BUS ADMITTANCE TEARING AND THE PARTIAL INVERSE

R. Bruce Shipley, Fellow

A. K. Ayoub, Member
Texas A&M University
College Station, Texas
Dorothy W. Coleman, Member
Tennessee Valley Authority
Chattanooga, Tennessee

A bstract - Tearing permits larger problems to be solved on a given

computer. It also permits solutions to parts of large network problems
when the entire solution is unnecessary. This is very important to power
organizations who find it increasingly impractical to solve an entire
interconnected power pool when only local answers are desired. The
partial inverse provides an excellent means for simplifying tearing
while yielding useful intermediate results that have physical and
electrical meanings. The tearing aspects presented in this paper have
not been previously described.

lents through zero impedance elements actually amounts to the simple

addition of admittances, and no interconnection matrix needs to be
formed.
THE PARTIAL INVERSE

The detailed operations for producing a partial inverse have been

discussed in reference [7], and its origin reviewed in Appendix A. The
physical significance of the individual elements of a partially inverted
admittance matrix is helpful in understanding-its use in analysis.
INTRODUCTION
Consider an admittance matrix of a power system having a row
and a column for each bus including the system neutral. This matrix is
Many network problems including short circuits, load flow, singular, and each row and each column adds to zero. The admittance of
economic studies, etc. require each member of a power pool to con- each network element is placed in the matrix in four positions; it is
sider the whole interconnected network when solving local problems. added to each of the diagonal position corresponding to the two busses
This has caused power system planners and others to demand larger at which it is terminated, and is subtracted from the respective offprograms and computers. So long as computers require finite costs and diagonal positions. Let this admittance matrix represent a single power
time to perform arithmetic operations there is almost no top limit to system out of a large pool. Figure (1) shows the tie busses (torn from
the cost of performing network analysis on expanding power systems. the large power pool) in the upper left corner. This arrangement is
The methods so far introduced and elaborated upon in the liter- used only to simplify the explanation.
ature under the title of "Diakoptics" alleviate this problem by tearing
If all the busses of the given system except the tie busses are
suitable tie elements of the given network so as to create a number of changed to the dimension of impedance, the result is the partially
subnetworks [I I - [5 1. To compensate for the torn elements a set of. inverted matrix of Fig. (2).
The explanation of the negative sign appearing in the lower right
injected currents is introduced at the nodes from which those elements
are removed. The mathematical basis of these methods follows from portion of Fig. (2) is given in Appendix A. This portion of the partial
Kron's orthogonal network approach [6], and involve a tie connection inverse is to the dimension of impedance and carries a negative sign.
matrix.
Each diagonal element of that portion represents the impedance from
Admittance tearing of busses presented in this paper is simpler that bus to all other admittance (retained) busses in parallel. The offthan the tearing of network elements and is general. Fewer tearing diagonal elements of the impedance portion represent mutual impedpoints are involved, the method follows simple circuit logic which may ances and yield the voltage change at one impedance bus due to a curbe easily described in power system parlance. Network equivalents are rent change at another.
obtained and interconnected through zero impedance elements.
TIE BUSSES
The busses to be torn can be the tie points between the suba
systems making up power pool. Ownership lines usually pass through
TIE BUSSES
Yi
busses. When ownership extends to a point on a transmission line, an
Y2
artificial bus may be created for the purpose of bus tearing. An
equivalent is determined for each subsystem with its torn busses to the
dimension of admittance and all its other busses changed to the
Y3
Y4
dimension of impedance. The equivalents, one for each subsystem, are
interconnected through zero impedance elements to yield an equivalent of the entire power pool in which only the torn busses are retained. The equivalent of the entire power pool may be used in this
Figure 1.
form or it may be used to obtain the equivalent of the entire pool as
seen from any of its member subsystems. In the latter case only the
lIE bUSSES
torn busses related to the subsystem under consideration are retained
and all others eliminated from the equivalent of the entire pool. The
TIE BUSSES
K
Y
result is then used with that particular subsystem to yield the desired
answers in a localized study, taking the other subsystems into account.
The "partial inverse" plays the role of obtaining the equivalents
in the above procedure. The interconnection of the subsystem equiva- z
Kt
_-

Paper T 72 139-9, recommended and approved by the Power System Engineering Committee of the IEEE Power Engineering Society for presentation at the
IEEE Winter Meeting, New York, N.Y., January 30-February 4, 1972. Manuscript
submitted September 15, 1971; made available for printing November 19, 197 1.
2042

_,

rY--

Figure 2.

