_05_EE394J_2_Spring12_Power_System_Matrices.

doc

Power System Matrices and Matrix Operations

Nodal equations using Kirchhoff's current law. Admittance matrix and building algorithm.
Gaussian elimination. Kron reduction. LU decomposition. Formation of impedance matrix by
inversion, Gaussian elimination, and direct building algorithm.
1.

Admittance Matrix

Most power system networks are analyzed by first forming the admittance matrix. The
admittance matrix is based upon Kirchhoff's current law (KCL), and it is easily formed and very
sparse.
Consider the three-bus network shown in Figure that has five branch impedances and one current
source.

ZA

ZC
ZB

ZE

3
ZD

I3

Figure 1. Three-Bus Network

Applying KCL at the three independent nodes yields the following equations for the bus voltages
(with respect to ground):
At bus 1,

V1 V1 V2
+
=0 ,
ZE
ZA

At bus 2,

V2 V2 V1 V2 V3
+
+
=0 ,
ZB
ZA
ZC

At bus 3,

V3 V3 V2
+
= I3 .
ZD
ZC

Collecting terms and writing the equations in matrix form yields

Page 1 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
1
1
+

ZE Z A
1

ZA

1
ZA
1
1
1
+
+
Z A Z B ZC
1

ZC

V1 0
1
V2 = 0 ,

ZC

1
1 V3 I 3
+
Z C Z D
0

or in matrix form,
YV = I ,

where Y is the admittance matrix, V is a vector of bus voltages (with respect to ground), and I is a
vector of current injections.
Voltage sources, if present, can be converted to current sources using the usual network rules. If
a bus has a zero-impedance voltage source attached to it, then the bus voltage is already known,
and the dimension of the problem is reduced by one.
A simple observation of the structure of the above admittance matrix leads to the following rule
for building Y:
1. The diagonal terms of Y contain the sum of all branch admittances connected directly to the
corresponding bus.
2. The off-diagonal elements of Y contain the negative sum of all branch admittances connected
directly between the corresponding busses.
These rules make Y very simple to build using a computer program. For example, assume that
the impedance data for the above network has the following form, one data input line per branch:
From
Bus

To
Bus

Branch Impedance (Entered

as Complex Numbers)

ZE

ZA

ZB

ZC

ZD

The following FORTRAN instructions would automatically build Y, without the need of
manually writing the KCL equations beforehand:
COMPLEX Y(3,3),ZB,YB
DATA Y/9 * 0.0/
1

READ(1,*,END=2) NF,NT,ZB

Page 2 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
YB = 1.0 / ZB
C

MODIFY THE DIAGONAL TERMS

IF(NF .NE. 0) Y(NF,NF) = Y(NF,NF) + YB
IF(NT .NE. 0) Y(NT,NT) = Y(NT,NT) + YB
IF(NF .NE. 0 .AND. NT .NE. 0) THEN

MODIFY THE OFF-DIAGONAL TERMS

Y(NF,NT) = Y(NF,NT) - YB
Y(NT,NF) = Y(NT,NF) - YB
ENDIF
GO TO 1
2

STOP
END

Of course, error checking is needed in an actual computer program to detect data errors and
dimension overruns. Also, if bus numbers are not compressed (i.e. bus 1 through bus N), then
additional logic is needed to internally compress the busses, maintaining separate internal and
external (i.e. user) bus numbers.
Note that the Y matrix is symmetric unless there are branches whose admittance is directiondependent. In AC power system applications, only phase-shifting transformers have this
asymmetric property. The normal 30o phase shift in wye-delta transformers creates asymmetry.
2.

Gaussian Elimination and Backward Substitution

Gaussian elimination is the most common method for solving bus voltages in a circuit for which
KCL equations have been written in the form
I = YV .

Of course, direct inversion can be used, where

V = Y 1 I ,
but direct inversion for large matrices is computationally prohibitive or, at best, inefficient.
The objective of Gaussian elimination is to reduce the Y matrix to upper-right-triangular-plusdiagonal form (URT+D), then solve for V via backward substitution. A series of row operations
(i.e. subtractions and additions) are used to change equation
I1 y1,1
I y
2 2,1
I 3 = y3,1

M M
I N y N ,1

y1,2

y1,3

y 2,2
y3,2

y 2,3
y3,3

L
L

y N ,2

y N ,3 L

y1, N V1
y 2, N V2

y3, N V3

M M
y N , N V N

Page 3 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
into
I1 y1,1
I'
2 0
I 3' = 0

M M
'
I N
0

y1,2

y1,3

y ' 2, 2

y ' 2,3 L
y ' 3,3 L

0
M

y1, N V1

y ' 2, N V2
y '3, N V3 ,

M M
y ' N , N V N

in which the transformed Y matrix has zeros under the diagonal.

For illustrative purposes, consider the two equations represented by Rows 1 and 2, which are

I1
I2

=
=

Subtracting (

y1,1V1
y 2,1V1
y2,1
y1,1

+
+

y1,2V2
y 2,2V2

+
+

y1,3V3
y 2,3V3

y1, N V N
.
y 2, N V N

Row 1) from Row 2 yields

y1,1V1
+
y1,2V2

y2,1
y2,1
I2
I1 = y2,1
y1,1 V1 + y2,2
y1,2 V2

y1,1
y
y
1,1
1,1

I1

+ L +
+ L +

y2,1

y1,3V3

y2,1
+ y 2 ,3
y1,3 V3

y
1,1

+ L +

y1, NVN

y2,1
+ L + y2, N
y1, N VN

y
1,1

The coefficient of V1 in Row 2 is forced to zero, leaving Row 2 with the desired "reduced" form
of
I 2'

= 0 +

y 2' ,2V2

y 2' ,3V3

+ L +

y 2' , N V N .

Continuing, Row 1 is then used to "zero" the V 1 coefficients in Rows 3 through N, one row at a
time. Next, Row 2 is used to zero the V 2 coefficients in Rows 3 through N, and so forth.
After the Gaussian elimination is completed, and the Y matrix is reduced to (URT+D) form, the
bus voltages are solved by backward substitution as follows:
For Row N,

( )

1
'
'
'
IN
= yN
IN
.
, N V N , so V N = '
yN, N

Next, for Row N-1,

1
'
'
'
'
'
IN
IN
1 = y N 1, N 1V N 1 + y N 1, N V N , so V N 1 = '
1 y N 1, N V N .
y N 1, N 1

Page 4 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
Continuing for Row j, where j = N 2, N 3,L ,2 ,

I 'j = y 'j , jV j + y 'j , j +1V j +1 + L + y 'j , N V N , so

Vj =

1
y 'j , j

(I 'j y 'j, j +1V j +1 L y 'j, N VN ) ,

which, in general form, is described by

Vj =

1 '
I j y 'j , k Vk .

y 'j , j
k = j +1

A simple FORTRAN computer program for solving V in an N-dimension problem using

Gaussian elimination and backward substitution is given below.
COMPLEX Y(N,N),V(N),I(N),YMM
C

GAUSSIAN ELIMINATE Y AND I

NM1

= N - 1

PIVOT ON ROW M, M = 1,2,3, ... ,N-1

DO 1 M = 1,NM1

MP1

= M + 1

YMM

= 1.0 / Y(M,M)

OPERATE ON THE ROWS BELOW THE PIVOT ROW

DO 1 J = MP1,N

THE JTH ROW OF I

I(J)

