You are on page 1of 7

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3354242

Loss-constrained economic power dispatch

Article  in  IET Proceedings - Generation Transmission and Distribution · October 1996


DOI: 10.1049/ip-gtd:19960465 · Source: IEEE Xplore

CITATIONS READS

4 669

1 author:

Aydogan Ozdemir
Istanbul Technical University
145 PUBLICATIONS   643 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

DiNODR (Distribution Network Oriented Demand Response) - ERIGrid Transnational Access User Project Project View project

SEEMBEET (Smart Electric Energy Management and Buildings Energy Efficiency Technologies) - Bilateral Project View project

All content following this page was uploaded by Aydogan Ozdemir on 10 January 2016.

The user has requested enhancement of the downloaded file.


Loss-cconstra ined econorn ic power dispatch

A Ozdemir

Indexing terms: Power systems, P-dispatch module, Transmission losses, Fuel costs

~ ~ ~ _ _ _ _ _ _ _ _ _ _
However, the decomposition principle incorporated
~

Abstract: A new P-dispatch module for the into approximated equations suffers from two basic
decomposed optiimal power flow method is obstacles related to real power transmission losses
presented by which undesired increases in (RPTL). The first one is the real power mismatches in
transmission losses are prevented via an Q-dispatch, resulting from the elimination of real
additional constraint imposed upon the module. power balance equations. These mismatches may cause
The constraint only requires the restriction of the unrealistic transmission loss values and should there-
final updates of the control variables and fore be compensated to achieve an exact loss minimisa-
therefore neither destroys the form of tion [7]. The second problem is the uncertainty of the
optimisation equations nor requires extra behaviour of P-dispatch in relation to RPTL.
considerable calculation time. However, it The scope of this paper is mainly concerned with
provides flexibility in the choice of handling RPTL and introduces a new formulation to overcome
either the minimum cost or the minimum loss the second obstacle stated above. The proposed formu-
problem. The effectiveness of the method is lation is based on the successive solution of the twin
demonstrated by applying it to two sample subproblems (P-dispatch and Q-dispatch) in an itera-
networks tive scheme to achieve optimal allocation of associated
control variables. As in the usual approach, Q-dispatch
is used for loss minimisation. The distinct future of the
method is the ability of controlling RPTL in P-dispatch
1 Introduction by means of an additional loss constraint imposed
Optimal power flow is a special AC power flow where upon the module. Consequently, undesired increases in
an appropriate obje'ctive function (such as fuel costs RPTL are avoided, while minimising the hourly fuel
and/or real power transmission losses) is minimised, costs (HFC).
while satisfying the nodal power balance equations and The additional loss constraint is expressed as a sensi-
physical restrictions. Real time operation of electric tivity relation between RPTL and P-dispatch control
power systems requires efficient formulations to pro- variables. Therefore, it is encountered as a parameter-
vide fast and reliable solutions as well as minimal com- type inequality constradnt and its presence or absence
puter memory. does not change either the nature of remaining equa-
The gradi'ent approach introduced by Dommel and tions or the procedure (of optiimisation. This second fea-
Tinney [l] can be ranked as the start of exact formula- ture brings flexibility into the choice of loss constraint.
tions. Since then, a number of articles about the solu- That is, depending upon the objective desired, P-dis-
tion of this nonlinear programming problem have been patch can either be used as a conventional suboptimisa-
published [2-61. Decomposition of the process into real tion module or as a loss-constrained suboptimisation
power optirnisation (P-dispatch) and reactive power module. Since the objectives are determined by the load
optimisation (Q-dispatch) was later recognised to be configuration and the generation schedule, the pro-
more efficient than the simultaneous optimisation of all posed method can effectively be used for various time
the variable,s [4-61. IFuel cost minimisation and trans- periods of a day or ai week. etc., exhibiting different
mission loss minimisation have been the usual objec- load and generation characteristics.
tives of P-dispatch and Q-dispatch, respectively.
Following this decomposition idea, various formula- 2 Decomposed optimal lpower flow
tions based on either linear or nonlinear programming
techniques h.ave been devoted to the reduction of com- Decomposed optimal power flow makes use of the clas-
putation time and computer storage requirements. Only sification of system variables and equality constraints
dominant system equations and variables have been due to the well known weak coupling between the vari-
ables in electric power systems. Mathematically, it can
encountered in the suboptimisation processes to
be stated as a successive solution of the following sub-
improve the computation time. In addition, equations
resulting from the optimisation conditions have been problems until a desired accuracy is attained:
approximated providing minimal computer storage
requirement!;. 2. I P-dispatch
0IEE, 1996 Minimise .fp(xp,
U,)