The portions of the partially inverted matrix designated as K and

are nondimensional and exist as off-diagonal elements between the
impedance and admittance portions of the matrix. They have the
following physical significance. A current injected at any impedance bus
will distribute to the admittance busses in accordance with the respective nondimensional elements. They are sometimes called distribution factors.
The upper left portion of the partially inverted matrix remains to
the dimension of admittance, is square, and is of particular interest. It
represents the equivalent of the system under consideration as viewed
from its tie points. It has the important characteristic that circuit
elements can be added or subtracted from it without changing any
other portion of the partially inverted matrix. This fact leads to the
simple idea of bus admittance tearing.

Kt

-YA A f YBBA-

The use of the admittance tearing method is best xpldlaiU uy adl*

example. Fig. (3) depicts three power systems, A, B and C, where
system A is tied to system B at busses i and j, and syste-m B is tied to
system C at busses 1, m and n. This particular inter rconnection is
chosen only to simplify the explanation. The method acttually does not
place any restrictions on interconnection.

IA

/~~~~~l"

t
YBBM

SYSTEM

lB

-1
jB
nC

Figure 3.
The tie busses have been torn and renamed. Note that each torn
bus can have any number of lines connecting it into its respective
system, and fewer torn points are introduced than would be the case if
the systems were torn through impedance elements.
Let the admittance matrix of each of the three syistems be constructed, containing only the network elements of the particular
system. Perform a partial inverse on each system admitta nce matrix retaining the torn busses to the dimension of admittanc(e, while other
busses are changed to the dimension of impedance. The rresult is shown
in Fig. (4). The admittance portions in each partially ir iverted matrix
then represent the equivalent circuit of the correspond ling system in
admittance form.
SYSTEM A
IAA

BIB

SYSTEM B
IB mB nB

sSYSTEM C
IC mliCnC

KCC

-z cc

cc

Figure 4.

yBBM

YBBC +YCC

m
n
The equivalent circuits of the individual systems may be inter-

ni

nB

Figure 5.
e) r a hina 1hi

SYSTEM
B

BUS ADMITTANCE TEARING

SYSTEM
A

I-

connected to give the equivalent circuit of the entire

interconnection.

In terms of the admittance portions of the partially inverted matrices

this merely means assembly and addition; (Appendix B). This results in
the admittance of Fig. (5).
Suppose that it is desired to calculate the bus impedance matrix
of system B taking into account the effect of the rest of the interconnected network. YBBA in the partially inverted matrix of system B
shown in Fig. (4) is replaced by (YAA + YBBA) in Fig. (5) to account
for the effect of system A, and YBBC in Fig. (4) is replaced by
(YBBC + YCC) in Fig. (5) to account for the effect of system C. All
the rows and columns of the resulting system B matrix that are to the
dimension of admittance are inverted by the partial inverse method to
give the desired bus impedance matrix. Note that for the bus impedance
matrix of system B to exist, it is necessary for at least one connection
to the system neutral to exist in any one of the three systems. If no
such connection exists, the complete bus impedance matrix of system
B would not exist.
If the bus impedance matrix of system C is desired taking the
effect of the rest of the interconnected network into account, a partial
inverse is performed first on the rows and columns corresponding to
busses i and j, (Fig. 5) leaving those corresponding to busses 1, m and n
to the dimension of admittance. Observe that the admittance values resulting in that portion of the partially inverted matrix (Fig. 5) represent the equivalent of the entire interconnected network when only
busses 1, m and n are retained and all other busses eliminated.
YCC of the system C matrix of Fig. (4) is replaced by the equivalent
thus obtained. Performing the partial inverse on the admittance portion
of the resulting matrix gives the desired bus impedance matrix.
The bus impedance matrix of system A taking the effect of the rest
of the interconnected network into account can be calculated in a
similar fashion. Observe that the bus impedance matrices calculated by
this procedure constitute identical parts of the large blus impedance
matrix of the entire interconnected power pool. The latter can be calculated, if desired, in several ways. One way is to establish the bus admittance matrix for the entire interconnected network and simplv invert
it. If sparsity is to be considered, the order of inversion affects the
amount of arithmetic involved. An alternative is to assemble the partially inverted matrices of the separate systems shown of Fig. (4) as
shown in Fig. (6) where the undesignated portions contain only zeros.
Now, the portions to the dimension of admittances in the matrix of
Fig. (6) may be inverted by the partial inverse technique to yield the
complete bus impedance matrix of the entire interconnected network.
Needles to say any portion of the bus impedance matrix can be calculated by selecting the proper portion of Fig. (6) when computer space
is a problem.
Generally problems do not require the entire bus impedance
matrix to be calculated. Neither is it a requirement that the individual
system busses be grouped as shown, although this greatly simplifies the

