal impacts on environment
. Expected deviations due to unsatisfiedload
4U Power Systgm Optimization

The stochasticeconomicemissionformulationis adoptedby consideringfuel cost coefficients,

emissioncoefficientsand load demandas randomvariables.The stochasticmodelsare converted
ro their deterministicequivalentsby taking their expectedvalues,with the assumptionthat all the
randomvariablesare normally distributedand statisticallydependenton each other. Expectedfuel
cost and expecteddeviationsare describedin Section6.3.1 and are given in Eqs. (6.42) and
(6.51),respectively.

Expected NO, emisslon

The emissioncurve can be directlyrelatedto the cost curve throughthe emissionrate per Mkcal,
which is a constantfactor for a given type of fuel. Therefore,the amountof NO.,emissionis given
as a function of generatoroutput Pi, which is quadratic,i.e.
NG
F z =I (d,1'+ei4+fi) (6.61)
i=l

where di, €i, and f, are emissioncoefficients.

A stochasticNO, emissionmodel is formulatedby consideringemissioncoefficientsand
load demand at random. Using Taylor's series and taking the expectation,the expectedNO,
emissionis obtainedas'
NG
F? (d,F,'+ AiFj+ ji + Vi ,w (4 ) + cov(e,,P,)+ 2P, cov(di, P)) (6.62)
i=l

where dr, Vi, and L are expectedemissioncoefficients.

Rewriting the above equationas
NG
F2 t ( l + C ' n * 2 R a , 4 C a C p ) d ,+F(,tt+ R r , r , C r , C p ) V ijF, li + (6.63)
i=l

where Co, and Cr, are the coefficientsof variation of random variablesd, and ei respectively.
R4p,is the correlationcoefficientof randomvariablesdi andP;. Rr,4 is the correlationcoefficient
of the random variablesei and Pi.
The expecteddeviationsdue to variancein power mismatchare consideredas the third
objectiveFr, to be minimizedand are given by Eqs.(6.51a)and (6.51b),

Equalityand inequalityconstraints
When the network configurationis fixed and the load demand is random, then the equality
constraintis imposedto ensurereal power balance.This is statedas
NG
Fo+FL-Zn-o (6.64)
t:l

The expectedpower generationis boundedas inequality constraint

f l m i n- <' I E- '<I f i m a x
- t ( i = 1 , 2 , ,. . . ,N G ) (6.6s)
where 4'in and 4maxare expectedlower and upper limits, respectively,of generatoroutput.
 Stochastic Multiobje ve Generation Sch

Expectedtransmissionloss
The transmissionpower lossesexpressedthroughthe simplified well known loss fi la expres-
sion as a quadratic function of the power generationare given by
NG NG
Py- I > PiBuPj (6.66)
i=l j=l

The power generationsPi arerandomvariablesdependenton eachother.84 are also onsideredas

inaccurateB-coefficients.The expectedtransmissionloss using Taylor seriesis nted as
NG NG NG NG_I NG
Fr= I L F,EuFj+
| 8,,uw(4)
+I Pj)
L rE,cov(P,, (6.67)
j=l
i=l i=l ,t=i+l

This, on simplification,can be written as

NC NG NG
Fr= I +) I rt*Rp,p,cr,cr,>lEuFi
e+c218,,1' (6.68)
j=l j=l j=l
j*i

The cieterministicequivalentof stochasticeconomicemissionproblem is defined b considering

three objectives:(i) expectedfuel cost, (ii) expectedNO, emission,and (iii) expec deviations
due to unsatisfiedload demandsubjectto expectedequality.

Minimize lFr,Fr,ri' (6.69a)

NG
subjectto I r, = Fo+F, (6.6eb)
i=l

f,min 3F, S{max ( l = 1, 2 , . . . ,N G ) (6.69c)

where Ft, Fz, and 4 are the expectedvaluesof the objective functions to be mini zedoverthe
set of admissibledecision variable F,

6.5.2 SolutfonApproach
The weighting method is used to generatethe non-inferior solutions of the tiobjective
optimization problem. In this method the problem is convertedinto a scalar optimi on problem
and is given below as

Minimize (6.70a)

subject to (6.70b)

E d n <- 'pI . <- f. i' m

' | t ax (l = 1, 2, ...,-NG) (6.70c)
3

Z * o =l , w * >o- (6.70d)
k=l
 426 Power System Optimization

where w1,?tE the levels of the weighting coefficients.This approachyields meanin

the DM when solved many times for different values of wr1* = l, 2, 3). The val
weighting coefficients vary from 0 to l.
'ro
solve the scalaroptimizationproblem defind by Eqs. (6.70), the Lagrangian
defined as

L(4, L) = (6.7r)
* * F , , + ^ ( r , +;Ft l,
where ,1,is Lagrangianmultiplier.
The necessaryconditionsto minimize the unconstrainedLagrangianfunction

(6.72a)

(6.72b)
where

G + C24+ ZRo,oCo,C
4) + (l + Ru,,Cu,C
e bi (6.72c)
",4
a E
= (t+
-
- - Ciri +2R44Co,C4)7,4
+ (t + R,,nC,,Cr,)
ei (6.72d)
dP,

+d 4 = zc,on.X
'i' R4r,c4cpjFj
'! ' (6.72e)
E,
aF, NG
= 2(r* cA)8,,1+ [t* R4,r,
c4cr,]EuFi (6.72f)
i )j=l
j*i

The scalar optimization problem is solved using the Newton-Raphson algorithm. The
si of the
formulated Hessian matrix is the same as that for the deterministic problem,
becau single
objective function is solved in both caseswith the same number of constraints.To impl
the
Newtonr-Raphsonmethod, the following equationis solved iteratively till no further im1
vement
in decision variablesis achieved.

[o"" v*I [.a"l=[-v"I

vullo^JL-o^J
LvT^ (6.73)

Hessianmatrix elementscan be obtained from the above equationsby differentiating

wi respect
to confiol variables,one by one.

# = * . r W . ^ ( # - r ) ( , = , ,,2N, G ) (6.74a)
 Stochastic Multiobjective Generation Sched, ling 427

dzt ( i = I , 2 ,. . . , N Gj ;= r , z ,. . . , N G ;* t) (p.7+a1
+ '+nI +
a F 4 = w z* @ ffii
a2r dFt -r ( i=1 , 2 , " ' , N G ) (6.74c)
ffi,= fr
* = Q (6.74d,)
ar
where

= 2e+ c,n* 2Rqp,cqcil

ai (6.74e)
W
0rF, =
2(l + C', * ZRo,r,Ca,C
I di (6.74t)
dF,'
ar4 =
zcrn 6.7ag)
dF,'
(i*i) (6.74h)

+ czflE;i (6.741)

a2F =
4 2Q+Rnr,Cncp)Eij Q*j) (6.74i)
E4E,;
Algorithm 6.3 can be used to generatethe non-inferiorsolutions.Use Eqs. (6.72) d (6.74)to
obtain the Hessianand Jacobianmatrix elements.

6.5.3 Test System and Results

The validity of the proposedmethodis illustratedon a six-generarorsamplesystem.1 fuel cost
and NO' emissionequationsare givenin Thbles6.20 and6.21 and the averageexpect transmis-
sion loss coefficientsare giv0n in Thble 6.22 tHill and Stevenson,19681.The operati limits of
the lgeneratorsare given in Thble 6.23.

Thble 6.20 Expectedfuel cost (Rsftr) equations

Ftt = 0.15247
4' + 38.53973
Pr + 756.79886
4, = 0.ros87F] + 46.r5s16F2+ 4s1.32513
Fir = 0.028$F? + 40.396.554+ r049.9s770
4o = o.o3s46F?+ 38.305fiF4+ 1243.53110
F,r = O.OzrttFl + 36.32782
Ps + 1658.5
6960
4u = o.or7zgF? + 38.2704r
F6 + t3s6.6sgzo
 428 Power SystemOptimization

Thble 6.21 ExpectedNO, emission(kgAr) equations

Fzt = o.oo4
D n, + 0.32761P1 + 13.85932
Fzz - o.oo4rs
F] + 0.32767P2 + 13.85932
Fzt = 0.C1683
Fr2 + 0.5a551P3 + 40.2669
Fz+ = 0.00683
F+2+ 0.54551Pa + 40.2669
Fzs = 0.00461
Ps2+ 0 . 5 1 1 1 P
6 s + 42.89553
Fru = 0.00461
Fo2+ 0 . 5 1 1 1 P
6 6 + 42.89553

Thble 6.22 Expected loss coefficients

0.002ct22 - 0.000286 - 0.000534 - 0.000565 - 0.000454 0. 103

-0.0002t86 0.003243 0.000016 - 0.000307 -0.000422 -0. 147
- 0.0005;33 0.000016 0.002085 0.000831 0.000023 -0. 70
- 0.000565 - 0.000307 0.00083
r 0.001129 0.0001
13 -0. 95
- 0.0004
54 -0.000422 0.000023 0.0001
13 0.000460 -0. 153
0.000103 - 0.000147 - 0.000270 - 0.000295 - 0.000153 0.

Thble 6.23 Operating limits

Generator Lower limit Upper limit

no. Pr-min(MW) pmax (Mw )

I 10 125
2 10 150
3 35 225
4 35 210
5 130 325
6 125 315

In addition, the following values of the coefficientsof variation and correlationcoeffici nts are
assumedin the study:

C n ,= 0 . 1 , C b i= 0 . 1 , C p i = 0 . 1 ; (t = I,2,..., 6)
R p , p=i 1 . 0 ( i = 1 , 2 , . . . , 6 i j = 1 , 2 , . . . ,6 i i * j )

Owing to the existenceof variancesand covariancesof the randomvariablesin the formul ion of
the probllem,an analysisof the variationsin objectives Fr and Fz seemsnecessary.In study,
the covarianceof bivariate random variablesis consideredpositive or negative. Si e the
covarianr;eis representedby correlationcoefficients,the correlationcoefficientsare vari from
-1.0 to 1.0 in stepsof 0.5. one pair of randomvariablesis consideredat a time, while rest of
random 'variables are considered independent of each other (uncorrelated). By taking the ights
'tu1,
w2, and b'3 &s 0.4, 0.3, and 0.3, respectively, the percentage relative deviations in FL F2

l--' - ,. .
."r-t
 Stochastic Multiobjective Generation Schedulifg

from their deterministicvalues with respectto Rn,n,Ru,n,Ro,n,R;4, srtd Rnr, Q*j) shown,
one b,y one, in Figures 6.7 and 6.8 respectively,for an expecteddemand bf ZOO w. It is
observ'edthat (i) there is an increasein the percentagerelative deviation in F, and
valueof Rnp,(i * /) is changedfrom a positivevalue to a negativevalue, (ii) there is decrease

1.5

U'

O
'o
q , ) 1
6q) r
a.
x
(.)

c
.9 0.5
(g
l<

C)
bo
E
o
o
o
0.)
tu

-0.5 0 0.5
Correlationcoefficients
---F Case1 ""4"' Case2
-'-c'-' C a s e3 --*-- C a s e4
Figure 6.7 Percentagedeviationin expectedcost with respectto correlationcoeffi

E
'A
16
E
()
!
(,)
()
()
a
5 1 4

Cg
-
1 a
€-) L /-
bo
Gl
!)

(u
O
c)

-1 -0.5 0 0.5
Correlation coeffi cient

# C a se1 --* -- C ase2 ----+ --- C ase3

-'-+'-' Case4 --tC- Case5
Figure 6.8 Percentagevariationin expectedemissionwith respectto correlationcoefficients.
430 Power System Optimization

in the percentagerelative deviation in \ and a small changein the percentagedeviati l l n Fz

as the valuesof Ro,r,al,;rdRu, are changedfrom positive to negativevalues,and (iii) Iere i s a
in the percentagddeviation in Fz and a small changein the percentagedeviati n i n
decreas;e 4
as the valuesof Ro,Oand R",n are changedin the above manner.

Minimum and maximum values of objectives

Minimr,rmvalues of the objectivesare obtainedby giuqg full weightageto one of the bjectives
and nelglectingthe others.When the given weight vahie is 1.0, it meansfull weightage is grvento
the objective and when the weightageis zero the objectiveis neglected.
T'heconflicting natureof the objectives,Fz and F3, will have maximum valuesw n 4 is
minimum. The minimum and maximum valuesof the objectivesobtainedin this way given
below when Pp is 700 MW.
pmn - 38932.88Rs/tr 4t* = 39964.13Rsftr
rnin
f - 49r.320kg/h Ffo - 582.766kgrh

4min = 5291.854
MW2 4** MW 2
= 5613.872

of optimal'best'solution
Deterrnination
(a) Case with dependent variables: In this case all the random variablesare nsidered
dependenton eachother and the weights w1, w2, arfldtil3 zta varied in the range 0.0 to in such
a way that their sum is 1.0. The percentagedeviation in objectives F1,F2, and F3 their
expectedminimum values are shownin Figures6.9, 6.10, and 6.11, respectively, with spectto

0.9
0.8

E 0.7
.5 0.6

.E o.s
a)
:@ 0.4
E
c)
0.3
o
L

d 0.2

0.1

0.1 0.2 0.3 0.4 0.5 0.6 0.7

Weight,wt
-ts 0.6 - -A- - 0.5 ---#-- 0.4
-.-+.-.0.3 --*-- 0.2 ----+---.0.1
Figure 6.9 Percentagedeviationin expectedcost with respectto weights.
 Stochastic Multiobjective Generation m8 431

16

l4

.e4 t z
q)

.E l0
' € 8

o 6
bo

(.) 4
o
fr
o
O.
a
L

0
0.1 0.2 0.3 0.4 0.5 0.6 0.7
Weight,w,
* 0.6 --^-.- 0.5 ----+---' 0.4
-.-(}'-. 0.3 --r(-- 0.2 ---.+-- 0 . 1
Figure 6.10 Percentagedeviationin expected
emissionwith respectto we hts.

3.5

J4
cA
Lr x
.= 2.5
o
C\'

O L
E
C)
bo
cl
c t s
c)
o
Lr
a)

0 0.1 0.2 0.3 0.4 0.5

Weight,w,
=r- 0.6 ---^--- 0.5 - . - - + - - - . 0 . 4
-.-..-.0.3 - - + e - - 0 . 2 -.-+-- 0.1

deviationin risk with respectto weights.

Figure 6.11 Percentage

the various combinationsof weights when the expecteddemand is 700 MW. T conflicting
nature of the objectives can be observedfrom these diagrams.The normalizedmembership
funr:tionp[, obtainedfrom membershipfunctions p(F), p(F), and p(U of objectve functions
 432 Power System OBtiriization

fr, F;, and 4 respectively,is shownin Figure 6.12 for each non-inferior solution. h e ' b e s t '
solutions are given in Table 6.24 and corresponcingto the schedules obtained, th expected
power generationsare given in Table 6.25 for expecteddemandsof 500, 700, and lvlW.

0.019

() 0 . 0 1 8

-q
U)
tr 0.017
()
a.)
F

0.016
c.)
N

tr
r<
0.015
z

0.014
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Weight,w,
--{- 0.6 ---*-- 0.5 + 04
- . - . . } . _ . 0 . 3- - r ( - - 0.2 + 0 1
Figure 6.12 Variationin normalized
m embershipfunctioriwith respectto we

(b) Case with independent variables: In this case all the random variables are consi red inde-
p e n d e n r teoaf c h o t h e r R 4 4 = R a , n =R " , n = R n r , = 0 . 0 ( =
i 1 , 2 , . . . , 6 ;j = 1 , 2 , ., 6). The
'best' lr,,O=
r;olutionsso obtainedare given in Tables6.24 and,6.25 for expecteddemandsof 700,
and 9Ct0MW respectively.

Thble 6.24 Best optimal solurions

Demand Fl F2 F3 PL
(Mw) (Rsih) (kg/h) (MW2) MW)

s00.0 28550.15 3r2.513 2674.s67 7.162

700.0 39070.14 528.441 540t. I 82 4.927
900.0 50807.24 864.060 9 t 10.655 .498
Case 2
500.0 28476.63 287.483 558.758 o.799
700.0 39010.74 493.977 rt14.285 9.669
900.0 50854.86 800.629 1861.073 .032
 Stochastic Multiobiective Generation Schedul

Table 6,25 Expectedoptimalgenerationschedules

Dentand P2 P3 P4 Fs P6
(Mw) (Mw) (Mw) (Mw) (Mw) (Mw)
CaseL
500.0 59.873 39.651 35.000 72.397 r85.241 125.000
700.0 85.924 60.963 53.909 t07.124 250.503 176.504
900.0 122.004 86.523 59.947 140.959 325.000 220.063
Case2
500.0 59.67r 4r.410 51.870 83.265 157.831 126.752
700.0 89.098 65.475 69.713 116.079 223.596 r75.709
900.0 124.610 92.965 87.093 1s2.743 286.843 220.777

The conventionaleconomic thermal power dispatch method allocatesgenerati schedules

to thrl individual thermal generatingunits basedupon deterministiccost function or N , emission
function, ignoring inaccuraciesand uncertaintiesduring estimation.Furthennore,the I demand
is co,nsidered constant,but in practiceit is random. Such generationschedulesresult in the lowest
cost or lowest emission,but the cost or emissionalways associatedwith such a ule has a
relapively large variance and covariance, which can be interpreted as risk measure.
Conventionaldispatch schedules,therefore,render the solution non-optimal a power
system analyst who would like to avoid risks. The feasibility of quantitativere tation of
inaccruraciesand uncertaintiesof the input data and power demand for the pow dispatch
problem in terms of probabitity and statistics,has been investigatedin this chapter. econoruc
dispa,tch problern is formulated as the multiobjective optimization problem, consideri expected
cost and risk measureas two conflicting objectivesto be minimized. The solution of such a
formulatedproblem is non-inferiordue to the conflicting nature of objectives.Fuzz set theory
has trcen applied to get the efficient optimal dispatch from the non-inferior set.
Through this stud!, a theoreticalbasis and methodologicalgrounding for optimal omical
dispartchproblem in a unified multiobjectivefrarnework are established. The proposed on
enabllesthe DM to consider the inaccuraciesand uncertaintiesin the econornicdispatc procedure.
It allows explicit trade-off between the operating cost of units and the risk levels, d provides
the efficient optimal solution from the non-inferior solutions'

6.6 MULTIOBJECTIVEOPTIMAL THERMAL POWER DISPATCH_

RISIVDISPERSIONMETHOD
Until recently, most optimization models have been formulated in terms of min mizing (or
maxjmizing) a single scalar-valued the order of the < y showsa
objectivefunction. Nevertheless,
definritefrend towardsformulatingsuch problemsin terms of multiple objectives.The
of m,anyobjectivesin the planningprocessaccomplishesthree major improvementsin e problem
solvi.ng.First, multiobjective programmingpromotes a more appropriaterole for decision-
maklLngprocess.Second,a wider rangeof alternativesis usually identified when a m Itiobjective
methrodology of a prob
is erhployed.Third, modelsas well as the analysts'.perception become
moro realistic if many objectivesare considered.
434 Power System Optimization

Since mathematicalmodels are an idealizationof actual system models, the vari system
responsesare bound to deviate. The magnitude of such deviations can be best ev luated by
sensitirvityanalysisbecause:
(rDThe stability of the optimal solution rnay be critically dependenton es in the
model parameters.
(t') Some parametersmay be controllableand, therefore,it is importantto know t effects
may result from changingtheir values.
G) Other parametersmay be estimatedmore accuratelyif the solution is critically pendent
on one of them.
'fhe
multiobjectiveformulationis amenableto sensitivityanalysisby specifyinga sensltlvlty
index, that is, a function which indicatesthe relative size of the perturbationsin the so ution due
to var:iationsin the parameters.The specifiedindex may be includedas an additional 1ecilve to
be minimized in a multiobjectivesystem.Then the problem is solved to find the solution
[Osyc:zka and Davies, 1984]. A,second way of applying sensitivity analysisto a mul iobjective
problem is to solve the system;without Eeatingthe sensitivityindex as an objective d then to
evalua,tethe sensitivity after a preferred solution of the original system is found ig and
Haime,s,1983;Kaunasand Haimes,1985].
.,4.classicaleconomic dispatchproblem is formulated as a multiobjective optimi ion prob-
lem consideringtwo non-commensurable objectivesto be minimized, namely the rting'cost
and the impacts on the environment.The formulated problem adoptsan g-constraint , which
allows explicit trade-offs betweenobjective levels for each non-inferior solution t rig and
Haimes, 19831.The effects of random variationsin the model parametersof the opti I solution
to nonlinear programming lead to a sensitivity measurecalled dispersion (O). Th index is
interpretedas a first-order approximationto the standarddeviation in the optimal sol on of the
nonlinearprograms.A sensitivitytrade-offV"Q (AppendixD) is consideredthat gives explicit
represr3ntationof the trade-dffs between the sensitivity and the objective levels. sults are
obtainr:d for two sample systems having three and six generators,respectively.

