IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, VOL. PAS-90, NO.

1, JANUARY/FEBRUARY 1971 103

Automatic Adjustment of Transformer and

Phase-Shifter Taps in the Newton Power Flow
NORRIS M. PETERSON, MEMBER, IEEE, AND W. SCOTT MEYER, MEMBER, IEEE

Abstract-In recent years, Newton's method has been generally t23

accepted as a superior algorithm for solving power flow problems.
But the introduction of generator var limits, area-interchange
control, load-tap-changing (LTC) transformers, and variable phase
shifting transformers has significantly increased the solution
times (compared with a fixed system). This paper shows how LTC
transformers and variable phase-shifting transformers may be
automatically adjusted by the Newton algorithm, thus complement-
ing a previous inclusion of the area interchange control variables
within the Newton algorithm [1]. Tests of the new method for
adjusting LTC transformers and phase shifters indicate a significant
overall saving in solution time.

Fig. 1. Sample system.

INTRODUCTION
FOR an N-node power system the nodal admittance power ing (LTC) transformers with tap settings t4 and t5, respectively;
flow equations can be formulated as these are to be adjusted to regulate E4 and E5 (the voltage
N magnitudes at nodes 4 and 5) at values E4reg and E5reg, re-
(Pk + jQA) = Ek E Em*
Ykm(1) spectively. 12s is a phase-shifting transformer with setting 4)23;
m=l it is to be adjusted to regulate the real power flow P28 in branch
where (2,3) to value p23reg
Present solution methods start by fixing t4, t5, and (P23 at
Pk + jQk injected power into node k reasonable values, and then solve the linearized power flow
Ezv voltage at node k equations (2a) for node-voltage corrections AEi and Ai..
Ykm negative of admittance between nodes k and m,
kXm AP1 H1l H14N14 E21
ykICi summation of all admittances terminating on AP2 H22N22 H232N2 H24N24 A62
node k lAE2
*
complex conjugate AQ2 J22L22 J23L23 J24L24
E2
= \-1. AP3 H32N32 H33A733 H35N35 A63
AE3
The well-established Newton power flow method involves J32L32 J33L33 J35L35 (2a)
repeated direct solution of a system of linear equations derived
from (1) for the node voltages Es. Early development and testing AP4 H41 H42N42 H44N44 H45N45 A584
were performed by Van Ness and Griffin [2], and their notation AE4
AQ4 J41 J42L42 J44L44 J4sL45
will be used throughout. With the development of optimally E4
ordered elimination methods by Sato and Tinney [3], the Newton AP5 H53N53 Hs4N54 H55N55 A35
power flow has become the generally superior power flow method Js3L53 Js4Ls4 J65L5s AE5
[4]. AQ5
To review the basic ideas and notation, consider the sample
system of Fig. 1. Node 1 is a generation node where real power
and voltage magnitude are constrained; nodes 2-5 are load H1m = OPk N1m
Nk7n = Em
nodes, and node 6 is the swing (reference) node having known
Aih5 51Em
voltage magnitude and angle. T4 and T5 denote load-tap-chang- d Qk aQk E
Jkm = -1dam L Qm
= (2b)
I
dOEm m
Paper 70 TP 160-PWR, recommended and approved by the Power
System Engineering Committee of the IEEE Power Group for
presentation at the IEEE Winter Power Meeting, New York, N. Y., Formulas for elements of (2), the Jacobian matrix, are given
January 25-30, 1970. Manuscript submitted July 29, 1969; made in Appendix I. Using present methods, problems containing
available for printing November 21, 1969.
N. M. Peterson was with the System Planning Department, LTC transformers and/or variable phase shifting transformers
Northern States Power Company, Minneapolis, Minn. He is now would be solved as follows. After each solution for AEi and i6s,
with Systems Control, Inc. Palo Alto, Calif. 94306. tap settings t4, t5, and 4)23 would be adjusted by a displacement
W. S. Meyer is with the University of Minnesota, Minneapolis,
Minn. 55455. technique (reminiscent of Gauss-Seidel methods) to better meet
104 IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, JANUARY/FEBRUARY 1971