2043

-ZAA

K AA

KAA

[41 H. K. Kesavan, M. A. Pai, M. V. Bhat, "Piecewise

Load Flow

Solution of the

Problem", IEEE Summer Meeting, 1971.

[5] M. A. Laughton, "Decomposition Techniques in Power Systems",
Vol. 115 IEEE, April 1968.
[61 G. Kron, "Tensor Analysis of Networks", John Wiley and Sons,
1939.
[7] R. Bruce Shipley, Dorothy Coleman, "A New Direct Matrix Inversion Method", Communications and Electronics, Nov. 1959,
pp. 568.

YBBA)

KBBA

YBBM

APPENDIX A

in

KtBBA ~ZBB
Z

Yt
BBM

Kt
BBC

Consider the set of linear equations:

KKBBC

(K
(BBC+YCC)

~~~Yl Y2

Kc

Ktc

expressing the performance of a network in the bus frame of reference,

and to be solved for the bus voltage vectors E 1 and E2.
If the values of the bus voltage vector E2 are not of immediate
interest, they can be eliminated and the number of equations reduced

-Zc

Figure 6.

illustrations. Frequently the partially inverted matrix is the desired

form and no further work need be done.

[6].
Thus:

IMPLEMENTATION
The procedures described in this paper have been incorporated
into a digital program that yields solutions to three phase short
circuits, load flow, including all line out cases, steady-state and transientstability, from a single converged case. Line out and stability cases
are approximations. Using computers available today it can be used to
study power pools as large as the U.S.-Canadian interconnection. This
is made practical by the tearing procedure which permits each power
system to study only its own portion of the interconnected network
while automatically taking into account all the other parts of the network without studying them in detail. This results in approximations
which can be improved to any desired accuracy. Accuracy is improved
by extending the area of study. This can be done in parts since any
member system can request that adjacent systems be included or that
the entire pool be calculated in detail. The individual member system
can make any local changes it wishes, which can then be incorporated
into its equivalent automatically for other pool members to use.

I1 Y1E1+ Y2E2

(1)

I2 =Y3E1 y4E2

(2)

-1
" y4 I 2

(3)

From (2):
E2

Y 3 E 1)

Substituting in (1) and rearranging:

I

= (Y

y2Y4 Y3) E + Y2Y42

(4)

which can be written in matrix form as:

Lf2I1 = (Yl 2 Yi Y3
I2
(5)

CONCLUSIONS
A method for the solution of very large power pools has been
presented which yields the same results as other techniques discussed
under the title of diakoptics. Simplification has been produced in that
no connection matrices need to be determined. Tearing and interconnection are accomplished by simply subtractinig and adding the
admittance portions of partially inverted matrices. The individual
portions of the partially inverted matrices have physical significance
and are useable as equivalent circuits. The amount of computations
is usually less than that required by other diakoptic methods due to the
fewer points of tear. Sparsity methods may be used if desired.

Rewriting (3) as:

Ep2-Y4

Y3

Y-l

-El

12
(6)
the process of eliminating E2 can be indicated by annexing (6) to (5):

REFERENCES
[1] H. H. Happ, "Diakoptics and Networks", Academic Press, 1971.
[21 A. Brameller, M. N. John, M. R. Scott, "Practical Diakoptics for
Electrical Networks", Chapman & Hall, 1969.
[3] A. M. Sasson, H. E. Brown, ';Deductions and Clarifications on the
Diakoptics Approach", 1971 PICA Conference Proceedings, pp.
433-439.
2044

En
11

(2
(7)

in which 12 and E2 have been exchanged

Compared with the complete solution form:

across

the equality sign.

EBC

EBC
ETCB =
where the components of EAB are Ei and Ej and these of EBC
E1, Em and En. These equations can be written as:

-i

(8)

II

ETBC

iA
jA ETAB
iB
B

equation (7) represents a "partial inverse" of the original bus admittance matrix.
An important implication of the partial inverse follows from
writing equation (4) as:

(1

- Y

ly4I2)

(Y1

Y2Y41Y3)

(. I

J lv

ETBA

3EBc

IB
mB ETBC
nB

(9)

which represents the equivalent of the given network with the busses
corresponding to the bus voltage vector E2 eliminated.
Comparing (9) to (7), the portion of the partially inverted matrix
remaining to the dimension of admittance thus represents the admittance matrix of the equivalent network involving the retained busses
only.
Note that the off-diagonal portions of the partially inverted