IEE Proceedings online no. 19960465 wrt up


Paper received 14th March 1996
subject t o g,(x,, up) = 0
The author is with the Istanbul Technical University, Electrical-Electron-
ics Faculty, 80191 G U ~ U S S LIstanbul,
~ U , Turkey h,(x,,u,) 20
IEE P r o c C e n e r . Transm. Distrib., Vol. 143, No. 5, September 1996
where up denotes the control vector comprisinggenera-
tor real power outputs except the swing bus ( P G )and
phase shifter angles (a). xp denotes the state vector
comprising bus voltage phase angles (6). The control
and state variables of the adjacent module are assumed = g,(x,,u,) = 0
dXn
to be fixed throughout the module. j; is the total Where the superscript 'T' denotes the transpose of a
hourly fuel cost of N G generators, which can be repre-
matrix and wI,(xp,uI,) denotes the penalty term for vio-
sented as a second-order polynomial function of gener- lated functional variables; i.e. slack bus active genera-
ator real power outputs tion (Per) and real line flows (T). Note that wp is in
NC:
fact not an explicit function of system variables xp and
, =
f p ( x pU,) a,P$3 + b,Pct + c, a,,b,, e, constants up. However, generalised generation distribution factors
7=1
require it to be transformed to a function of generator
(2) real power outputs [8] and so thus to a function of sys-
Equality constraints are real power balance equations tem variables. The solution procedure for P-dispatch is
of nonslack buses (see Appendix 9.1). Physical restric- given in Fig. 2; where L' is a constant and r is an appro-
tions upon real power generations and real power priate function of [dLp/dup] evaluated according to the
transmission line flows form the inequality equations hp conjugate gradient method [7].
( X p up).
from Q -dispatch
2.2 Q-dispatch
Minimise f, ( x, , U,)
real power flow
wrt uq
(3)
subject to g,(x,, U,) = 0 I t
h,(x,,u,) 2 0
Where uq denotes the control vector comprising genera-
tor voltage magnitudes (VG), VAR injection values of
shunt capacitorireactors (Qc) and tap values of tap
changing transformers (t). The state vector xq com-
prises load bus voltage magnitudes (V,) and generator
reactive power outputs (QG). The control and state var-
iables of the adjacent module, U,, and xp, are assumed
to be fixed throughout the module. ,fq is chosen as
active generation of the swing machine. Fig.2 Flow chart of P-dispatch
fq(X,,U,) = Pl(X,:U,) (4)
Equality constraints are reactive power balance equa- The solution procedure for Q-dispatch can be
tions of load buses (see Appendix 9.1). The optimisa- obtained similarly. Since the final form of equations do
tion procedure for the decomposed optimal power flow not yield significant changes, they will not be repro-
is given in Fig. 1. duced again.

t
Q 4 Loss variation in P-dispatch

Most of the formulations depending on the decomposi-


initial power flow
tion principle use H F C and RPTL as the objective
functions for P-dispatch and Q-dispatch, respectively.
I Q-dispatch
Since the control variables of P-dispatch, i.e. real
power generation of nonslack buses, are assumed to be
fixed throughout the Q-dispatch, optimum loss condi-
tions also correspond to lower fuel costs. That is, reac-
tive dispatch not only acquires an optimal loss but also
lowers the cost, although it is formulated to achieve the

