Lecture 20 Power Engineering - Egill Benedikt Hreinsson 1

Sparse Matrices in Power Flow

Calculations
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 2

Sparse Matrices?

Any matrix that has a high proportion of its

elements equal to zero is a sparse matrix.

•Most large matrices that arise

in engineering applications are
sparse.
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 3

The Sparsity of a Matrix - A definition

A sparse matrix: A matrix, where most of the elements

are zero, but a few are different from zero

A full matrix: A matrix, where all the elements are non-

zero

The sparsity = Number of elements =0

⋅ 100 %
of a matrix Total number of elements
 Why and What Are Sparse Matrices in
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 4

Power Systems?
• Matrices in power systems are often very large and very sparse!
• A large matrix: is for instance of dimension 100x100,
1000x1000 or 10000x10000!
• A sparse matrix: A matrix where, for instance, 99% of the
elements are zero, but 1% are different from zero.
• Examples of these matrices are the Ybus matrix and the Jacobi-
matrix, which both are used in power flow calculations.
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 5

Ybus Matrix Composition

• The diagonal elements of the Ybus
We review the rule for matrix consist of the sum of the (series and
building the Ybus matrix: shunt) admittances connected to the bus in
question
⎡ y11 y12 y1n ⎤ • These elements are always
⎢y y ⎥ present!
y1n ⎥ • The off-diagonal elements of the Ybus
Ybus = ⎢ 21 22
matrix consist of the negative value of the
⎢ ⎥ series admittances connecting the 2 buses
⎢ ⎥ in question. These elements will be zero
when no direct connection is between
⎣ y n1 y n 2 ynn ⎦
buses. (The Ybus matrix is very sparse!)
• These elements are present
only when there is a link (a line
γ ik or transformer)
yik = yik e
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 6

The Sparsity of a Matrix: Example 1:

An example: Assume, in an electrical power system, each bus is on the
average connected to 1.5 other buses. Further assume that we have a 100 bus
system. We get a 100 by 100 matrix with 100 · 100 elements.

With the above assumption in the Ybus matrix there are 100 diagonal elements ≠ 0 and
150 elements above the diagonal and 150 elements below the diagonal ≠ 0. The
sparsity of this matrix will therefore be:

100 ⋅ 100 − 150 − 150 − 100

Sparsity = 100 ⋅ 100
Sparsity = 96,0 %
We can say that the matrix is 4% full.
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 7

The Sparsity of a Matrix: Example 2:

Similarly for a 1000 bus system we get a matrix with 1000 · 1000 elements. We
have for the same assumption of how each bus is interconnected for the Ybus matrix 1000
diagonal elements ≠ 0 and 1500 elements above the diagonal and 1500 elements
below the diagonal ≠ 0. The sparsity of this matrix will be:

1000 ⋅1000 − 1500 − 1500 − 1000

Sparsity = 1000 ⋅1000
Sparsity = 99, 6%

We can perhaps say that the matrix is 0,4% full.

Lecture 20 Power Engineering - Egill Benedikt Hreinsson 8

The Sparsity of the Jacobi-Matrix

The J1 sub-matrix : The J3 sub-matrix
∂Pi n
∂Qi
= − ∑ Vi Vk yik ⋅ sin (δ i − δ k − γ ik ) Vi V j yij ⋅ cos (δ i − δ j − γ ij )
n

∂δ i k =1, k ≠ i ∂δ i
= ∑
j =1, j ≠ i

∂Pi ∂Qi
= Vi Vk yik ⋅ sin (δ i − δ k − γ ik ) i≠k = − Vi Vk yik ⋅ cos (δ i − δ k − γ ik ) k ≠i
∂δ k ∂δ k

The J2 sub-matrix : The J4 sub-matrix :

∂Pi ∂Qi
= 2 Vi yii ⋅ cos ( −γ ii ) + ∑ V j yij ⋅ cos (δ i − δ j − γ ij ) = 2 Vi yii ⋅ sin ( −γ ii ) + ∑ V j yij ⋅ sin (δ i − δ j − γ ij )
n n

∂ Vi j =1, j ≠ i ∂ Vi j =1, j ≠ i

∂Pi ∂Qi
= Vi yik ⋅ cos (δ i − δ k − γ ik ) k ≠i = Vi yik ⋅ sin (δ i − δ k − γ ik )
∂ Vk k ≠i
∂ Vk