6.6.1 MultiobjectiveOptimizationProblemFormulation
In the multiobjective optimization problem formulation, two important non-co
objectivesin an electrical thermal power systemare considered.These are economy
mental impacts.The multiobjectiveoptimizationproblem is defined as

N
Ivlinimize F{P'1= I
i=l
(aiPiz+ bi4 + c;) (6.75a)

Ivlinimize F}(P) = I
i=l
(d,Pi'*e,P,+fi) (6.7sb)

subjecil to
I e -(Po+P) =0
i=l
(6.75c)

p,*tn< Pi < P,** (i = 1,2, ...,M (6.7sd)

 Stochastic Multiobjective Generation Scheduling 435

where
A p O i, and c; 31tacost coefficientsof the ith generatinggnit
ffid fi are emissioncoefficients
d; ' €i'
Pp is the power demandto be met
as
P1 is the transmissionlosses,which are approximatedin terms of B-coefficients
N N

PL- PiBijPj
:, I
i=l j=l

prninis the lower operatinggenerationlimit

Pi'n"* is the upper operatinggenerationlimit
N is the number of generatorsin the power system
ectslon
F{p), Fze) are the objective functionsto be minimized over the set of admissible
vectorP.

6.6.2 The e'ConstraintMethod

nstraint
To gene:ratenon-inferior solutions to the multiobjective oplimization problem, the t-c(
of the objective f nctions
method is used [Haimes, Ig77). For the- e-constraintmethod, one
To more
constitutesthe primary objective function and all other objects act as constraints.
problem as de ined by
specific,this procedureis implementedby replacingone objective in the
Eq. (a.5) with one constraint.
(6.t6a)
Minimize lrr(P)]
F j ( P )s e 7( i = 2 ) (6.76b)
subjectto
N
6.76rc)
Ie-Qo+P)=o
i=1
p m i n< P i < 4 * * ( i = 1, 2 , . . . ,M (6.76d)

for w ich the

where ,rr.is the maximum tolerable objective level. The value of €, h chosen
solut on. The
objectiye constraintsin problem defined by Eqs. (6.76) are binding at the optimal
functi Fr(P)-
level of e; is varied parametricallyto evaluatethe impact on the single objective
into an unc( strained
To find the solution, the constrainedproblem of Eqs. (6.76) is converted
problenr.The generalizedLagrangianL is formed as
(-r )
L- Frr)*Ln@z,;)-82)-ltlI -PL-pol (6.77)
\td "t )

that 2 is grange
where Lrz and lt are the Lagrangemultipliers. The subscript t2 denotes
function is r(P).
multipl:ierassociatedwith the secondconstraint,where the prime objective
The necessaryconditionsfor the optimal solution of Eq. (6.77) are

aL = a4(P)t- x aFre)u( ,-%) - o (6.78a)

E a4-atutzq-',[r An)
 Power System

AL = F z ( P ) - t z = O (6.78b)
oLt,

aL D +h (6.78c)
f r = P o + ' Pd L
=l
-LP,=0
The optimal solution must satisfy the above Kuhn-Tuckerconditionsbesidesthe follow ng main
condition:
)42(Fz(P) - ez) - 0.0; )42 > 0.0 (6.78d)

The Lagrange multipliers related to the objectives, as constraints may be zeto or non- . The
set of rlon-zero Lagrange multipliers correspond to the non-inferior set of solutions. e set of
non-zerc Lagrangemultipliers representsthe set of trade-off ratios betweenthe principle bjective
and each of the constrainingobjectivesrespectively. The systemgiven by Eq. (6.77) is ved for
R values of E2 using the Newton-Raphson method. Only those values of L$ > 0 which
corespond to the activeconstraintsFi(P) = ei, r = L,2, ..., R are considered,sincethey ong to
the norr-inferiorsolution [Haimes, t977).
Irr the problem, the initial value of s2 is taken such that tz ) Fjtn and E2 1 F2 . Since
cost and emission are of conflicting nature,the value of objective Fy will be maximum, hen the
valueof F, obiectiveis minimum and vice versa.So, minimum and maximum valuesof 1 and F2
are obtained by performing economic dispatchand minimum emission dispatch separa ly.
Similarly, for more than two objectives,the trade-off functions LU Q + il, can be enerated
-
with respect to the fth prime objective, by varying e; while other e* (k /) are fixed at some
level. l'he ith prime objective can be replacedby the yth objective and the solution rocedure
is repeated for more information. Algorithm 6.2 can be implemented for the ge tion of
non-infbrior solution.

6.6.3 ParameterSensitivity
In practice, the values of the objective and the constraintsof the problem defined by . (6.76)
will actually depend on the decision variable P;. Usually each equation is formulated in terms of
the decision variablesand other parameters.Therefore,the values of the objective and straint
functions depend not only on the values of the decision variables but also on thc alues of
coefficirentsdi, br, ci, dr, €i,,f, and Bii (i ='1, 2, or these
coeffic:ientsare approximatedby estimation, statisticalaveraging,past experience,and there-
fore pr,oneto error.
Irr the light of the above consideration,-themultiobjeCtiveoptimizattonproblem as given by
the prc,blemdefined by Eqs.(6.76) is reformulatedby assumingthat, in addition to decision
variables Pi(i = I, 2, ..., N), the objective function is dependenton a set of'cost x;(i =
1,2,..., 3M) and the constrainedobjectivealong with the problem constrain.r,s is depen t o n a
technology parameter set yi! = I,2, ..., m).
I\4inimize lFr(P,x)l (6.79a)
subject to \(P,y)sei U=2) (6.7eb)
4(!, Y)= u (6.79c)

.d
 Stoclmstic Multiobjective'Generatiort Schedulilry

where
ct,..,cNlT
lQ-..,ay, b1,..,b7r1,
v- fr,..rftu, Btt,
ld1,..dy, €1,..,€p1, 812,.-.,Bruru]'
N
q ( P , ) ) I= a . - P L (6.7ed)
i=l

u = Po and m = 3N + (N(ru+ l))12

6.6.4 l?isk Index and Sensitivity Trade-offs

Supposethe problemdefined by Eqs. (6.79) admitsa local solutionP* anclsatisfiesthe su iciency
conditionsfor a strict local minimum. Then, for a sufficientlysmall neighbourhoodabou z, the
optimal l'* to problem (6.79) is the same[Rarig and Haimes,1983] as the solutionto:
Minimize lF(P, x)l (6.80a)
subject to g(P, !) = z 6.80b)
where the set of binding constraintsset F2(P,y) and q(P, y) by vector equation 8(P, y) 4 \ 4 -

lerlu)').1lhe dependenceof the optimal solutionF* to the problemof Eqs. (6.80)on the pa eter
setsx, y, z ca.nbe expressedby writing
F* = F* (*, y, z) = F* (w, z) (6.81)
wherew = f, * I y lr"
The parametersetsr and y are the approximatedvaluesof the actual unknown
Keeping this in view, x and y can be treatedas random variables.Therefore,wi dte
random variableswith meansand variancesas

frr = E(wi) (6.82a)

o,2=var(w;) wherei - I,2, ...,3N + m- 6.82b)

If it is arssumedthat e is a deterministicvariable,then the optimal solution F. to the roblem
' n t | (

definedby Eqs. (6.80) can be consideredas a function of the random variableW ; , 1 . 9 . f = F. (w)-
By ignoring higher terms, Taylor seriesexpansionabout the mean is

F*=p.(*).* (wi-wi) (6.83)

t#]
whereL '= 3N + m.
Tal,;ingthe expectedvalue and varianceof the aboveexpression,a first-orderapproxmatron
to the mean values and variancesof random variableF* is
E(F-) = F.(w)

var(F*)= o! (6.84)
l[#)'
where thrl partial derivatives are evaluated at the point w i = wi
 Power System Optimization

A new scalar sensitivity variable to be the standarddeviation of F* is defined.This can be

interpretedas a measureof dispersionof the optimal solution F* due to uncertarntl in the cost
and technology parameters.
By definition:

O- (6.8s)

Since tj colrespondsto a particularvalue of zi, the trade-offbetweenthe sensitivity d objective

leve,ls,aCyae;is obtainedby differentiating(6.85) with respectto zi, i.e.

ao= [ ar".)l f
_ tT ^?(aF.')
a, l*"tlm;JtrrilJj
d (6.86a)

where
V*F* = LL* Vrg I VrFl (6.86b)

(6.86c)

where /<is number of binding constraints.

For simplicity, the sensitivity trade-off ratio can be evaluated as

a4
a4 = -
dti
where P *o (6.87)
acl aC, dtj
dti
As the numberof objectivesincreases, the trade-offsof sensitivitywith respectto obj, tive levels,
ti, ire determined.Thesetrade-offsare the componentsof the vector z;, which are cal ulatedfrom
Eq. (6.86a).The calculatedtrade-offsare submittedto the decisionmaker to select preferred
solution.

Derivativesof Y*F' and y ,rF

In order to calculate V * *,
and Y ,*F considerthe following optimization problem:
"F
Minimize F(p, x) (6.88a)
subjectto g(p, y) = z (6.88b)
In order to satisfy the second-order
sufficiencyconditions,the following system equations,
the t;o-calledKuhn-Tuckernecessaryconditionsare to be solved:

VpF (P, x) + ).Yp8(P,)) = 0 (6.89a)

g(P, !) = z (6.8eb)
If thresecond-ordersufficiencyconditionsare satisfiedthen the optimal solution P* an the corres-
ponding LagrangeMultiplier )", satisfyingthe optimization problem as defined by Eqs (6.88),are
 slochastic Multiobjective Generation scheduling 439
implicitJlydefined as function of y, x, and z in Eqs. (6.89a) and (6.89b). So p* and 1' can
be
wntten its
*
P = p*(y,x, z) (6.e0)
A,*= ^,*(J, x, z) (6.e1)
Using thesefunctionalrelationships, V*f andY*fcan be calculated in a straightforward
manner.
This is rlone by using the chain rule to successively differentiate the system of Eqs. (6. 9a) and
(6.89b) with respect to the vector variables y, x, and z. Since w = Iy *l', to obtain the desired
|
quantities, vrF* and Vr*t', it is required to calculate VrC, Y*f ,yoFI ,andvuf .

Cafcufations tor Y,F- and YoF'

The optimal objectivevalue F* to the problem of the systemof Eqs.(6.88) can be w tten as
F* = F(,P-).Differentiatingthis expressionusing the chain rule with iespectto y gives

V r F *= Y p F Y r P * (6.e2)
Next difflerentiating
the expressiong(P*, !) = z yields
VpgVyP*+Vr8=0 (6.e3)
Multiply Eq. (6.89a)by VuP*to obtain
YpF YrP*+ L.Yrg VrP* = 0 (6.e4)
MultiplyEq. (6.93)by t to obtain
L. YrF YrP*+ t Vrg = o (6.es)
After solving Eqs. (6.94) and (6.95), the following can be obtained
YpF YrP* - A) Yyg (6.e6)
EquatingEqs. (6.96) and Eq. (6.92), gives

Y rF* - L*Yrg (6.e7)

Finally, ffansposingEq. (6.97)and differentiating
with respectto z by successive
appli tion of
the chain rule yields

Y ryF.= (Vyg)r1Vrt1r + (6.e8)

Calculallions tor Y ,F.and Y ,*F.

Assume that the optimal solution to problem in Eq. (6.88) dependsonly on the parame SEtS
x and 3, whereasall other parametersare kept fixed. Then, the optimal objective value can be
written as
F* = F(P", x)
Using the chain rule, differentiatethe optimal objective function with respectto x, to

V;F*=YpF%P-+VrF
 Power SYstemOPtimization
problem efined b,y
Kuhn-Tucker necessary conditions for an optimal solution to the
(6.88) are
YpF+1,V6=9 (6.100)
8(P) = 0 (6.101)

In theseequations,p* and A,* aredefinedimplicitly as functionsof x and z. Multiplying

(6.100)
by V,p*, and differentiatingEq. (6.101)with respectto,r yields two new set of.e(uat ons
vpFvrP-+tv6V,P*=o (6.102)
V P g V T P=* 0 (6.103)

Substitutingthe result of Eq. (6.103)into the Eq. (6.102),we get

YPFVTP-=0

this result into Eq. (6.99),gives

Subs,fituting
Y r F *= Y * F (6.104)

Diffe,rentiatingthe aboveresult with respectto 4, using chain rule yields

Y uF* + Vr(V ,F)r Y rP*= Y pf Y rP* (6.10s)
for Y*l* and V.*F*
CombiningEqs. (6.97),(6.98),(6.104)and (6.105),the following expressions
are obtained
v*F- - I vrF-] = h- vy8| vrF-l
[VrF- (6.106)
and

(6.107)

The implicit function theorem assuresthat in some neighbourhoodof zero, there ex a unique
L* - h*(z),P* = P.(z)to thesystemEqs.in (6.100)and(6.101).Moreover, *(z)
solurfion exists a
neighbourhood the functionP-(z) and
of zero suchthat for any z in neighbourhood satisfy
the second order sufficiency conditionsfor a strict local minimum.
So V.P*and YrAlcan be expressedas

[o.".-l_l v r r (vps)rl[ol
- Lvrrr (6.108)
Lv.r.J o llrl
PurringEq.(6.106) in Eq. (6.107),the requiredexpressionis obtained.
 Stochastic Multiobjective Generation Scheduling 441

6.6.5 TgstSystemand Results

Resultsare obtainedfor two samplesystemsto illustratethe applicabilityof the meth
Case 1: The method is applied to a three-generatorsample system including transmisson loss.
Expected generator characteristics are given below:

Fuel characteristics ($/h) are:

F r r = 0 . 0 1 0 P+? 2 . 0 P 1
+ 10.0
F r z =0 . 0 1 2 P+1 1 . 5 P+2 2 0 . 0
Ftt= 0.00aPl+ 1.8P3 + 20.0
NtC,,emission characteristics(kg/h) are:

F z t = 0 . 0 0 0 6 5 7 0 -P0?. 0 5 4 9 7 P+1 4 . 1 1 1
Fzz= 0.00059 I6P3- 0.05880P2 + 2.593
Fzt= 0.0004906P? - 0.05014h+ 4.268
The B-coefficients of transmissionlosses(MW-l) are:

Btt= 0.0002725Bzz= 0.0003090 Bn = 0.0032295

Bn= Bzr ,=-0.0000351
Bn= Btr= -0.0003679
Bzt= Btz= -0.0000565
Table 6.26 displays the results obtainedfor particularvalues of the standarddeviati 6;-
Cr,, xfri, (i = L, 2, ..., 24) at various objective levels. The values of coefficients of vari tion for
cost and emissionparametersare assumed0.IVo. Coefficientsof variationsfor transmi ion loss
parameters are assumed to be zero, because of small variations in these parameters [Pa i et al.,
19831.lt'his table depictsthe meanvaluesF1 and F2, dispersionQ of F1 aboutthe mean Iue, the
sensitivjitytrade-off ratio of the function Fr at the dispersionand trade-off )"p along ith non-
inferior optimal generationschedules.C) conveysthe significant information to decisio maker,
i.e. the larger the value of C), the greater the possibility that the actual solution eviates
significantlyfrom the nominal solution.In the caseof 150 MW there is dispersiono 74.64r
about the mean cost 390.852 $/h and sensitivity trade-off ratio is 0.0708. About the cost
390.705,$/h, there is dispersionof 35.041and the ratio is 0.0643.The decisionmaker w ll prefer
the schr:dule of cost 390.705 $/tr which has comparatively small dispersion. The decisi maker
will pre'ferthe next schedule,corresponding to the cost 390.603$/h (5th set) rather than the cost
389.10$/h (6th set) becauseof a smalldispersion. It is apparentthat eachstepin the red ction of
emissionof NO, becomesincreasinglyexpensivewith more power losses.Table 6.27 ives the
non-inferiorschedulescorresponding to the resultsin Table 6.26.
From the resultsof minirnumcost and minimum emissionshownin Thble6.28,it is rved
that thereis an increasein the expectedfuel cost by 1.161$ftr and the reductionin the xpected
emissionis 0.039 kg/h for 150 MW. For 200 and 250 MW the expectedcostsincrease y 2.022
and 3.251 $/h and the expectedemissionsreduceat the rate of 0.08 and 1.141kg/h ively.
It is observedfrom the preferredexpectedsolutionsgiven in Table 6.29 that for a red ctlon ln
1 kg of' NO, per hour, the costsare increasedby 1 $/h for 150 MW 1.77 $th for 200 , and
3.02 $Ar for 250 MW respectively.
 Power System OPtimization

Table 6.26 Results

Demand Sr. Fl F2 trr, o aFtt&

( : M W) no. ($/n1 Ge/h) ($rkgi
150.0 I 390.852 7.t48 320.677 74.&l 0.0708
2 390.705 7.r55 142.071 35.041 0.0ft3
-J 390.656 7.160 92.093 23.060 0.0588
4 390.623 7.165 54.196 r3.90r 0.046s
5 390.603 7.r70 24.080 6.752 0.0285
6 3 8 9 l. 1 0 7.192 l89.l6l 48.066 0.2269
200.0 I 551.639 8 .r 3 8 345.8M 163.159 0.0335
2 551.306 8.160 r99.028 r04.528 o.0927
3 551.225 8.165 187.008 98.854 0.0994
4 5 5l . 1 0 4 8.170 85.359 44.513 0.0569
5 55r.06r 8.175 63.1Lz 33.432 0.0521
6 55r.026 8.180 43.707 23.763 0.0448
7 550.998 8.r85 26.56r 15.381 0.03s2
8 550.997 8.190 rr.238 8.685 0.0244
9 550.134 8.330 336.504 167.771 0.58ss
250.0 I 135.869 r0.503 133.873 tr4.75r 0.0540
2 735.4r4 r0.525 r1.343 20.r03 0.0412
3 735.4r3 r0.526 l 1.025 15.74r 0.0339
4 735.402 10.528 4.082 12.474 0.0304
5 735.3r9 ro.529 53.768 5r.9& 0.0898
6 735.t93 10.539 65.702 63.461 0.1090
7 734.878 10.555 1r9.829 115.424 0.1591

Table 6"27 Non-inferior schedulescorrespondingto the results given in Ta 6.26

Generationschedules(MW)
Demand- Sr.
(Mw) no. Pl P2 P3

150.0 I 48.2r9 58.024 51.688

2 49.566 57.M7 50.560
49.824 57.&l
-t
J 49.932
4 50.062 57.816 49.39
5 50.285 57.974 48.841
6 50.358 59.224 46.390
200.0 I 68.645 77.694 66.869
2 72.379 74.r10 66.372
3 72.605 74.246 65.810
4 71.807 76.238 &.r03
5 72.075 76.212 63.724
(Contd.)
 Stochastic Multiobiective Generation Scheduling

Table 6.27 (Contd.)