= I(J) - Y(J,M) *YMM * I(M)

THE JTH ROW OF Y, BELOW AND TO THE RIGHT OF THE PIVOT

DIAGONAL
DO 1 K = M,N
Y(J,K) = Y(J,K) - Y(J,M) * YMM * Y(M,K)
1

CONTINUE

BACKWARD SUBSTITUTE TO SOLVE FOR V

V(N)

= I(N) / Y(N,N)

DO 2 M = 1,NM1
J

= N - M

BACKWARD SUBSTITUTE TO SOLVE FOR V, FOR

ROW J = N-1,N-2,N-3, ... ,1

V(J)

= I(J)

JP1

= J + 1

DO 3 K = JP1,N
V(J)

= V(J) - Y(J,K) * V(K)

Page 5 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
3

CONTINUE
V(J)

= V(J) / Y(J,J)

CONTINUE
STOP
END

One disadvantage of Gaussian elimination is that if I changes, even though Y is fixed, the entire
problem must be re-solved since the elimination of Y determines the row operations that must be
repeated on I. Inversion and LU decomposition to not have this disadvantage.
3.

Kron Reduction

Gaussian elimination can be made more computationally efficient by simply not performing
operations whose results are already known. For example, instead of arithmetically forcing
elements below the diagonal to zero, simply set them to zero at the appropriate times. Similarly,
instead of dividing all elements below and to the right of a diagonal element by the diagonal
element, divide only the elements in the diagonal row by the diagonal element, make the
diagonal element unity, and the same effect will be achieved. This technique, which is actually a
form of Gaussian elimination, is known as Kron reduction.
'
Kron reduction "pivots" on each diagonal element y m
,m , beginning with y1,1 , and continuing
through y N 1, N 1 . Starting with Row m = 1, and continuing through Row m = N - 1, the

algorithm for Kron reducing I = YV is

1.

Divide the elements in Row m, that are to the right of the diagonal, by the diagonal
'
element y m
,m . (Note - the elements to the left of the diagonal are already zero).

'
Im

2.

'
with
Replace element I m

3.

'
Replace diagonal element y m
,m with unity.

4.

Modify the Y ' elements in rows greater than m and columns greater than m (i.e. below
and to the right of the diagonal element) using

'
ym
,m

'
y 'j , k = y 'j , k y 'j , m y m
, k , for j > m, k > m.

5.

Modify the I ' elements below the mth row according to

'
I 'j = I 'j y 'j , m I m
, for j > m.

6.

Zero the elements in Column m of Y ' that are below the diagonal element.
Page 6 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
A FORTRAN code for Kron reduction is given below.
COMPLEX Y(N,N),V(N),I(N),YMM
C

KRON REDUCE Y, WHILE ALSO PERFORMING ROW OPERATIONS ON I

PIVOT ON ROW M, M = 1,2,3, ... ,N-1

NM1

= N - 1

DO 1 M = 1,NM1

MP1

= M + 1

YMM

= 1.0 / YMM

DIVIDE THE PIVOT ROW BY THE PIVOT

DO 2 K = MP1,N
Y(M,K) = Y(M,K) * YMM
2

CONTINUE

OPERATE ON THE I VECTOR

I(M)

= I(M) * YMM

SET THE PIVOT TO UNITY

Y(M,M) = 1.0

REDUCE THOSE ELEMENTS BELOW AND TO THE RIGHT OF THE PIVOT

DO 3 J = MP1,N
DO 4 K = MP1,N
Y(J,K) = Y(J,K) - Y(J,M) * Y(M,K)
4

CONTINUE

OPERATE ON THE I VECTOR

I(J)

= I(J) - Y(J,M) * I(M)

SET THE Y ELEMENTS DIRECTLY BELOW THE PIVOT TO ZERO

Y(J,M) = 0.0

4.

CONTINUE

CONTINUE

LU Decomposition

An efficient method for solving V in matrix equation YV = I is to decompose Y into the product
of a lower-left-triangle-plus-diagonal (LLT+D) matrix L, and an (URT+D) matrix U, so that YV
= I can be written as

LUV = I .
The benefits of decomposing Y will become obvious after observing the systematic procedure for
finding V.
It is customary to set the diagonal terms of U equal to unity, so that there are a total of N 2
unknown terms in L and U. LU = Y in expanded form is then

Page 7 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
0
l1,1
l
2,1 l 2,2
l3,1 l3,2

M
M
l N ,1 l N ,2

0 1 u1,2
0 0 1

l3,3 L
0 0 0

M
O
M M
M

l N ,3 L l N , N 0 0
0
0

L
L

L u1, N y1,1

u 2,3 L u 2, N y 2,1
1 L u 3, N = y3,1

M
O
M M
0 L
1 y N ,1
u1,3

y1,2
y 2, 2
y3,2
M
y N ,2

y1, N
y 2, N
y3,3 L y3, N .

M
O
M
y N ,3 L y N , N
y1,3
y 2,3

L
L

Individual l and u terms are easily found by calculating them in the following order:
1.

Starting from the top, work down Column 1 of L, finding l1,1 , then l 2,1 , then l3,1 , ,
lN,1 . For the special case of Column 1, these elements are l j ,1 = y j ,1 , j = 1,2,3, ,N.

2.

Starting from the left, work across Row 1 of U, finding u1,2 , then u1,3 , then u1,4 , ,

u1, N . For the special case of Row 1, these elements are u1,k =
3.

k 1

, k = 2,3,4, ,N.

l j,m u m,k , j = k , k + 1, k + 2,L, N

, Column k ,2 k N .

m=1

Work across Row (k = 2) of U, finding u 2,3 , then u 2,4 , then u 2,5 , , u 2, N , using
yk, j
uk, j =

5.

l1,1

Work down Column (k = 2) of L, finding l 2,2 , then l3,2 , then l 4,2 , , l N ,2 , using

l j ,k = y j , k
4.

y1,k

k 1

l k , m u m, j

m=1

l k ,k

, j = k + 1, k + 2, L , N , Row k ,2 k ( N 1) .

Repeat Steps 3 and 4, first for Column k of L, then for Row k of U. Continue for all k =
3,4,5, ,(N1) for L and U, then for k = N for L.

The procedure given above in Steps 1 - 5 is often referred to as Crout's method. Note that
elements of L and U never look "backward" for previously used elements of Y. Therefore, in
order to conserve computer memory, L and U elements can be stored as they are calculated in the
same locations at the corresponding Y elements. Thus, Crout's method is a memory-efficient "in
situ" procedure.
An intermediate column vector is needed to find V. The intermediate vector D is defined as
D = UV,
so that

Page 8 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
d1 1 u1,2
d 0 1
2
d 3 = 0 0

M
M M
d N 0 0

L u1, N V1
u 2,3 L u 2, N V2
1 L u3, N V3 .