matrix shown in equation (7) are nonreciprocal, differing only in sign
The matrix can be made
for a linear bilateral network where Y 3= Yt.
2
symmetrical in that case by attaching a negative sign to the current
vector 12. Thus:
LI

are

IC
mC

ETBC

nC
L.e.:
-1
E

Now for the

old

= At

power to

new

be invariant, the currents will transform

as:

-1

new

and the admittance matrix

-El

Ynew

=A-

old

as:

- A

-1

YoldA

t-

-Y 12

YAAtYB8A

(10)

YBBM

APPENDIX B

The assembly of the equivalent admittance matrices of Fig. (4)

corresponds to the performance equation:

iA
jA

YTAB

YBBA YBBM

ETBA

I'TBC

yt YBRC
YBBM

ETBC

IB

mB
nB

~CC

E_-TAB

YAA

i B ITBA
jB

B8BC

BBM

IC 'TCB

_YcC

mC

nC

Thus the interconnection of the separate equivalents of the individual systems to form the equivalent of the entire interconnected
network can be accomplished by simple addition and the explicit
formation of At-l is not required.

Discussion

ETCB

where the subscripts attached to the bus current and voltage vectors
correspond to the tie busses designations after tearing, and the primes
indicate equivalent bus currents related to the actual bus currents of
the system.
The interconnection of the equivalents is accomplished by zero
impedance links between the two parts of each torn bus (61, yielding:

ETAB

EAB

ETBA

EAB

B. F. Wollenberg (Leeds & Northrup Company, North Wales, Pa.

19454): The discussor agrees with the authors on the desirability of
using a partially inverted Y-matrix for power system studies. This
arises through the need to have an equivalent circuit available in
Y-matrix fotm for use with a Newton ioad flow program. In addition
to solving load flows with an equivalent circuit attached, there is a
desire to be able to solve a load flow Without an equivalent circuit
attached and couple it to the solved case when needed.
Our programs have been set up primarily for real time dispatch
computer load flow studies, and use the author's techniques in the
following manner. Load flow data for neighboring systems is stored on
bulk memory with provision for updating it if changes are known.
Upon demand a program calculates a partial inverse of the type shown
by the authors in Figure 2. The calculation does not use the technique

Manuscript received February 17,

2045

1972.

shown by the authors in Appendix A, instead, an equivalent sparsity

programmed "partial triangularization" scheme is used. In this scheme
the busses are ordered as shown below, the nontie bus rows are zeroed
up to the diagonal as in normal Gaussian elimination while the tie bus
rows are zeroed only for the nontie bus columns. The remaining lower
right corner of the Y-matrix is the desired equivalent circuit Yequivalent as shown.

Nontie
Busses

Tie
Busses

Nontie
Busses

ya

Yb

Tie
Busses

Yc

Yd

REFERENCE
[8] Happ, H. H., "Multi-Level Tearing and Applications", IEEE Paper
Scheduled For Presentatioh at the IEEE Summer Power Meeting
(submitted September 15, 1971).

Y-Matrix of Neighboring Systems Before Reduction

Nontie

Buisses

tion that tearing through buses, which for the case in Fig. 3 involved
fewer tearing points than tearing through lines, could not be accomplished in Diakoptics. Tearing through buses can readily be acconmodated iin Diakoptics with no special provisions required and represents
the case where tie branches have zero impedance.. A three-area numerical example that has one tie branch whose impedance is zero is given
on pp. 258-263 of Reference 1. They correctly state that in Diakoptics
a tie connection matrix (Intersubdivision Matrix) is required whose
function it is to account for the interconnection of the subdivisions.
The tie model (or modelsl,8) is also used in Diakoptics to transmit the
contribution of all active sources, from one area to another, or am1ong
many areas, whereas the case with sources in other areas has not been
discussed in the paper.

Tie
Busses

Nontie
Buisses
Tie
Busses

ecii
quiv

Resulting Equivalent Y-Matrix After "Partial Triangularization"

The resulting equivalent Y-matrix thus formed is stored back on
bulk memory after removing high impedance lines. Forreal time applications it is desirable to obtain a solved load flow matching telemetered
flows at the system tie points and fitting all.other information telemetered from within the system (the state estimation problem). After