Generationschedules(MW)
Demand Sr.
(Mw) no. Pl P2

6 72.337 76.175 63 7 r
7 72.591 76.t35 63 37
8 72.836 76.094 l8
9 70.576 84.195 53 29
250.0 I 92.289 99.35r 75 75
2 96.169 94.r76 77 40
3 96.077 94.456 76 52
4 96.025 94.723 76 l 6
5 96.803 92.978 78 25
6 97.251 92.792 77 32
7 97.804 92.894 76 39

Table 6.28 Minimum economicand emissiondispatch

Generationoutput (MW) Power FueI

Demand loss cost
(Mw) Pl P2 P3 (Mw) ($n1 s/h)
Economic clispatch
150.0 s0.& 58.21 48.03 6.8 390.592 .t79
200.0 73.84 75.90 61.46 ll.l2 550.938 .212
250.0 98.14 94.36 73.63 16.13 735.209 .592
Minimum emissiondispatch
150.0 47.46 55.67 56.r2 9.25 391.753 .140
200.0 67.31 76.48 71.11 r4.87 552.960 .r32
250.0 88.62 98"32 83.91 20.85 738.460 .451

of results
Table 6.29 Comparison
Demand Dispatch Fl F2
(Mw) ($n; (kg/tt)

150.0 Minimum cost 390.592 7.t79 80

Minimum emission 39r.753 7.140 .25
Combined 390.603 7.r70 .10
200.0 Minimum cost 550.938 8.212 I .12
Minimum emission 552.938 8.132 I .87
Combined 550.977 8.r90 I .64
735.209 r0.592 .l .13
250.0 Minimum cost
Minimum emission 738.460 10.451 .85
Combined 735.402 10.528 .48
M Power SYstemOPtimization

to illustratethe methodand to evaluatethe mpromised

Casie2z A largesystemis considered
solutionfrom the combinedogonomic-emission dispatch.The expected generator c racteristics
are given below:
($/tr)are:
Fuel characteristics
Ftr = 0.005P? + 2'00P1+ 100'0
Fn= 0.0lOP7+ 2.00P2 + 200.0
Frt= 0.020P? + 2'00P.+ 300'0
Fu= 0-0CBP7 + 1.95P4 + 80.0
F r s = 0 . 0 1 5 P+3 1 . 4 5 P+5 1 0 0 ' 0
Frc= 0.010Pe + 0.95P6+ 120.0
(kgft) are:
NO, emissioncharacteristics
2P?- 0.05497
Fr = 0.000657 P1 + 4.111
- 0.05880P2
Fzz= 0.00059n6P? + 2.593
- 0.05014P3
Fzt= 0.0004906P? + 4-268
Fz+=0.0003780P?- 0.03150P4+ 5.526
-
Fzs=0.0004906P?0.05014P5 + 4-268
LBPA- 0.05548P6
Fzs= 0.0005 + 6.132
Tfte of variation
aregivenin Thble6.30.The valuesof coefficients
B-coefficients emission,"
and areassumed
B-coefficients to be O.lvo.

Table 6.30 B-coefficientsx tO2 (UW-l)

Cr.0200 0.0010 0.0015 0.0005 0.0000 0.0030

c).0010 0.0300 -0.0020 0.0001 0.0012 0.0010
0.0015 -0.0020 0.0100 0.0010 0.0010 0.0008
0.0005 0.0001 0.0010 0.0150 0.0006 0.0050
0.0000 0.0012 0.0010 0.0006 0.0250 0.0020
0.0030 0.0010 0.0008 0.0050 0.0020 0.0210

Table 6"31 displaysthe mean,cost,emissionand lossesobtainedfrom minim cost and

minimum emissiondispatchprocedures.The combined dispatch of economic emi sion is also
presentedin Table 6.31. The generationschedulesare shown in Table 6.32, c ponding to
dep,ictedresultsin Table 6.3L.
This section incorporatesthe sensitivity measureinto multiobjective optimi tion, whish
gen.eratesnon-inferior optimal solution with respect to the objective level utility and the sensitiv-
itv index. The index providesuseful informationabout the distributionof the opti I solutionin
the p;esence of variations in the model parameters defining the problem. The decis on maker is
able to analyze the sensitivity information conveniently, since sensitivity index is a ar-ualued
quarntity,regardlessof the number of objectives.The most important characteristic f sensitivity
ind,exis that a sensitivitynade-off,is calculatedat each non-inferiorpoint. This all the decision
maker to know the trade-offsbetweenthe objective-,levelsand parametersensitivity.
 Stothnstic Multiobjective Generation Sch

Thble 6,31 Resultsfor six-generatorsystem

Dispaitch Demand Cost

Mininrum cost 200.0 1290.071 21.665

400.0 \813.l 56 24.782
600.0 2452.050 37.672
Mininrum ernission 200.0 1322.448 20.222
400.0 t901.441 20.757
600.0 2643.M3 28.601
Combineddispatch 200.0 t291.41r 21.100
400.0 1818.828 23.036
600.0 2462.224 33.749

Table 6.32 Generationschedulescorrespondingto the results given in Thble 6.

Dem"and Pl P2 P3 P4 Ps

Minimumcost
200.0 29.886 r5.560 8.193 55.416 28.222 6 5 . 18 6
400.0 78.397 40.818 2r.732 133.535 45.451 88.959
600.0 12i.878 66.914 36.060 2t2.450 63.M9 113,216
IMinimumemission
200.0 30.638 37.356 35.994 22.T33 36.194 39.514
400.0 57.015 66.398 7r.747 68.100 71.222 72.634
600.0 84.025 95.916 108.806 rr4.795 106.622 105.969
dispatch
Combinred
200.0 31.863 21.725 12.158 45.M5 30.2r5 60.856
400.0 63.349 5r.969 31,.384 120.118 52.952 88.463
600.0 rt2.823 79.026 48.674 187.398 74.r79 116.628

llhe proposed method suffers from the drawback of computationalefficiency f the e-

consffarintmethod when there are severalobjectives.The number of solutions of the lem in
the non-inferior set increasesexponentiallywith the number of objectives.There exists itional
difficulties of evaluation of derivativesbefore calculating the sensitivity trade-off. large
systems,there is computationburdenfor which the only recourseis to calculatethe vatives
numerlically.
lihe results summarizedin this chapterreinforce the following premisesupon hich the
study 'wasbased.
' Risk assessmentmust be consideredas an integral part of the multiobjectiv decision
making process.
,, tJncertaintyin model parameterscan be quantified in terms of sensitivity,where e-offs
among cost, sensitivity of model parameterscan be evaluatedwithin a mul objective
risk-baseddecision-makingprocess.
 Power System Optimization

index methodcan be modified to incorporatesome specifi attributes

The Risla{Dispersion
associatedwith the power dispatchproblem.

6.7 ISTOCHASTICMULTIOBJECTIVESHORT.TERM
IIIYDROTHERMAL
SCHEDULING
The sr:hedulingproblem requiresan appropriateobjective function such as minimum neration
cost or transmission lossesor minimum pollutionlevel, etc. In essence, the hydrothermalcheduling
problem can be visualized as a multiobjectiveone. A simple and robust solution me ology for
a class of multiobjectiveshort-rangefixed-headhydrothermalproblems is undertakn which
makes use of fuzzy set theory.First of all, a short-rangefixed-headhydrothermals uling
problem is formulatedin a unified multiobjectiveframework consideringstochastic st, NO,
emission, SO2 emission,and COz emissioncurves for thermal power generation nits and
uncertaintyin systemload demand.The expectedvaluesof thermalt'uelcosts,NO, emis o n s , S O 2
emissionand CO2 emission,over whole of the planningperiod are the four conflicting bjectives
to be rninimized.Further,the expectationof the squareof the unsatisfiedload becauseo possible
varian,ce of generator outputs over whole of the planning period, is incorporated another
objective to be minimized. Basically, the solution procedure for the multiobjective prob is based
on the generationof non-inferiorsolutions.The weighting techniqueis used to gen ate non-
inferior solutions,which allows explicit trade-offbetweenobjectivelevels.Exploiting fuzzy set
theory [18], a cardinalpriority rankingof the non-inferiorsolutionsis definedthat maxi izes the
satisfactionof all the objectivesand is utilized to find the best compromisingsolution from the
non-inferior solution set. The practical viability of the multiobjective hydrothermal heduling
proble:mhas been demonstrtedon three samplesystems.

6.7.1 StochasticMultiobjectiveOptimizationProblemFormulation
Consideran electric power systemnetwork having N thermal generatingplants and M hydro
plants,where M + N is the total numberof generatingplants. The basic problem is find the
activepower generationof eachplant in the systemas a function of time over a finite ti period
from 0tto T.

Stochrasticthermal model
The objective function to be minimized is the total systemoperatingcost, representedb the fuel
cost of thermal generation,over the optimizattoninterval.

rt= ( a i P+i zb i P i * , , > ) a , (6.10e)

l[*
where
ct;, b;, and c; are the cost coefficientsof the rth generatingunit
1/ is the numberof thermalunits
?^is the total planning period
-of
l\ stochasticrnodel of function Jt is formulated by considering errors in cients
input-output characteristics and load demand during each subinterval as random va les. Any
 Stochastic Multiobjective Generation F'ht!!!!E

possible deviation in coefficients of input-output characteristics and load demand fro their
,*p""t"d values are manipulated through the randomness of random variable, P;. A str tic
model is reducedto its deterministicequivalentby taking its expectedvalue. Presumin that

ranciomvariables are normally distributedand statistically independent,the expectedva ue of

operatingcost comes out to be

Jr = a'
+d,+ a; vartnrl)
@,1'+ biFi . rl 0 )
I [*
where
Qi, ,bp and di, are expected cost coefficients of the ith thermalunit
p
t I
jrsthe expectedpower gcnerationby the ith thermalunit.

The varianceof power Pi is given as

var (P;)= COF,' ( i = 1, 2 , . . . ,I 9 . 1 11 )

where C4 is the coefficient of variatronof random variable Pi

The zero value of the coefficientof variationimplies no randomness,in other words, t is
en as
completecertainty about the value of the randomvariable. Equation (6.110) can be rewri

Jt = (tr* czo)a;r,, al))a,

+b,F,. .r12)
I [}
Stochasrticemission models
conce on
Thermal lpowerstationsare a major causeof atmosphericpollution becauseof high
causedby them. The emissioncurves for a thermal plant can be directly rel ted to
o f [ottuteints
i
tor for
the cost curve through,theemissionrate per MJ (1 Btu = 1055.06J), which is a constant
I terms
a given type or gradeof fuel, thus yielding quadraticNO', SOz,and CO2 emissioncurves
of ermal
of active po*", generation.The aim is to optimize the NO', SOz, and CO2 emissions
plant witlh full utilization of water availableduring the optimizationperiod.
NO, emissionobjectivecan be definedas

12- (dtiPi' + ruf

+e,,P; 6 . r1 3 )
I [* "
where dn, eti, and ,fi; are NO" emissioncoefficientsof the ith thermal unit'
SO,2emissionobjectivecan be statedas

rt = (d,,P,, +rzi)]d'
+e,,F, 6.1r4)
I [*
where d.2i,€2;,nd fzi are SO2emissioncoefficientsof the ith thermal unit.
u8 Power System OPtimization

obje-'ive can be describedas

T(N ,,
. )
rq= J > @2,42
+ ezi4+ fti) ( 6 . 1l s )
| la,
o \i=t )

wherc dti, (ti, andft;, are CO2 emissioncoefficientsof the ith thermal unit.
Development of stochasticmodels of functions Jr, ,I3, and J4 are adopted after much
cogitation that thermal generationsand load demandduring each subintervalare variables.
Any possible deviation of NO' SOz, and CO2 emission coefficients and load deman from their
expected values is managed through the random power generation P;. Presuming t random
varierblesare norrnally distributed and statistically independent, the expected value of N r emlsslon
is esrtimatedas

J2 = + aLiFi
@,,F,' + Iti+ d,, var(P,ll]a (6.116a)
I [*
wherc d1;,V1;,and fr; are expectedNO.. emissioncoefficientsof the ith thermalunit
Substituting Eq. (6.111)into Eq. (6.116a),

T(u )
i, = J- o | > [(r* cl) duFtz
+a,,F,
+frilla, (6.116b)
\i=t )
The expectedvalue of SO2emissionbecomes

it = (d,,F,'
+d,,8+Iz,+d, var(piD)" (6.117a)
I [X
where d2;, e2i, and j2,, are expectedSO2 emis5ioncoefficientsof the lth thermal uni
Equation (6.Ll7a) can be rewrittenas

Jz= [(r* c'o)az,Ez

+a,,F,
+ Izir)r, (6.117b)
I [l
The expectedvalueof C02 emissionbecomes
- r(ru
f l *
io= JII @r,1'+e,F,+Iti+7o (6.1l8a)
o \;=t
whene d3i, d3i, and /3,, are expectedCO2 emissioncoefficientsof the ith thermal uni
Eq. (6.118a)can be rewrittenas

i4= +voF,.,,,])
* cA)i,,F|
[rr (6.1r 8b)
I [l
 Stochastic Multiobjective Generation Scheduli

Stochastic hydro modet

In a short-range hydrothermal scheduling problem, there is insignificant fuel cost incu in the
operationof hydro units [Rashid and Nor, 1991]. The input-outputcharacteristicof a hydro
generatroris expressedby the variation of water discharge q(t) as a function of power tput Pj
and net head h. Accordin! to Glimn-Kirchmayer model, the dischargeis

Q j= K Q ( h )r ( P ) ( / = 1 , 2 , - . - ,l A (6.11e)

where
Q, T are functions of head and hydro generations,respectively
K' is a constant
Izf is total number of hydro units.
For a large capacity reservoir,it is practical to assumethat the effective head is constant
over th,eoptimizationinterval.In the caseof constanthead, fih) becomesconstant,Eq. . 1 1 9 i)s
rewritternas

Qj = K' r(P) ( j = I , 2 , . . .M
, ) (6.r20)

where K' becomesthe new constantand is formed by the multiplication of K and ). Each
hydro plant is constrained by the amount of water available for the optimization interv i.e.

J
0
Q 1d t = R i (j = L,2, ...,Itf) (6.r2r)

where R; is the predefinedvolume of water availablefor the 7th hydro plant.

The perforrltanceof hydro Q1rs representedby

Q i = x 1 P , 2 + Y+
f 12 1 ( / = L , 2 , . . . 1, 4 ) (6.r22)

where xi, !i, and zi are the dischargeccefficientsof the 7th hydro plant.
Slincethe thermal generationsand load demand are random, the hydro gene ons also
become random in view of the load demand constraint as given by Eq. (6.124). A tochastic
model of function Qi is developedby deeminghydro generationand dischargecoeffici ts during
each subintervalas iandom variables.Any possibledeviation in dischargeand load de and from
the expected value is manipulatedthrough the randomnessof hydro generator Pj. ince the
randonnvariables are assumednormally distributed and independent,the expected value of
dischar:gebecomes

( / = 1 , 2 ,- . .M
, ) (6.123a)
4 i = T t F l+ y i F i + 2 , + 7 , v w ( P )

where ii, yj, and Zj are the expecteddischargecoefficientsof the 7th hydro plant.
Eq. (6.1'23a)can be rewrittenas

4 i = . Q + C f ; ) i i F+ty' r F , + 2 , ( / = 1 , 2 , . .W
., (6.r23b)
45tD Power System OPtimization

Equalityand inequalityconstraints
(i) The expected load demand equality constraint is
M+N
sr
= (6.r24)
L P' Po+Pt
i=I

"'"
,ois the expected load demand during the interval

PL is the expected transmission loss during the interval

(ii) The expected limits are imposed as

P--min
<Fi <P.max (i = 1, 2, ..., N + IuI) (6.12s)
where
^tn interval.
P, is the expectedlower limit of the. ith generatoroutput during
Prt* is the expectedupper limit of the ith generatoroutput during interval.

Expectedtransmissionlosses
A ,coffirrlonapproach to model transmissionlosses in the system is to use the Kron's pproximated
loss formula through B-coefficients
gg
Pr= L PiBUPj (6.126)
L
i= 1 j=L

The evaluationof B-coefficientsis very sensitiveto operatingconditionsso B-c fficients will

also become random becausegenerationlevels are random. With normally distri uted random
variables, the expectedtransmissionlossesare
M+N M+N M+N M+N M+N
Pr= (6.127)
i=l i=l r=l i=l tr.l

Covarianceof power P; and Pr.is given by

c o v ( P ;P, ) = R4r,CnCr,F,Fi (i-7,..,M +N; i--I,..,M+N; i*i) (6.128)

whrere Rp,p,are the correlationcoefficientsof P; and Pp random variables.

SubstitutingEqs. (6.111)and (6.128)into Eq. (6.127),
M+N N+M N+M
Fr=
j=l i=l ./=l
jii

or
N+M N+M
(6.12e)
Pt=
I \F,ruF,
j=l
i=l
 Stochastic Miltiobjective Generation Schedul 451

Tii = 1t + Cf,) Bii

7,j = (l + R."- CpCpj) Bij (i*j)

Expecteddeviations
Generatoroutputs P; are treatedas random variables,and the stochasticmodel is co erted into
its deterministicequivalentby taking its expectedvalue. So, the solutionwill provi only the
expectedvaluesof power generations. By virtue of the aboveconsideration,
therewill mismatch
in load demand. The variance of a random variable quantifies the degree of uncertaint associated
with the mean value of the random variable.The active power loss, the systemfu I cost and
emissioncurves are quadraticfunctionsof decision variable P;, and their variancesq ntify the
degree of uncertainty associated with their expected values. So, the expected mi h can be
estimiltedthrough minimization of the squarederror of the unsatisfiedpower demand, l e

E- (6.130)

where P; is the actual power generationrequiredto meet the load, which is consi random.
Using Eq. (6.124),Eq. (6.130) can be rewritten as

E_ (6.131)

This rcn simplification reducesto

M+N-I M+N
1-
") var (P,) + (6.r32)
i=| j=i+l

Eqs" (6.1Il) and (6.128) into Eq. (6.132),

Subst:ituting

r (u+w M+N-I M+N

Js=
JI
0 \i=M
c 2 a 4 2 +I
i=l
\zRr,ricr,Cr,
j=i+l

I _ (6.133)
"5

where;
S;i= e,
SU= Rr,r,Cr,Cr, (i #j)
452 Power SYstemOptimization

optimizationformulation
Multiorbjective
(a) expectedoperatingc t, (b) the
Multiolbjective optimtzationproblem is framed considering: the
NO, ernission,(c) the expectedSO2emission,(d) the expectedCO2 emission thermal
"*p""t.ri variables P; from
unitr, and (e) the expectedrisk associatedwith possibledeviationof random
loar demand
their respectiveexpectedvaluesover the optimizationinterval to meet the expected
in eactr interval. Each hydro plant is constrained by expected amount of water available or draw-
down 1n the interval. Mathematically, the multiobjective optimization problem is defin d a s

I4inimize li r, ir, ir, i o,i rl' (6.134a)

M+N
subjectto ZP'=P r + P ,
i=l
(6.134b)

M
t -
I q' J , d t - RJ ; ( i = 1 , 2 , . . . ,l v l ) (6.13ac)
J
0

mrn < P. < p. max (i=1,2,...,M+N) (6.134d)

^P.
r - ' I - ' l

The abrovemultiobjectiveoptimizationproblem can be redefinedin discreteform as

T
I4inimize Jl = ,LJT '- o,Flnu,r,+-,))
(cr* c?,) (6.135a)
k=l [i
T
Itdinimize J2 s.L '- [r,*c?,)d,,F,'*d,,F,.
,]) (6.135b)
k=l [l
\
\rT * v,F,* ]r,]l
dziP,'
[tr* cf,)
lrlinimize J3 (6.135c)
.L
k=l

T
lylinimize J4 T
.L
'- [r,* c?,) * v,,F,*,f
d,,P,' (6.135d)
k=I [*
lvlinimize Js= (6.135e)

M+N
(6.13s0
subject to
Lj = l F,o= Fp1,-F11

I
k=l
t*Tir = Rj (/ = 1,2, ..., It4) (6.1359)

p_?t"sF,r<F,.flu* (6.135h)
 Stochastic Multiobjective Generation Scheduli

6.7.2 SolutionProcedure
To generate
the non-interior optimizationproblem,the weighting
solutionsto the multiobjective
methodis applied,In this method,the problemis converted into a scalaroptimizatio as given
below':

Minimize (6,136a)

subjectto (6.136b)

T
( j = 1 , 2 ,. . . ,I u [ ) (6.136c)
L ' 0 4 , 0 =Rj
k=l

Fftn s P,[* ( i - - 1 , 2 ,. . . ,M + M (6.r 36d)

5

Lk=1,* o =t (w* > 0) (6.136e)

where wp zte levels of the weighting coefficients.