M
O
M M
0 L
1 V N
u1,3

Since LUV = I, then LD = I. Vector D is found using forward-substitution from

0
l1,1
l
2,1 l 2,2
l3,1 l3,2

M
M
l N ,1 l N ,2

l3,3

l N ,3 L

0 d1 I1
0 d 2 I 2

0 d3 = I3 ,

M M M
l N , N d N I N

which proceeds as follows:

From Row 1, l1,1d1 = I1 , d1 =

1
l1,1

( I1 ) ,

From Row 2, l 2,1d1 + l 2,2 d 2 = I 2 , d 2 =

1
l 2,2

(I 2 l2,1d1 ) ,

From Row k, l k ,1d1 + l k ,2 d 2 + L + l k , k d k = I k , d k =

k 1

1
I k lk , j d j .

lk , k
j =1

Now, since D = UV, or

d1 1 u1,2
d 0 1
2
d 3 = 0 0

M
M M
d N 0 0

L u1, N V1
u 2,3 L u 2, N V2
1 L u3, N V3 ,

M
O
M M
0 L
1 V N
u1,3

where D and U are known, then V is found using backward substitution.

An important advantage of LU decomposition over Gaussian elimination or Kron reduction is
that the I vector is not modified during decomposition. Therefore, once Y has been decomposed
into L and U, I can be modified, and V recalculated, with minimal work, using the forward and
backward substitution steps shown above.

Page 9 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
A special form of L is helpful when Y is symmetric. In that case, let both L and U have unity
diagonal terms, and define a diagonal matrix D so that
Y = LDU ,
or
0
1
l
1
2,1
Y = l3,1 l3,2

M
M
l N ,1 l N ,2

L 0 d1,1
L 0 0
1 L 0 0

M
O 0 M
l N ,3 L 1 0
0
0

0
d 2,2
0
M
0

0 1 u1,2
0 0 1

0 0 0
d 3,3 L

M
O
M M
M
0 L d N , N 0 0
0
0

L
L

L u1, N
u 2,3 L u 2, N
1 L u 3, N .

M
O
M
0 L
1
u1,3

Since Y is symmetric, then Y = Y T , and LDU = [LDU ]T = U T D T LT = U T DLT . Therefore,

an acceptable solution is to allow L = U T . Incorporating this into the above equation yields
0
1
l
1
2,1
Y = l3,1 l3,2

M
M
l N ,1 l N ,2

L 0 d1,1
0 L 0 0
1 L 0 0

M
O 0 M
l N ,3 L 1 0
0

d1,1l 2,1
d 2,2

d1,1l3,1

d 2,2 l3,2 L

d 3,3

d1,1l N ,1
d 2,2 l N ,2
d 3,3l N ,3 ,

d N , N

which can be solved by working from top-to-bottom, beginning with Column 1 of Y, as follows:
Working down Column 1 of Y,
y1,1 = d1,1 ,
y 2,1 = l 2,1d1,1 , so l 2,1 = y 2,1 / d1,1 ,
y j ,1 = l j ,1d1,1 , so l j ,1 = y j ,1 / d1,1 .

Working down Column 2 of Y,

y 2,2 = l 2,1d1,1l 2,1 + d 2,2 , so d 2,2 = y 2,2 l 2,1d1,1l 2,1 ,

y j ,2 = l j ,1d1,1l 2,1 + l j ,2 d 2,2 , so l j ,2 = y j ,2 l j ,1d1,1l 2,1 / d 2,2 , 2 < j N .

Working down Column k of Y,

Page 10 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc

yk , k =

k 1

k 1

m =1

m =1

lk , m d m, m lk , m + d k , k , so d k , k = y k , k

l k , m d m, m l k , m ,

k 1

k 1

y j , k = l j , m d m, m l k , m , so l j , k = y j , k l j , m d m, m l k , m / d k , k , k < j N .

m =1
m =1

This simplification reduces memory and computation requirements for LU decomposition by

approximately one-half.
5.

Bifactorization

Bifactorization recognizes the simple pattern that occurs when doing "in situ" LU
decomposition. Consider the first four rows and columns of a matrix that has been LU
decomposed according to Crout's method:
l1,1
l
2,1
l 3,1

l 4,1

=
=
=
=

y1,1
u1,2 = y1,2 / l1,1
y 2,1 l 2,2 = y 2,2 l 2,1u1,2
y 3,1 l 3,2 = y 3,2 l 3,1u1,2
y 4,1 l 4,2 = y 4,2 l 4,1u1,2

u1,3 = y1,3 / l1,1

u 2,3 = y 2,3 l 2,1u1,3 / l 2, 2
l 3,3 = y 3,3 l 3,1u1,3 l 3,2 u 2,3
l 4,3 = y 4,3 l 4,1u1,3 l 4, 2 u 2,3

u1,4 = y1,4 / l1,1

u 2, 4 = y 2,4 l 2,1u1,4 / l 2,2

.
u 3,4 = y 3, 4 l 3,1u1,4 l 3,2 u 2,4 / l 3,3

l 4, 4 = y 4,4 l 4,1u1,4 l 4,2 u 2,4 l 4,3 u 3,4

The pattern developed is very similar to Kron reduction, and it can be expressed in the following
steps:
1.

Beginning with Row 1, divide the elements to the right of the pivot element, and in the
pivot row, by l1,1 , so that
y1' ,k =

2.

y1,k
y1,1

, for k = 2,3,4, , N.

Operate on the elements below and to the right of the pivot element using
y 'j ,k = y j ,k y j ,1 y1' ,k , for j = 2,3,4, , N, k = 2,3,4, , N.

3.

Continue for pivots m = 2,3,4, , (N - 1) using

'
ym
,k
'
, k = m + 1, m + 2, L , N
y m,k =
'
ym
,m

followed by
'
y 'j ,k = y 'j ,k y 'j ,m y m
,k , j = m + 1, m + 2, L , N ; k = m + 1, m + 2, L , N ,

Page 11 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
for each pivot m.
When completed, matrix Y' has been replaced by matrices L and U as follows:
l1,1 u1,2
l
2,1 l 2,2
' l
Y = 3,1 l3,2

M
M
l N ,1 l N ,2

u1,3

u 2,3 L
l3,3

l N ,3 L

u1,5
u 2,5
u 3,5 ,

l N , N

and where the diagonal u elements are unity (i.e. u1,1 = u 2,2 = u3,3 = L = u N , N = 1 ).
The corresponding FORTRAN code for bifactorization is
COMPLEX Y(N,N),YMM
C

DO FOR EACH PIVOT M = 1,2,3, ... ,N - 1

= N - 1

DO 1 M

= 1,NM1

FOR THE PIVOT ROW

2
C

NM1

MP1

= M + 1

YMM

= 1.0 / Y(M,M)

DO 2 K

= MP1,N

Y(M,K)

= Y(M,K) * YMM

CONTINUE

BELOW AND TO THE RIGHT OF THE PIVOT ELEMENT

DO 3 J

= MP1,N

DO 3 K

= MP1,N

Y(J,K)

= Y(J,K) - Y(J,M) * Y(M,K)

CONTINUE

CONTINUE
STOP
END

6.

Shipley-Coleman Inversion

For relatively small matrices, it is possible to obtain the inverse directly. The Shipley-Coleman
inversion method for inversion is popular because it is easily programmed. The algorithm is
1.

For each pivot (i.e. diagonal term) m, m = 1,2,3, ,N, perform the following Steps 2 - 4.

2.

Kron reduce all elements in Y, above and below, except those in Column m and Row m
using

Page 12 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc

y 'j , k

y 'j , k

'
y 'j , m y m
,k

y m, m

, j = 1,2,3, L , N , j m; k = 1,2,3, L , N , k m .
1

3.

'
Replace pivot element y m
,m with its negative inverse, i.e.

4.

