Advanced Transformer Control Modeling in An Optimal Power Flow Using Newton's Method

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

2911 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO.

I , FEBRUARY 2000

Advanced Transformer Control Modeling in an


Optimal Power Flow Using Newton's Method
E. Acha, Hugo Amhriz-PCrez, and C. R. Fuertc-Esquivel

Abstract-This paper reports on advanced transformer within the context of load flow solutions [6]-[8], is in daily use
modeling facilities snitahle for large-scale optimal power flow in most UK distribution companies. However, three-winding
stndies. The new transformer models are developed from first transformer models for load flow studies do not seem to be
principles and incorporated into an existing Newton-based op-
timal power flow cnmputcr program for highly robnst iterative highly developed and with the growing interest in thc use of
solutions. A three-winding transformer model with tap ratios FACTS devices at the transmission and distribution levels 191,
in all three windings is shown to be a general case for existing [IO], it is most important that fully fledged three-phase trans-
two-winding transformer models and the classic load tap-changing former models, with loaded tertiary windings, he made avail-
and phase-shifting transformer models. The newly developed able in the open literature.
transformer models add a great deal tu software flexibility and
are amenable tu morc realistic electric energy studies. This is This paper now reports on morc advanced transformer models
partly due to the transformer models being fitted with complex than the ones currently available in the open literature 111-181.
tap changers in each winding and a nonlinear representation of Comprehcnsiveuess and flexibility have bcen our over-riding
the magnetizing branch. The threc-winding transformer model concerns while developing these new models. Considering the
interfaces easily with reactive power plant models, e.g., static var need l o conduct realistic energy studies in a deregulated envi-
compensators.
roilincut, these models havc been developed within the context
Index Ternis-FACTS, load tap-clianging transformers, of optimal power flows: a subject of considerably more engi-
Newton's method, optimal power fluws, phase-shifting trans- neering and mathematical complexity than load flows.
formers.
A generalizcd, three-winding transformer model is developed
from first principles, which caters for both LTC transformers
I. INTRODUCTION and phase-shifting transformers. The combined model takes the
OWER system trans€osmer models for studies at the form of a nodal admittance matrix equation where complex tap
P fundamental frcqaency have been in existence for many ratios on the primary, secondary, and tertiary windings of the
years [1]-[4]. These transformer representations incorporate
transformer are represented explicitly. The magnetizing branch,
which wider saturated conditions is nonlinear, is included to ac-
load tap-changing (LTC) and phase-shifting control modeling
count for the core losses. The generalized transformer model has
facilities and are suitable for large-scale load flows [11-[31
and optimal power flows (OPF) 141, 151 studies. However, been incorporated in an optimal power flow algorithm, leading
there is growing concern among practicing cngineers that to very robust iterative solutions since the optimization process
these transformer models may have various shortcomings 161 is carried out viaNewton's method. It should be stressed that the
algorithm is a fully coupled method, and contrary to methods
including model inflexibility and a lack of magnetizing branch
representation. These models were developed bearing in mind bascd on decoupled formulations, it does not run into prob-
the structure of high-voltage transmission circuits, e.g., the lems when solving networks containing branches with high R/X
tap-changing facilities are assumed to be on the primary side of ratios. The model is very flexible and can be set to simulate
the transformer [:11-[5]. different operating modes with ease, and in a highly efficient
fashion. The efficiency of the solution stems from the fact that
Following deregulation of the electricity supply industry in
the UK, there have been repeated calls by experienced engi- the systctn dimensions remain unchanged throughout the itera-
tivc solution; there is no need to recalculate the elimination se-
neers that the transformer magnetizing branch be modeled in
power flow algorithms in order to conduct more realistic clec- quence in the Gaussian elimination proccss, regardless of con-
trol requirements.
tric energy studies 161. Good progress has been achieved in
this direction and a two-winding trandormer model, developed It is shown in the paper that more realistic studies of ac-
tive and reactive power losses at the system level are possible
with the use of the ncw optimal power flow transformer models.
Manuscript lrceivcd July 21, 1998: revised Dccember 8, 1998. The work of A company power network [ I l l , consisting of 166 nodes, 236
H. Amhriz-PBm' was financially Supported by CONACYT and CPE, M6xico. trallsmission lines, and 11 transformers, is used for this purpose,
E. Acha and H. Ambriz-Pfrez are with the Virtual FACTS Research I.abo-
ratow Dcnarlmeol of Electronics and Electrical Eneineerine.
I I ". thc Univcrsilv of
It is 'Iso shown that the more general transformer
Glasgow, Scotland, UK. model is amcnahle to easy interfacing with static var compen-
C. R. Fuertc-Esquivelis with the Virtoal FACTS Research Laboratory, Depar- SatOr(svc)models, leading to optilnal coordination strategies
tamento de Ingenieria EICctrica y Olectrhica, Institulo Tecnol6gica de Morelia,
MCxico. involving the transformer's tap-changing control and the SVC's
Publisher ltem Identifier S 0885-8950(00)01886-l. firing angle control.
o x ~ ~ - ~ ~ s o ~ uDo 2000
~ ~ oIEEE
.oo
ACHA et a,,: ADVANCED TRANSFORMBR CONTROL MODFI.ING IN AN OPIIMAL POWER BLOW USING NHWTON'S METHOD 291