This approachyields meaningfulresult to the decisionmaker when solved man times for
differ,entvaluesof w1r(k= L,2,..., 5). Weightingfactors\r1,arradeterminedbasedon relative
importanceof various objectives,which may vary from place to place and utility to ility.
The simplest way to a#ive at the required equationsis to use the calculus of anatlons.
Each consffaintequationis associatedwith an unknown multiplier func'uionknown as Lagrange
multiplier. The augmentedobjectivefunction is

5 , t M
L- Z *oio*) +F'o
viair+Lr'[u'- I,,4 (6.137)
k=l k=l [,f, Zal ,/=N+l

where:
vr.is water conversionfactor of the 7th plant
/4 is incrementalcost of power deliveredin the system.
The <iptimality conditidns are describedby taking the partial derivatives ugmented
objectivefunction with respectto the decisionvariables.

i , , % * L , f % - t )^ ) =o ( i=r, 2 , . N
. . ), (6.138a)
A,,rdF,o,,"*[a";

, ! g o * w" < J g * L", @ - r )

,'dP^o =o , ; m =j + A D
( j = r , z ,. . .M (6.r38b)
dP^o [DPrr )
T

I
k=L
ttQ i * -R j = Q ( J = 1 , 2 ,. . . M
, (6.138c)
 Optimization

For+r*-ZF,o-o (6.138d)
i=l

Inherently,theseequationsare nonlinear.The Newton-Raphsonmethod has been appli to find

factoris modifi to satisfv
vi waterconversion
4r e =1,2,..., N + lA, )"1,forall theI intervals.
the available water constraint. Suppose the initial values of control variables Pik,,4"1r, vj are
known. The updated values of control variables in the next iteration are
- F,f * Llo
F,.o:"* (i = l, 2, ...,N + M; k - 1,2, -.-,D
A.f* = LP+ LJ4 (k = I ,2, ...,T)
,l'* - ,f + A,v1 U = 1,2, ..-,I[)
Any small changein control values from their previousvaluescan be obtainedas gi below:

'rffi+Looffi')
ouu.[H
(6.139a)

a2F,
aP*ka

. L,i =-["#+,3,rH * L.o(#')]

.(#- r)a,r-
H
; = i + N ; k = 1 , 2 ,. . . ,T )
( j - - 1 , 2 , . . . , Mm (6.l3eb)

y[&.-,) (-
= -l %o+F&
N+M \
(6.139c)
?r 1EP,* )*tr i=l )

LF^,= (j=I,2,...,M; in=j +Af (6.13ed)

I['-#-')
Substitutingthe values of derivativeswhich are evaluatedabout initial values in Eq. 6 . 1 3 9 a )
N+M
sr
(2tra , + 2towsSi; + ZLoo7,,)M,n + L Mrt + (r(,?- 1)A
Qtrws Si + ZLooru)
j=I
j*i

=' (r*,rn+ Fi*l,us . ^*trfl - u]

zstiFfrl (6.139e)
[- I
 Stochastic' Multiobjective Generation ^Sc

zruFf

= wtlt+ czPai+Z *,1t + c'PIi,

li

5
n i - . f
F i = w 1 b , *L * i t i t
j=2

Equation (6.139e) cap be simplified by realrangingthe tenns, i.e.

N+M
f A

(2t4u;+ )atow5S,, 7,,)M* + \

+ ZAP* E,o
{zr6*rSu+ ZLooTi)
't=l
j*i

( i - 1 , 2 ,. ,

To simptify calculat{ons,neglect all the terms with S,i and Tij, i * i. So, neglectig the term,
N+M

\ {Zro*sSu+ 27il Mjo, the above equationbecomes

i=L
j*i

t( M+N
(2toa,+ ztowrf Ti) lrik = -l tolza,rfl+ Fi I ws
i, + 2fuor ( ik

L \. j=l
1,

LPu ( i = L , 2 ,. . . ,M

or

LF,* (6.140)

where
A4= t{Au,Plr+gt+ Li*) (6.1a0a)
rpsSi,+ trff,,1
X4= 2(t11ai+ (6.140b)

J=T
456 Power SYstem OPtirnization

Substitutingthe valuesof derivativesin Eq. (6.139b)'

+ Zvo,t
(zt owrS,n^ M**
oCi,Ti + zLf;T,un)
N+M
+ A4r+ (K,?r- l) A,tr+ toQy,Ff* )Av;
l=l
I*m
M+N I
Flo*)y)+ - 1)
Loo1rr:l^o
I t,,F,r*vltoQy, J
I
l=l

Ti= 1t+Cr;*,
M+N
K|o=| zr,,,rfl
l=l

The above equation can be simplified by rearranging the terms,-i.e.

N+M

(zt rw5sr,,* 2vo,t ) En * + |

j + LitooT^n
oc2r,7 {2,o*ss,,/ + 27,; ap,r
l=l
I*nt
't
+ t)
+ (r? + Av,) t1,Qy1F,9,p + @1, (Kl* - 1)I
+ a^hl,)
I
( / = L , 2 , . . . , Mm
; =/+N; k=L,
N+M

Neglectingthe term simplification,

',=:^

n) LF*r
rr + 2vlt oC'r,ii + zLDoTn
(2t1,w5S

S^tFt*+ v]"ntpQTiFlo * );) + 1i'* (

( / = 1 ,2 , . . . , M i * = j + N ; ft=1,
or
-L^r,
^n 0-KI*; 1;"* -t*(ZTiF,9,o+!)v]"*
Af ^1=

or
e - K?,i h";* - E1,v]"* - L^k (6.141)
M^k
Yio
where
Ej* = tr(2TiFlo+ y,) 6.l{la)

6.141b)
vt
jk - 1rc7,ii + hooT^*)
5s*^ + voit
2(t1rw
 Stochastic Multiobjective Generation Schedul

I-,k- trwsy zs*jFfk (6.14lc)

j=l

Substitutingthe valuesof derivativesin Eq. (6.139c),

N+ M (t' t+ u \ ( N+M \
=-[";**F:o
zriErt-r)M'r Ff ( f t= 1, 2 , . . . ,D
E t; ] )
N+M N+M
- =
.I tt KlolLFp For,* F&-I &-' ( k - 1 , 2 ,. . . ,D
,l=l i=l

Substituting Eqs. (6.140)and (6.141)into the above equatiOtr'

i ,t - K,oo,
r =
tt - I - o-
[rT" ",g
, \ x i k
'1
)
.X,,-"n,I L"fn (l - Kl"i - E jkvlew-
Yio

( N + M \

l F o o + P &I - P ; ' l ( f t =r ,2 r . . . , 7 ' , m = j + /v)

\ i = l )

m=j+N)

co L"f* - I
M
Di1,v]* - Fr (6.r42)
j=l

where

(r - Kl,r)z (6.r42a)
C*=
Yio

Dj* = (6.r42b)

( 1- , K , ? ) A i r
Ft= (6.r42c)
x,*
 System Optinization
( 6 .l 3 9 d ) ,
Substituting the values

(/=

i suur,itutingEq. (6.141) into the aboveequation,

T K,or) A.";* - Ei1,v]"* - L^k

S r
ti* to4lo /+/V)
L Yjr
k=l

or
T
(6.r43)
T Dir,L"fn-Hi']"*
ik=l
=oi (i=

where

T F n
Lr --
,'j S "ik"ik 6.143a)
.L Y,,.
k=l J'r

oj=vj-i,ralo+iry 'ik
6.143b)
k=l k=l

From Eq. (6. I42), we find the value of i.['*, i'e'

K = +.$
(6.144)
" Ly* ?yle*
ck'?rck'l

Substitutingthe value of l.['* into Eq- (6.143)'

( - M \
T

i r r o -r(l * r . i +Ckv rt* l - t r u i " *= o i ( i = r , Z , . . . m

,LJ
, M=;i + M )
k=l t i=L )

D i * Dyi x1 . * l - " r r t =
eo* j - Z + ( i = r , 2 , . .M. ,; * = i + M (6.145)
c k ' )

(6.145)can be written in matrix form, i.e.

IgiMxM lvlu,1' fRi)u*t

where

Qii =
E ( j = I,2, "', Il'l) (6.145a)
 Stochastic Multiobjective Generation Schedulin

e' i tL= ft W
r
( j = t , 2 , . . . , ,IM
= ;1 , 2 ,. . . ,M ; j * l) (6.145b)
tv
k=l k

T
R j =o j - > T ( j = r , 2 , "w' , (6Ja5c)
k=l

Here only matrix of M x M is to be solved to calculatevjnt*, and A,t'* can be com from
Eq. (6.144).When the valuesof url'* and 21"* .areknown then n4olt - 1, 2,..., M) a LF*
(/ = 1,,2, ..., M; m = / + /f) can be computedtiom Eqs. (6.140)and (6.141),respectiely. The
detailed algorithm iiss elaborated below:

Algorithm 6.4: (Generationof Non-Inferior Solution for Multiobjective Hydrothe Sched-

uling Using Approximate Newton-Raphson Method
1. Read the number of thermal units N, the number of hydro units M, the n mber of
sub-intervals T, expected cost coefficients, expected emission coefficients, xpected
B-coefficients, expected discharge coefficients, expected demand and pre- pecified
available water, coefficient of varianceand correlation coefficients,etc.
2. Set iterationfor non-inferior solutions,k = 1.
3. Incrementcount of non-inferirx solutions,k = k + l.
4. If (k > Kl GOTO Step 17.
5. Feedor generatethe weights,w1(i- 1,2, ..., IA.
6. Calculatethe initial guessvaluesof F,.f (i = 1, 2, ..., N + M, l,f and vf f = l, n.
7. Start the iterationcounter,r = 1.
8. Compute the variables,K:k, Ai* from Eq. (6.1a0a), Xik from Eq. (6.140b), :jk from
Eq. (6.L4la), X;r from Eq. (6.141b),Cpfrom Eq. (6.142a),D;pfrom Eq. (6.142b F1 from
Eq. (6.14',2c),H1 from Eq. (6.143a), and Ol from Eq. (6.143b).
9. Compute vrlewby solving the following simultaneous equations using the Gau elimi-
nation method
[Q1ilu,ufv]u,t = fRlu*t
10. Checkthe convergence
if Iv;n'*-,jo | < E r thenGOTOStep16.
11. Compute1.t"*from Eq. (6.144).

=*.i+,;"*
L+"* tvk
j=l ^

12. CalculateLP;1,(i= 1,2,..N + M) usingEqs.(6.140)and (6.141).

13. Calculatethe new valuesof F#t*'

4f'* = F,l* LF,o (i = 1,2,"', N + M; k = I,2, "', D

14. Set limits correspondinglyas

; if Pf"* t P#u*
tDnew
ik _
; if P;f"*< P,.fln
; otherwise
460 Power SYstemOPtimization
whose limits have been set either to lower or
Disallow to participatethe generation'
to zero.
upperlirnitsbysettingtherelatingcoefficients
15. i , > IT, then GOTO SteP5,
e l s er = r * 1 ,
;0
rik r[-new
ik

= Llr"*
Lo* (k = 1, 2, ..., D
vf = Vjn"* (J = 1,2, "', ll)

GOTO SiteP12 and rePeat.

valuesand loss
16. Record:it as non-inferiorsolutionand calculatethe objective
Step 3.
17. StoP.

6.7.3 Decision Making

judgment,it is naturalto assu e that the
Consideringthe imprecisenatureof the decisionmaker's
function. The fu: setsare
decision makegnuy have fuzzy or imprecisegoals for each objective
representt e degree
defined by equations called the membershipfunctions. These functions
1993]'Thr member-
of membershipin some fuzzy setsusing valuesfrom 0 to 1 [Klir and Folger,
y taking
ship value 0, lndicates incompatibilitywith the sets,while t meansfull compatibility'
with rate of
accountof the minimum and maximum valuesof eachobjectivefunction together
increaseof menrrbership satisfaction,the decisionmaker must detectmembershipfuncti n t"t(Ji)
dec sing and
in a subjec,tivemanner.Here it is assumedthat p(/;) is a strictly monotonic
continuousfunction defined as
1 ;i; s 7,.'t"
ji* -j,
; 7,.*t"< J-,< 7,.*u* (6.146)
t r t ( iJ) =
7,'t"* - 7,'ttt
o ;ji2.4*u*
where
F 0) is membershipfunction of objective, Ji
are minimum and maximum valuesof the fth objective, respectively.
],.nttn,.ii.max
n-inferior
1) a
The value of membership function suggestshow far (in the scale from 0 to
value;
(non-dominated)solutionhas satisfiedthe i i objective.The sum of membershipfunc on
computed in order to measurethe acco ishment
F0) (i = 1,2, ...,5) for all the objectivescan be
solution
of each solution in satisfying the objectives. The accomplishment of each non-dominal
solutionsby normalizingtts acco plishment
can be rated with respectto all the K non-dominated
over the sum of the accomplishments of K non-dominatedsolutionsas follows [Tapiaat Murtagh,
1 9 9I l :
)

pB= K i=1 5
T pO!) (6.r47)

I
k=l
\ u<i!'t
i=l

where pj is cardinal priority of the kth non-inferiorsolution.

 Stochastic Multiobjective Generation Schedu

The function 1tp in Eq. (6.147) can be treatedas a membershipfunction for noR inated
solutions in a fuzzy set and representedas fuzzy cardinal priority ranking of the non minated
solutions.The solutionthat attainsthe maximummembershippB, in the fuzzy set so ed can
be chosenas the best solution or the one having the highestcardinal priority ranking
Maxtp* , k = L,2, ...,Kl (6.148)

6.7.4 Test Systems and Results

Testsystem 1
A short-rangehydrothermalload schedulingproblem of 24 hours duration has been dertaken.
The entire optimization period has been divided into 24 intervalsand each interval is o one hour.
The systemunder study consistsof one thermalplant and one hydro plant [Rashidand oa 19911.
The expected croSt,NO, emission, SOz emission, COz emission of thermal plant and discharge
characteristics of hydro plant are given below:

of thermalplant ($ft):
Expected,costcharacteristic

/r r(r) = 0.001991
4' + 9.606Fr@ + 373.7
Expected .NO, emission characteristic of thermal plant (kg/h):
jrr(t) = 0.006483 - 0.7902t
4' ?ft> + 28.82488
of thermalplant(kgA):
ExpectedSO2emissioncharacteristic
jrr(t) = 0.00232Pr'+ 3.84632Pr(t)
+ 182.2605
of thermalplant (kg/h):
ExpectedCOz emissioncharacteristic
jor?) = 0.084025F12
- 2.9445484
Pr(r)+ r37.7043
(Mm3ztr):
Expectedhydroplant characteristics

qz@ = 2.19427x 10-5F]<r> - 2.5709x 10r Pz(t) + 1.742333

R, = 72'4797Mm3

(MW-l):
ExpectedB-coefficients

Brr = 0.00005,Er, = Er, = 0.00001,Bz, = 0.00015

If the coefficient of variation is zero and randomvariablesare uncorrelatedto each othe , then the
problem is considereddeterministic.The random variablesare uncorrelatedonly if the orrelation
coefficient is zero.

Effectof varianceand covariance

Owing to the existerice of the variances and covariance of random variables in the f ulation
of the problem, an analysisof variationsin objectives /r seemsnecessary. Actually, rty
properties are known from past history or can be estimatedvia the Monto Carlo imulation
techniques[Sen Gupta, 1972]. There is a need to use exact values of coefficientso variation
and correlation coefficients as and when re;quired.But, here in the study assumedvalues of
these coefficients are chosen. Since variance is representedby the coefficient of v ation, the
462
wei htalge to
coefficientsof variation are varied from OVoto LjVo in stepsof |Vo. By giving full
all objective one b y
one objective and neglectingothers,the effect of varianceis obtainedon
SO2 emir ;ion and
one. The percentagerelative deviationsin total expectedcost, NO' emission,
ien'ts of
COz emission from their respectivedeterministic values with respect to the coeffi
vanauon a,re shown in Figure 6.13. It is observedtrom thls tlgure mat tne retatlvePc
lragc

deviationtrf all the objectivesincreasesas the varianceincreases.The effect of varianceis

more
on the operating cost as comparedto NO' SO2,and CO2 emissions,respectively.

5
4.5 t
4
c
3.5
/ /
r ./
/ /
3
0.) .J .I-
o
bo
2.5 a /
. . ] t ' '

2 ,' ,.,-t''' - - -- ** ''

o
o
C) 1,5
1
0.5

0 dl 0J2-- O.os o.o+ 0.05 0.06 0.07 0.08 0.0e U

Coefficient variation

-+ - cost -'-A'-' Minimumexpected NO, emiss )n

Minimum exPected
--+-- MinimumexpectedSO, emission ---+--- Minimum expectedCO2emiss rfl

Figure6.13 deviationin expectedminimumcost, NOr emission,SO2eln

Percentage
and CO2 emissionwith respectto coefficientsof variation,respectively

By st:tting all weights equal to 0.20, the effect of varianceon all objectivesis :ved
efVet

simultaneot rsly- The percentagerelative deviationsin total expectedcost, NO" emiss.on, S Soz
O
emissionarrd CO2 emissionfrom their respectivedeterministicvalues, with respectto co :ffi ents
cient
of variation are shown in Figure 6.13. It is observedfrom Figure 6.14 that the relative p )rce;ntag tage
on cost bec )ome
mes
deviationsof all the objectivesincreaseas the varianceincreases.The effect
smallerconrparedto NO' SO2and COz emissionswhen equal importanceis given to all c bje< :tivet
tves.
Further, the: effect of varianceon water conversionfactor is also considerable.
The r:ovarianceof bivariaterandom variablescan be consideredpositive or negalve The
covarianceis representedby correlationcoefficients.The correlation coefficients are va ied fron from
-1 .0 tc l.t) in steps of 0.2. The percentagerelative deviations in total expected c NO
NOr
ernission,SiO2 emissionand COzemissionfrom their respectivedeterministicvalueswil r()spec ;pect
to correlation coefficient(Rr,r, (i f i) are shownin Figure 6.15. The weights wy w2, w2 w 4r Bllt and
w5 are takr)n as 0.25,0.25,'(i.25,0.25 and 0.0 respectively.It is examinedthat (i) th re is rs an
a,
increasein the percentagerelative deviationsin total expectedcost Jr tr the value of R, Pj i i ++j t)
is changedfrom a negativevalug to a positive u+t, (ii) there is a decreasein the p :ntag
rtage
relative de'riations in total expectedNO, emission Jz as the value of Rnp,Q/ il is charse! fror from
a negativevalue to a positive value, and (iii) there is a very small effect on SO2emissio:r . /3r, ZlIl
and
 Stochastic Multiobjective Generation Scheduli

15
t4
t3
t2
11
q
g
l0
Cg
9
o)
!
()
8
@
6J 7
O
(-) 6
(.) 5
4
3
2
1 -:l=-fi:#+-:+-+:1:
0.05 0.07 0.09 0.11 0.13 0.15 0.17
Coefficient of variation

-+- Expectedcost --+-- ExpectedNO, emission

- - +" - ExpectedSO, emission ---+--- ExpectedCO, emission
""x---- Waterconversion
factor

Figure 6.14 Percentagedeviationin expectedminimumcost, NO, emission,SOz sslon,

COz emission,and water conversionfactor with respectto coeffi nts of
variation,respectively
when all weightsare set equal to 0.2.

COz emission J 4 objectives.From this study,it has been observedthat the existence f random
variablesgives a significanteffect to each objective either consideredindividually or in the
multiobjectiveframework.

Minimum and maximum values of objectives

To computethe membershipfunction, tt(J) of Ji objective,thereis a needto find the minimum
and maximum values of that objective.Minimum objective values are obtainedby g ving full
weightageto one objective and neglectingothers.When the assignedweight value i s 1 . 0 ,i t
meansthat full weightageis given to the objectiveanciwhen the assignedweightageis zero,the
objectiveis neglected.Maximum objectivevalues are obtainedby exploiting their c nflicting
nature.If an objective i, is in conflict with anotherobjective i j then the i j objective will have
the maximum value correspondingto the minimum value of Ji objectiveor vice vers
Owing to the conflicting natureof objectives, ir, f a, and is will have maxim m values
when Jr is minimum. The objective i3 will have maximum value when Jz is mini um. The
obtained minimum and maximum objective values are given below for uncorrelat random
variableshaving ll%o variance.

/tt'n = 9645077 $ ./r"* = 97475.09$

 System OPtimization

1.5
tr
'E
o
r.zs
'5
a)

o l
bo
cd
d

$ o.zs
0)
or

-0.4 -0.2 0 0.2

Coefficient of correlation

---t--- Expectedcost --*-- NO, emission

Expected
"-'+"" SO2emission -'-+'-'
Expected ExpectedCO2emission

Figure6.15 Percentagedeviationin expectedcost, NOr emission,SOz emission,and Coz

emissionwith respectto coefficientsof correlation,respectively.