'
Multiply all elements in Row m and Column m, except the pivot, by y m
,m .

'
ym
,m

The result of this procedure is actually the negative inverse, so that when completed, all terms
must be negated. A FORTRAN code for Shipley-Coleman is shown below.
COMPLEX Y(N,N),YPIV
C

DO FOR EACH PIVOT M = 1,2,3, ... ,N

DO 1 M

= 1,N

YPIV

= 1.0 / Y(M,M)

KRON REDUCE ALL ROWS AND COLUMNS, EXCEPT THE PIVOT ROW

AND PIVOT COLUMN

DO 2 J

= 1,N

IF(J .EQ. M) GO TO 2
DO 2 K

= 1,N

IF(K. NE. M) Y(J,K) = Y(J,K) - Y(J,M) * Y(M,K) * YPIV

2
C

CONTINUE

REPLACE THE PIVOT WITH ITS NEGATIVE INVERSE

YPIV

= -YPIV

Y(M,M)

= YPIV

WORK ACROSS THE PIVOT ROW AND DOWN THE PIVOT COLUMN,

MULTIPLYING BY THE NEW PIVOT VALUE

DO 3 K

= 1,N

IF(K .EQ. M) GO TO 3
Y(M,K) = Y(M,K) * YPIV
Y(K,M) = Y(K,M) * YPIV

CONTINUE

CONTINUE

NEGATE THE RESULTS

DO 4 J

= 1,N

DO 4 K

= 1,N

Y(J,K)

= -Y(J,K)

CONTINUE
STOP
END

Page 13 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
The order of the number of calculations for Shipley-Coleman is N 3 .
7.

Impedance Matrix

The impedance matrix is the inverse of the admittance matrix, or

Z = Y 1 ,
so that
V = ZI .

The reference bus both Y and Z is ground. Although the impedance matrix can be found via
inversion, complete inversion is not common for matrices with more than a few hundred rows
and columns because of the matrix storage requirements. In those instances, Z elements are
usually found via Gaussian elimination, Kron reduction, or, less commonly, by a direct building
algorithm. If only a few of the Z elements are needed, then Gaussian elimination or Kron
reduction are best. Both methods are described in following sections.
8.

Physical Significance of Admittance and Impedance Matrices

The physical significance of the admittance and impedance matrices can be seen by examining
the basic matrix equations I = YV and V = Y 1 I = ZI . Expanding the jth row of I = YV yields

Ij =

y j ,k Vk

. Therefore,

k =1

y j, k =

Ij
Vk

,
V m = 0, m =1, 2,L, N , m k

where, as shown in Figure 2, all busses except k are grounded, V k is a voltage source attached to
bus k, and I j is the resulting current injection at (i.e. flowing into) bus j. Since all busses that
neighbor bus k are grounded, the currents induced by V k will not get past these neighbors, and
the only non-zero injection currents I j will occur at the neighboring busses. In a large power
system, most busses do not neighbor any arbitrary bus k. Therefore, Y consists mainly of zeros
(i.e. is sparse) in most power systems.

Page 14 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
Induced Current at
Bus j

Applied Voltage
at Bus k
Power System
+

A
Vk

Ij

All Other Busses

Grounded

Figure 2. Measurement of Admittance Matrix Term y j , k

Concerning Z, the kth row of V = Y

z j, k =

I = ZI yields Vk =

Vj
Ik

zk , j I j

. Hence,

j =1

,
I m = 0, m =1, 2,L, N , m k

where, as shown in Figure 3, Ik is a current source attached to bus k, Vj is the resulting voltage at
bus j, and all busses except k are open-circuited. Unless the network is disjoint, then current
injection at one bus, when all other busses open-circuited, will raise the potential everywhere in
the network. For that reason, Z tends to be full.
Induced Voltage
at Bus j

Applied Current at
Bus k
Power System

+
V

Ik

All Other Busses

Open Circuited

Vj
-

Figure 3. Measurement of Impedance Matrix Term z j , k

9.

Formation of the Impedance Matrix via Gaussian Elimination, Kron Reduction,

and LU Decomposition
Gaussian Elimination

An efficient method of fully or partially inverting a matrix is to formulate the problem using
Gaussian elimination. For example, given

Page 15 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc

YZ = I ,
where Y is a matrix of numbers, Z is a matrix of unknowns, and I is the identity matrix, the
objective is to Gaussian eliminate Y, while performing the same row operations on I, to obtain
the form

Y Z = I ,
where Y' is in (URT+D) form. Then, individual columns of Z can then be solved using backward
substitution, one at a time. This procedure is also known as the augmentation method, and it is
illustrated as follows:

Y is first Gaussian eliminated, as shown in a previous section, while at the same time performing
identical row operations on I. Afterward, the form of Y Z = I is
y1,1

0
0

M
0

y1,2
'

y 2,2
0
M
0
1
I'
2,1
I 3' ,1

M
I '
N ,1

y1,3
'

y 2,3 L
y '3,3 L

y1, N z1,1

y ' 2, N z 2,1
y ' 3, N z 3,1

M M
y ' N , N z N ,1

M
0

O
L

0
0
I 3' ,3 L
0

M
O
M
'
'

IN
,3 L I N , N

I 2' ,2
I 3' ,2
M

'
IN
,2

0
0

z1,2

z1,3

z 2,2

z 2,3

z 3,2

z 3,3

z N ,2

z N ,3 L

z1, N
z 2, N
z 3, N =

M
z N , N

L
L

where Rows 1 of Y' and I' are the same as in Y and I. The above equation can be written in
abbreviated column form as
y1,1

0
0

M
0

y1,2
'

y 2,2
0
M
0

y1,3
'

y 2,3 L
y ' 3,3 L
M
0

O
L

y1, N

y ' 2, N
y '3, N [Z1

M
y ' N , N

Z2

Z 3 L Z N ] = I1'

I 2'

'
,
I 3' L I N

where the individual column vectors Zi and Ii have dimension N x 1. The above equation can be
solved as N separate subproblems, where each subproblem computes a column of Z. For
example, the kth column of Z can be computed by applying backward substitution to

Page 16 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc

y1,1

0
0

M
0

y1,2

y1,3

y ' 2,2
0
M
0

y ' 2,3
y ' 3,3
M
0

L
L
L
O
L

y1, N Z1, k I1' , k

y ' 2, N Z 2, k I 2' , k
y ' 3, N Z 3, k = I 3' , k .

M M M
'
y ' N , N Z N , k I N
, k

Each column of Z is solved independently from the others.

Kron Reduction

If Kron reduction, the problem is essentially the same, except that Row 1 of the above equation
is divided by y1' ,1 , yielding

y1,1

0
0

M
0

y1,2

y1,3

y ' 2,2
0
M
0

y ' 2,3
y ' 3,3
M
0

L
L
L
O
L

y1, N Z1, k I1' , k

y ' 2, N Z 2, k I 2' , k
y ' 3, N Z 3, k = I 3' , k .