the desired operating conditions: zero when no branch (k,m) exists for the network, but 2) C6km
and Dkm are also zero when branch (k,m) does not represent
t,new = told + c(Ereg - 3= 4,5 the LTC transformer that is regulating node m. Each LTC-
.2,new = o23°ld + d23 (P23reg - P23). (3) regulated-node column in (4a) thus usually has only one nonzero
off-diagonal C6km and/or Dkm term.
Acceleration factors C4, c5, and d23 are empirically determined, The nonzero block pattern of the Jacobian matrix in (4a) is
often resulting in slow and/or unpredictable overall convergence like that of (2a), so elimination may proceed in the usual order
for the problem. (based on network topology) [3], [4]. Appendix II details
formulas for the C6km and Dkm elements, and it is seen that only
AUTOMATIC ADJUSTMENTS minor changes from the elements in (2a) exist. As a result,
Continuing with the sample system, t4, ts, and 023 can be programming modifications are minor, and the changing of
"automatically" adjusted if included as variables, rather than LTCs between fixed and variable types when LTC limits are
as constants, in the constraint equations. The two variable types encountered during the course of the solution is easily done.
require two procedures: automatic LTC tap changing and Automatic Phase-Shifter Adjustments
automatic phase-shifter adjustments.
Phase shifter b23 in Fig. 1 is typically operated to constrain
Automatic LTC P23 at a value p23reg. But since P23 is not a regular problem vari-
An LTC such as T4 is operated so that t4 is adjusted (within able, a variable switch as was done for LTCs is not possible.
limits) to constrain E4 at value E4reg For this mode of operation 423 enters as an extra problem variable, and the linearized
E4 is a constant and t4 should replace it in (2a) as a variable. relation AP23 = P23reg - P23 is an additional constraint equation.
Hence, any LTC-regulated node i has its voltage E? replaced The resulting linearized Newton equations appear in (5a).
as a variable by ti. Incorporating these changes, (2a) is modified Note that both LTCs and the phase shifter will now be auto-
to become (4a): matically adjusted.
i- * 1
AP1 T
Hill
;r
H140 Aa1
AP2 H22N22 H23'N23' H24C24 E2,23 A32
AE2
AQ2 J22L22 J23'L23' J24D24 F2,23 aE2
AP3 H32'N32' H3aN33 H35C35 Es,23
AEs
AQ3 J32'LS2' J33Ls3 Fs3,23 E3
AP4 H41 H42AN42 H450 (5a)
A34
At4
AQ4 J41 J42L42 J450 t4
AP5 H53N53 H540 H5C655 6A5
At5
AQ5 Js3Ls3 J540 J55Ds5 t5
_ AP23 - H23,2'N23,2' H23,31N23,3I _LA\423 -

AP1 1H1l H140

AP2 H22N22 H23N23 H24C24 A32 Ek,km Pkm
AE2
AQ2 J22L22 J23L23 J24D24
'Fk,kjm (5b)
As2 b,,m
4Q
a
=-
AP3 H32N32 H3sN33 H35C35 m9k
AQ3 J32L32 J33L3s J3sD3s
AEs Terms in the Jacobian matrix that are associated with the
(4a)
E3 phase shifter are distinguished by primes and as Ek, and Fk, km
AP4 H41 H44C44 H450 A84 in (5a). Detailed formulas for their calculation are provided in
AQ4 J41 J44D44 J450
At4 Appendix III.
t4 Test Results
AP5 H58N53 H140 H1ssCs Initial tests of the automatic LTC adjustment scheme in-
At5
AQ5 J53L53 J540 J55D55 dicate typically a 2 to 1 speed advantage over the old (present)
t5 method, if generator var limits are not imposed. Fig. 2 shows
convergence behavior for different solution techniques applied to
Ckm = tm a 347-node 572-branch system with 91 generator nodes and 8
- tm
LTC transformers. All generators were assumed to have infinite
var limits, and no area interchange control was applied. The same
D -m= {m (4b) problem was solved in three different ways:
atm
1) as a "fixed system," with all transformer tap settings
The terms Ctm and Dkm replace the usual terms Ntm and permanently fixed (no LTC transformers);
Lkm, respectively, if node m is regulated by LTC transformer 2) by the old method, using the present NSP Newton power
Tm whose tap is tm. Because the variable t, is not a node variable flow program; the 8 LTC transformer taps were adjusted using
in the same sense as Em 1) C7km and Dtm are not only always (3) (tap adjustments started after the third iteration);
PETERSON AND MEYER: AUTOMATIC ADJUSTMENT OF TRANSFORMER AND PHASE-SHIFTER TAPS 105

regulated line flows, it was observed that convergence behavior is

103- I W II similar to that for the corresponding fixed system. Sotutions are
found in four or five iterations from a flat start.