Fig. 1
(ges
Flow chart of decomijosed optimal power dispatch
former objective. However, P-dispatch may result
either in a decrease or an increase in RPTL while mini-
mising fuel costs. In particular, the systems consisting
of generators possessing extremely different cost coeffi-
3 P-dispatch cients may yield unacceptable loss increments.
Let us now try to express loss variation in terms of
The set of necessary conditions of the real subproblem system variables. The sum of the net real bus powers of
stated by eqn. 1 can be obtained by differentiating the an N-bus power system gives the RPTL.
augmented Lagrangian function
L,(X,. u p >A), = f P ( X P >U,) + A%; &>, U,) +Wp(Xp, u p )
(5)
with respect to xp, up and hp Lagrangian variables, i.e.

388 IEE ProcGeizer Trunsn?. Distrib., Vol. 113, No 5. SepiemOrr 1996


where PGl and P,, denote the real power generation subvector corresponding to phase shifter angles (a),A,,
and consumption at bus i, respectively. Suboptimal val- is the portion of the Lagrange variables corresponding
ues of the module yield a loss increment, i.e. to generator buses, and hp2 is the portion of the
Lagrange variables corresponding to load buses. How-
ever,

[$1
agp, 1 i - j = 2 , 3 , . . . ,N G
The partial derivatives in eqn. 8 can be derived from
eqn. 7,
2~ = = {0 otherwise (19)
(see Appendix 9.1) and
aR,PTL - dRPTLdP1 - dP1
-~- ~- - -
i3X, ap, ax, ax, (9)

~ = 2 , , 3. . . , N G 1 = 1 , 2, . . . ,IVa
can be calculated according to Appendix 9.2. Hence,
where I denotes a vector whose entries are all ones and
Sp,-a denotes the sensitivity of the slack bus active gen-
eration with respect to phase shifter angles (see Appen-
IT
Auy2

(21)
dix 9.2). Introducing eqns. 9 and 10 into eqn. 8 results Or in terms of corresponding control variables,
in

Since real power balance equations are included in the Eqns. 21 and 22 shows the sensitivity of RPTL to the
optimisatiort process, optimal values of up and xp do control variables of P-dispatch.
not produce a residual term for these equations. That
is. 5 Loss constrained P-dispatch

The conventional P-dispatch module of a decomposed


optimal power flow ininimises the fuel costs, while
Hence, resulting in either an increase or a decrease in RPTL.
However, the change in transmission losses expressed
by eqn. 21 or eqn. 22 can be imposed as an additional
constraint to prevent an undesired increase in RPTL.
This imposition can be done by restricting the updates
of control variables, i.e.

i = 2 , 3 , . . . ,N G
(14) [aUP212=A(2i=Aai if ;~ST,-,.,AP+SP,-,
And hence,
i = 1,2,...>ATCk
{ 0 ot Ierwise

(24)
If eqns. 13 and 15 are introduced into eqn. 11 In another words, loss constraint is held as a parameter
inequality constraint and therefore it does not destroy
the nature of the optimisation equations. Moreover, it
does not affect the computation time significantly. But,
it can effectively be used to control the change in trans-
+ [IT s;l-a]}.nu, mission losses as well as providing flexibility in the
choice of implementation.
(16) 6 System studies
If hp is derwed frorn the first equation of eqn. 6 and
replaced by the related term in eqn. 16 The proposed formulation is programmed and tested
by solving several test systems. Among them, the
results for the IEEE 14-bus test system and the IEEE
30-bus test system are illustrated here.
If control variables are classified, then the loss incre- The line diagram alf the 14-bus system containing
ment can be expressed as three generators, three tap changing transformers and
two parallel capacitive reactors is given in Fig. 3. Nec-
essary data of the test system are given in Tables 1 and
2. The results are suminarised in Table 3. The columns
CDOPF and LCDOPF correspond to the results for
conventional decomplosed optimal power flow and
where upl denotes the subvector corresponding to non- decomposed optimal power flow possessing the loss-
$lack generator real power outputs (&), up2denotes the constrained P-dispatch module, respectively.
I E E Pro< -Gener Trumm I h t r i h V d 143 No S Septemhcr I994 389
The most significant result is the saving of 0.58MW 4.78MW, respectively. The results of LCDOPF is an
in transmission losses (corresponding to 6.5%) with an 80.39$/h (10%) increase in HFC for the sake of a
increase of 2.8 $/h in hourly fuel costs (corresponding 5.OMW (510/0) saving in RPTL.
to 0.25%). A 4