M M M
'
y ' N , N Z N , k I N
, k

Because backward substitution can stop when the last desired z element is computed, the process
is most efficient if the busses are ordered so that the highest bus numbers (i.e. N, N-1, N-2, etc.)
correspond to the desired z elements. Busses can be ordered accordingly when forming Y to take
advantage of this efficiency.
LU Decomposition

Concerning LU decomposition, once the Y matrix has been decomposed into L and U, then we
have
YZ = LUZ = I ,

where I is the identity matrix. Expanding the above equation as L[UZ ] = I yields
0
l1,1
l
2,1 l 2,2
l3,1 l 2,3

M
M
l N ,1 l N ,2

l3,3

l N ,3 L

0 uz1,1 uz1,2
0 uz 2,1 uz 2,2
0 uz 3,1 uz 3,2

M M
M
l N , N uz N ,1 uz N ,2

Page 17 of 33

uz1,3

uz 2,3

uz 3,3

uz N ,3 L

uz1, N
uz 2, N
uz 3, N = .

M
uz N , N

_05_EE394J_2_Spring12_Power_System_Matrices.doc

1
0

M
0

0
1
0
M
0

0
0
1
M
0

L
L
L
O
L

0
0
0

M
1

The special structure of the above equation shows that, in general, UZ must be (LLT+D) in form.
UZ can be found by using forward substitution on the above equation, and then Z can be found,
one column at a time, by using backward substitution on UZ = [U ][Z ] , which in expanded form
is
1 u1,2
0 1

0 0

M
M
0 0

L u1, N z1,1

u 2,3 L u 2, N z 2,1
1 L u 3, N z 3,1

M
O
M M
0 L
1 z N ,1
u1,3

0
uz1,1
uz
2,1 uz 2,2
uz 3,1 uz 3,2

M
M
uz N ,1 uz N ,2

10.

0
0
uz 3,3
M
uz N ,3

z1,2

z1,3

z 2,2

z 2,3

z 3,2

z 3,3

z N ,2

z N ,3 L

z1, N
z 2, N
z 3, N =

M
z N , N

L
0
L
0
L
0 .

O
M
L uz N , N

Direct Impedance Matrix Building and Modification Algorithm

The impedance matrix can be built directly without need of the admittance matrix, if the physical
properties of Z are exploited. Furthermore, once Z has been built, no matter which method was
used, it can be easily modified when network changes occur. This after-the-fact modification
capability is the most important feature of the direct impedance matrix building and modification
algorithm.
The four cases to be considered in the algorithm are
Case 1.

Add a branch impedance between a new bus and the reference bus.

Case 2.

Add a branch impedance between a new bus and an existing non-reference bus.

Case 3.

Add a branch impedance between an existing bus and the reference bus.

Case 4.

Add a branch impedance between two existing non-reference busses.

Direct formation of an impedance matrix, plus all network modifications, can be achieved
through these four cases. Each of the four is now considered separately.
Page 18 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
Case 1, Add a Branch Impedance Between a New Bus and the Reference Bus

Case 1 is applicable in two situations. First, it is the starting point when building a new
impedance matrix. Second, it provides (as does Case 2) a method for adding new busses to an
existing impedance matrix.
Both situations applicable to Case 1 are shown in Figure 4. The starting-point situation is on the
left, and the new-bus situation is on the right.
Situation 1

Situation 2

Bus 1

Power System
N Busses
Zadd

Bus (N+1)

Zadd
Reference
Bus

Reference
Bus

Figure 4. Case 1, Add a Branch Impedance Between a New Bus and the Reference Bus
The impedance matrices for the two situations are
Situation 1

Z1,1 = Zadd ,

Z
N , N NxN
Situation 2: Z N +1, N +1 =

[0 0 L 0]1xN

.
M

0 Nx1
[Zadd ]1x1

The effect of situation 2 is simply to augment the existing Z N , N by a column of zeros, a row of
zeros, and a new diagonal element Zadd. New bus (N+1) is isolated from the rest of the system.
Case 2, Add a Branch Impedance Between a New Bus and an Existing Non-Reference Bus

Consider Case 2, shown in Figure 5, where a branch impedance Zadd is added from existing
non-reference bus j to new bus (N+1). Before the addition, the power system has N busses, plus
a reference bus (which is normally ground), and an impedance matrix Z N , N .

Page 19 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc

Bus j

Induced Voltage
at Bus k

Zadd

Injected Current at
New Bus (N+1)

Power System
+

I(N+1)

Vk
-

Figure 5. Case 2, Add a Branch Impedance Between New Bus (N+1) and Existing NonReference Bus j
Since all of the current I N +1 injected at new bus (N+1) flows through Zadd and into Bus j, the
original N power system busses cannot distinguish between current injections I j and I N +1 .
Therefore,

Vk
V
= k , k ( N + 1) ,
I N +1 I j
meaning that impedance matrix elements z k , j and zk , N +1 are identical for busses k ( N + 1) ,
and that the effect on the impedance matrix is to augment it with an additional Row (N+1) that is
identical to Row j.
Likewise, since bus (N+1) is an open-circuited radial bus stemming from bus j, a current injected
at another bus k creates identical voltage changes on busses j and (N+1). This means that

V N +1 V j
=
, k ( N + 1) ,
I k
I k
so that the effect on the impedance matrix is to augment it with an additional Column (N+1) that
is identical to Column j.
Next, due to the fact that all of injection current I N +1, N +1 passes through new branch impedance
Zadd, the relationship between V j and VN +1 is

V j
V N +1
=
+ Zadd ,
I N +1 I N +1
so that the (N+1) diagonal term of the impedance matrix becomes
z N +1, N +1 = z j , j + Zadd ,

Page 20 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
where z j , j is the jth diagonal term of Z NxN .
Summarizing, the total effect of Case 2 is to increase the dimension of the impedance matrix by
one row and column according to

Z N , N NxN

Z N +1, N +1 =
Row j of Z N , N 1xN

[Column j of Z N , N ]Nx1

[z j, j + Zadd ]1x1

Case 3, Add a Branch Impedance Between an Existing Bus and the Reference Bus

Now, consider Case 3, where a new impedance-to-reference tie Zadd is added to existing Bus j.
The case is handled as a two-step extension of Case 2, as shown in Figure 4.6. First, extend
Zadd from Bus j to fictitious Bus (N+1), as was done in Case 2. Then, tie fictitious Bus (N+1)
to the reference bus, and reduce the size of the augmented impedance matrix back to (N x N).
Step 1

Step 2

New Bus (N+1)

Bus j

Bus j

Zadd
Power System

Power System

Zadd

Figure 6. Case 3. Two-Step Procedure for Adding Branch Impedance Zadd from Existing Bus j
to the Reference Bus
Step 1 creates the augmented Z N +1, N +1 matrix shown in Case 2.
V N +1 = Z N +1, N +1 I N is

Z N,N
V Nx1
NxN
=

Row
of
j
Z
=
0
N , N 1xN
REF

The form of equation

[Column j of Z N , N ]Nx1 I Nx1

[z j, j + Zadd ]1x1

,
I REF

where V REF , z j , j + Zadd , and I REF are scalars. Defining the row and column vectors as R j
and C j , respectively, yields

V Nx1 Z N , N NxN
V
=
REF = 0 R j 1xN

[ ]

[C j ]Nx1

I Nx1

.
z j , j + Zadd
I

REF

1x1

At this point, scalar I REF can be eliminated by expanding the bottom row to obtain

Page 21 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc

I REF =

[R j I Nx1 ]1x1
z j , j + Zadd

Substituting into the top N rows yields

V Nx1 = Z NxN I Nx1

1
1
C j R j I Nx1 = Z NxN
C jRj

+
z j , j + Zadd
z
Zadd
j
j
,

) I Nx1 ,

or
'
V Nx1 = Z NxN
I Nx1 .