7y" = 14591.82
kg Jfl'* = 15169.05
kg
tj-" = 44337.36
kg /j"o = 44665.53kg
J;'" = 245810.70
kg J fu* - 252472.20
kg
j{ * = 46805.98Mw2
"ry" = 46044.05
lvrw'

of the optimalor 'best' solution

Determination
First,the optimalor bestsolutionis foundfor only two objectivesI *d 7r. ft weigh l'U1and
"
lr2, respectively,are varied in the range0.0 to 1.0 so that their sum is 1.0. The weights,W3,W4,
and.)v5&re taken zero to neglect the other objectives, i3, 7a, and J=5,respectively.The random
variablesare assumedindependentof each other with I}Vo variation. Total expectedcost u/1,and
NO, emission ir, are tabutatedin Table 6.33, for eleven possible weight combinatio which
correspondsto non-inferiorsolutions.The membershipfunctions p(/-1) and lt(J) of Jr and Jz
objectives,respectively,are also presentedin Table 6.33. The expectedcost Jt rises and the
expectedNO, emission Jz declines,when w1 weightageto expectedcost is reducing. Similar
trend has been explored in the membership functions of "I1 and J2 objectives. It can be luded
from Figure 6.L6 that if two objectives are in conflict then their membership functions also in
 Stochastic Multiobjective Generation Scheduli

Table 6.33 Totat expectedcost and NO, emissionscorespond to non-inferior

only two objectivesare considered(wl = 'il4= ws = 0)

Sr w1 Jl J2 t t ( Jr ) tt(Jz)
no. ($) (ke)
1 1.0 0.0 96028.66 14839.02 1.0000 0.0000 0.07057
2 0.9 0.r 96032.16 r4774.06 0.9966 0 . 11 2 5 0.07827
3 0.8 0.2 96044.09 r4707.rr 0.9851 0.2284 0.08564
4 0.7 0.3 96067.08 14638.60 0.9628 0.3471, 0.09244
5 0.6 0.4 96104.& 14569.26 0.9265 0.467r 0.09835
6 0.5 0.5 9616r.45 14500.20 0.8716 0.5867 0.10292
7 0.4 0.6 96243.90 lM33.l3 0.79r9 0.7029 0.10549
8 0.3 0.7 96360.7r 14370,68 0.6789 0.8110 0 . 1 0 5l 5
9 0.2 0.8 96523.80 r4316.77 0.5213 0.9044 0.10061
10 0.1 0.9 96750.05 14277.39 0.3026 o.9726 0.08999
11 0.0 1.0 97063.07 14261.55 0.0000 1.0000 0.07057

3 0.6
E
a
v)
q)
2 0.4
c)
-1

0.4 0.5 0.6

Weight,w,
---l--- Mernbershipfunctionof expectedcost
--*-- Membershipof expectedNO, emission

Fisure6'16 of membershipfunctionsof expectedcost and NO, emissio with

[T;*ll" wetght,w1,whergW2=1.0 - w1and ws- w4- ws= O.

conflict or vice-versa.Figures6.L7 and 6.18 show that the expectedcost is in conflict with the
expectedSO2 emissionand risk level, respectively.
 Power System OPtimization

0 .9 8 5

0.97

(J

0 .9 5 5

.a
() 0 .9 4
L

C)

0.925

0.91

o.8esd 0.1 0.2 0.3 0.4 0.5 0.6 o.1 0.8 0.9
Weight,w,

--l--- Membership function of expected cost

--*-- Membershipfunction of expectedSO, emission

Figure6.17 Variationof membershipfunctionsof expectedcost and SOz n with

respectto weight, w1,where ws= 1 - W 1a n d W 2 =W q =W s =0 .

o
3 0.6

(a

c)
€ 0.4
q)

-l-- Membership function of expected cost

--*-- Membership function of variance

Figure6.18 Variationof membershipfunctionsof expectedcost and variance power

generationwith respectto weight,lyt, where ws= I - LV1and W2= Ws w t = Q .

ii.-. .
 stochastic Multiobjective Generation scheduling 467
The non-inferiorsolution that attainsthe maximum membershippB, is
distinguishecas the
best solution among the non-inferior solutions.The weight combination,presented
a serial
number7 in Table 6.33 gives the maximum valueof i.e. 0. I0504g and thereforeprovi
lt\, the
best or preferredweight combination.
The non-inferior solutions for 126 differentsimulatedweight combinationsare ge
consideringall, the objectivessimultaneously.Non-inferior solution that acquiresthe
ma lmum
membershippj, is chosenas the best solutionand is furnishedin Thble6.34. Ttre best
lutions
are securedfor distinct values of coefficientsof variationand correlationcoefficientsin lverse
situationsand are conferredin Table 6.34.

Table 6.34 Best expectedoptimal schedulesfrom non-inferior set

Cn Rnp, Cost NO, SOz COz Risk

emission emission emission
($) (ke) (ke) (ke) (Mw2)
1 0.01 0.0 96386.43 r4360.t2 44156.03 242124.3 460.4161 0. r656
2 0.05 0.0 96480.82 14442.96 442t4.s8 243249.0 I 1 5 1 6 . 6 6 0. 1787
3 0.10 0.0 96814.30 14686.73 44410.25 246600.r 46125.73 0. 168
4 0.10 1.0 96735.22 14726.10 44386.10 247042.8 88802.83 0. 1561
5 0.10 -1.0 96802.52 14687.49 44405.48 246601.r 3455.506 0 2648

The weight combinations and the water conversion factor v2, cotrssponding to best sch ules
are depicted in Thble 6.36. For case one, the achieved expected generation schedules of 24 hours
have been furnishedin Table 6.35. For each sub-interval,expeotedtransmissionloss FL, incre-
mental cost L, and expecteddischarge qz, are exhibitedin Table 6.35. The attained uality
constraintLPD,during each sub-interval,is also providedin Table6.35 which showsthe
of the obtainedsolutions.
Conventional economic short-term fixed-head hydrothermal power dispatch method al tes
generationscheduleto the individual generatingunits basedupon deterministiccost functi n and
load demand,ignoring inaccuraciesand uncertainties.Such generationschedulesresult n the
lowest expectedtotal cost, but this cost is also associatedwith a relatively large varianceth can
be interpretedas risk measure.Moreover,in power systemoperationplanning, there exist m Itiple
objectives to be attained, which conflict with each other and ard subject to a mutual interf It
means that any one objective can be improved only at the expenseof other objectives. n the
multiobjective framework, the analysisof hydrothermalshort-rangefixed-headis undertakr with
explicit recognition of uncertaintiesin productioncost, NO' SO2 and COz emissionsa load
demand.

6.8 STOCHASTIC
MULTIOBJECTIVE
LONG.TERIJI
HYDROTHERMAL
SCHEDULING
A modern power system may consist of several thermal, conventional hydro power lants
l
connected to various load centres through a lossy transmission network With the insigni cant
incrementalcost involved in hydro generation,the problem of minim izing the operational of
a hydrothermalsystemcan be reducedessentiallyto that of minimizing the iuel cost for al
 Power SystemOptimization

Table 6.35 Expectedscheduleof each sub-intervalfor caseone given in Table 6. and

Thble 6.36

k Pot Pu Ptp Pz* No* Qz*

(Mw) (Mw) (Mw) (Mw) (Mw) m3/h)

1 455.0 9.90335 10.81529 259.7265 206.0888 0.000013 .621406

2 425.0 9.4t2t8 9.57364 237.7585 196.8151 -{.000001 .541795
3 415.0 9.24936 9.17863 230.4520 r93.7266 -0.000021 .516119
4 407.0 9.11942 8.86938 224.6126 19r.2567 0.000061 .495888

5 400.0 9.00595 8.60373 2r9.5074 r89.0963 -o.000033 .4784r1

6 420.0 9.33072 9.37495 234.1042 r95.2707 0.000047 .5289M
7 487.0 r0.43174 12.23392 283.2397 2r5.9942 0.000034 .710609
8 604.0 12.40413 18.26157 369.9315 252.330r 4.000023 .074704
9 665.0 13.45848 2r.93765 4r5.587s 271,.3502 -o.000034 .288394
r0 675.0 13.63308 22.57575 423.1026 274.4732 {.000053 .324999
lt 695.0 13.98378 23.88217 438.1586 280.7235 0.000032 .399545
12 705.0 14.15988 24.55056 445.6997 283.8509 -o.000040 .437487
13 580.0 tr.99425 16.91630 352.0551 244.8612 4.000019 .995131
14 605.0 12.42i26 1 8 . 31 8 8 6 370.6774 252.&r5 0.000011 .078074
15 616.0 12.61010 18.95552 378.8879 256.0677 {.000048 .rr5442
16 653.0 13.24963 2r.18517 406.5808 267.6043 {.000013 .245054
17 72r.0 14.44270 25.64107 457.7835 288.8576 0.000021 .499r24
18 740.0 14.78024 26.96996 472.1622 294.8078 {.000010 .573808
19 700.0 14.07176 24.2r5r0 44r.928r 282.2870 0.000013 .418460
20 678.0 r3.68556 22.769t5 425.3588 275.4104 {.000019 .336067
21 630.0 12.85130 1,9.78314 389.3524 260.4306 0.000120 |
.16377
22 585.0 t2.07942 17.r9r90 355.7754 246.4165 4.000059 .011500
23 540.0 rt.3r724 t4.79958 322.3689 232.4306 0.000024 .868128
24 503.0 10.69769 12.97995 295.4278 220.9521 0.000093 .756874

whereA,Pp= (4 + F ) - Ft-F,

Table 6.36 Weight combination and water conversion factor corresponding to the sc edule
given in Table 6.34

(wy w2, w3, r,14,w5) $/Trrlm3l

1 (0.4,0.
L,0.2,0.2,0.r) I 432
2 (0.5,0.2,0.
1,0.1,0.1) 11 0.2r9
3 (0.4,0. 1,0.I )
1,0.3,0. l 1 .390
4 (0.6,0.
1,0.1,0.1,0.1) 11 . 1 6 8
5 (0.5,0. I)
L,0.2,0.1,0. I 5.220
 Stochastic Multiobjective Generation Scheduling 469

plants under the constraintsof the water availablefor hydro generationin a planned period.
Mostly, hydrothermaloptimal schedulingis achieved,with .the asiumption,that the water nflows
to the reservoirsand the load demandsare known with completecertainty.However,thi is not
true.
The availability of limited amount of hydroelectric energy, as stored water in the system
reservoirs,
makesthe optimaloperationcomplex,because
it createsa link betweenan rating
decision in a given stage and the future consequencesof this decision. Further, it is im ible to
have perfect forecasts of the future inflow sequenceand the load variation during a given riod.
Therefore, for long-term storage regulation, it becomes necessaryto account for the rando nature
of the load and river inflow and so a stochastic representationof these must be used.
Most of the algorithmsincorporateuncertaintiesin the system load demand an water
inflows, but choosea deterministiccost function for thermalgeneratingunits. A major s rce of
uncertaintyin optimal dispatchis that associatedwith cost coefficients[Dhillon er al., 19931.
Howeverwith the increasing,concernrecentlygiven to the environmentalconsiderations illon
et al., 1993; Dhillon et al., 19941,a revised generation scheduling for the hydrothermal power
systemis requiredthat meetsthe constraintsof availablewater at hydro plants and load emand
for power while accountingfor both cost and NO, emission.
Fuzzy sets were first introducedin solving power systemlong-rangedecisionmakin prob-
lems. Fuzzy decisionmaking theoriesattemptto deal with the vaguenessor fuzzinessinh nt in
subjective or imprecise determinations of preferences,constraints, and goals. Thpia and urtagh
[1991] put up a methodologyfor solving a decisionmaking problem involving a multipli ity of
objectivesand selectioncriteria for the best compromisedsolution.
The intent of this section is to provide a technique that allows scheduling of lon -range
hydrotherrnalsystemprobabilisticallyconsideringstochasticcost and NO, emissioncur es for
thermal power generationunits and uncertaintyin load demand and reservoir water i flows.
However,there is a growing trend towardsformulatinga multiobjectiveoptimizationp lem
[El-Hawary and Ravindranath, 1991], so, the approach is developed by formuiating hydrot ermal
scheduling as multiobjective optimization problem. The expected fuel costs and NO, e i s s i o n
over whole of the planning period are consideredas two conflicting objectives.The form lation
also incorporatesany possibledeviationsin generationsover whole of the planning period as the
third objectiveto be minimized.The weightedminimax techniquelKlir and Folger, 1993] i used
to generatethe non-inferiorset by convertingthe problem into a scalaroptimization To
reduce the complexity of the problem, interval-wisedecompositionis carried out. Eac sub-
problem is separatelysolved by using the conjugategradient method to obtain the timal
discharge[Parti, 1987]. In each subinterval, thermal generationsare calculatedby a sim lified
technique,which reducesthe economicdispatchproblem into an equivalentlosslessproble . The
methodis lesstime consuming.A numericalexampleof a power systemconsistingof th hydro
and four-thermal plants is solved and the results are presented.

6.8.1 StochasticMultiobjectiveOptimizationProblemFormulation
In this section,the multiobjectiveswith equalityand inequalityconstraintsconcerningthe power
systemoptimizationproblem are described.The importantobjectivesare consideredhere, e
1. Economicoperations
2. Minimal impacts on environment
3. Expecteddeviationsdue to unsatisfiedloads.
470 Power System Optimization

The stochasticformulation is adoptedby consideringfuel cost coefficients,NO, mission

coefficients,load demand and water inflows into reservoirsas random variables.Water inflows
into reservoirsof various hydro plants are assumedto be statisticallyconelatedduring same
subintervalsbut independentat differentsubintervals.The stochasticmodels are converted to their
deterministicequivalentsby taking their expectedvalues,with the assumptionthat all the random
variablesare normally disributed. A hydrothermalsystemis consideredwith N thermaland hydro
plants.The problem is visualizedas an M stagedecisionprocessby subdividingthe lanning
period into M subintervals.

Expected fuel cost

The aim is to optimizethe runningcost of thermalstationswith full use of water availab during
the optimizationperiod.The objectivefunction,which is fuel cost of the thermalplant, is
to be approximatedby a quadraticfurtctionof generatdrpower output and is given as

Fr = (ai(Pi^)z
+ b,P,^*',,], (6.14e)
E []
where a;, b;, and c; &ra cost coefficients. P;' is the thermal power generationduring he mth
subinterval.
A stochasticmodel of function F1 during the mth subintervalis formulatedby co idering
cost coefficientsand load demand,during the nth subintervalas random variables.The xpected
value of fuel cost function rnay be obtainedthroughexpandingthe function using Tayl s series
about the mean. The obtainedexpectedfuel cost during the nth subintervalis represen by
N
Fr"= I la,1F,\' +6,P,^*V, + a, vw (P,^)+2P,^ cov(ai,Pi^)+ cov(b,,P,^)] Sltr (6.150)
i=l

where
Pi^ is the expectedvalue of thermalgeneratoroutput during the nth subinterval.
d,, 6;, and d, *. the expectedcost coefficients.
Equation(6.n50)can be rewrittenas
1V
Fr^= lei (F,^)' + Bi F,^* q ] slrt (6.151)
j=l

where
At' = [1.0 + (C(Pi\)2 + 2R(ai,Ph C(a) C(Pi\]Ai
Bi" = [ .0 + R(bi, Pi) C(bi) C(Pi) bi
C(Pi^), C(a), and C(b) are the coefficientsof variation of random variables P{, ai and br
respectively.R(a;, Pin')is the correlationcoefficientof randomvariablesai and P;^, and b i , P i )
is the correlationcoefficientof randomvariablesb; and P;''.

ExpectedNOxemission
Only thermalpowerstationsaremajorcausesof atmospheric of high oncent-
pollutionbecause
ration of pollutants caused by them. -{-heNO, emission curve for thermal power plan can be
 Stochastic Multiobjective Generation Scheduling 47r
directly relatedto the cost curve through the emissionrate per MBtu, which is a consta factor
for a given type of fuel. The aim is to optimize the NO, emissionof thermal stations ith full
use of water availableduring the optimizationperiod.The amountof NO, emissionis gi n a s a
function of the generatoroutput P;^, which is quadratic.

Fz- f ti (di(p;*)z
+eip,^*
rll re (6.rs2)
;L,J I
where di, ei, ffid fi are emission coefficients.
A stochastic model is formulated by considering emission coefficients and load and as
random.Using Thylor's seriesand taking expectations,
the expectedNO, emissionfor mth
subintervalis obtainedas
N
- L
Fi - >Wt(1^f+a,F,^
+ fi + cl, var(Pi*)+21^ cov(d,,p,*)+ cov (e;,pi^;] tgltr (6.153)
i=l

where 7,, Zr, *d j, are the expectedemissioncoefficients.

Rewriting the above equationas
N -
r; = I tfr,r(F,^)'+Ef p,^* j,luiln (6.154)
i=l

where

,i = [.0 + (C(Pnz + ZR(d;,Pi\ C(d) C(Pi))di

E{ = [1.0 + R(ei,P;") C(e) C(Pi')]Ai

C(d;) nd C(e) are the coefficientsof variationof random variables,d; and e;, res tively.
R(du P,t") is the correlationcoefficient of random variables d; and Pi.
R(ei, Pi') is the conelation coefficient of randgrnvariables ei and Pi.

Expected deviations
Since generatoroutputs P!' of hydro and thermal plants are treated as random variabos, the
expecteddeviationsare proportionalto the expectationof the squareof unsatisfiedload mand
during the nth subinterval.These expecteddeviationson the whole of the planning pe od are
conSideredas an objectiveto be minimized.The expecteddeviationsduring the rzth su nterval
are representedas

MW2 6.1ss)

where
PI is the expectedload demandduring the ruth subinterval
F; are the expectecitransmissionlossesduring the mth subinterval
Z i s the total number of hydro and thermalplants.
472- Power SYstemOPtimization

This on simPlificationreducesto:
T T T
u* (Pi^) (6.1s6)
F;'= i
i= I i=l j=l
j*i

are independentof each other uring the

It is assumedhere that--thermaland hydro generations
differentsubintervals.