SUMMARY AND CONCLUSIONS

102 A new method has been presented for the automatic adjust-
ment of LTC and phase-shifting transformers within Newton's
method. Experimentally observed convergence rates appear
1. - Fixed system with similar to those for fixed systems (requiring about five iterations),
no LTM C transformers. thereby representing typically a 2-to-i improvement over exist-
Or-
'-I^ ing displacement procedures. Such improvement is based on
IV ignoring generator var limits-the inclusion of which would
ZW2-Iir further complicate any comparison in unpredictable, system-
Old m
w
8 LTC transformers. dependent ways-but it is believed that the new method offers
substantial improvement for many if not most power flow prob-
0
lems. It is especially practical because its incorporation into
I-a- 100~
i________.__ New method for system with existing Newton power flow programs requires (generally) only
z 8 LTC transformers. minor modifications.
w
n
0.
z
x
\\ \
APPENDIX I
FORMULATION OF CONVENTIONAL JACOBIAN TERMS
0
For an electrical network in the steady state, phasor node
w voltages, injected currents, and admittances are complex values
-IC
w
cc written respectively as
i \' s~~~OLD METHOD
-j
0 -2 '0~~NEWI METHODi-I\ Em em + jfm
=
Cl)
or
4t Im = am + jbm
IL FIXEDH | >
0 SYSTEM j Ykm = Gkm + jBkm. (6)
z
0 The injected node-power equation is
!4 N
Co (Pk + Q) Ek E Em*Ykm*.
= (7)
m=l
Elements of the Jacobian matrix (2a) are, for k # m
am + jbm- (em + jfm)(Gkm + jBkm)
2^,^ 4 15 6 8 9;0Alftl11I 2 15
ITERATIONS
Hkm = Lkm = amfk- bmek
Fig. 2. Solution plot.for 347-node system. Nkm = -Jkm = amek + bmfk (8)
and the diagonal elements are
3) by the new method, with automatic adjustment of the 8 Hkk = -Qk -
BkkEk2
LTC transformer taps as in (4a).
An examination of Fig. 2 shows that convergence using the Lkk Qk= -
BkkEk2
new LTC adjustment scheme is virtually identical to the well- k= Pk + GkkEk2
known quadratic convergence for the fixed system. This is
contrasted with the much slower, linear conivergence of the old Jkk = Pk - GkkEk2 (9)
method (curve 2). The exactness of the new method is also worth
emphasizing; LTC-regulated nodes have their voltage main- APPENDIX
tained at "exactly" the desired values.- This is in contrast to the II

old method (curve 3), where taps are continually changed after TERMS ASSOCIATED WITH LTC TRANSFORMERS
each Newton iteration until all changes in (3) are less than 10-3
pu; then taps are permanently fixed (iteration 12 of Fig. 2), and a If node m is regulated by LTC transformer Tm (with tap tin),
final solution is obtained after an additional iteration (iteration the conventions shown in Fig. 3 can be assumed. Tm affects the
13). Although of course taps on the actual transformers are dis- self- and mutual admittances of nodes k and m by
crete, at least one is now able to "exactly" solve the problem
being formulated.
Ykm' = (Gkm0 + j1Bkm') = -Y

As of June 1969, the testing of automatic phase-shifter ad- Ykm = Ymk = tmYkm0
justments by the new method has been confined to small test
systems only. By inserting four phase shifters into the 20-node Ymm =
YmmI + t?m2y

NSP test system and finding solutions for different values of Ykk = YkkI+ yO. (10)
106 IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, JANUARY/FEBRUARY 1971

.Okm

k 1- -->
m

km km
Fig. 3. Representation of ETC transformers. Fig. 4. Variable phase-shiftinig transformer representation.