Table 1: IEEE 14-bus test system impedance and line


charging data
I I l l l 4 lo& I 8-
Line From To
Line impedance Line charging Tap setting
no bus bus

R Lp.u.1 XLp.u.1 B Ip.u.1 m i n max


~

1 1 2 0.01938 0.05917 0.0528


2 2 3 0.04699 0.19797 0.0438
3 2 4 0.05811 0.17632 0.0374
4 1 5 0.05403 0.22304 0.0492
5 2 5 0.05695 0.17388 0.0340
6 3 4 0.06701 0.17103 0.0346
7 4 5 0.01335 0.04211 0.0128 -
8 5 6 0.00000 0.25202 0.0000 0.90 1.10 Fig.3 IEEE ICbus test system
9 4 7 0.00000 0.20912 0.0000 0.90 1.10
10 7 8 0.00000 0.17615 0.0000 7 Conclusions
11 4 9 0.00000 0.55618 0.0000 0.90 1.10
Depending on the system parameters and the load
12 7 9 0.00000 0.11001 0.0000
schedule, P-dispatch of decomposed optimal power
13 9 10 0.03181 0.08450 0.0000
flow methods may result in an increase in transmission
14 6 11 0.09498 0.19890 0.0000 losses. In particular, the systems consisting of genera-
15 6 12 0.12291 0.25581 0.0000 tors exhibiting highly different cost coefficients yield
16 6 13 0.06615 0.13027 0.0000 unacceptable loss values.
17 9 14 0.1271 1 0.27038 0.0000 This undesired increase is avoided by an additional
18 10 11 0.08205 0.19207 0.0000 loss constraint imposed to restrict the updates of
19 12 13 0.22092 0.19988 0.0000
control variables. Since this additional constraint is
expressed as a parametric inequality constraint, the
20 13 14 0.17093 0.34802 0.0000
proposed formulation neither changes the original form
of equations nor requires a significant calculation time.
The IEEE 30-bus test system is the most common But, it achieves less transmission losses with an
one and is used widely in power system optimisation acceptance of a slight increase in fuel costs. Moreover,
problems. Necessary information is presented in [6] and flexibility in the choice of objective function is
will not be reproduced here. H F C and RPTL values provided. That is, the formulation allows the
for CDOPF are 802.06$/h and 9.78MW, respectively. possibility of holding either the minimum cost or the
Corresponding values for LCDOPF are 882.45$/h and minimum loss problem.

Table 2: IEEE 14-bus system load data, cost coefficient data and physical
restrictions

Bus Loaddata Cost coefficient Physical restrictions

LP.U.1 Lp.u.1 Lp.u.1


1 0.000 0.000 50 245 105.0 1.00 1.10 0.5 2.0 -0.2 1.0
2 0.217 0.127 50 35 1 44.4 1.00 1.10 0.2 1.0 -0.2 0.5
3 0.942 0.190 0.95 1.10 0.00 0.25
4 0.478 0.040 0.95 1.10
5 0.076 0.016 0.95 1.10
6 0.112 0.075 50 389 40.6 1.00 1.10 0.2 1.0 -0.2 0.5
7 0.000 0.000 0.95 1.10
8 0.000 0.000 0.95 1.10 0.0 0.25
9 0.295 0.166 0.95 1.10
10 0.090 0.058 0.95 1.10
11 0.035 0.018 0.95 1.10
12 0.061 0.016 0.95 1.10
13 0.135 0.058 0.95 1.10
14 0.149 0.050 0.95 1.10