Expected transmission losses

well-known loss formula are giv by
The power tralsmissionlossesexpressedthroughthe
T T
P i " B i i P fM w (6.1s7)
P[= LI
j=l j=I

where B4s arethe B-coefficients'

randomv ables.The
The power generations'Pf' during the mth subintervalare dependent
obtainedusing the Taylor's
expectedtransmissionlossesduiing the mthsubintervalare
independentof each other unng the
the assumptionthat thermal and hydro generationsare
s.
different subinterval
T T T T _ I T
LpL. ^= E , , v u ( P , ^ ) + ) Mw (6.1s8)
II Ij = l F , * E : F ]i"= l* f L z
i=l j=i+l
E r c o ' ( P i ^ , P n
i=l

Expected water storage equation

the storageat the endof the mth sub terval can
If all hydroplantsareon differentwaterstreams,
be obtainedfrom the equation,
(6.1se)
xln*t= xl" + Jj"- o;" - srn' (/ = 1,2,...,L)
where
Xr.' is water storageat the 7th reservoirduring the mth subinterval
Jj" is water inflow into the 7th reservoirduring the rlth subinterval
the jth reservoirduring the nrth subinterval
** i, spillage from
jth turbine during the nth subinterval'
ej" is water dischargethrough the
With the ions that
Spillage occurs only when the reservoirstoragelimit is exceeded.
are uncorrelatedto e other, the
water inflows and water storageduring the mth subinterval
expectedvalue of Eq. (6.159)is given as

Xf*'-- xf *if -0f - sf (i = r,2,..,

L) (6.160)

where
'T"
;::;:ffT:'L
:i:ff#:::i ,ffi T:,'l:il:'
11,'::ffi
The corresPondingvariancesare:
- EKXf*t - x;*')'l
var(xl"*t) U = 1,2,..,
L)
 Stochastic Multiobjective Generation Scheduling 473

On simplification, the previous equation can be rewritten as

var(Xf+l)= var(Xi^)+ var(Jj') ( i = l, 2, ...,L) l6r)

Similarly,covariance
is
x{*t) - EI6f*t - xf*')6f*t - V{*')l
cov(xjn*', ( / = 1 , 2 ,. . . L
, ; k = I,2, ...L
, i j /k)
After simplification,covarianceis
cov(xj'*t,xtf*t) = cov(xf',xr{l+ cov(Ji^,
Jt{) (/ = 1,2,...,L; k= r,2, ..-,Li j * k) ( r62)

Expected hydro generation

The averagehydro generationduring any subintervaldependson the water dischargethrou the
turbine and on the averagehead, which is also a function of the storage.The average ydro
generationduring the rnth subinterval is given by
Pf*N= hiU + 0.59;QXf + 4" - Qi^ - Srr)l (Qi'- 0 (i = 1,2,..., L) ( 163)
where
hi is the basic head of the 7th hydro plant
g; is the water headcorrectionfactor to accountfor variationin head with storageof jrh
hydro plant
l-ti it the non-effectivewater dischargeof the 7th hydro plant.
Since water inflows and water storageare random variables,so hydro generationswill
random. Expectedhydro generationof the rzth subintervalcan be written as

Piir = hj U+ 0.5s,QXf + rf - Of - sUQf - Fi) ( j = 1 , , 2 ,. . - ,L ) r64)

Varianceof hydro power is given as follows:

= EIQfrw-F,T*)'l
var(Pfrls) (i = 1,2,...,L)
On simplification,the aboveequationcan be written as

- nfQi^ - h)' s/ [var(xj")+ 0.25var(Jj\)

var(Pl*N) (J'= r,2,..., L) 16s)
is givenas below
of hydrogeneration
Covariance

cov(Pf*,,u,PT*i= hjh*(0;' - k) Q{ - pt) BiBt [cov(Xrrn,X{) + 0.25 cov(Ji*,Jn]

( j = I , 2 , . . . , L ; k = L , 2 , . . . , Lj ;f k ) ( 6166)

Aggregating the above equations, the hydrothermal multiobjective optimization prob ls

defined below:

M M
Minimize ,I Fy,\ (6. 67a)
m=l m=I
-474 Power System OPti^rzotign

subject to: (a) expectedload

demandconstraintfor the rrrthsubinterval

'IT F'' -F;' - pi' - o (6.167b)

,4'',
(b) expectedstoragecontinuityconstraint

v'i*t= x';'*i"" -Qi' - S'i (i = r,2,.-.,

L) (6.167c)

(c) total expected volume of water available constraint

M M M
ytn+l -x|-Lti' * I e | ' * I t i ' = o (j = 1 , 2 ," ' , L ) (6.l67d)
m=l nt=l m=l

(d) expected hydro generation equatton

rli* - hiu+0.5s; - t'ti)

(zxi"*ii' -Q';'- sjr')l(Q';' ) (6.161e)

(e) expected output of thermal plants

( i = I , 2 , - . . ,M (6.167f)
4 m i n{ F , u ' < 4 t t *

(f) water discharge limit

oi* 3Q'i'<oi ( j = 1 , 2 ,-, . . L

,) (6.167g)

(g) expected forebay limit of reservoir

XrT'ns x j' -< "Xl T A X ( j = 1 , 2 ,. - . L

, ) (6.167h)

6.8.2 Optimal GontrolStrategy

The scalarizedform of the optimizationproblem is obtainedby using the weigh mlnlmax
method and is given below:

Minimize (6.168a)
:[]
3
subjectto (6.l68b)
L*o-t
ft=l

where w1 is the level of the weightingcoefficients.All other equalityand inequalit constralnts

are sameas mentionedin the problemof the systemof Eqs. (6.167).
This approachyields meaningfulresultto the decisionmaker when solvedoften br different
valuesof w{k - l, 2,3). The scalarized optimizationproblemis solvedby forming t Lagrangian
function, which is obtained by augmenting the objective function with various equalit constraints
expressedin terms of expected values, through dual variables as follows:
 Stochastic Muttiobiective Generation Scheduling 475

L-ili,*ru['+Ll'
.L,l I Lt
nr=l L t=l

L
- I tii [Fji* -t,j (r + o.ssi eV|" *ij" - Q';'- s;r'>)<Q'j'- ! 1))
j=l

* t'li Gf" - x!' - ii'*ai +si)

r ( M M M \ l
+ L Lo,luf" - x', .r69)
j:l \ m=l m=l nt=l ) )

During each subinterval,the control variablesare the water dischargesthrough the turbi es of
hydro plants.The reservoirstorageand the hydro power generationat the end of the subi terval
are obtainedfrom Eqs.(6.160)and (6.163),respectively. Irrespective of the hydro generatr s the
thermal generationssatisfythe power transferconstraintto achievethe minimal fuel c t . T h e
dual variablesLf , ryj, and ).fi are obtainedby equatingthe partial derivativesof the I-a gian
function with respectto the dependentvariablesto zero.

* - # + r ; ( # - t ) = o ( i =r , 2 , 1 r ) r70)
#=l
= o ( i = t ' 2 '' L ) 6 . t 7t )
#--tii+fi[+-')
t\;' - h\i + Mi hi siQi' - t-r) = o 6.n2)

X,llj = 0, since the equationsassociatedwith this set of dual variablesare redundant.

AL = - -2Q';' + p)]
aQi, iQxl' + Jf s;'
tii + ).aj T)i hi[t + o-ss

aor\ \'lPfirn,pillr,)l
[*(u*\- ei]il)*I
*wt'LaQi' z*(cou . 4'(*\
',!.rji,)r,...,,,
k=t J
6.n3)
where

Qi' - F)lvarof )+0.25var(Ji')) (i = 1,2, ---,

Qj|il] = zn?ej L) 6 . I t4 )
#[var
-p1)[cov(Xi',Xf)+ 0.25-cou(J'j',J'l')]
= h,ttogig1,Qf
1Pj']x,PUilJ
#[cou
(j=l'2'"''L;jtk) 6.r7s)
L
aPI = Bi+N.i*ru;fu
a r ,,n.-r
+
fva,r(Pfr")] ' +[cov (Pi]*,
2 ru,.Mk+N 11 PI]ilJ
aor r r
aQi ft=l aQi'
(j = l'2' "'' L) 6 . t 76 )
476 Power System Optimizatton

The dual variablesLai are to be adjustedto maximize the Lagrangianfunction under constraints
of other optimality conditions.The correspondinggradientvector is

. M M M
AL = x f * t- x j - I
6L ai
i f * Z O f* I s . f ( i = r , 2 , . .L. , (6.t77)
m=l m=l m=l

The upper and lower limits on the control variables are taken care of by making se variables
equal to the respective bounded values whenever such limits are violated. For he dependent
variables, these limits can be considered by augmenting the cost function throug the Powell's
Penalty function.

Algorithm 6.5 StochasticMultiobjective Long-Term Hydrothermal Scheduling

To generatethe non-inferiorsolutions,the stepsof the algorithm are given below:
1. Set weightsw1 (k = 1,2, ..., n at desiredvalues.
2. Assumea setof ),a1for7 = I,2, ...,L.
-
3. Set ru I, kf= 0, var(Xi\= 0, cov(Xj*,Xt\= 0 ( j = 1,2, ..., L; k = I, Li j *k)
Oi" <i= 1,2,...,L).
4. Assume
5. Calculatehydro generationsP1*, var(Pt*),cov(P1',Pn ( j = I,2, ..., L; k = l r 2 , , . , ' ,L i
l = j + N ; n = k + N ; j * k ) f r o m E q s . ( 6 . 1 6 4 )( ,6 . 1 6 5 a
) n d ( 6 . 1 6 6 )r,e s p e ively.
c
6. Calculatethermalgenerations Pi* Q = 1,2,..., /f) andLi by solvingthe fo lowing set of
equations.
J
rl

L** ( i = 1 , 2 ,. . . ,N
k=l

The above set of equations is solved by a simplified method [Osyczka and avies,1984]
which is a less time-consuming algorithm. The method reduces the econ mic dispatch
problem into an equivalent losslessproblem.
7. M *d Mi t* j = 7,2, ...,L aresolvedusingEqs.(6.111)and(6.172), tively.
8. Calculatethe gradientvectorfrom Eq. (6.173).
If the optimality conditionsare achievedwithin the prescribedaccuracy then GOTO
Step 9, else adjust the expected water discharges using the conjugate g ent method,
and GOTO Step 5.
9 . Calculatethe expectedvalues,variancesand covariancesof storagefrom Eqs. (6.160),
(6.161)and (6.162),respectivelyfor all the hydro generarors.
1 0 . If (m > IvDthen GOTO Srrep11,
else set m = m + I and GOTO Step 4.
11. Check the convergence using the gradientvector given by Eq. (6.177).It nvergence ls
not achieved,then adjust 2a; using the steepestascentmethod.

L+j= )'+j+ (j = 1,2, "', L) andGoro Step3.

"h
 Stochastic Multiobjective Generation Scheduling 477

t 2 . Calculate the overall cost, emission and risk objectives.Calculate the me bership
function lr(Fr)with k = l, 2, ..., K, from Eq. (6.149).
1 3 .Check if the schedulewith new weight combinationsis required.
If 'Yes' then modify the weightsand GOTO Step 2 and repeat
else stop.

Efficient algorithm to solve the thermal problem during the sub-interval

A simple and efficient economic dispatch algorithm is used. The algorithm reduces the onomic
dispatchinto an equivalentlosslessproblemfrom which solutionscan be easily obtai ed. The
The first step of the algorithmis to calculatean initial guess.It is obtainedby solving th follow-
ing losslessproblem for the rnth subinterval.Time dependenceis suppressedhere for si plicity.

2ai4 + Fi - )"1 (6.178)

N
\r.r
= PD (6.r7e)
LP,
j=l

where
A t = ( A ; + D i + (c(Pi))2

P i - ( B i + Ei)
From Eq. (6.178)
Lr-F,
Pi (6.180)
2a,
After summing Eq. (6.180) over i, and on simplification,

Lt= N
(6.18r)
s.(-/ 2d,,
r
'
i=l

Equations(6.180) and (6.181)determinethe analyticalsolutionfor the losslesscase. e avail-

ability of an analyticalsolutionto Eq. (6.181)significantlyincreasesthe efficiencyof the al orithm.
Considernow the lossy case.The equationto be solved has the form:
3

I
k=L
* oFt"* - Ll ( i = 1, 2 , . . . ,M (6.r82)

N
new= F, + F:"* (6.183)
4
i=1

These are nonlinear equationsin P,s and /,1 and can only be solved iteratively,.Let 21old P-loto
=
(f 1,2, ..., N) be approximated solutionsto Eqs.(6.182)and (6.183).Herethe aim is to nd new
approximations
1,1"*- llta + 6)4

f,new= P--old
+6F,
 Power Syste

The expressions ,l,1n"f and flnew (i = 1,2, ..., M) depend on the transmission loss a used.By
tp first order
Thylor'sexpression

pT* = Flrd.* -p-ord)

#(4n"*

Frl"* - o raPTto*.'S$
62rpotd
o L 16
L
'''f . n e w-*
1
40to)
dP, a4
rtt
fr aPiaPi
.L .lFtE

Similarly, the obj ve furdctionscan be approximatedto first order.

a 4 a 4 ' 3 ry(4n"*-4o"
ry=ugo.$
T=ff.hffi(4"'*-P,"'")
aE',
Putting thesevaluesin Eqs. (6.182) and (6.183),the losslepsproblem can be

?o;1! + Fi ,=Ll"* (d= I ,2, ...,lD (6.184)

N
t.{-/ 'rF,*:=F; (6.18s)
d=l

where

"/V
Pi=F;'f+F,
i=l

Pi= 1"'* Q

a' =dL - 44
To retain the classicf fora and at the same time to improw convergence,it is possi e to include
only the ith term of sumnaption,i.e.

.. [;2P-old
^^r^l d'F,o'u .. II
1'l'o
^L AP:
| +(4n"* -4"t0)l
J
 Stochastic Multiobjective Generation Scheduling 479

6.8.3 SampleSystemStudy
A long range hydrothermal problem with four thermal and three hydro electric p ants is
consideredhere. Ttre expectedincremerrtalfuel cost and NO, emissionequationof plants
consideredhere are given below:
Expectedincrementalcost equations($/TvtW-h)
are:

A)l/aPr^ - 0.1Pt* + 2.5

d)'rrlAP{ = o.LzP{ + 2.6

la4* - 0.16p3,"+ 2.9

Dloig

dl'rqlaP{ = 0.16P{ + 2.8

Expected NOr emission equations (kglh) are:

Fu(P() = 0.00767
(,.P1")2
+ 0.80507
Fr^ + 363.7048

Fzz(P{) = 0.00167
(:,Pz^)z
+ 0.80507Fr^+ 363.7048
- r.24885Fr^ + L37.370r
F23(\*) = 0.01378(,P3'")2
- r.24885
Frq(P{) = 0.01378(:,P;')2 P{ + 137.370t

The averagetransmission areas follows(MW-l):

losscoefficients

Er,= Er, = 0.0005,Er=

. , 8 4 4= 0 . 0 0 0 8 ,E r , = B u u= 0 . 0 0 0 7

E r , = 0 . 0 0 0 9\ i, i = 0 r . 0 ( i = L , 2 , . . . , 7j;= 1 , 2 , . . . , 7 ; i * j )

The following valuesof coefficientsof variationsand of correlationcoefficientsare ass

the study.
C ( a i ) = C ( b ) =C ( P i ) = C ( d i , i )C=( e )= 0 . 1 ( i = 1 , 2 , . . .N
, ; m=L,2,. ta.
-
R(ai, Pi^) R(bi, Pi*) = R(ei,P;*.)= R(di, Pi^) = R(Pf , Pj') = I.0
( i = I , 2 , . . . ,N ; j = L , 2 , . . . N
, i i *jt m= l, ..., IIO
Actually, probability propertiesare kno'wn from past history or can be estimatedvia the Monto
Carlo simulation techniques.Data for the three hydro plants are presentedin Table 6.37. ex-
pectedwater inflows and the standarddeviationsfor 12 subintervalsare given in Table6. 8. The
correlation coefficientsbetweenthe water inflows for the samesubintervalare assumedto be 0.5.
The overall expectedincrementalcost, emissionand risk associated are shownin T e 6.39
for different weight combinations.The normalizedmembershipfunction valuesare also p
in Table 6.39 that correspondto non-inlleriorsolutions.Each non-inferiorsolutionis opti . A s
the partial derivativesof Lagrangianwith respectto control variablesare achievedequiv ent to
zero within the prescribedaccuracyfor each subinterval,the availablewater is used in full br the
whole of planningperiod by adjusting lfa; so that Eq. (6.177) becomesequivalentto zero within
the prescribedaccuracy.
 Power SystemOptimiz.ation

Thblle6.37 Hydro plants data

Hydro plants

ft-basic head (metres) 0.98 0.50 0.75

t-water head correction factor 0.004 0.002 0.004
l-non-effective w ater di scharge 0.0 0.0 0.0
Xr-initial storage (cu.m/sec-month) 100.0 r50.0 100.0
X'* | -,ftnal storage(cu.m/sec-month) 100.0 150.0 150.0
Maximum allowable discharge(cu.m/sec) 73.0 75.0 80.0
Minimum allowable dischuge (cu.m1sec) 3.334 4.353 3.543
Maximum hydro generation (MW) 89.03 57.61, 75.0
Minirnum hydro generation (MW) 0.0 0.0 0.0

Thble 6.38 Expecteddemand,water inflows, standarddeviationsand assumeddi harge

Sub Demand WaterinJl'ows Standard deviation Assumed ischarge

int. (Mw) (cu.m/serc) (cu.m/sec) (cu. sec)
rI l rI2 Jf.s o(J) o(Jz) o(J) Qt Qt
1 200.0 0.8 23.0 0.8 0.08 2.30
0.08 10.8 0 10.8
2 2r0.0 10.0 17.0 10.0 1.00 r.70
1.00 20.0 .0 20.0
3 205.0 11 . 0 t2.0 I 1.0 1,10 r.20
1 .1 0 2r.0 0 2r.0
4 180.0 I 1.9 10.0 11.9 1.19 1.00
1.19 2r.9 0 2r.9
5 195.0 r7.0 20.0 r7.0 t.70 2.00
r.70 27.0 3 0 27.0
6 200.0 18.5 40.0 i8.5 1.85 4.00
1.85 28.5 5 0 28.5
1 220.0 27.6 53.0 27.6 2.76 5.30 2.76 37.6 630 37.6
8 204.0 43.8 63.0 43.8 4.38 6.30 4 . 3 8 3 3 . 8 530 33.8
9 189.0 56.4 62.0 56"4 5.64 6.20 s.64 46.4 5 0 46.4
10 199.0 40.3 48.0 40.0 4.03 4.80 4.03 30.3 38 0 30.3
11 207.0 30.1 55.0 30.1 3.01 5.50 3.01 20.r 45 0 20.r
L2 198.0 46.3 48.0 46.3 4.63 4.80 4.63 36.0 3 80 36.3

The deterministicincrementalcost is given in Tabie 6.39. The optimal expected ules

for two weight combinations(0.8, 0.0,,0.2) and (0.6, 0.2, 0.2) areshownin Tables6.40 d 6.4r,
respectively.
The expected discharges for the corresponding weight combinations are shown in Ie 6:42.
It can be observedfrom Figure 6.19 that the relative percentagedeviation in total- ted
incremental cost with respect to the varlueof the total incremental cost obtained from de rministic
schedule is considerable, about ZVo. The deviation of total expectecl incremental t from
deterministictotal incrementalcost varies as the weight combinationsare changed.The variation
in weight combination also gives tlhe change in the expectedgenerationscheduleof each
subintervalwhich can be observedby comparingthe schedulesdepictedin Tables6.40 nd 6.42,
 S'tochastic Multiobjective Generation Scheduling 481
Thble 6.39 Variation in the overall expectedincrementalcost, NO,
emission,risk, and
transmission losses
Sr. no. wl Wt W3 Fl F2 F3
Deterministic 6800.129
I 1.0 0.0 0.0 6954.13r r25rr.94 1877.853 834
2 0.8 0.0 0.2 6954.910 12500.69 r875.r23 104
3 0.6 0.0 0.4 6964.954 12484.r3 1872.207 308
4 o.4 0.0 0.6 6974.38r 12457.82 r870.487 466
5 0.2 0.0 0.8 7001.399 12406.4s 1867.762 621
6 0.7 0.1 0.2 6967.032 12456.69 r874.578 2r8
7 0.6 0.2 0.2 6974.552 12402.58 r872.526 5s8
8 0.5 0"3 0,2 7026.460 12336.67 1877.206 87

2.8

o
2.6

ttt-ttttI
$?
q)
oo 2.4
tr
c)
(J
L.
o

2.2

2
0.0 0.4
Wbightvu3, whenwz=0.0

Figure 6.19 Percentagerelativedeviationof expectedcost from deterministic

value of
with respectto weights.

respectively:The expecteddischargeof each subintervalis also changedas the weight combiration

is changed as shown in Table 6.43. The overall expectedincrementalcost and expected
. NOr
emission are of conflicting nature. It can be observedfrom Figure 6.20 that the membe
hip
function of cost decreaseswhereasthe rnembershipfunction oi wO, emission increases.
The
weight w3 is fixed to 0.2 and the weights w1 and w2 ?ra varied in the iung" of 0.0 to 0.1
in such
a way that the sum of weightsis 1.0. T'he overall expectedcost is also in conflict
wit the
variance of generationmistnatchfor whol,eof the planning period and is showh
in Figure .21.
482 Power System OPtimization

Thble 6.40 Expected generation schedule when weights 1ttr w1= 0.8, w2 = 0.0, and w

Sub int. Expectedthermalgenerations Expectedhydro generat

D D
Pl P2 P3 P4 15 t6 P7

I 55.1656 46.2359 33.4042 33.4042 8.9242 20.1,417 .53448

2 53.4951, 44.7988 32.3410 32.3410 19.7942 16.0753 I .9887
3 5r.9912 43.5057 31 . 3 8 3 8 31 . 3 8 3 8 2r.4292 t2.8346 1 .0834
4 47.0929 39.2989 28.2657 28.2657 16.4695 r0.0243 I .2158
5 45.7565 38.1525 27.4149 27.4149 24.2367 1 6 . 1 513 I .8258
6 42.8291 35.643r 25.54r1 25.54rL 25.6652 27.8963 .8856
7 40.2646 33.4469 23.9r8r 23.9181 37.8828 35.7316 .8465
8 36.sA8 30.2881 2r.5664 21.5664 38.2481 30.7843 .327r
9 27.2141 22.3022 15.6054 15.6054 46.6932 28.3645 .5962
10 29.4934 24.2449 t7.0576 t7.0576 54.1909 24.2649 .&r4
11 25.3906 20.7490 14.4436 14.4436 63.674r 33.4735 8310
t2 20.4548 16.5504 rr.2983 r1.2983 69.9625 25.6950 .3836