such a solved load flow is obtained, the equivalent circuit is attached or
coupled as shown by the authors in Figure 5. As part of the coupling
procedure, the net injections at the tie busses are adjusted to compensate for the flows in the equivalent circuit lines. The result is a solved
load flow matching the telemetered boundry tie flows, with an equivalent circuit attached. This load flow is now suitable for various contingency and other security studies which require an equivalent circuit
to properly model the effects of connections to neighboring systems.

Albert M. Sasson (American Electric Power Service Corp., New York,

N. Y. 10004): The ideas contained in this paper provide excellent
means for efficient computation of Z-bus matrices corresponding to
areas,of a larger system and to the solution of various network problems on a per area basis. These comments will present ideas that, to this
discusser, clarify some of the points made in the paper.
Fig. (4) presents the partially inverted matrices of the three area
example used in the paper. In matrix equation form,

ITAB

YeaTeL

or the branches as generated from the elements of its

matrix,
processed as part of the system studied.
The procedure the authors suggest is useful when Z matrices of
adjacent systems are to be calculated, since the partial inverse as an
intermediate step yields the equivalent for the adjacent system and
accepts the equivalent of the adjacent system. The dimension of each
partial inverse is the internally retained buses plus ties. We assume that
the next step in the authors' procedure is a row and column elimination, which completes the initial partial inversion, with the result that
all tie axes disappear in each area's Z matrix.
In the introduction, the authors have compared their method to
that of Diakoptics and have claimed certain advantages for their
method. We will address ourselves brieftly to that comparison. Diakoptics provides for both the passive representation of the total interconnected network and the representation of all sources in the neighboring areas. In Diakoptics, we are interested in obtaining a solution of
the combined or total system (A+B+C) as shown in Figure 3, considering the sources (P,Q, current, voltage, etc.) in all areas. The
authors have not discussed the presence of sources in the adjacent and
other areas in the paper; they may wish to elaborate on it in the closure.
It appears doubtful that the method the authors present has advantages
over Diakoptics when considering the overall solution of large scale
system problems, load flow, etc. The authors implied in the introduc-

KtAA 'A +

YAA

ETAB

(1.2)

YBBA ETBA + KBBA IB + YBBM ETBC

(2.1)

EB

KtBBA

(2.2)

ITBC

= yt
BBM ETBA + Kt
BBC B +YBBC ETBC
= YCC ETCB+ KCC 'C

Ec

ETBA - ZBB IB+ KBBC ETBC

KCETCB - ZCCIC

(2.3)

(3.1)

(3.2)

These equations form three groups which do not reflect the presence
of each other. A torn node appears in two equations. For example, a
torn node between areas A and B, will have an equation in (1.2) and in
(2.1). The voltages ETAB are equal to ETBA and can be used interchangeably. The two currents in ITAB and ITBA represent a division
of the real current entering nodes AB, so that,

IAB
EAB

Yequ=(Yl Y2 Y4 Y3)
The

(1.1)

ITBA

ITCB
G. K. Carter and H. H. Happ (General Electric Company, Schenectady;
N. Y. 12305): This paper concerns itself with the inclusion of passive
network equivalents into the model of the network under study. At
present, it is common practice when conducting network studies such
as load flow or fault studies to include passive equivalents of the networks of the outlying regions. These equivalents are usually generated
using an admittance network reduction technique which gives a matrix
with only tie axes present:

= -ZAA IA +KAA ETAB

EA

='TAB + 'TBA
=

ETAB ETBA
=

(4.1)
(4.2)

Between B and C, the relations are similar,

IBC

EBC

= ITBC '+TCB
= ETBC = ETCB

(4.3)

Equations (4.1 )-(4.4) show the relation between the torn equation (1)(3) and the equations of the whole system. The current equations (1.2),
(2.1), (2.3) and (3.1) will be rewritten as,

Manuscript received February 28, 1972.

"'TAB

'TAB KtAA 'A = YAA EAB

ITBA ITBA - KBBA 'B = YBBA EAB + YBBM EBC
BBC 'B =6I3BM EAB + YBBC EBC
I"TBC = TBC
I TCB= ITCB - KCC IC = YCC EBC
=

(5.2)

(5.3)
(5.4)