390 IEE Proc:Gener. Transni. Distrib., Vol. 143, No. 5. September 1996
Table 3: Summary of the results for the IEEE 14-bus test 8 NG, W.Y.: ‘Generalized generation distribution factors for power
system system security evaluations’, IEEE Trans., 1981, PAS-100, pp.
1001-1005
Initial
Variable Limits Final states 9 Appendixes
state
_ _ _ ~
Lower Upper CDOPF LCDOPF 9. I Equality constraints and solution
p.u. 0.50 2.00 2.1 136” 1.6303 1.4832
algorithms
0.20 1.00 0.4000 0.6647 0.6499 Equality constraints fior suboptimisation modules are
0.20 1.00 0.2000 0.3839 0.5399 real and reactive power balance equations at appropri-
p.u. -0.20 1.00 0.0301 0.0722 -0.0285 ate buses. Decoupled Newton-Raphson load flow solu-
-0.20 10.50 0.6665’ 0.2409 0.41 13 tion algorithm applied for solving these equations,
-0.20 0.50 0.3080 0.4537‘ 0.3624 constitutes the successive sollutions o f the sparse, line-
0.0 10.25 0.0000 0.0532 0.0427 arised matrix equations
0.0 0.25 0.0000 0.0445 0.0371
p.u. 1.00 1.10 1.0700 1.1000 1.0936
1.00 1.10 1.0500 1.0791 1.0792
0.95 1.10 0.981 1 1.0252 1.0224
0.95 1.10 1.0022 1.0462 1.0431 where
0.95 1.10 1.0151 1.0642 1.0603
1.00 1.10 1.0500 1.0629 1.0592 [ g p ] i =PG’-PL~-X Y[Gi.icos(Sj-Si)-B,, sin(Sj--Si)]
0.95 1.10 1.0055 1.0591 1.0527 3 W )
0.95 1.10 1.0055 1.0664 1.0589 (26)
0.95
0.95
0.95
1.10
1.10
1.10
0.9964
0.9978
1.0196
1.0404
1.0364
1.0454
1.0333
1.0295
1.0398
, QG,-QL,-K
[ g q ]= x:V,
3E{[}
[G,, sin(S,-&-B,, cos(S,-S,)]

0.95 1.10 1.0321 1.0475 1.0436 (27)