Expanding C j R j gives
z1, j
z
2, j
z 3, j z j ,1

M
z N , j

z j ,2

z1, j z j ,1
z z
2, j j ,1
z j ,3 L z j , N = z 3, j z j ,1

z N , j z j ,1

z1, j z j ,2
z 2, j z j ,2
z 3, j z j ,2
M
z N , j z j ,2

z1, j z j ,3
z 2, j z j ,3
z 3, j z j ,3
M
z N , j z j ,3

L z1, j z j , N
L z 2, j z j , N
L z 3, j z j , N ,

O
M

L z N , j z j , N

'
so that the individual elements of the modified impedance matrix Z Nx
1 can be written as

z i' , k = z i, k

zi, j z j , k
z j , j + Zadd

, i = 1,2,3,L, N ; k = 1,2,3,L, N .

Note that the above equation corresponds to Kron reducing the augmented impedance matrix,
using element z N +1, N +1 as the pivot.
Case 4, Add a Branch Impedance Between Two Existing Non-Reference Busses

The final case to be considered is that of adding a branch between existing busses j and k in an
N-bus power system that has impedance matrix Z N , N . The system and branch are shown in
Figure 4.7.

Page 22 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
Zadd

Ib

Ib

Bus k

Bus j
Power System
with N Busses

Ik + Ib

Ij - Ib

Ij

Ik

Reference Bus

Figure 7. Case 4, Add Branch Impedance Zadd between Existing Busses j and k
As seen in the figure, the actual current injected into the system via Bus j is I j I b , and the
actual injection via Bus k is I k + I b , where
Ib =

V j Vk
, or V j Vk = (Zadd )I b .
Zadd

The effect of the branch addition on the voltage at arbitrary bus m can be found by substituting
the true injection currents at busses j and k into V = Z N , N I , yielding
Vm =

z m,1 I1

z m, 2 I 2

+ L z m, j I j I b

z m ,3 I 3

z m, k ( I k + I b ) + L +

+ L +

z m, N I N ,

or
Vm =

zm,1I1 +

zm, 2 I 2

zm,3 I3

+ L zm, j I j

L +

z m, N I N

+ L +

+ L +

zm, k I k

(zm, j zm,k )Ib

For busses j and k specifically,

V j = z j ,1 I 1
V k = z k ,1 I 1

z j ,2 I 2

z j ,3 I 3

z k ,2 I 2

z k ,3 I 3

+ L +

+ L +

(
(

z j, j I j I b
zk, j I j Ib

z j ,k (I k + I b ) + L +
z k , k (I k + I b ) + L +

Then, V j Vk gives

Page 23 of 33

)
)

+ L +

+ L +

z j, N I N
.
z k,N I N

_05_EE394J_2_Spring12_Power_System_Matrices.doc

)(

V j Vk = z j ,1 z k ,1 I1 + z j ,2 z k ,2 I 2 + z j ,3 z k ,3 I 3 + L + z j , j z k , j I j I b + L

(z j, k z k , k )(I k + I b ) + L + (z j, N z k , N )I N

Substituting in V j Vk = (Zadd )I b and combining terms then yields

0 = z j ,1 z k ,1 I1 + z j ,2 z k ,2 I 2 + z j ,3 z k ,3 I 3 + L + z j , j z k , j I j + L

(z j, k z k , k )I k + L + (z j, N z k , N )I N (z j, j + z k , k z j, k z k , j + Zadd )I b .
All of the above effects can be included as an additional row and column in equation
V = Z N , N I as follows:

[Col j of Z N , N Col k of Z N ,N ]Nx1

Z N ,N
[V N ]Nx1
NxN
[0]
= Row j of Z

R
ow
k of Z N , N
N ,N
Nx1

1xN

[z j, j + z k ,k z j,k z k , j + Zadd ]Nx1 .

[I N ]Nx1
[ I ] .
b 1x1

As was done in Case 3, the effect of the augmented row and column of the above equation can be
incorporated into a modified impedance matrix by using Kron reduction, where element z N +1, N +1
is the pivot, yielding
z i' , m = z i, m

(zi, j zi, k )(z j, m z k , m )

z j , j + z k , k z j , k z k , j + Zadd

Application Notes

Once an impedance matrix is built, no matter which method is used, the modification algorithm
can very easily adjust Z for network changes. For example, a branch outage can be effectively
achieved by placing an impedance, of the same but negative value, in parallel with the actual
impedance, so that the impedance of the parallel combination is infinite.
11.
c
c
c

Example Code YZBUILD for Building Admittance Matrix and Impedance Matrix
program yzbuild.

10/10/04.

implicit none
complex ybus,yline,ycap,ctap,yii,ykk,yik,yki
complex lmat,umat,unity,czero,ytest,ysave,diff,uz,
1 uzelem,zbus,elem,alpha
integer nb,nbus,nfrom,nto,j,k,ipiv,jdum,kdum,jm1,kp1,l
integer nlu,ipiv1,nm1,ny,nbext,nbint,iout,mxext,mxint
integer jp1,ipp1,ngy,ngu
real eps4,eps6,eps9,pi,dr,qshunt,r,x,charge,fill,ymag,umag

Page 24 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc

dimension ybus(150,150),nbext(150),nbint(9999)
dimension lmat(150,150),umat(150,150),ysave(150,150),
1 uz(150,150),zbus(150,150),unity(150,150)
data mxint/150/,mxext/9999/,nbus/0/, nbint/9999 * 0/, iout/6/,
1 nbext/150 * 0/,czero /(0.0,0.0)/,eps4/1.0e-04/,
2 eps6/1.0e-06/,eps9/1.0e-09/
data ybus /22500 * (0.0,0.0)/
data zbus /22500 * (0.0,0.0)/
data uz
/22500 * (0.0,0.0)/
data unity /22500 * (0.0,0.0)/
c
pi = 4.0 * atan(1.0)
dr = 180.0 / pi
open(unit=1,file='uz.txt')
write(1,*) 'errors'
close(unit=1,status='keep')
open(unit=1,file='zbus_from_lu.txt')
write(1,*) 'errors'
close(unit=1,status='keep')
open(unit=1,file='ybus_gaussian_eliminated.txt')
write(1,*) 'errors'
close(unit=1,status='keep')
open(unit=1,file='unity_mat.txt')
write(1,*) 'errors'
close(unit=1,status='keep')
open(unit=1,file='zbus_from_gaussian_elim.txt')
write(1,*) 'errors'
close(unit=1,status='keep')
open(unit=6,file='yzbuild_screen_output.txt')
open(unit=2,file='ybus.txt')
open(unit=1,file='bdat.csv')
write(6,*) 'program yzbuild, 150 bus version, 10/10/04'
write(2,*) 'program yzbuild, 150 bus version, 10/10/04'
write(6,*) 'read bus data from file bdat.csv'
write(6,*) 'number
b'
write(2,*) 'read bus data from file bdat.csv'
write(2,*) 'number
b'
c
c
c

read the bdat.csv file

1 read(1,*,end=5,err=30) nb,qshunt
if(nb .eq. 0) go to 5
write(6,1002) nb,qshunt
write(2,1002) nb,qshunt
1002 format(i7,5x,f10.4)
if(nb .le. 0 .or. nb .gt. mxext) then
write(6,*) 'illegal bus number - stop'
write(2,*) 'illegal bus number - stop'
write(*,*) 'illegal bus number - stop'
stop
endif
nbus = nbus + 1
if(nbus .gt. mxint) then
write(6,*) 'too many busses - stop'
write(2,*) 'too many busses - stop'
write(*,*) 'too many busses - stop'
stop