branch. However, in the inductive path the relationship between


the current and the voltage is dictated by the RMS V-1 char&-
terislic, which under saturating conditions becomes nonlinear.
Rclationships in the ideal transformers 1, 2, and 3:

(3)
0 Thc currents across the impedances Zp,Zs,and Z t are:

while at the center of thc core, the following relationship holds:


0 = I\ + 1; + I ; - l o = 7 ' 1 l i + l J i l z + W113 - l o . (7)
Substituting (4)-(6) into (7) gives:
,"'l'
()-----
u;v,> w;vl
%, Z S zt

Schematic representation and equivalent circuit of


Fig. 1
tranSfurlllCL
G thnc-winding
It must be notcd that T ; , Ut, and Wt are scalar quantities
since they are the products of l'v'l';, TJV U:, , and Wv W ; ,
respectively.
11. ADVANCED
TRANSFORMER MODELING Putting (4)-(6) and (8) in matrix form is shown in (9) at the
A. The Three-Winding Trumformcr bottom of the next page. Equation (9) represents the transformer
shown in Fig. 1. However, it is possible to find a reduced equiv-
Based on the equivalent circuit of the three-winding trans-
alent matrix that still models the transformer correctly while re-
former shown inFig. l(a), a generalized nodal admittance model
taining only the external nodes p , s, and 1. This is done by means
can be derived for the three-winding transformer. Each winding
of Gaussian elimination a s shown in (IO) at the bottom of the
is represented as the series combination of a resistance, leakage
next page where
reactance, and ideal transformer. Furthermore, each winding is
provided with a complex tap changing mechanism to allow for A = T:,Z,Zt + +
U:,ZpZ:t W $ % , Z , %,Z,ZtYo. +
phase-shifting facilities. Thc magnetizing branch of the trans-
former is included to account for the iron core losses.
The ideiil transformer in the primary winding has complex tap B. The Two-Winding Transformer Model
ratios Tv : 1 and T I : 1in series with an impedance Z, where The nodal admittancc representation of a two-winding trans-
l'v = T +
Y = t jcy = 'TLyl. The ideal transformer in the sec- former can be easily obtained by introducing simplifying as-
ondary winding has complex tap ratios IJv : 1 and U , : 1 in se- sumptions in (IO). For instance, when the tertiary winding, I ,
ries with an impedance Z,, where IJV = 0; = u+jP = U L p , does not cxist, the row and column corresponding to node t
Similarly, the ideal transformer in the tertiary winding has com- become redundant and they are removed from ( I 0). Moreover,
plex tap ratios W v : 1 and !NI : 1 in series with an impedance the tap ratios WV and Wr become zero and the impedance 8 1
+
Z1,where WV = W*, = w jy = W L $ . The magne- takes an infinite value. Hence, applying the CHospital differen-
tizing branch of the transformer is represented by the admittance tiation rule, the nodal admittance matrix equation representing
+
Y o = Go jl3,. In the paper, bold type represent complex the two-winding transformer is arrived at:
quantities.
The resistive path of the magnetizing branch is directly re-
lated to the iron losses and its conductance G o draws a cur-
rent that varies linearly with the voltage ~icrossthe magnetizing
292 IEEE TI<ANSACTIONSON POWEK SYSTEMS. VOL. IS. NO. I , FRBRUARY 2000