Thble 6.41 Expected generation schedule when weights ire ld1 = 0.6, w2= 0.2, and

Sub int. Expected thermal generations Expected hydro generat

Pr P2 P\ P4 Ps P6 P7
1 52.2826 44.2199 35.1298 35.1298 9.6734 20.3345 .0045
2 50.9489 43.0630 34.2878 34.2878 20.0005 16.1455 1 .r094
3 49.4997 41.8055 33.3729 33.3729 21.5401 12.8879 I .1457
4 44.4189 37.4065 30.1650 30.1650 16.8690 10.1635 I .M46
5 43.1654 36.3222 29.3'735 29.3735 24.4937 16.2737 1 .9634
6 40.2784 33.8265 27.5.503 27,5503 25.8547 27.9998 .9763
7 37.8151 31.6991 25.9,946 25.9946 37.9125 35.7750 .8340
8 34.3518 28.7rr2 23.81J70 23.8070 37.8655 30.7354 .07&
9 24.8012 20.4902 17.7'73r 17.773r 46.6259 28.4318 .4912
10 27.956r 23.2028 19.71665 19.7665 52.8726 23.7285 .5602
11 21.5810 t7.7245 15.7"382 15.7382 66.5194 33.6934 .3091
L2 20.1380 r6.4862 14.8'263 14.8263 65.5869 25.4539 .8206

The membershipfunction of cost decreasesbut the membershipfunction of risk i when

weights w1 and w3 vte varied and weigtrt w2 is set to. zero.
The minimurn and maximurn values of the functions cost, emission and risk are lected
from the non-inferiorsolutions.It also dependson the choice of the .operatoror decisionmaker,
i.e. the range in which he is seekingthe solution.
The operatorcan selectthe best operatingpoint correspondingto the weight com nations
among the non- nferior
 Sto'chasticMultiobjective Generation Scheduling 483

Thble 6"42 Expecteddischargefor different combinationof weights

Sub int. wt = 0.8, ,Nz=0.0, w3= 0.2 wt = 0.6, W2= 0.2, w3= 0.2

Qt Qz Qt Qr Qz
I 6.5585 31 . 1 8 3 6 7.2424 7.1148 3r.4895 7.
2 14.77r0 25.2078 15.6412 14.9532 25.3325 15.739
3 16.2284 20.3914 17.0027 16.3512 20.4918 17. I
4 12.5828 16.rrzt 14.329r 12.9281 16.3529 t4. I
5 1,8.5799 26.2228 :20.1538 18.8530 26.4567 20. 98
'.2t.4299 20.0018 45.985s 2t.
6 19.7593 45.7307
'30.9220 29.4813 59.3567 31.
7 29.296r 59.1508
8 29.0027 50.7031 28.8706 50.7470 29.7 2
45.6405 "29.8922
:i6.4336 33.6552 45.8687 36.46 I
9 33.5294
10 37.5595 38.24M 35.1948 36.7977 37.4751 34.223
11 44.865r 52.2372 38.3460 47.1684 52.6787 39.87 l
t2 50.6312 39.7570 ,+6.9679 47.7347 39.4613 44. 37

0.8
ct)

-Y
t.
-t-
H 0.6 / z - \
\
,.i \\

(A

3tr 0.4
a)
.l'
G4

0.2

0
0.8 0.7 0.6 0.5
W:ight rll1, when wl=0.2

--l-- Cost --*-- Emission

natureof objective,
Figure6.20 Conflicting cost,and expectedNO' emission
expected

solutions. The weight combination presentedat serial number 5 in the Table 6.39 giv the
maximumvalue of p#.The numberof iterationsfor the discussedalgorithmdependson the i tial
guessof dischargeand dual variable)"ai.t\n initial''guessof dischargecan be made by usin the
local variation method.
 Power System Optimization

--l--

---lrf
at, , \
tr
o
- a - '

H 0.6 --
,7 x
Ea
Lr

-8 0.4
E
o
r<<
a

0.2

0.4
w3,whenwr-- 0.0
Weight,

--f -- Cost --+-- Risk

Figure 6.21 Conflictingnatureof objectives,expectedcost, and risk.

6.9 MULTIOBJECTIVE USING

THERMALPOWERDISPATCH
NEURALNETWORK
ARTIFICIIAL (ANN)
Many complex, real-world problemsare characteizedas decisionmaking problemswi multiple,
conflicting and non-commensurable objectives.The objectivesare said to be in con ct, if the
level of one objective can be improved only at the expenseof others.Thermal pow dispatch
problem is also one of them, becausea large-scaleelectric power systempossesse multiple
objectivesto be achieved,namely economicoperations,reliability, security,and minim I impacts
on environment.The main purposeof the optimal power dispatchproblem has so tar mainly
confined to minimi ze the total generationcost of a power system.However,in order t meet the
environmentalregulationsenforced in recent years, emission control has become of the
importantoperationalobjectives.Here, the amountof NO, emission,which is in prop ion to the
active power output of a generator,is selectedas an evaluationcriteria and the minim emission
is ,soughtwithin a small region around an economically feasible operatingpoint.
System security is anotheressentialfactor in power system operationand also in system
planning.To be specific,it is importantto limit line flows within the prescribeduppt bounds.
Then a security inOexas a function of overloadswill be minimrzedby some preventi e control
actions.The use of a sensitivity method to calculateflows in system security and nttngency
analysisstill remains popular,even though fast load flow algorithmshave been deve ped. The
reasonsare its linearity, the speedycomputation,its simplicity and its accuracy. inherent
propertiesmake it widely acceptablefor power system appli,cations.
Decision rmakingis the processof choosingsome possiblecourse of action ong the
variousalternatives.In analyzingsuch problems,becauseof the multiplicity of objectiv and the
variety of alternatives,it is often desirableto obtain complete knowledge of the decisi maker's
 Stochastic Multiobjective Generation Scheduli 485

global preferencestructure explicitly representedby a prescriptive decision model, uch as a

multi-attributeutility function basedon the decisionmaker'sprior articulatedpreference. ut some
studiesindicate that the existing multi-attributeutility theory and methodshave som inherent
disadvantages. Fuuther,in a complex phenomenon,the determinationof a functional s ture of
formal representation for generalpreferencesis usually difficult. So the assessment
of decision
maker underlyingmulti-attributevalue function using the existingmethodsis not an y task.
The recent advancesin ANN's make it possibleto developand apply the new meth logy and
technology to extract decision rules from available information in the setting of decisio making
with multiple ob.jectives.
fut ANN consistsof many highly interconnected, simple and similar processing lements,
called neuronsoperatingin parallel to perform useful computationtasks such as gnizing
pre-programmed or learnedpatterns.A neuralnetwork can be usedto classifypatternsby selecting
the output which representsan unknown input pattern,.in the event where an exact i ut-output
relationshipis not easily defined. T'he approachcan also be used to recognize rns of
behavioursin decisionmaking [Wang and Malakooti, L992], where exact functional rel ionships
are not easily defined. A multilayer AIVN with an error backward propagation learniqg le is the
most popular adaptiveneuralnetwork model basedon supervisedlearningstrategy[Dhil et al.,
1 9 91 l.
Actually, thereare severalinaccuracies and uncertaintiesin the input informationw ich lead
to deviationsfrom the optimal value.For a more realisticapproach,the electricenergys stem has
been representedas a network characterizedby random variablesand investigatedby umerous
researchers at variouslevels [Parti et a1.,1983;El-Hawary and Mbamalu, 1989].In this tion, a
classicalstochasticthermalpower dispatchas a multiobjectiveproblemis formulatedwi explicit
recognition of inaccuracies and uncertainties in the system data. The stochastic beh viour of
random variablesis known in advancefrorn past history.Variancedue to mismatch in g eration
is incorporatedas anotherobjectiveto be minimrzed.Some authors[Tapia and Murta r991,;
David and Rongda, I99Il suggestformulating a multiobjectivedecision problem as a fuzztl
programmingproblem whereinthe aspirationlevels for the objectivesare expressedby of
fuzzy set membership functions. So, multiobjectives are reformulated as fuzzy sets, k ping in
mind that higher the membership the greater the satisfaction with a solution. The fuzzy ts more
accuratelyrepresentthe operationalconstraintsof the power system.Therefore,transmision line
flow constraintsare included in the formulation by using the fuzzy set theory. A th -layered
neural network ir; trained to achievethe desired level of objectivesusing the backp pagatlon
algorithm, when jrnputto the network is known in advance.The network is trained for sample
system consistingof three generators.The range of objective level is provided for the decision
maker by performingminimum cost and emissionalgorithms.The emissionwill be maxim m at the
schedule,when cost is minimum and vice versa,becausethe objectivesare in conflict. rnrmum
cost generationschedulesare consideredas inputs for variouspatternsfor different load ds.

ProblemFormulation
6.9.1 StochasticEconomic-Emission
In this section,ttre multiobjectiveswith the equality and inequality constraintspertain g to the
power systemoptimizationproblem are described.The importantnon-commensurable o jectives,
taken into accountare economicoperation,minimal impactson environment,and expect devia-
tions due to unsattisfied
load. The stcchasticmultiobjectiveformulationis adoptedby c sidering
fuel cost coefficients,NO, emission coefficients, and load demand as random varia les. The
stochasticmodels are converte$to their deterministicequivalentsby taking their expec values,
486 Power SYstem OPtimization
;t l-

tic
on
ler

reliable solution strategY.

Transmissionline security
bus I to bus 7 must satisfythe foll wing
The active power flow in transmissionline ,n connecting
constraint:
.186)
F#" 3 Pr^sP# ( m = I , 2 , . . . ,N L )

NG
Fr^ .187)
i=l

and
Dt-j,r= AI-j,i * Dt-j,n
.187a)

where
D pi,n= GGDF for line /-7, due to the slack generatorR

NG
Pr,,-\ e,-1,,1
i=l
i*R .187b)
NG

I1
i=l

due to shift of genera

Ar-j,,=golef&tion shift distribution factors for line l-7,
generator i

Xt_t-Xj_, ( .187c)
Xri
and &; is the reactanceof line r-j.
where xp;, xi_iare erementsof trre bus reactancematrix,
to consider three obj ctives:
The stochastic economic emission problem is extended
generationmi hto
(i) expectedfuel cost,(ii) expectedNo, emission,and (iii) varianceof
meerrhe expectedloaddemandwithin (a) ttreexpectedgeneration limits and (b) the tra slon
as
line flow bounds. The multiobjective optimization problem is defined
 Stochastic Multiobiective Generation Schedulin

(6.188a)
Minimize

(6.188b)
subject to 1
( t = 1 , 2 ," ' , N G ) (6.188c)
4 m i n< 1 < f , n a x

F;" 3Fr^SFff (m= 1,2,...,

NL) (6.188d)

functions, n mely the

where Fr, Fz, and n are the expected values of the three objective
over the set of missible
operatingcost, emissionand deviations,respectively,to be minimized
(6.68) respectively'
decisionvariable Fi and are defined by Eqs. (6.45), (6.63), and
NG -1+e,l
+
C,)A,E2 (l
( 1+ R u , , C r , c flbi (6.188e)
4 = E [ ( r +cC2^
' n+ZR^,C-
*ZR*4Co,Cr,)o, '
i=l

NG
Fz= I + I,J
(1+ R,,nc,,cflai\
ttt + C, +ZRo,oco,c4)V,1'+
(6.188f)
i=l

NG NG NG

n=>,e#*Lii=l i=l j=l

Rr,r,C4CPjPiPi (6.188g)

j*i

NG NG NG
(6.188h)
FL= 0+c?)8,,1'+) I tt +Rp,Pic4cr)FiEtiFi
i=1 j=l j=l
j*i

where
d,, b;, and ci are the expected cost coefficients

Ii are the expectedNOr emissioncoefficients

- r - n
di , €;, 31116-
8,, are the exPectedB-coefficients
(where -r may be ai, br, d, ei, Pi)
C, is the coefficient of variation of random variable x
y (where x and y may ai, bi, d;,
R ,, is the aorrelation coefficient of random variables x and
ei, P)'

6.9.2 MembershiPFunctions
by fuzzy sel using the
In an approximatereasoning,logical decision making can be defined
the membershi' functions.
operating condirtions.The fuzzy iets are defined by equationscalled
sets using the ues from
These functions representthe degree of membership in some fazzy
while I mea full com-
0 to l. The mernbershipvalue 0 indicatesincompatibility with the sets,
patibility. when neither is true, a value between 0 and I is taken.
488 Power SystemOptimization

Objectivefunctions tt(Fx)
By takingaccount,of the calculatedindividual minimum and maximum of each objectiv turr-
tion togetherwith the rate of increaseof membershipof satisfaction(Figure 6-22), the d slon
maker must determine his membership function lt(F), in a subjective manner. Here, it is a umed
ttratLt(F) is a strictlymonotonic decreasing and conttnuous function with respect to 4 and is
givenbelow:
I ; Fr 3 4min

p(F) = ;ry'n< Fk .Fo** 6 . 18 9 )

0 : F r 2 4max

Fk
Fmrn
rk F,lu*

functionof the objectives.

Figure6.22 Membership

Transmission line tlow tt(Pr^)

constraintsof the power systemFuzzy
the operational
The fuzzy setsmore accuratelyrepresent
membershipmay have a variety of shapesbut for simplicity, here the line flow cons ts are
representedby the triangular membershipfunction (Figure 6.23). Mathematically,the rp
function is defined as

P,^ m ( Pt^n
'm

p-min> P- >F:
rm rm rm

[t(Pr^) = 6.1e0)
F;^rFr^rFt*
-P-to
P^
^ >
- ' l ^
I M

where Pf^ is the mean of the lower and upper limits of line flows.
 Stochastic Multiobjective Generation Schedul

E min
T ? " ' T
Ec
t m t m

Figure 6.23 Membership

functionof line flow.

Generatorlimits IL(P)
The fuzzy sets more accuratelyrepresentthe operationalinequality constraintsof the power
system.Fuzzy membershipmay have a variety of shapesbut for simplicity, here th generator
limits are representedby a rectangularmembershipfunction (Figure 6.24). Mathe tically, the
membershipfunction is defined as

0
(
4 <4'"*
P,mintP-ltPrmin+AP-'

It(4) = I p , * "+ M , > F i > 1 ^ * - / P . (6.1ex)

(
I

I P;*o - M, r1, P,'*

t
0 F,>1^*
where A{ is the rangein which the membershipfunction varieslinearly,as decidedby decision
maker.

6.9.3 Performancelndex
The objective functions are reformulatedas fuzzy sets and eacn line flow constrain defines a
fuzzy region of acceptability.Keeping in mind that higher the membershipvalue the ter is the
solutiort,the line flow constraintsare viewed as objectivesthat maximize the membershi functions
for each line. The multiobjectiveoptimizationproblem of Eqs. (6.188) can be rewritt n a s

Maximize[pG), lr(F), tt(\), p(Pr)] W = l, 2,...,NL, F(F) (i = l, 2, ...,N)r (6.r92a)

NG
subjectto Fo+FL-I A - o (6.rezb)
i=1
 Power SYstem OPtimization

F,lnt" Fr*"* AF, Fi*+ AF P-,P*

Figure 6.24 Membershipfunctionof generatorlimits.

problt by
A multiobjective optimization problem can be changed to a scalar optimization
reference( get)
defining a global function which minimizes the deviationsfrom the so-called
in space of the ob.ective
objective [Osyczka and Davies, 1984]. Any reasonableor desirable point
reference objective. The scalar
is chosen by the decision maker and may be considered as the
is given bel w :
optimizationproblem obtainedfrom the above statedmultiobjectiveproblem
NL+NG+3

Ir4inimize E-+ Zrr,o- t'i)' .193a)

j=l
NG
193b)
subject to Fo +FL - >F,=o
i=l
i to an
To solve such optimization problems,the original constrainedproblem is transformed
auxiliary unconstrainedproblem,whoseminimum is the sameas the minimum of the origin pro-
constrained blem
blem [Sasson,Lg6g].By applying the Zangwill's method [Sasson,1969], the
form and is
of Eqs. (6.193) is transformedinto an unconstrained given below:

NL+NG+3
E = + L w , ( . )u- ) 2 + 6.194)
j=l

where h, it the target membershipfunction for the 7th function and R is a constant'

6.9.4 Structureof ANN

structure of the human brain. The netw k i s a
I The design of ANNs is based on the neural
neurobiological odels.
i, massivelv interconnected dynamic system with interacting parts based on
It consists of a large network of processors called neurons which are affanged in lay' and
connected to each other by means of information channels called interconn'ections. Th input
neurons,forming the input layer, receivedata from the outside world. The output neurons'
 Stochastic Multiobiective Generation Scheduling

the output layer, send information to the user. The hidden neuronsform the hidden I yers in
between, and store the information obtainedthrough training.
A schematicof a three-layeredfeedforwardneural network model is shown in Fig 6.25.
There is a connectionstrength,synapsesor weight associatedwith each connection.Each neuron
can have multiple inputs while there can be only one output. The inputs to a neuron uld be
from externalstimuli or could be from the output of the other neurons.In the simplest each
neuron producesits output by computingthe inner product of its input and
signe.l(s) ociated
weightswhich is passedthrougha nonlinearfunction as shown in Figure 6.26. One monly
used nonlinearmonotonicfunction is the sigmoidal one, which can be defined as follo

1.0 (6.1es)
flx) - 1.0+ exp (-x)

Outputs

Outputlayer

frt, wLz
Hidden
layer

Input layer

Inputs I
neural network.
Figure 0.25 Schematicof feedforwardthree-layered

n"ti =,!r*iiPi

model of neuron.
Figure 6.26 Mathematical
F-
4gZ Power SYstem OPtimization

their performance by learnin new

A crucialpropertyof thesenetworksis their ability to improve I

I
by modifying the interconnection strengthsamong n( rons i
information.This is accomplished
permits htgh computa ional
according ro some prescribed rules. Further, the network architecture
processing. It is also due t this
rare ro be obtained through the rnassively parallel distributed
kind of processingthat such networkshavea greatdegreeof robustness or fault tolerancetr local
famousbei g the
damages.Variouslearningalgorithmshave been recentlyintroduced,the most
backpropagationalgorithm.

Algorithm
6.9.5 BackproPagation
Backpropagation learningalgorithrnfinds the valuesof all of the weightsthat minimi the
prese'ntt, the
function using a methodof gradientdescent.That is, after each patternhas been
error gradient tc ards
error on that pattern is computed and each werght rs moved down the
value for rhat pattern. The error is actually defined by Eq. (6.194), wh.ere h'
is the
its minimum
component f the
target objective for the 7th component of the output patteru and /rl.) is the 7th
pattern. The nett rk is
actual objective produced by the network representationwith the input
specifiedas

Fj=fi(net1)- .1e6)
_
I+e
+

ne!
\rN wirPr .197)
^L
k=l
/ to be adusted.
where 141is the weight on connectionfrom unit 7 at layer (/-1) to unit k atlayer
is used is
To obtain a rule for adjustingweights,the gradientof E with respectto )t r
representedas follows:
AE = .1e8)
6iFo
dwp

of k is a recursiveprocessthat startswith the outputunit, and is give by

The determination
)tr
6;= f ( 1 - Pri ') rP i 6.ree)
dPi'
that re< ve no
The backpropagation rule comes from its assignment of deltas to hidden units
influ e the
direct feedback from the training patterns in the outside world. These'deltas actually
tor idden
rnodificationof weights to connectionsleading to the hidden units. The delta term
of the
units.for which thereis no specifiedtarget,is determinedt'ecursivelyin termsof delta tern:
connections. A unt in an
units to which it is directly connectedto and the weights of these
arbitrary hidden layer is given bY

6j= r,0- P--

)> 5 t w t j 6.200)
k

The rule for adjusringweightscan be obtainedby using Eq. (6.198)and is given by

,fo*' = n6iFr + awit 6.201)

 stochastic Muttiobiective Generation schedu

4 is the learning rate Parameter

the effectof previousweightchan
constantto determine
a is the momentum

6.9.6 SamPleSYstemStudY
The appticabilityof the methodis demonstratedon a samplethree-generatorpower s whose
expected cost and emission characteristicsare given in Tables 6.43 and 6.44, pectively.
ExpectedB-coefficientsfor transmissionloss are depictedin Table 6.45. Table 6.' showsthe
exiected GGDFs (Secrion3.10.5).In addition,the valuesof the CVs and CCs are tak as 0.1 and
1.0, respectively,for all ranciomvariables.