Page 25 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
endif
nbext(nbus) = nb
nbint(nb)
= nbus
ycap
= cmplx(0.0,qshunt)
ybus(nbus,nbus) = ybus(nbus,nbus) + ycap
go to 1
5 close(unit=1,status='keep')
c
open(unit=1,file='ldat.csv')
write(6,*)
write(6,*) 'read line/transformer data from file ldat.csv'
write(6,*) ' from
to
r
x',
1 '
b'
write(2,*) 'read line/transformer data from file ldat.csv'
write(2,*) ' from
to
r
x',
1 '
b'
c
c
c

read the ldat.csv file

10 read(1,*,end=15,err=31) nfrom,nto,r,x,charge
if(nfrom .eq. 0 .and. nto .eq. 0) go to 15
write(6,1004) nfrom,nto,r,x,charge
write(2,1004) nfrom,nto,r,x,charge
1004 format(1x,i5,2x,i5,3(2x,f10.4))
if(nfrom .lt. 0 .or. nfrom .gt. mxext .or. nto .lt. 0
1 .or. nto .gt. mxext) then
write(6,*) 'illegal bus number - stop'
write(2,*) 'illegal bus number - stop'
write(*,*) 'illegal bus number - stop'
stop
endif
if(nfrom .eq. nto) then
write(6,*) 'same bus number given on both ends - stop'
write(2,*) 'same bus number given on both ends - stop'
write(*,*) 'same bus number given on both ends - stop'
stop
endif
if(r .lt. -eps6) then
write(6,*) 'illegal resistance - stop'
write(2,*) 'illegal resistance - stop'
write(*,*) 'illegal resistance - stop'
stop
endif
if(charge .lt. -eps6) then
write(6,*) 'line charging should be positive'
write(2,*) 'line charging should be positive'
write(*,*) 'line charging should be positive'
stop
endif
if(nfrom .ne. 0) then
nfrom = nbint(nfrom)
if(nfrom .eq. 0) then
write(6,*) 'bus not included in file bdat.csv - stop'
write(2,*) 'bus not included in file bdat.csv - stop'
write(*,*) 'bus not included in file bdat.csv - stop'
stop
endif
endif
if(nto
.ne. 0) then
nto
= nbint(nto)

Page 26 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
if(nto .eq. 0) then
write(6,*) 'bus not included in file bdat.csv - stop'
write(2,*) 'bus not included in file bdat.csv - stop'
write(*,*) 'bus not included in file bdat.csv - stop'
stop
endif
endif
yline = cmplx(r,x)
if(cabs(yline) .lt. eps6) go to 10
yline = 1.0 / yline
ycap = cmplx(0.0,charge / 2.0)
c
c
c

the line charging terms

if(nfrom .ne. 0) ybus(nfrom,nfrom) = ybus(nfrom,nfrom) + ycap
if(nto
.ne. 0) ybus(nto ,nto ) = ybus(nto ,nto ) + ycap

c
c
c

shunt elements
if(nfrom .ne. 0 .and. nto
1 ybus(nfrom,nfrom) + yline
if(nfrom .eq. 0 .and. nto
1 ybus(nto ,nto ) + yline

c
c
c

.ne. 0) ybus(nto

,nto

) =

transmission lines
if(nfrom .ne. 0 .and.
ybus(nfrom,nto ) =
ybus(nto ,nfrom) =
ybus(nfrom,nfrom) =
ybus(nto ,nto ) =
endif
go to 10

c
c
c

.eq. 0) ybus(nfrom,nfrom) =

nto .ne. 0) then

ybus(nfrom,nto )
ybus(nto ,nfrom)
ybus(nfrom,nfrom)
ybus(nto ,nto )

+
+

yline
yline
yline
yline

write the nonzero ybus elements to file ybus

15 close(unit=1,status='keep')
write(6,*)
write(6,*) 'nonzero elements of ybus (in rectangular form)'
write(2,*) 'nonzero elements of ybus (in rectangular form)'
write(6,*) '-internal-external-'
write(2,*) '-internal-external-'
ny = 0

do j = 1,nbus
do k = 1,nbus
ysave(j,k) = ybus(j,k)
if(cabs(ybus(j,k)) .ge. eps9) then
ny = ny + 1
write(6,1005) j,k,nbext(j),nbext(k),ybus(j,k)
write(2,1005) j,k,nbext(j),nbext(k),ybus(j,k)
1005 format(2i5,3x,2i5,2e20.8)
endif
end do
end do
close(unit=2,status='keep')
fill = ny / float(nbus * nbus)
write(iout,525) ny,fill

Page 27 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
525 format(/1x,'number of nonzero elements in ybus = ',i5/1x,
1'percent fill = ',2pf8.2/)
c
c
c

bifactorization - replace original ybus with lu

nm1 = nbus - 1
do ipiv = 1,nm1
write(iout,530) ipiv
530 format(1x,'lu pivot element = ',i5)
ipiv1 = ipiv + 1
alpha = 1.0 / ybus(ipiv,ipiv)
do k = ipiv1,nbus
ybus(ipiv,k) = alpha * ybus(ipiv,k)
end do
do j = ipiv1,nbus
alpha
= ybus(j,ipiv)
do k = ipiv1,nbus
ybus(j,k) = ybus(j,k) - alpha * ybus(ipiv,k)
end do
end do
end do
write(iout,530) nbus
write(iout,532)
532 format(/1x,'nonzero lu follows'/)
open(unit=4,file='lu.txt')
nlu = 0
do j = 1,nbus
do k = 1,nbus
ymag = cabs(ybus(j,k))
if(ymag .gt. eps9) then
nlu = nlu + 1
write(iout,1005) j,k,nbext(j),nbext(k),ybus(j,k)
write(4,1005) j,k,nbext(j),nbext(k),ybus(j,k)
endif
end do
end do
fill = nlu / float(nbus * nbus)
write(iout,535) nlu,fill
535 format(/1x,'number of nonzero elements in lu = ',i5/1x,
1'percent fill = ',2pf8.2/)
close(unit=4,status='keep')

c
c
c

check l times u
write(iout,560)
560 format(1x,'lu .ne. ybus follows'/)
do j = 1,nbus
do k = 1,nbus
lmat(j,k) = czero
umat(j,k) = czero
if(j .ge. k) lmat(j,k) = ybus(j,k)
if(j .lt. k) umat(j,k) = ybus(j,k)
end do
umat(j,j) = 1.0
end do