where
A = TZZ. + U $ Z P + ZpZ.Yo.
It must be noticed that owing to the flexibility of the two-
winding transformer model in ( l l ) , it is possible to assemble
a transformer model that represents the transformer equivalent
circuit shown in Fig. l(b) by using three of these two-winding
transformer models. An example of how this can be achieved is
given below, where suitable parameter selection is used for each
one of the two-winding transformer models involved.
.
I -
Fig. 2. Equivalent circuit of a LTC transformcr.

+
TranLformer I-Uv, 1 = UI, 1 = 1 j 0 , Z., 1 = 0,und We arrive at a nodal admittance matrix representation that
Y o . 1 = Yo: is fully equivalent to the nodal admittance model of the
three-winding transformer as shown in (15) at the bottom of
the next page.

($1
Transformer 2-ll;, 2 = U,, 2 = 1 + j 0 , Z s ,2 = 0, and
C. The Clussical Transformer Model
In the open literature it is common practice to represent the
LTC transformer as an impedance connected in series with an
Y0,Z = 0: ideal autotransformer of tap ratio 1 : 1. Fig. 2 shows the equiv-
alent circuit of an LTC connected between nodes p and 8 . The
transformer tap-changing capabilities are taken to be on the tap
connected to node p [1]-[.5].
The classical LTC model can be obtained by introducing
Transformer 3-lJv, 3 = U,, 3 = 1 + j o . Z a ,3 = 0, and
simplifying assumptions i n the two-winding transformer model
Y o , 3 = 0:
+
given by (11). If T ; = T I = t j 0 , 1Jv = U I = 1 j 0 , +
Y O= 0, Z p = 0, and R, = Z, a nodal transfer admittance
model for the LTC shown in Fig. 2 can be obtained

Furthermore, by selecting the following topology constraints:


ACHA cf al.: AIIVANCBO TRANSPORMER CONTROI. MODELING IN AN OPTIMA[. POWER FLOW USING NI1WTONS METHOD 291

Phase-shifting transformer niodels similar to those available A. Power Equations


in open literature [1]-[5] can also be obtained from (11) Based on ( l l ) , the following active and reactive power
by suitable parameter selection, i.e., ITV = 7'; = I L d, = equations can he written for a two-winding transformer
cos d, + j sin 4,//v = U ; = ILOo,Y o = 0,8, = 0, and connected between node m and node n :
Z , = Z . The nodal transfer admittance modcl is:
+
P,,&= ViGnm V,V, (c,w.171 cos(4,, - 4,)
I I -(cos d, + j sin 4) + O m , siIl(O,,A - B,)) (18)
I QM = - Kiflnzm+ L V n (G,,, sin(8,,, -8
,)
-&m COs(Om - Of%)) (1%
= K;(LV,K + ( ~ , , , , ~ ~ ~ ( n 8,)
- ,,
+U,,,,, ~ i i i ( 8 ~4
- ,)~ (20)
TllANSFORMBR MODELSFOR OFF NEWTON'S
111. ADVANCED
METHOD Qn = - V:Bnn + V,V, (G,,,, sin(O,h - 0,)