The power flow equations for no(les k an-d m are 2) The off-diagonal terms in the row of the Pk1reg constraint
N equation are
(Pk + iQk) = Ek2Ykk* + EkEm*tmYkkm * + Es Ei*yki* Hkmm = - Hkm,k = Hkm
i/k
Nkm,m = Nkm.
(Pm + iQm) = Em2(tm2yO * + ymm' *) N1m, k = Nkm + 2Ek2Y COS (15)
07am
N 3) The terms in the column of the Pk1reg constraint equation
+ EmEk*tmYkmO * + Em E Ei*Ymi*. (11) are calculated using (5b). The off-diagonal terms are
t=1
am + jbm = (em + jfm)(Gkm + jBkm)
Differentiating (11) as in (4b), the off-diagonal terms where Ek,km -amfk + bmek= Hkm
= -

ki # m are
Fk,km = +amek + bmfik = Jkm
am0 + jbmO = (em + ifm) (Gk1f + jBkm') ak + jbk = (ek + jfk)(Gmk + jBmk)
Ckm = (amIek + bm0fk)tm = Nkm Em,km = +alkfm - bkem = Hmk
Dkm =
(am0fk -
bm0ek)tm =
Lkm (12) Fm,km = -akem - bkfm =Jmk (16)
and the diagonal terms are and the diagonal term is
Cmm = -2Gkm'Em2tm2 + (ak0em + bkofm)tm Ekm,1km= Ek,km. (17)
Dmm = 2B1m0Em2tm2 + (aOfm - bklem)tm. (13) REFERENCES
[1] J. P. Britton, "Improved area-interchange control for Newton's
method load flows," IEEE Trans. Power App. Syst., vol. PAS-
88, pp. 1577-1581, October 1969.
APPENDIX III [2] J. E. Van Ness and J. H. Griffin, "Elimination methods for
load flow studies," AIEE Trans. (Power App. Syst.), vol. 80,
TERMS ASSOCIATED WITH VARIABLE PHASE-SHIFTING pp. 299-304, June 1961.
[3] N. Sato and W. F. Tinney, "Techniques of exploiting the
TRANSFORMER sparsity of the network admittance matrix," IEEE Trans.
Power App. Syst., vol. 82, pp. 944-950, December 1963.
If nodes k and m are the terminal of the km phase-shifting [4] W. F. Tinney and C. E. Hart, "Power flow solution by Newton's
method," IEEE Trans. Power App. Syst., vol. PAS-86, pp. 1449-
transformer and if the phase shifter is on the k side of the Y 1460, November 1967.
admittance, the phase-shifting transformer can be represented
as shown in Fig. 4.
The admittance relationships are
y = Y(Cos 0klm + j sin Okm)
Ykm = Gkm + jB&-m
Discussion
-Y [COS (0k.m - 'k.) + ij ll (0km -
Okm)]
Ymk = Gmk + jBmk
(Itndian Institute of Science, Bangalore, India):
D. K. Subramanian
- -Y[cos (0m1k + Pkim) + i in (Omk + 'km)] iinproved Newton's method of incorporating
The authors have
additional autom.atic facilities that had been hampering the conver-
gence. This will aid in the -ise of these modified methods for on-line
Ykk = Ykk + " load flows. The authors' opinions are sought on the following poin-ts.
Whein the and powers corrected by tap changers
Ymm = Ymm +- Y (14) and1) phase shiftersvoltages
are not fixed at particular values but are kept
within specified limits, can this method be suitably modified to
The calcutlationi of tlhe termls of the Jacobian matrix (5a) incorporate the changes? Also, can miniinization methods minimizing
that are affected by the presenice of a plhase-shifting transformer losses be used in conjunction with these methods?
can be divided into three categories. 2) From Fig. 2 it is fouind that both old and new methods con-
1) The terms in the anid m rows of the Jacobian
k matrix verge with nearly the saimie speed up to an accuracy of 0.1. So for
shown as primes in (5a) are calculated by (8) using the admit-
tances given in (14). Manuscript received February 9, 1970.
PETERSON AND MEYER: AUTOMATIC ADJUSTMENT OF TRANSFORMER AND PHASE-SHIFTER TAPS 107