To get the impedance matrix of area B taking into account the

effect of areas A and C, it can be done by replacing equations (2.1) and
(2.3) by equations (7.1 ) and (7.2) derived in this section.
Adding equation (5.1) and (5.2) and separately (5.3) and (5.4),
Manuscript received February 18, 1972.

2046

(5.1)

" TAB + ITBA

IAB

IBC

(YAA + YBBA)

yBBM EBC

'TBC + "TCB = YBBM

(YBBC + YCC) EBC

(6.1)

EAB
(6.2)

Extracting the term KnBA ID contained in "TAB of (6.1), (see 5.2),

and the term K c F onPained in ITBC 6f (6.2), (see 5.3) and
passing them to t righl hand side,
I'AB = "TAB + 'TBA = (YAA + YBBA) EAB
+ KBBA IB+ YBBM EBC

"'BC

=
+

'TBC + "TCB = YtIBM EAB + KBBC 'B

(YBBC+YCC) EBC

(7.1)

(7.2)

Equations (7.1) and (7.2) contain the same right hand variables as
(2.1 )and (2.3) which they replace in equation (2). The partial inverse of
(2) is then continued to get the impedance matrix of area B. It should
be noticed that the resulting impedance equation is,

III[AB

FEAB]

IB

:iB

EBC

LEBCJ
Wh-ile EA and EBC are the actual voltages of the boundary nodes,
I"AB an
are not actual system currents. However, they are rerelated to actual currents. For example, the term I"AB can be traced
back through equations (7.1), (5. 1 ) and (4.1) as follows:

B["BC

Y"AB
"'AB
I"AB

I TAB + 'TBA

'TAB KAA 'A + 'TBA

(9.1 )
(9.2)

'AB-KtA'A

(9.3)

where IAB are the injections at the boundary nodes and IA are the
injections at nodes within area A. The same can be done for I'13C.

"'BC = 'BC - KCC IC

(9.4)

To get the impedance matrix of area C taking into account the

effect of areas A and B, equation (3.1) is replaced by (11) derived in
this section.
Eliminating EAB from (6.2) by replacing it with EAB from (6.1),

""BC = "BC- tBBM (YAA

+ YBBA)

"AB

-YBBC + YCC - BBM (YAA + YBBA)-1 BBM) EBC

Extracting the term KCC IC contained in I'BC of (6.2), see (5.4),
and passing it to the right hand side,
IV

'BC

=TCB + "I'BC

t'A (YAA + YBBA) 'AB

YBBM

BBC + YCC - YBBM (YAA + YBBA)1lBBM)

(11)
EBC+KCCIC

Equation (I 1) has the same right hand variables as (3.1) and replaces it.
The partial inverse of (3) is then continued to get the impedance
matrix of area C. As pointed out in connection with equation (8), it
should be noted that the impedance equation is,

[EC

=[ZC]

and to use (12) in network problems

injections,

[Lc
IIVC

(12)

has to be related to physical

t
t
IIV
- BC - K BBC 'B YBBM (YAA + YBB A)
BC IIAB KkAI A-KBBA'B
-

The above sections show the basic ideas of the paper derived by
direct algebraic approaches using only Kirchoff's laws. This emphasizes
once more that Kron's invariance of power transformation, while
correct, is not necessary and is completely implied in Kirchoff's laws
when used merely to alter the form of the equations of a system. When
the power invariance is used to go from one reference frame to another
in the B-constants derivation, it becomes necessary as it is part of the
object of a transformation which also include some arbitrary relations
between currents, voltages or loads.
Equations (8), (9), (12) and (13) will be considered to discuss the
applications of the method and its difference with diakoptics. Suppose
area C is to be studied and one has a current vector for-all nodes in the
system and (12) is to be computed. First IJ2YC would have to be computed from (13) using the injections of alFnodes outside and on the
boundary of C, Then the matrix multiplication. (i2) can be performed.
This procedure is somewhat analogous to the diakoptics procedure
where an intersubdivision matrix equation, corresponding to (13),
would first have to be used before using the area equations.
With diakoptics you do not find sections of the impedance matrix
of the whole system as with the proposed method. However, diakoptics is an efficient way of solving a linear problem broken into many
areas. With the proposed method, it appears to be less efficient to solve
exactly a large problem broken into many areas when you want results
for all areas. One would have to get the impedance matrices for each
area, which imply in general many partial inversions of Fig. 5 into its
various sections (as also implied in equation (I1)) followed by the
completion of the partial inversions originally made on each area.
Then, the reflection of the injections outside of each area have to be