The transformcr models presented above have been incor- COS(O,, -on&)) (21)
porated in an OPF Newton method. Owin& to the very large where p,, , p,, Q m , Q7& are the active and reactive power flow
number of variables involved in the three- and two-winding injections at buses and 1 1 , v,,
v, B,,, H are, ~the
transformer models, their implementation requires consider- magnitudes and phase angles at buses
and n , respectively,
able analytical and programming work, i.e., the three- and
two-winding models have 144 and 48 first and second order B. LuRmnRjan ~. Funcrion
derivative terms, respectively.
The two-winding transformer state variables are combined
After careful consideration, keeping maximum model flex-
with the network nodal voltage magnitudes and phase angles
ibility and miniinum software complexity as our overriding
in a single frame of reference for a unified optimal solution via
coiicerns, it was decided to cast in software the two-winding
Newton's method. The transformers state variables are adjusted
transformer model, ( I I ) , as opposed to thc Lhrce-winding trans-
automatically so as to satisfy specified power flows, voltage
former model, (9) or (IO), or the classical LTC model, (16). In
magnitudes, and Kuhii and Tucker optimality conditions 1121.
this way, modcl functionality is kept since the three-winding
The first stcp in finding the optimal solution is to build a
transformer can be assembled, at the data network building
Lagrangian function for the power flow mismatch equations
stagc, by a suitable combination of three two-winding trans-
at nodes m and 11. They are explicitly included in the OPF
former models. It should also be mentioned that the flexibility
Newton's inethod as equality constraints given by the following
gained by upgrading from the classical LTC to the two-winding
equation:
LTC, does not really increase execution time by any significant
margin when compared to the overall solntion. For instance, Lrnn(z,A)
in cases when the two-winding model is used to represent a
classical LTC, i.e., one-winding, the only overheads incurred
= + pum I'Gm) + A Q m ( Q m Q U m - Q G m ) +
will be when evaluating nodal admittance matrix ( I I ) a s
opposed to (16). The actual OPF solution is not affected
+ +
+ AprA(1\ 1'11~i - pen) AQ,(Q, Q D , ~- Qun) +
(22)
since the equivalent transformer is still a two-node branch
representation, unnecessary derivative terms are not evaluated where I'n,, Po,, Qnm,Q D , ~are active and reactive power
and sparsity techniques are uscd throughout the iterative loads at nodcs i n and n, respectively. PG,,,, P,,,, Qo,, QG,,
solution. are scheduled active and reactive power generations at nodes
294 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. IS,NO. I,FEBRUARY 2000

m and n,, respectively. A p m , Ay, AP,~, A Q ~ are Lagrange where


multipliers at nodes n% and n, respectively. The vector of state
variables is 3: = [V O]', where V and U are the voltage
magnitudcs and angles, respectively. The superscript t indicates
transposition.
The two-winding transformer modcl may be used to study w,m =
voltage magnitude regulation at either the sending or the re-
ceiving node of the LTC transformer. In order to achieve this,
it is only necessary to add the second derivative term of a large,
quadratic penalty term at the appropriate location in the lin-
earized system of equations. The first derivative terms of the
quadratic penally functions are evaluated and addcd to the cor-
responding gradient element [SI. This point is discussed further
in Section 111-C below.
Alternatively, the two-winding transformer model may be
used as a phase-shifting transformcr to regulate the active
power flowing froin node nL to node n. In the OPF formulation
this condition is expressed as an equality constraint which re-
mains activc throughout the iterative process unless one wishes
this constraint to be deactivated. Thus operating condition is
represented by the following Lagrange equation:

L ( r , A) = L " p m n , - I:peclfied) (23)


I 32L
where, ,A, is the Lagrange multiplier associated with the active
power flowing from node rib to node n. Y&ifird is the required
value of active power flow across the phase shifter.

C. Linearized System of Equafiona


Thc lincarized system of equations for minimizing the La-
grangian function via Newton's method is given by [SI,1121:

WAz = -9 (24)

where matrix W contains second partial derivatives of the La-


grangian function L ( r , A) with respect to the state variables z
andLagrangemultipliers A. The gradient vectorg is [Vz VA]',
which consists of first partial derivative tcrms. Az is the vector
of correction terms, given by [Az AA]. Also, Az = z(') -
z(~-'),where R is an iteration counter.
The representation o€LTC and phase-shifting transformers in
thc OPF algorithm requires the matrix W to be augmented by
one row and one column per transformer. Under this condition
either T or U and 4 or p arc extra state variables which be-
come incorporated in the OPF formulation. Furthermore, if the
phase shifting transformer is controlling active power flow then
the dimensions of matrix W are increased further, one row and
one coliinin is required for each phase shifter that is controlling
power flow.
An expanded version of (24) corresponding to the case when
the two-winding LTC transformer has tapping facilities only in
the primary winding, i.e., T : 1, is:

Wmm wmn
W,,, W,,,,
W,/.m Wnl
WmT
W.T]
W7,r
[ A%
Aztz] = -
AZT
gna
[;;] (25)
ACHA d i 01,: ADVANCBD TRANSWIIMER CONTROL MODBIlNG IN AN OPTIMAL POWER PLOW USING NBWTONS METHOD 295

and Once (24) has been assembled and combined with matrix W
and gradient vector g of the entire network then a sparsity-ori-
= A A Q m It
A k AAh, (36) ented solution is carried out. This process is repeated until a
small, prespecified tolerance is reached. It must be noticed that
= [As, av,,AA,,, ax, 11 (37) thisprocedurecorrespondstocasesinwhichtheLTCandphase-
shifting transformers are operated in standard control mode. The
LTC transformer is set to control the nodal voltage magnitude
AHT= [A/']. (38) at either node m or node n, whereas the phase-shifting trans-
former is set to control the active power flowing from node m
If the tapping facilities of the two-winding LTC transformer
to node 11,
are on the secondary winding rather than the primary winding, In OPF solutions, variables are normally adjusted
i.e., rJ : 1, then the tapping variable U will replace the vari- aLltomatically during the solution process in order to reach the
able T in (30)-(32), (351,and (38). The other equations remain
best operating point of the electricalpower system,
unchanged. if the phase-shifting transformer is not set to control a fixed
For the case when the two-winding transformer has phase- amount of active power flowing from node ni to node n then
shifting facilities only in the primary winding, i.e., l L 4 : I, an matrix is suitably modified to reflect this operating condi-
expanded version of (24) is: tion. This is done bv addine the second derivative term of a large
I ~

(infinite), quadratic penalty factor to the diagonal elements cor-


responding to multipliers A,, . The first derivative term of the
quadratic penalty function is added to the corresponding gra-
dient elements.
Also, if the LTC transformer is not controlling the voltage
magnitude, the quadratic penalty factor is not used. The voltage
magnitude is updated throughout the iterative process as given
by the solution of (24) until it reaches a prespecified tolerance.

D. The Inequality Constraints Set


In the OPF formulation, voltage magnitude and active power
limits are included in the inequality constraints set The mul-
tiplier method [13] is used to handle this set. Here, a penalty
term is added to the Lagrangian function L ( z , A), which then
becomes the augmented Lagrangian function. Variables inside
bounds are ignored while binding inequality constraints become
part of the augmented Lagrangian function and, hence, become
enforced.
The handling of LTC and phase-shifting transformer in-
equality constraints, which enter the OPF solution in the form
of extra state variables, can be carried out by means of the
following generic function:

@i(dZ), Pi)

(43)

AH^ = [ A 4 AA,, 1" (44)

If the phase-shifting facilities of the two-winding transformer where


are on the secondary winding rather than the primary side, i.e., si(.) is the state variable value, i.e., tap ratios T or U and
l & : I, then the tapping variable 'p will replace the variable phase-shifting angle 4 or 'p.
-
4 in (40)-(44). Also, it must he remarked that (26)-(29), (33), gi are the maximum and minimum limits of the extra state
(34), (36), and (37) are valid for both LTC and phase-shifting and variables, i.e., upper and lower tap ratio ceilings and
transformers.' phase-shifting angle ceilings.
The derivative terms corresponding to inequality constraints p; is a multiplier term and
are not required at the beginning of the iterative solution, they (1 is a penalty term that adapts itself in order to force
are introduced into matrix equation (24) only after limits are inequality constraints within limits while minimizing
enforced. the objective function.
296 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. IS, NO. 1, FEBRUARY 2000

Cwe Active Reactive Active Power Active


Power Power Oeneratian Power
Generation Generalion cost Lasses

4 1.09
1.08

Pig. 3. Modified AEP 1Cnodcs tcst nctwork.