a. system with ranges specified for controlled voltages and powers, -I E

if the old method is used up to an accuracy of 0.1 and then the new
method is resorted to, will the computations be reduced? > >
3) In cases where phase shifters are employed on a line with the
voltages of the buses connected to that line kept at fixed values, 0 0:
uL U.a LL
00
JLt:
can phase shifters then be eliminated completely from the system
sr

equations? 0 0 0 0

I thank the authors for making available a copy of the paper. tL L L

CONTROL EQUATION

ROW FOR PNET k SHADED AREA

ROW FOR 0NETk IS STRUCTURE
OF JACOBIAN
ROW FOR PNET m WITHOUT CONTROL
EQUATION
ROW FOR QNET m
W. F. Tinney and W. L. Powell (Bonneville Power Administration,
Portland, Oreg.): This important paper describes an improved
ROW FOR Q N ET m
wz//////// ///A u U )
scheme for the control of transformers and phase shifters in Newton's Z ADDED NON-ZERO TERM
method of power flow solution. A similar scheme has been imple- E1 TERM AFFECTED BY ADDED TERMS
mented in the Bonneville Power Administration's new 2000-node
power flow program. This new scheme, which reduces the overall Fig. 5. Structure of Jacobian matrix with one control equation.
solution time for average problems by about 50-percent, further
establishes Newton's method as the definitive algorithm for power
flow solution.
The paper shows the modifications of the Jacobian matrix for the % CONTROL EQUATIONS
transformer control scheme. A practical implementation must also
consider tap limits, simultaneous control of a bus by both a trans-
former and a reactive source, and the orientation of a transformer's JACOBIAN MATRIX
WITHOUT CONTROL
variable-tap side with respect to the controlled quantity. EQUATIONS
Our experience indicates that the handling of tap limits deserves
careful attention. The derivation of the Jacobian matrix does not
take tap limits into account. As a result, the Newton adjustment may Fig. 6. Jacobian matrix with several control equations.
cause transformer taps to go beyond their limits. If these excessive
adjustments are curtailed, the compensating nature of the remaining
dependent variable adjustments is destroyed. The aggregate effect has a nonzero diagonal, its elimination can take place anywhere in
of such curtailments may induce greater error in the system than the ordering that is advantageous from the standpoint of program-
existed prior to the adjustment. This adverse effect may be partially ming convenience or efficiency.
alleviated during the back substitution by replacing the computed The programming for single-criterion control equations that
adjustments with smaller truncated adjustments and allowing their augment the original system and have nonzero diagonal terms (e.g.,
effects to be propagated to the remaining adjustments. Limiting the a phase shifter controlling the power flow through itself) can be
size of the error to be corrected in one Newton adjustment cycle is accomplished with very little disturbance of the basic algorithm.
another way to prevent tap ratios from exceeding their limits by One control equation of this type is shown added to the original
large amounts. system in Fig. 5. Note that its row and column have symmetrically
The control of a bus voltage by the simultaneous adjustment of a located nonzero diagonal terms only in positions corresponding to
transformer and a reactive source is an actual operating situation regular problem variables associated withsince nodes k and m. The
that can cause difficulty in the power flow solution. Such controls significance of this structure is twofold: 1) the added equation
operate under priorities: the reactive source is used first, then the has a nonzero diagonal term its elimination can always be performed
transformer tap adjustment. In order to circumvent difficulties without pivot search; 2) only those elements with a dot are directly
in this situation it has been found expedient to introduce a tem- affected by the elimination of the control equation. Since these
porary constraint on reactive flow through the transformer (similar dotted elements are nonzero from the beginning, no new entries are
to the phase-shifter control of real power through itself). This causes created when the control equation is first in the order of elimination.
the system to first utilize the reactive sources most effectively. These properties are true for any number of control equations of this
When a reactive source reaches its limit, the transformer reactive type. Within the system of control equations, no elimination process
flow constraint is removed and direct transformer control is as- is necessary since the relevant submatrix in the Jacobian matrix is
sumed. diagonal, as shown in Fig. 6. The matrix need not be actually aug-
The transformer and phase-shifter adjustments are examples of mented. Upon reaching the rows for PNET k, QNET k, PNET m, QNET m.
single-criterion controls. A control variable is adjusted to main- the dotted elements are modified to eliminate the control variable.
tain another functionally dependent variable at a specified value. The correction for this variable is found after the completed back
Examples other than those used thus far can be envisioned and may substitution from the control equation.
be of potential value. The equations for the direct control of such
adjustments through the Newton algorithm can be derived quite
readily, but their implementation can become exceedingly difficult.
Implementation strategies depend upon whether the controlled
variable is a regular system variable or a new variable and whether Slobodan T. Despotovic (Electrotechnical Research Institute Nikola
the control equation has a nonzero diagonal term or not. If the Tesla, Belgrade, Yugoslavia): The introduction into of LTC trans-
controlled variable is a system variable and the control equation formers and variable phase shifting transformers any current
has a nonzero diagonal, then the control equation can be substituted method of solving power flow problems significantly increases the
for the normal equation. If its diagonal is zero, the equation can be solution time. This paper gives a new and very simple method for the
substituted but its order of elimination may have to be changed from automatic adjustment of such devices within the Newton power
that of the equation it replaces and the strategy for its elimination flow algorithm and indicates that significant overall saving in solu-
can become very complicated. This would be the situation for a tion time is thereby possible.
transformer controlling a remote node voltage. If the controlled The experimentally obtained results given in Fig. 2 show that
variable is not a regular system variable, the original system must be convergence is almost identical to that for the fixed system having no
augmented by the control equation. If the augmented equation LTC transformers. In this way, the new method for automatic