made through operations similar to (9.3), (9.4) and (13), for which additional matrices have to be used.
It thus seems that the proposed method is less efficient than
diakoptics for problems where there are injections in all busses. For a
load flow problem the whole process is iterative with a new current
vector produced at each iteration that would have to be reflected down
to each area. In all these problems, if a Z-bus approach must be used,
either diakoptics should be used or the proposed method or other
methods should be used to generate the entire Z-busmatrix.
On the other hand, if current injections are zero outside of the
area of interest, as in fault problems, then the proposed method is
very efficient because the reflections are trivial.
A final comment on the use of the method to generate the impedance matrix of.the entire interconnected network. An explanation
of Fig. 6 will first be giyen. Consider the second row of Fig. (6). This is
obtained from (7.1 ) and (9.3). Replacing I"AB of (7.1 ) from (9.3) and
passing the KtAA IA term to the right hand side produces the desired
result. The current vector of Fig. (6) is thus made up of the true injections at each node. If computer space is a problem, the various
sections of the matrix can be found by the proposed method. However, this implies repeated re-start of the partial inverse technique
towards the desired section. It appears here that the use of sparsity
techniques to get Z-bus starting from the entire Y-bus would be more

efficient.8

The author's comments on this discussion would be appreciated.

This has been a very enjoyable paper to read.

REFERENCE

[81 A. M. Sasson, "A method for Z-bus matrix generation for fault
studies," Paper No. 70 CP 704-PWR, IEEE Summer Power
Meeting, 1970.

R. Bruce Shipley, A. K. Ayoub, and Dorothy W. Coleman: We wish to

thank the discussors for their interest in our paper and their worthwhile
discussions of the method it presents.
As to Mr. Wollenberg's comments, the method he describes for obtaining an equivalent is faster and requires less computer memory than
the partial inverse if an equivalent is the end product. He obtains the
terminal conditions from telemetered data and thus has a converged
load flow with which to begin his studies. It appears he does not have
the ability to "rerun" the total network as Dr. Happ would, or as we
could choose to do.
With regard to the discussion by Messrs. Carter and Happ, we
would like to point out that we -generally view tearing, diakoptics and
equivalents as part of the same procedure with the same objectives and
consider them all as techniques rather than any real fundamental differences. As such, we agree with the view that bus tearing does amount
to the removal of zero impedance elements connecting buses to themselves. Although the paper did not discuss the presence of sources out-

(13)

2047

Manuscript received April 25, 1972.

side the area of interest explicitly, it does not actually ignore them.
Rather than attempt to include a complete derivation here, we would
like to refer to Dr. Sasson's discussion.
Dr. Sasson's discussion also points out the similarity between element tearing and bus tearing. We agree with his comment that Kron's
invariance of power transformation is not necessary for the derivation
and are completely implied in Kirchhoff's laws in this case.
As pointed out in the paper, programs are available to solve simultaneously for short circuits, load flow and approximate transient
stability. Once the total system is in balance, a power company may

study only its detailed system. Changes may be made within the detailed
system without reconverging the entire network if inaccuracy in the
division of flows in tie lines is acceptable. In practice, each subsystem
will be available to the computer; therefore, any numnber of subsystems
can be "rerun". Only the output of its own system would be given to
a company.
A comparison of this method with that of diakoptics as far as
efficiency is concerned is not available. However, to study large networks by the method does require many partial inverses.

ITERATIVE LINEAR AC POWER FLOW SOLUTION FOR FAST APPROXIMATE OUTAGE STUDIES
Norris M. Peterson
Systems Control, Inc.
Palo Alto, California

Donald W. Bree, Jr.

Systems Control, Inc.
Palo Alto, California

William F. Tinney
Bonneville Power Administration
Portland, Oregon