1.05

E. Initial Conditions >o 1.03

Similarly to the load flow problem, good initial conditions are


1.01
mandatory in OPF solutions. Extensive use of the models pre- 7 8 1410 9 131211 6 4 3 5 2 1
sentcd in the paper indicates that the following initial conditions Nodes
are amenable to very robust iterative solutions:
a) Voltage magnitudes and angles of 1 and 0, respectively, Fig,4. Nodal voltage profilcs Cor three diffcrcnt scenarios.
are selected for all nodes.
b) Tapping and phase-shifting positions of I and 0, rcspec- Tapping ftacilities exist at Nod-6 and a SVC is connected to
tively, are selected for all the transformers. Nod-7. The parameters of both transformers are identical and
c) A lossless economic dispatch that includes gencrators' the windings contain no resistance.
limits is used as the initial condition for the active powcr In order to illustrate the impact of LTC's on system opera-
schedule. tion, three different scenarios are presented below. The first one
d) Activc and reactive powers Lagrangc multipliers of 1 and assumes that the three taps in both transformers remain fixed at
0, respectively, are selected for all the transformer nodes. their nominal position, e.g., T , = IJ, = W, = 1 L 0'. The
e) Active powcr flows Lagrange multipliers of 0 are selected second sccnario assumes that the tap changer in LTCl is ac-
for all the phase-shifting transformers. tive and the tap changer in LTC2 remains fixed at its nominal
value. The third scenario considers the case when both LTC's
IV. TESTCASES are active.
The number of iterations taken by the OPF solution to con-
The two-winding transformer model, given in ( I 1). was cast verge to the specified tolerance, i.e., for all the variables,
in software in an OPF algorithm developed under the OOP was 5 iterations for the first two cases and 7 iterations for the
philosophy. The extended algorithm has been applied to the third case. Table I shows the active and reactive power gener-
solution of a large number of power networks of different
ations, costs and losses for the three scenarios. Improved s o h -
sizes. Some of these networks can been found in the open
tions are obtained for the last two cases compared to the first
literature 1141. Our optimal solutions agree very well with the case, owing to the influence of the tap changers. For instance,
results given in thosc refercnces. in the last case, the generation cost was reduced by 0.416 $/hr
In this section, we present OPF results which relate to the and the power losses were reduced by 3.8%,with respect to the
standard, 14-nodes test network and to a real life power net-
base case.
work, consisting of 2172 nodes, 2294 transmission lines, and
Fig. 4 shows the nodal voltage profiles for the three cases.
768 transformers [I 11. However, for the purpose of this paper Significant diflerences can be observed between these profiles.
only 166-nodes are explicitly modeled, with the rest of the net-
The system voltage profile improves slightly when the first LTC
work represented as an external equivalent.
is in operation. However, when two LTC's are in operation, the
system voltage psofile gets much improved, and the overall ef-
A. Small Test Network fcct is an important reduction in active power losses, as well as
The standard AEP 14-nodes system has been modified to a reduction in reactive power generation. It must be noted that
include two three-winding transformers in the locations shown large voltage variations cxist at the transformer nodes.
in Fig. 3. The first transformer is connected to nodes Nod-4, Table 11 gives examples of voltage control interactions be-
Nod-8, and Nod-9. Tapping fac es exist at Nod-8 and a tween the transformer tapping facilities and two kinds of VAR
synchronous condenser is attached to the node. The second compensating plant. When the voltage magnitude is set to 1 p.u.
uansformer is connected to nodes Nod-5, Nod-6, and Nod-7. at node Nod-7, reactive power is drawn from the system. The
ACHA er 01.: ADVANCBO TRANSFORMER CONTROL MOOELING IN AN OPTIMAL POWHI< Fl.OW USING NEWTONS METHOD 2%

TABLE I1 TABLE V
LTC'S AND VAR COMPENSATINGPI.ANT INTERACTIONS NETWORK WIT11 ELEVEN TAP-CIIANGINO
TRANSFORMERS

0.900 135.66 0.949 26.00


0.912 134.58 1.014
0.961 I 134.67 1.061 I 40.00

TABLE 111
CHARACTPRISTICS Or; THE TESTSYSTEMS. WHFRE APG AND RPG
GENERAL
ARE NDDES WITH ACTIVEAND REACTIVE POWER GENERATION, RESPECTIVELY System Information Magnetising Branch Representation

Network Nodes Total Controllable


None I Linear I Non-Linear
Total I APO I WO Branches Transformers
166Nodes 166 22 24 236 I1