Page 28 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc

do j = 1,nbus
do k = 1,nbus
ytest
= czero
do l = 1,k
ytest = ytest + lmat(j,l) * umat(l,k)
end do
diff = ysave(j,k) - ytest
if(cabs(diff) .gt. eps4) write(iout,1005) j,k,nbext(j),
1
nbext(k),diff
end do
end do
c
c
c

form uz (urt + diag)

write(iout,536) nbus
536 format(1x,'uz pivot = ',i5)
uz(nbus,nbus) = 1.0 / ybus(nbus,nbus)
nm1 = nbus - 1
do kdum = 1,nm1
k
= nbus - kdum
write(iout,536) k
uz(k,k) = 1.0 / ybus(k,k)
kp1
= k + 1
do j = kp1,nbus
uzelem = czero
jm1
= j - 1
alpha
= 1.0 / ybus(j,j)
do l = k,jm1
uzelem = uzelem - ybus(j,l) * uz(l,k)
end do
uz(j,k) = uzelem * alpha
end do
end do
write(iout,537)
537 format(/1x,'nonzero uz follows'/)
open(unit=8,file='uz.txt')
nlu
= 0
do j = 1,nbus
do k = 1,nbus
ymag
= cabs(uz(j,k))
if(ymag .gt. eps9) then
nlu
= nlu + 1
write(iout,1005) j,k,nbext(j),nbext(k),uz(j,k)
write(8,1005) j,k,nbext(j),nbext(k),uz(j,k)
endif
end do
end do
fill = nlu / float(nbus * nbus)
write(iout,545) nlu,fill
545 format(/1x,'number of nonzero elements in uz = ',i5/1x,
1'percent fill = ',2pf8.2/)
close(unit=8,status='keep')

c
c
c

form z

Page 29 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
open(unit=10,file='zbus_from_lu.txt')
do kdum = 1,nbus
k
= nbus - kdum + 1
write(iout,550) k
550 format(1x,'zbus column = ',i5)
do jdum = 1,nbus
j
= nbus - jdum + 1
zbus(j,k) = czero
if(j .ge. k) zbus(j,k) = uz(j,k)
if(j .ne. nbus) then
jp1 = j + 1
do l
= jp1,nbus
zbus(j,k) = zbus(j,k) - ybus(j,l) * zbus(l,k)
end do
end if
end do
end do
write(iout,565)
565 format(/1x,'writing zbus_from_lu.txt file')
do j = 1,nbus
do k = 1,nbus
write(10,1005) j,k,nbext(j),nbext(k),zbus(j,k)
write( 6,1005) j,k,nbext(j),nbext(k),zbus(j,k)
end do
end do
close(unit=10,status='keep')
c
c
c

check ybus * zbus

write(iout,5010)
5010 format(/1x,'nonzero ybus * zbus follows')
do j = 1,nbus
do k = 1,nbus
elem
= czero
do l = 1,nbus
elem
= elem + ysave(j,l) * zbus(l,k)
end do
ymag
= cabs(elem)
if(ymag .gt. eps4) write(iout,1005) j,k,nbext(j),nbext(k),elem
end do
end do
c
c
c

copy ysave back into ybus, and zero zbus

do j = 1,nbus
unity(j,j) = 1.0
do k = 1,nbus
ybus(j,k) = ysave(j,k)
zbus(j,k) = czero
end do
end do

c
c
c
c

gaussian eliminate ybus, while performing the same operations

on the unity matrix

Page 30 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
nm1 = nbus - 1
do ipiv = 1,nm1
write(iout,561) ipiv
561 format(1x,'gaussian elimination ybus pivot = ',i5)
alpha
= 1.0 / ybus(ipiv,ipiv)
ipp1 = ipiv + 1
c
c
c

pivot row operations for ybus and unity

do k = 1,nbus
if(k .gt. ipiv) then
ybus(ipiv,k) = ybus(ipiv,k) * alpha
else
unity(ipiv,k) = unity(ipiv,k) * alpha
endif
end do

c
c
c

pivot element
ybus(ipiv,ipiv) = 1.0

c
c
c
c

kron reduction of ybus and unity below the pivot row and to
the right of the pivot column

1
1
c
c
c

do j = ipp1,nbus
alpha
= ybus(j,ipiv)
do k = 1,nbus
if(k .gt. ipiv)
ybus(j,k) = ybus(j,k) - alpha * ybus(ipiv,k)
if(k .lt. ipiv)
unity(j,k) = unity(j,k) - alpha * unity(ipiv,k)
end do
end do

elements directly below the pivot element

do j = ipp1,nbus
alpha
= ybus(j,ipiv)
ybus(j,ipiv) = ybus(j,ipiv) - alpha * ybus(ipiv,ipiv)
unity(j,ipiv) = unity(j,ipiv) - alpha * unity(ipiv,ipiv)
end do
end do

c
c
c

last row
write(iout,561) nbus
alpha
= 1.0 / ybus(nbus,nbus)
do k = 1,nbus
unity(nbus,k) = unity(nbus,k) * alpha
end do
ybus(nbus,nbus) = 1.0
ngy = 0
ngu = 0
write(iout,562)
562 format(/1x,'writing gaussian eliminated ybus and unity to files')
open(unit=12,file='ybus_gaussian_eliminated.txt')
open(unit=13,file='unity_mat.txt')

Page 31 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc

do j = 1,nbus
do k = 1,nbus
ymag = cabs(ybus(j,k))
if(ymag .ge. eps9) then
write(12,1005) j,k,nbext(j),nbext(k),ybus(j,k)
ngy
= ngy + 1
endif
umag = cabs(unity(j,k))
if(umag .ge. eps9) then
write(13,1005) j,k,nbext(j),nbext(k),unity(j,k)
ngu
= ngu + 1
endif
end do
end do
close(unit=12,status='keep')
close(unit=13,status='keep')
fill = ngy / float(nbus * nbus)
write(iout,555) ngy,fill
555 format(/1x,'number of nonzero elements in gaussian eliminated',
1' ybus = ',i5/1x,
1'percent fill = ',2pf8.2/)
fill = ngu / float(nbus * nbus)
write(iout,556) ngu,fill
556 format(/1x,'number of nonzero elements in gaussian eliminated',
1' unity matrix = ',i5/1x,
1'percent fill = ',2pf8.2/)
c
c
c

back substitute to find z

do k = 1,nbus
zbus(nbus,k) = unity(nbus,k)
end do
do jdum = 2,nbus
j = nbus - jdum + 1
write(iout,550) j
do kdum = 1,nbus
k = nbus - kdum + 1
zbus(j,k) = unity(j,k)
jp1 = j + 1
do l = jp1,nbus
zbus(j,k) = zbus(j,k) - ybus(j,l) * zbus(l,k)
end do
end do
end do
write(iout,566)
566 format(/1x,'writing zbus_from_gaussian_elim.txt file')
open(unit=11,file='zbus_from_gaussian_elim.txt')
do j = 1,nbus
do k = 1,nbus
write(11,1005) j,k,nbext(j),nbext(k),zbus(j,k)
write( 6,1005) j,k,nbext(j),nbext(k),zbus(j,k)
end do
end do
close(unit=11,status='keep')

Page 32 of 33

_05_EE394J_2_Spring12_Power_System_Matrices.doc
c
c
c

check ybus * zbus

write(iout,5010)
do j = 1,nbus
do k = 1,nbus
elem
= czero
do l = 1,nbus
elem
= elem + ysave(j,l) * zbus(l,k)
end do
ymag
= cabs(elem)
if(ymag .gt. eps4) write(iout,1005) j,k,nbext(j),nbext(k),elem
end do
end do
stop
30 write(iout,*)
write(*,*)
stop
31 write(iout,*)
write(*,*)
stop

'error in reading bdat.csv file - stop'

'error in reading bdat.csv file - stop'
'error in reading ldat.csv file - stop'
'error in reading ldat.csv file - stop'

end

Page 33 of 33