Manuscript received February 12, 1970. Manuscript received February 13, 1970.
108 IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, JANUARY/FEBRUARY 1971

adjustment of LTC transformers presents significant improvement In the NSP program we have chosen to limit the magnitude of all
for the solution of power flow problems. variable changes during the back-substitution step, with present
Incorporation of the new technique into the Newton algorithm
requires only minor modifications, yet gives significant saving
bounds being |AVI and |Atl < 0.1 pu, and IA5I < 30°. For certain
systems these restrictions might retard convergence during the
in solution time. I compliment the authors for incorporating the early stages of the solution; but they seem to increase the probability
new technique into existing Newton power flow programs. of solving other cases that would otherwise diverge or oscillate due to
the imposition of excessive variable corrections. Naturally the en-
countering of (or backing off from) tap limits requires the use of
transformer-type switching analogous to the case for generators,
where reactive power limits are involved.
We have chosen to handle the regulation priority question (where
Norris M. Peterson and W. Scott Meyer: The authors wish to thank both a generator and an LTC transformer regulate the same node
the discussers for their constructive comments and interest. voltage magnitude) a little differently from the way suggested by the
Mr. Subramanian's suggestion of using the proposed procedure discussers. Although we assume that a generator will be given first
to maintain node voltage or power flows within certain predeter- priority to regulate the node voltage magnitude, we have chosen to
mined limits would require a reversal of the type switching and hold the associated LTC transformer tap constant so long as gen-
priority schedules usually employed. The LTC transformer and erator reactive limits are not reached (rather than constrain the
phase-shifter taps would then remain fixed until voltage or power reactive power flow). The LTC transformer tap would only become a
flow inequality constraints were violated, at which point the proper variable if and when the generator reaches its reactive limit.
control equations would be substituted or added to adjust the taps Mr. Tinney and Mr. Powell are absolutely correct concerning the
to maintain the appropriate limits. It must be noted that such a ordering of the phase-shifter equations; this recognition, included in
scheme permits a wide range of solutions for a given power flow [6], is important for maximum elimination efficiency. The suggested
problem and appears to be unnecessary due to the rapid convergence order is clearly preferable to that shown in (5a), where fill-in during
of the proposed new method. the elimination generally will occur.
The proposed method can be used in conjunction with minimiza- It has been found necessary to develop logic for the NSP program
tion or optimization techniques, provided that the voltage and/or to handle special system configurations such as series and parallel
power flow regulation thereby provided is indeed a constraint to the LTC transformer connections. The series condition occurs when two
problem. For such cases, the proposed Newton power flow solutions or more LTC transformers are connected in series and one LTC
fit naturally into gradient optimization procedures [5]. But if such transformer regulates the nonregulated terminal of another LTC
regulation is not desired, the proposed adjustment procedure is of transformer. The parallel condition occurs when a node is regulated
course inapplicable. The LTC regulation of node voltage magnitude by two or more LTC transformers that are connected to different
will not, in general, produce the minimum possible system loss, nodes. The inclusion of such features makes it possible to handle
for example. The same would generally be true of constraining the general topology.
real power flow through phase shifters, which is generally not A matter which we feel requires investigation is the area of remote
compatible with general economic dispatch. control by the proposed method (i.e., the control of a line flow or a
A question was also raised by Mr. Subramanian about the com- sum of line flows by a remote phase shifter, or the control of a node
parative rates of convergence during early stages of the solution. voltage magnitude by a remote transformer). Although there is no