TABLE 1V
NETWORK WIT11 NO CONTROLLABLE TRANSIORMERS voltage regulation capabilities. As expected, there is an in-
crease in the reactive power demanded from the system when
the magnetizing branch is included in the study. However,
the increases are smaller than those shown in Table 1V. The
reason is that in this case additional control variables become
available, which minimize reactive power losses by regulating
nodal voltage magnitudes in an optimal fashion.
transformer tap hits its lower tap position, i.e., 0.9, and the SVC 3J Phase-Shifting Transformers: Now the eleven trans-
operates in the inductive region, i.e., tiring angle is 135.66'. formers in the network are assumed to be two-winding,
phase-shifting transformers. Table VI presents results where
To maintain a voltage magnitude of 1 p.u. at Nod-8 requires
that the synchronous condenser supplies 26 MVAR and that the it can be observed that the phase shifters reduce active power
transformer tap is set at 0.949. If higher voltage magnitudes are generation costs, active power losses, and system reactive
required at these nodes, e.g., 1.05 and 1.1 p.u., then the SVC is power requirements, compared to the cases when no regulating
set to operate in the capacitive region. In fact, the synchronous transformers or tap changers are used.
condenser hits its upper reactive limit and the transformer tap It can be observed in cases I), 2), and 3) that the inclusion
goes above the nominal tap position, i.e., 1 p,u. of the magnetizing branch in the study does not affect the gen-
eration costs by much. However, the magnetizing branch is re-
B. Real-Life Utility Power Network sponsible for increasing the active power losses and the reiic-
tive power generated by the system; average increases of 1.4
Table 111 shows the general characteristics of the utility net-
and 2.3% are observed, respectively. As far as active power
work used in this paper [111.
losses are concerned, the linear and the nonlinear representa-
The eleven controllable transformers in the network are
tion of the magnetizing branch yield identical results. However,
three-winding transformers. Synchronous condensers are the nonlinear representation becomes important when assessing
connected to the tertiary windings of all the transformers to
the system reactive power generation.
provide additional reactive power contributions.
I ) Unregulated Transjormers: Table IV presents cilscs i i i
V. CONCLUSIONS
which the 11 transformers have no voltage regulating capabili-
ties. The transformers' tapping and phase-shifting capabilities The paper has introduced advanced transformer control
are on the primary windings. modeling facilities within the context of optimal power flow
Three different magnetizing branch representations are used solutions. Tlic newly upg,:ided OPF algorithm uses Newton's
below to assess their impact in OPF solutions. The magnetizing method to solve the linearized system of equations, leading
branch corresponds to a practical power transformer whose data to very robust iterative solutions for networks o f any size
is given in [15]. The generation costs are almost the same in all and with a large number of controllable and noncontrollable
three cases. However, some differences are observed in the ac- two- and three-winding transformers. Moreover, the mnlti-
tive power losses and in the reactive power generated by the pliers method is used to handle the LTC and phase-shi tmg
system. The incrrase is due Lo lhe tl-ansformcis' magnetizing transformer control variables, leading to highly flexible and
branches. Convergence was obtained in 5 iterations in every efficient, constrained OPF solutions. Since the algorithm is
case. a fully coupled Newton method, as opposed to a decoupled
2) R e p l u t e d Transformers: Table V shows similar results one, no convergence problems havc been observed when
to the previous case but now the eleven transformers have solving networks with branches containing high R/X ratios. Its
298 IEEE TRANSACTIONSON POWER SYSTEMS, VOL. 15, NO. I. FEBRUARY 2000