0.95 1.10 1.0240 1.0415 1.0369
where P,, P,, are real power generation and consump-
0.95 1.10 0.9895 1.0225 1.0162
tion at bus i, Q,, are reactive power generation
deg. - -4.3980 -3.0000 -2.8590
-
.-
and consumption at bus i, Vi,aiare voltage magnitude
-11.5130 -9.6230 -9.2680
- -8.9 240 -7.1280 -6.5630
and angle at bus i, G,, By are real and imaginary parts
- -7.5590 -5.9320 -5.3350 of the (i, j)-th entry of the bus admittance matrix, Y =
- -1 1.7860 -8.1560 -6.1740 G + jB, AVIV, A6 are voltage magnitude and angle
- -1 1.5970 -9.3610 -8.47 00 increment vectors and {I} signifies the set of buses con-
- -1 1.5970 -9.3610 -8.47 00 nected to bus i.
- -13.0430 -10.5880 -9.5220 The entries of partial derivative matrices are
- -1 3.1090 -10.4480 -9.2210
- -12.3630 -9.4400 -7.8390 [ J l ] i j = [J4],, 1& V , [ C i j sin(Sj - S i ) + Bij cos(6j - S i ) ]
- -12.7390 -9.1460 -7.2350 for i # j
- -12.8230 -9.3420 -7.4930
- -14.0320 -1 1.0700 -9.6730 [Ji]ii = Qi + y2Bia
0.90 ‘1.10 0.9620 1.0618 1.042 1 [J4]tt = -Qt + y2&
0.90 ‘1.10 0.9780 0.9623 0.9628
(28)
0.90 ‘1.10 0.9690 0.9600 0.9602
where Qi denotes the reactive power of the ith bus. In
Cost [$/hl 1159.391 1134.444 1137,2380 practical power systems the following assumptions are
Loss [MVVI 12.340 8.890 8.3130
- always valid:
Number of iterations 3 2
Maximum power mismatch 0.0001 0.0001
Maximum constraint violation 0.0037 0.0
*denotes the constraint violation
Hence the partial derivatives can be approximated as
References [J1]27 = [J1]32 [J4],, = [J4]p V,V,Bt, cos(S7 - 6%)
for 1, #
DOMMEL. H.W., and TINNEY, W.F.: ‘Optimal power flow
solutions’, IEEE Trans., 1968, PAS-87, pp. 1866-1 876 [J1127.[J41,2 K2B1.,
SASSON, A.M., and MERRIL, H.M.: ‘Somc applications of (30)
optimization techniques to power system problems’, Proc. IEEE,
1974, 62, (7,) pp. 959-972 eqn. 25 possesses symmetric, sparse and diagonal dom-
ARVANITLDIS, N.V., and ROSING, J.: ‘The use of objective inant matrices [Jl] and [J4]. The optimally ordered tri-
functions in real power dispatching’, IEEE Trans., 1971, PAS-90, angular bifactorisation method is chosen for solving
pp. 1742-1751
SHOULTZ, R.R., and SUN, D.T.: ‘Optimal power flow based
eqn. 25.
on P-Q dec:omposition’, IEEE Trans., 1982, PAS-101, pp. 397.-
node i
405
BURCHETT, R.C., HAPP, H.H., VIERATH, D.R., and WIR-
GAU, K.A.: ‘Developiments in optimal power flow’, IEEE Trans.,
1982, PAS-101, pp. 406414
LEE, K.Y., PARK, Y.M., and ORTIZ, J.L.: ‘Optimal real and
reactive power dispatch’, Electr. Power Syst. Res., 1984, (7), pp.
20 1-2 12
OZDEMIR, A.: ‘Approximate sub-optimum techniques for opti-
mal power flow’, Electr. Much. Power Sy.rt., 1993, 21, pp. 267- ideal transformer
279 Fig.4 Generalisedphase shifter

IEE Proc.-Gener. Trunsm. Dirtrih , Vol 143, N o 5, September 1996 391


9.2 Phase shifters = y2t2 g,, - t.Ky [gzJ cosy - b,, siny]
A generalised phase shifter, in which both voltage mag- (32)
nitude and phase angle vary, is given in Fig 4 Kirch- PJZ= Real { ~ : . T J z }
hoff Laws and power balance equations result
- = v;” g2, ~ t KV, [gzg cosy + b,, sin y]
I,, = (V, - t V,e J c r hyz,
)
apt __ apt,
sin y + b,, . cos 71
-
I zJ - -t e-Jak 73z = t’y,, V,
- t y z J e-Jak.V3 (31) __ - __ = t.K V, [g,,
dai, doli,
where overlined quantities denote complex voltages and (33)
currents, ak denotes the phase shift, t = t ( q ) denotes 3
aak
= % = t KV,[gzg siny
(?ai,
- b,, cosy]
the tap value and y, = gi,- J 6, denotes phase shifter where y = 6, 6, + a,<.If the tap value is 1 and phase
~

admittance Real line flows dnd bus power sensitivities shifter admittance is approximated to be purely induc-
can be derived as tive, the sensitivities reduce to

Pz, = Real { V,”


.TZJ}

392 IEE Prm-Cener. Trunsm. Distrih.. Vol. 143, No 5, September I996

View publication stats

You might also like