The old method is similar to that described in [4], where the problem theoretical reason why remote regulation by the proposed method
is treated as a fixed system (LTC taps fixed) until after the third should not be possible, we have not investigated its practicality and
iteration. It is thus only from the fourth iteration on that the old feel that three areas may cause difficulty.
method is actually solving the problem that was posed, and com- 1) The inclusion of remote regulation requires that the renumbering
parisons must be made for this portion of the solution process. It strategy be modified to insure that a nonzero diagonal exists (follow-
should be remembered that the computational effort per iteration ing elimination to the left of the diagonal). For the case of remote
for the new and old methods is nearly identical. voltage regulation by a transformer, nodes forming a path between
Mr. Subramanian's final point concerns the removal of phase the remotely controlled node and the controlling transformer, plus
shifters. If voltage magnitude is maintained constant at one terminal the transformer terminal that terminates this path, must all be
this could indeed be accomplished by opening the element across the eliminated before the remotely controlled node. Analogous con-
ideal phase shifter and inserting two generators having opposite siderations apply to the phase-shifter case. The development of
injected power outputs. The reactive generation requirement could efficient program logic for the general case is simply stated in words,
be calculated to maintain the voltage magnitude at the voltage- but it appears to be quite involved.
controlled node. But such elimination is not completely general in 2) The altered node numbering (as above) will degrade the
that it requires that a phase shifter control both the real power flow solution efficiency. Inclusion of remote control by phase shifters
and either the reactive flow or the voltage magnitude at the terminal does not allow the elimination order as outlined by Mr. Tinney and
node. Further, such an approach would not significantly affect Mr. Powell, which is most efficient; the general case is similar to
problem storage requirements (see comments by Mr. Tinney and that associated with area-interchange equations [1]. The use of
Mr. Powell), and would not in general converge any faster than the transformers to remotely regulate voltage magnitude adds two
proposed scheme. Remote regulation also would not be possible additional nonzero blocks to the original Jacobian matrix for each
using the suggested approach. such transformer; remote regulation in both cases thus results in an
The authors are especially grateful to Mr. Tinney and Mr. Powell asymmetrical nonzero block pattern for the Jacobian matrix. The
for their very relevant comments. We are gratified to find that inclusion of many remote controls might significantly degrade the
the Bonneville Power Administration has experienced solution time problem sparsity and the efficiency of the overall algorithm.
reductions similar to those stated in the paper and confirmed by our 3) Remote regulation should in general require changes in the
experimental and production runs. The authors concur that tap control variable larger than would be required for local regulation.
limits and the other points mentioned must be efficiently handled in This might increase convergence difficulties normally associated
order to make the method practical for production programs; these with constraint limits.
points are considered in the following paragraphs.
Certainly the first and most important consideration is that
of constraint limits. As the discussers state, Newton's method cannot REFERENCES
directly take limits into account; if large tap changes At occur during
the back substitution, it is possible for the resulting transformer [5] H. W. Dommel and W. F. Tinney, "Optimal power flow solu-
taps to greatly exceed their predefined limits, thereby resulting in tions," IEEE Trans. Power App. Syst., vol. PAS-87, pp. 1866-
large AP and AQ residuals when the tap limits are actually imposed. 1876, October 1968.
[6] H. W. Dommel, W. F. Tinney, and W. L. Powell, "Further
developments in Newton's method for power system applica-
tions," Paper 70 CP 161-PWR, presented at the IEEE Winter
Manuscript received March 30, 1970. Power Meeting, New York, N. Y., January 25-30, 1970.