prowess has been illustrated by numeric examples. In general, 1121 D. G. Luenberger, Introduction to Linear and Nonlinear Progrummbig:
the solution of networks containing LTC's and phase shifter Addison Wesley Publishing Co., 1984.
1131 D. F! Bertsekas, Constmined Optimization LagrangeMelfiplier Method:
transformers is achieved in the same number of iterations as Academic Press, 1992.
networks with no controllable transformers i.e., 5-1 iterations. [I41 R. Divi and 1. A. Kesavan, "A shifted penalty function approach far
A new and generalized three-winding transformer model optimal ladflow," IEEE Tranr. on Power Apparatus and Sy.stcms, vol.
PAS-101,no. 9, pp. 3502-3512, Scpt. 1982.
with complex tap rations in each winding has been developed 1151 H. W. Dommel, A. Yan, and R. J. 0. De Marcana, ''Case studies for
from first principles. The model is shown to be a general case for electromagnetic transients," Internal Report, Department of Electrical
existing two-winding transformer models and the classic LTC Engineering,University of British Columbia, Vancouver,Canada, 1983.
and phase-shifting transformer models. The model includes
the effect of the magnetizing branch of the transformer, which
under saturating conditions becomes nonlinear. This allows
for more realistic electric energy studies, where the nonlinear E. Aehr graduated from the University of Michoacdn, Mexico. in 1979 and
obtained the Ph.D. degrcc from tho University of Canterbury, New Zealand, in
magnetizing branch is shown to be responsible for increasing 1988. He WBS a postdoctoral Fellow at the Universities of Toronto, Canada and
active power losses and reactive power system requirements Durham, England. He is a Senior Lecturer at the University of Glasgow, Scot-
by 1-2% and 2-3%, respectively. Also, the three-winding land, where he Icctwes and conducts research an power electronics applications
in power systems. He is Chairman of the interuniversity Glasgow-Strathclyde
LTC transformer model interfaces easily with SVC models, Centre far Economic Renewable Power Delivery and is also Head of the FACTS
leading to optimal coordination strategies between the two Research Laboratory.
plant components.

REFERENCES
[11 G. W. Stagg and A. H. El-Abiad, Computer Metliod,r in PowerSy.stems Hugo Ambrie-Perez received the B.Eog. and the M.Sc. degrees from the Insti-
Analysis: McGraw-Hill, 1968. tuto Politdcnico Nacional, MCxico, in 1987 and 1992, respectively. He worked
[2] 0. Elgerd, Electric Enerfy Systemr Theory-An Introduction: McGraw- in the planning dcpartmcnt of the Mexican National Utility for four years. Hc is
Hill, 1971. currently working toward the P h D degree at the University of Glasgow. Scat-
[3] N. M. Petersoli and W. Scott-Meyer, "Automatic adjustment of trans- land.
former iind phase shifter taps in the Newton power flow," IEEE Tronc
on Power Apparatus and Systems, vol. PAS-90, no. I, pp. 103-108,
Jan./Feb. 1971,
[41 C. M. Shcn and M. Laughton, "Determiitcionafoptimum power-system
operating conditions undcrcanstraints," Proc. ofIEE, vol. 116, no. 2, pp. C. R. Facrte-Esquivel received thc B.Eng. degree from the lnstituto Tcc-
225-239, Feb. 1969. nol6gico de Morelin, Mexico, in 1990, the MSc. degrcc from the lnstituto
[51 D. I. Sim, B. Ashley, B. Brewer, A. Hughes, and W. F. Tinney, "Optimal Politecnicu Nacional, Mkxico, in 1993 and the Ph.D. degree from the
power flow by Newton approach," IEEE Trans. on Power Apprrrubsnnd University of Glasgow, Scotland, in 1997. He is an Assistant Professor at
, PAS-lO3,ilo. 10, pp. 2864-2880, Oct. 1984.
S y ~ t e m svol. the Instituto Tecnol6gica de Morelia. His main research interests lit on the
161 Tchnicui Informolion for Efingineerx International Computers Limited. dynamic and steady-state analysis of FACTS and CUSTOM POWER, real-time
Leeds, UK, July 1993. electromegnetic transients.
171 C. R. Fuerte-Esquivel end E. Acha, "Newton-Raphsan algorithm for
the reliable solution of large power networks with embedded FACTS
devices," in IEE Prm-Gener, Tranrm.Disfrib.,vol. 143, Sept. 1996,
pp. 447454.
[XI -, "A Newton-type algorithm for the control of power flow in elec-
trical power ncworks," I€€€ Trans. on Power Syrtems, vol. 12, no. 4,
pp. 14741480, Nov. 1997.
[91 N. H. Hingormi, "Flexibly AC transmission systems," IEEE Spectrum,
pp. 40115, Apr. 1993.
1101 -, "Introducing CUSTOM POWER," IEEE Spectrum, pp. 4 1 4 8 ,
June 1995.
[II] E Aboytes imd G. Arroyo,"Security assessment in the operation of lon-
gitudinal power systems." IEEE Trans. on Power Systernr, vol. I,no. 2,
pp. 225-232, May 1986.

You might also like