A bstract - A fast approximate method is presented for solving

the ac power flow problem for line and generator outages. The method
is significantly more accurate than any linear approximation and significantly faster than the Newton-Raphson method for an approximate
solution. The method has applications in system planning and operationls where approximate ac power flow solutions are acceptable. The
method is applicable to system planning for rapid location of design
criteria violations and it is particularly well adapted for system operation use as an on-line security monitor. Efficiency is achieved through
decoupling of real and reactive power equations, sparse matrix methods,
an experimentally determined iteration scheme and the use of the
matrix inversion lemma to simulate the effect of branch outages.

INTRODUCTION
The power flow program is indispensible for the analysis of ac
power systems. Although improved solution methodsl2'3 and faster
computers have extended its capabilities, this gain has been largely
offset by the increased number and size of studies that must be made.
For many planning and operations applications, an exact ac power flow
solution with a detailed output is not necessary. For these applications
it is much more efficient to use a fast approximate solution method
and to limit the output to include only the most essential information.
This paper describes an approximate ac power flow solution
method, the Iterative Linear Power Flow (ILPF), that is particularly
well suited for the steady-state analysis of both the real and reactive
effects of a series of contingent line and/or generator outages. Unlike
other approximate methods which are based on strictly linear models4,
the ILPF gives a good approximation of the non-linearities of the power
flow equations and, therefore, is significantly more accurate. Also, unlike some other approximate methods, it includes reactive as well as
real power and it does not depend on the precalculation and storage of
linear distribution factors. The ILPF is fast, sufficiently accurate for
the intended applications and does not require excessive computer
Paper T 72 140-7, recommended and approved by the Power System Engineering Committee of the IEEE Power Engineering Society for presentation at
the IEEE Winter Meeting, New York, N.Y., January 30-February 4, 1972. Manuscript submitted September 15, 1971; made available for printing November 19,
1971.

memory. It is a composite and extension of methods used in other

but their combination and use for outage simulaapplications
tion presents new problems. The ILPF can be used to rapidly identify
from a iarge set of outages those which cause violations of specific
criteria. The method also gives a very good approximation of the magnitude of the violations. For system planning the ILPF can be used to
identify the problems that exist at the next level of contingent outage.
As a complement to a standard ac power flow program, the ILPF can be
used to scan a large set of problems to determine those that require
exact solutions and detailed outputs. For system operations, the ILPF
can be used for assessment of security either off-line, as in the planning
application, or on-line as part of an automatic security monitoring
scheme.
The paper is concerned mainly with the solution algorithm rather
than the input-output which will vary depending on the particular
application. The work is a result of a -cooperative research contract
between the Bonneville Power Administration and Systems Control, Inc.

NOTATION
The general quantities used are defined as follows:

Pk + i Qk = complex power injection into node k

Ykm = Gkm + j Bkm = complex elemnents of the nodal admittance matrix

Vk = the voltage magnitude of node k to datum

Ok= the voltage phase angle of node k to reference

Other symbols are defined as they are introduced. Matrices and vectors
are enclosed in brackets, e.g., [VI. Complex quantities are indicated by
a superbar, e.g., V. Complex conjugate is indicated by an asterisk,
e.g., V*. The summation notation mmeJk means m runs over the set of
nodes of class p adjacent to node k.
The term contingent outage is used to define a single branch or
generator outage for which a solution is to be simulated. The term
solution cycle is used to describe the solutions for a complete set of
contingent outages.

2048