Tbble 6.43 Expectedcost coefficientsand generatorlimits

Plant Ai bi ci P,t* 4*
no. ($/Iuw2h) ($fvtwtrl ($n; (Mw) (MW)

1 0.006 5.506 264.634 225.0 40.0

2 0.016 5.268 rs4.298 240.0 20.0
3 0.005 5.627 26t.186 114.0 20.0

Thble 6.44 Expected NO' emission coefficients

PIarrt di €i fi
(kslTr4w2h) (kg/Iawh) (kg/h)
no.

I 0.00625 -0.56699 62.3672s

0.00626 -0.56699 62.36725
2
3 0.00348 0.36551 165.12520

Thble 6.45 Expected B-coefficients x L02

-0.003506 -0.036788
0.02725r
-0.003506 0.030896 -0.005653
-0.036788 -0.005653 0.32295

Thble 6.46 ExPectedGGDF

Line Dr*,r Dr*,2 Du,t

Line
no. t-k
0.5456 -0.4544 -0.4544
t-2
0.1548 0.1548 -0.3225
2-5
0.0621 o.062r -0.4154
54
0.1378 0.1378 -0.38s0
24
0.0081 0.0081 -0.9919
4-3
494 Power System Optimization

Input for the network

The input for the networkis obtainedby performingthe minimumexpected power
dispatcir.MinimumexpectedNO, emissiondispatchproblemcan alsobe usedto get the for
the network. The schedule obtained from the expected economic dispatch and minimum pected
emissiondispatchare givenin Table6.47 for differentexpecteddemands.

Thble 6.47 Expectedthermal load dispatch

Generation schedule (MW) Cost SrcN

Dernand (MW)
FD 4 P2 P3 E tsn'l (ke/h)

Minimum cost dispatch

150.0 84.4225 40.345r 28.1497 1599.194 8.092
200.0 114.6325 52.6536 38.0150 1948.139 15.879
250.0 r45.4r29 65.3569 47.6014 2319.782 58.73s
Minimumemissiondispatch
150.0 66.4080 66.0222 25.1528 1615.615 79.379
200.0 90.0697 89.0362 25.1528 198r.723 .269
2s0,0 108.7398 r06.2855 42.706t 2361.683 .105

Rangeof obiectivelevelsfor decisionmaker

and minimum ssron
4*n *d 4*" are obtainedby solving the expectedeconomicdispatch
dispatch separatetry.As economy, environmental impacts and risk are mutually c flicting
objectives,thereforeF2 will have maximum valueratthe schedule,when 4 is minimum. ert

the range of objective levels may be decided by the experienceddecision maker the
expectedsolution trend.

Neuralnetworkdesign
A three-layeredANN is formed with three neurons in input, hidden, and output layers net-
work is trained with backpropagationalgorithm to achieve the target values given in e 6.49,
whereasthe input given is ob',ainedfrom the minimum cost dispatch and is depicted in le 6.48.
The valuesof n and a arechosenas 0.000005and 0.5 respectively. The valueof R is its
5m. The strengthof connections as weightsis gtven in Table6.50.
The choiceof the number of neuronsis a difficult task.But goodresultsareachie if the
numberof neuronsis equalto the numberof patterns.The numberof iterationsdependon initial
weights and the value of n. The choice of membershipfunction is again a crucial point be the
objective functions are in conflict. Choice becomeseasy if the solution trend is known pnon.
A theofetical basis and methodology for optimal dispatch problem in a uni multi-
objective frar.neworkis established.The chapter also investigatesthe feasibility of q itative
representationof inaccuraciesand uncertaintiesof the input data and power demandfor power
dispatch problem in terms of probability and statistics.An artificial neural network lis
establishedto capturethe optimal generationdispatchfor power system operationsrvith ultiple
conflicting objectives
 Stochast,ic Multiobiective Generation Scheduling 495

Table 6.48 Real Pow€r line flows

Mirtinrum cost disPatch Minimum emission dispat

Line Power dennnd (MW) Power demand (MW)

l+ 150 200 250 150 200 250

L-2 14.9368 2r.3436 28.0090 -2.8563 -2.7454 8,3733
2-5 r0.2357 13.6360 17.2757 t4.0502 19.6138 9.5132
54 -3.9453 -5.3916 -6.6848 -0.0841 0.6739 4.3870
24 6.3553 8.4162 10.7r75 10.5488 t4.9969 3.1886
4-3 -26.9r10 -36.352r -45.5085 -23.4983 -45.5085 .6r84

Table 6.49 Targetmembershipfunctions

Expected power demand

Menbership
function of 1s0(Mw) 200 (Mw) ffry)
Fl 0.85 0.9 0.89
F2 0.5 0.46 0,46
F3 0.7 0.62 0.44
PTr 0.36 0.45 0.4
PTz 0.83 0.7 0.r2
PTt 0.54 0.7 0.1
PT+ 0.83 0.7 0.2
PTs 0.83 0.8 0.2
Pl 1.0 1.0 1.0
P2 1.0 1.0 1.0
P3 1.0 1.0 r.0

Thble 6.50 Weights

J wjt W;'t

Layer 1
I 72.r7046 -32.42737 - t 7 .75860
2 12.1,5827 -90.55445 .64217
3 14.84998 -r07.18450 .38823
Layer 2
I 2.62108 -40.59315 .89007
2 1.22125 42.6s099 .73229
3 0.73992 -59.48106 .13109
-

496 Power SYstem OPtimization

The synthesizedneural networks are very robust and possessgeneralizationcapabiity of

examples.The main advantageof the neural networks is that they are independentof fun
tional
form and insensitiveto parameterperturbation.Although training requiresconsiderabhcomp oll,

determining the optimal alternativecan be executedon personal computers. This featu also
facilitatesthe applicationsof the new technologyto me decisions.

REFERENCES
Books
Anillaga, J. and C.P.Arnold, ComputerAnalysisof PowerSystems,JohnWiley & Sons,Sin
I 990.
Christensen,G.S. and S.A. Soliman,Optimal Long-TermOperationof Electric Power S
PlenumPress,New York, 1988.
Elgerd,O.I., Electric EnergySystemsTheory:An Introduction,2ndedn.,Tata McGraw-Hill 1 9 8 3 .
El-Hawary,M.E. and G.S. Christensen,OptimalEconomicOperationof Power Systems,Ac tc
Press,New York, 1979.
Frederick Soloman, Probability and StochasticProcesses,PrenticeHall, New Jersey,1987-
Haimes,y.y., Hierarchical Analysisof Water ResourceSystem:Modeling and Optimi on of
I-arge Scale Systems,McGraw-Hill, New York, 1977'
Kirchmayer,L.K., EconomicOperationof PowerSystems,Wiley EasternLtd., New Delhi, 9 5 8
Klir, G.J. and B.yuan, Fuzzy Setsand Fuzzy Logic: Theory and Applications, Prentice- l o f
India, New Delhi, 1997.
Klir, G.J. and T.A. Folger, Fuu! Sets,(Jncertaintyand Information, Prentice-Hallof India, t993.
Kosko, B., Neural Networks and Fuzzy Systems:Dynamical Systems Approach to chine
Intelligence, Prentice-Hallof India, New Delhi, 1994.
Kusic, G.L., ComputerAided Power SystemsAnalysis,Prentice-Hallof India, New Delhi, 986.
i.s and
Mahalanabis,A.K., D.P. Kothari and S.I. Ahson, ComputerAided Power SystemAnaly
Control,TataMcGraw-Hill, New Delhi, 1991.
Modern PowerSystemAnalysis,TataMcGraw-Hill, New Del 1989.
Nagrath,I.J. and D.P.Koth a"J^,
Nagrath,I.J. and D"P.Kothari, PowerSystemEngineering,TataMcGraw-Hill, New Delhi, 994.
Osyczka,A. and B.J. Davres,Multicriterion Optimizationin Engineeringwith FORTRANP
Ellis Horwood Ltd., 1984.
papoulis,A., probability, Random Variablesand StochasticProce.tses,Tata McGraw-Hil , New
Delhi, L99l
Rao, S.S., Optimization:TheoryM Applications,Znd edn., Wiley EasternLimited, New Delhi,
1987.
Sen Gupta,J.K., StochasticProgramming,North Holland, 1972.
Analysis,McGraw-Hi l, New
Stagg,G.W. and A.H. Ei-Abiad, ComputerMethodsin Power'systems
Delhi, 1968.
 Stochastic Multiobiective Generation Sc

Stevenson,W.D., Elements of Power System Analysis, 4th edn., McGraw-Hill' N York,

1982.
Wood,A.J. and B. Wollenberg,Power Generation,Operationand Control, John Wiley, N w York,
1984.
Zurada,J.M., Introductionto Artificial Neural Network, Jatco Publishing House, Delhi,

Papers
Blaszczynski, G.M., Sensitivity study of the economic dispatch, Proceedings of PICA rence,
New Orleans,LA, LT.S.A.,L975.
Brar, Y.S., J.S. Dhillon, and D.P. Kothari, Genetic-fuzzylogic basedweightagepatternfi multi-
objective load dispatch, Asian Journal of Information Technology,Vol. 2, No. 4, pp.
-373,
Oct.-Dec.,2003.
Brar, Y.S., J.S. Dhitlon, and D.P. Kothari, Multiobjective load dispatch by fuzzy log c based
weightagepattern, Electric Power SystemResearch.Vol. 63, pp. 149-L60, 2002.
Chowdhury, N. and R. Billinton, Risk constrainedeconomic load dispatch in intet nnected
generatingsystems, IEEE Trans.on Power Systems,Vol. 5(4), pp. L239-L247, 1990.
Cohon, J.L. and D.H. Marks, A review and evaluationof multiobjectiveprogramming hniques,
WaterResourcesResearch,Vol. ll(2), pp. 208-220, 1975.
David, A.K. andZ. Rongda,An expert systernwith fuzzy setsfor optimal planning, IEEE rans. 04
Power Sysrens,Vol. 6(1), pp.59-65, 1991.
Dhillon, J.S. and D.P. Kothari, The surrogateworth tradeoff approachfor multiobj thermal
power dispatchproblem, Electric Power SystemResearch,Vol. 56, No. 02, pp. 103-1 0, Sept.,
2000.
Dhillon, J.S., S.C. parti, and D.P. Kothari, Multiobjective optimal thermal power dispatc , Int. J.
Electric Power & EnergySystens,Vol. 16, No. 6, pp. 383-390, Dec., L994'
Dhillon, J.S., S.C. parti, and D.P. Kothari, Stochasticeconomicemissionload dispatch,Electric
Power SystemResearch,Vol. 26, pp. 179-186, 1993.
-term
Dhillon, J.S., S.C.parti, and D.P. Kothari, Fuzzy decision making in multiobjective I
schedulingof hydrothermalsystem,Int. J. Electric Power & Energy Systems, Vsl. 23 No. 01,
pp. L9-29, Jan., 2001.
Dhillon, J.S., S.C. Parti, anciD.P. Kothari, Multiobjective Decision Making in Stochastic
Dispatch, Electric Machines an^dPower ,Sys/ems, Vol. 23, pp. 289-30I, 1995.
Dillon, T.S., R.W. Martin, and D. Sjelvgren,Stochasticoptimizationand modellingof large
hydrothermalsystemsfor long-termregulation,Int. I. Electrical Power & Energy Systems,
Vol. 2(l), pp. 2-20, 1980.
Dillon, T.S., S. Sestito,and S. Leung, Short terrn load forecastingusing an adaptiveneuralnetwolh
Int. Journal of Electrical Power & Energy $ysrems,Vol. 13(4), pp. I8GI92. 1991.
Dopazo,J.F.,O.A. Klitin, and A.M. Sasson,Stochasticload flows, IEEE Trans.on Power
Vol. 94(2),pp. 299409, L975.
and Systerns,
 495 Power SYstem OPtimization

Edwin, K.W. and R.D. Machate, Influence of inaccurateinput data on the optimal sh -term
operation of power generationsystems,IFAC Symposiumon Automatic Control in Power
Generation,Distribution and Protection,Fretona, Souilr Africa, 1980.
El-Hawary,M.E. and G.A.N. Mbamalu,A comparisonof probabilisticperturbationand de nistic
basedoptimal power flow solutions,IEEE Trans.on PowerSystem.s, Vol. 6(3),pp.l l 105,
r99r.
El-Hawary,M.E. and G.A.N. Mbamalu,Stochasticoptimal load flow using Newton-
iterative technique,Electric Machinesand Power,Sysrerus,
Vol. 15, pp. 371-380, 1988
El-Hawary,M.E. and G.A.N. Mbamalu,Stochasticoptimal load flow using a combined quasi-
Newton and conjugategradienttechnique,Int. Journal of Electrical Power & Energy stems,,
Vol. ll(2), pp. 85-93, 1989.
El-Hawary,M.E. and K.M. Ravindranath, A generaloverviewof multiple objectiveoptima power
flow in hydrothermalelectric power systems,Electric Machinesand Power Systems, t9,
pp. 313-327, 1991.
Glimn, A.F., L.K. Kirchmayer,G.W. Stagg,and V.R. Peterson,
Accuracyconsiderations
in nomic
dispatchof powensystems,A.I.E.E.Trans.on PowerApparatusand Systems, Vol. 75, ( -ilD,
pp. ll25-ll3l, 1956.
Haimes, Y.Y. and W.A. Hall, Multiobjectivesin water resourcesystemsanalysis:The S gate
worth trade-off method, WaterResourcesResearch,Yol. l0(4), pp. 61 5424, 1974.
Hannan, E.L., Linear programmingwith multiple fuzzy goals,Fuuy setsSystem,Yol. 6, p 235-
249,1991.
Hill, E.F. and W.D. Stevenson,A new method of determiningloss coefficients,IEEE
Power Apparatusand systems,Yol. 87(7), pp. 1548-1553,1969.
Hsu, Y.Y. and C.C. Yang, Designof artificial neural networksfor short-termload forecastin . Part
II: Multilayer feedforwardnetworksfor peak load and valley load forecasting, IEE Proc eding,
Part C, Vol. 138(5),pp. 41H18, 1991.
Kaunas,J.R. and Y.Y. Haimes, Risk managementof groundwatercontaminationin a multi-
objectiveframework,WaterResourcesResearch,Yol. 21(11), pp. 172l-1730, 1985.
Kothari, D.P. and I.J. Nagrath, Optimal stochasticscheduling of hydrothermal systems using
discretemaximumprinciple,JournAIInstitutionof Engineers(lndia),Yol.6L,pp. 22-26 1980.
Leberling,H., On finding compromisesolution in multicriteriaproblemsusing the fuzz
operator,Fuzzy,SersSystem,YoI.6, pp. 105-118,1981.
Lee, K.Y., Y.T. Cha, and J.H. Park, Short-termload forecastingusing an artificial neural n ork,
IEEE Trans.on PawerSystems, Vol. 7(1), pp. lZ+132, lg9L.
Leite da Silva,A.M., R.N. Allan, S.M. Soares,and V.L. Arienti, Probabilisticload flow c
network outages,IEE Proceedings,Part C, Vol. 132(3),pp. 139-14s, 19g5.
Mazumdar,M. and C.K. Yin, Varianceof power generatingsystemproductioncosts,IEEE
on Power System.s, Vol. 4(2), pp. 662467, 1989.
Meliopoulos,A.P., X. Chao, G.J. Cokkinides,and R. Monsalvatge,Transmissionloss eval
ation
basedon probabilistic power flow, IEEE Trans. on power System^r,Vol. 6(1), pp. 3 7t ,
1991.

h
 Stochastic Multiobjective Generation Schedu

Miranda, V. and J.T. Saraiva,Fuzzy nrodellingof power systemoptimal load flow, IEEE Trans. on
Power Syslerns,Vol. 7(2), pp. 843-849, 1992.
Stochasticgenerationexpansionplanning by meansof
Mo, B., J. Hegge,and I. Wange.nsteen,
stochasticdynamic programming,IEEE 'Trans.on Power Systems,Vol. 6(2), pp. , lggl.
Nanda, J., D.P. Kothari, and K.S. Lingamurthy,A new approachto economic and lnlmum
emissiondispatch,Journal Indian Institute of Science,Yol. 67, pp. 249-256, 1987.
stochasticdynamic pro
Neto, T.A.A., M.V.F. Pereira,and J. Kelman,A risk-constrained g
approachto the operationplanning of hydrothermalsystems,IEEE Trans, on Power pparatus
and Systerus,Vol. 104(2), pp. 273-279, 1985.
Ouyang, Z. and S.M. Shahidehpour,A hybrid artificial neural network-dynamicpro mmlng
approachto unit commitment,IEEE Trans.on Power System^s, Vol. 7(l), pp.236-24 1992.
Park, D,C., M.A. El-Sharkawi, and R.J. Marks II, An adaptively trained neural network, I Trans.
on Neural Networks,Vol. 2(3), pp. 334-345, 1991.
Parti, S.C., D.P. Kothari, and P.V.Gupta,Economicthermalpower dispatch,Joumal of nstitution
of Engineers (India), Vol. 63(EL-Z), pp. 126-132, 1983.
Parti, S.C., Stochasticoptimal power generationscheduling,Ph.D. (Thesis),TIET, Pati 1987.
Pereira,M.V.F., Optimal stochasticoperationsschedulingof large hydroelectricsy ms, Int.
Journal of Electrical Power & Energy Systems,Yol. 11(3), pp.16l-169, 1989.
Rarig, H.M. and Y.Y. Haimes, RisklDispersionindex method, IEEE Trans. on Systems,Man and
Cybernetics,Vol. L3(3), pp. 317-328, 1983.
Rashid,A.H.A. and K.M. Nor, An efficientmethodfor optimal schedulingof fixed-head ydro and
thermal plants, IEEE Trans.on Power Systems,Vol. 6(2), pp.632436, 1991.
Rau, N.S. and C. Necsulescu,Probabilitydistributionsof incrementalcost of prod tion and
production cost, IEEE Trans. on Power Apparatus and Sysfens, Vol. 104(12), pp. 3 93-3499,
1985.
Sasso4,A.M., Combined use of the Powell and Fletcher-Powellnonlinear programmin methods
for optimal load flows, IEEE Trans.on PowerApparatusand Systems,Vol. 88, pp. 1 30-1535,
1969.
Sasson,A.M., Non-linear programming solutions for load-flow, minimum loss and onomic
dispatchingproblems,IEEE Trans.on Power Apparatus and Systems,Yol.88(4), pp. 399409,
1969,
Sherkat, V.R.,R. Campo,K. Moslehi,andE.O.Lo, Stochastic long-termhydrothermal mization
for a multireservoirsystem, IEEE Trans. on Powq Apparatus and Systems, 104(8),
pp. 204V2050,1985.
Su, C.C.and Y.Y. Hsu, Fuzzydynamicprogramming:an applicationto unit commi nt, IEEE
Trans.on PowerSystem.s, Vol. 6(3),pp. L23t-1237,1991.
Tapia, C.G. and B.A. Murtagh, Interactive fuzzy programming with preferencecriteri in multi-
objective decision making, ComputersOperationsResearch,Vol. 18(3), pp. 307-31 , 1991.
Tsuji, A., Optimal fuel mix and load dispatchingunder environmentalconstraints,IEEE Trans.on
Power Apparatus and Sysrens,Vol. 100(5), pp. 2357-2364, 1981.
500 PowerSy opt

Vemuri,V.,Multiple objectfe optimizationin waterresourcesystgms,WaterResources

Re
Vol. 10(1),pp. 4U8, 1J974.
optimal energy dispatch,IEEtr Trans. on,
t22t-3228,1981.
reuralnetworkfor r4ultiple criteria decisionn
l), pp. 15l-167, 1992,.
electricalpowersysfems, EngineeringOptimt

with severalobjectivefut
linearprogramming
8.