The elements of Ybus are also in the Jacobi matrices

Conclusion: If Ybus is a sparse matrix, the Jacobi matrix is also sparse
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 9

Random Sparse Matrices

A Random Sparse Matrix

Sparse matrices can

be of different nature
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 10

IEEE TEST systems

• IEEE test systems are standardized power systems to test
research theories in power system analysis
• We will look at different sparse Ybus matrices for different
IEEE test systems along with their one-line diagram:
– 14 bus
– 30 bus
– 57 bus
– 118 bus
– 300 bus
• These test systems can be found at:
– http://www.ee.washington.edu/research/pstca/
– (http://www.pserc.cornell.edu/matpower/)
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 11

IEEE 14 BUS
TEST CASE
0

10

150 5 10 15
nz = 54
Use the SPY function in Matlab
 Lecture 20 Power Engineering - Egill Benedikt Hreinsson 12

IEEE 30 BUS TEST

CASE
5

10

15

20

25

30

0 5 10 15 20 25 30
nz = 112

Use the “SPY” function in Matlab to obtain the graphical

representation of the sparse matrix
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 13
IEEE 57 BUS TEST
CASE
0

10

20

30

40

50

0 10 20 30 40 50
nz = 213

Use the SPY function in Matlab

Sparsity 93%
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 14

IEEE 118 bus test system

0

20 (Ybus)
40

60

80

100

0 20 40 60 80 100
nz = 476
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 15

IEEE 300 bus test system

0
(Ybus)
50

100

150

200

See details for the one-line 250

diagrams for the 300 bus system

on the following slides 300
0 50 100 150 200 250 300
nz = 1118
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 16

300 bus IEEE test system (1) one line diagram

Lecture 20 Power Engineering - Egill Benedikt Hreinsson 17

300 bus IEEE test

system (2)
one line diagram
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 18

300 bus IEEE test system (3) one line diagram

Lecture 20 Power Engineering - Egill Benedikt Hreinsson 19
Sparsity structure of a 108 bus Icelandic
system Ybus matrix
0

10

20

30

40

50

60

70

80

90

100

0 10 20 30 40 50 60 70 80 90 100
nz = 346
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 20
Solutions to the power system problem with
sparse matrices
• For large systems (and power systems are
large) we need to exploit the sparsity of the
system matrices
• Therefore solutions to power flow problem with Newton’s method
involves solution to a set of linear equations rather than direct
inversion of matrices
• We use Gauss elimination as a standard set of solving large
linear systems of equations where we exploit simultaneously the
sparsity structure of the system matrix (Jacobi)
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 21
Sparse Matrices and Solutions to Sets of Linear
Equations in Power Systems
⎡ a1,1 a1,2 a1, n ⎤ ⎡ x1 ⎤ ⎡ b1 ⎤
⎢a a2,2 ⎥
a2, n ⎥ ⎢ x2 ⎥ ⎢b2 ⎥
⎢ 2,1 ⋅⎢ ⎥ = ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢⎣ an ,1 an ,2 an , n ⎥⎦ ⎢⎣ xn ⎥⎦ ⎢⎣bn ⎥⎦

To solve: Find a vector of

unknown x-es. Other quantities
are known or the a-s and b-s are
known
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 22

Matrix Notation in Linear Equations

A⋅x = b
⎡ a1,1 a1,2 a1,n ⎤ ⎡ x1 ⎤ ⎡ b1 ⎤
⎢a a2,n ⎥⎥ ⎢x ⎥ ⎢b ⎥
⎢ 2,1 a2,2
A= x = ⎢ 2⎥ b = ⎢ 2⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢⎣ an ,1 an ,2 an ,n ⎥⎦ ⎣ xn ⎦ ⎣ bn ⎦

A is a known matrix
x is a vector with unknown
variables. b is a known vector.
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 23
Solution methods for linear equations based on
numerical analysis
Direct methods

Gaussian elimination
Gauss-Jordan elimination
LU factorization
Cholesky factorization

Iterative methods

Jacobi iterations
Gauss-Seidel iteration
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 24

Solutions to a Set of Linear Equations

• If we can transform the matrix to a triangular form,
it is possible to solve the equations
• “Elementary Operations” can be carried out
without changing the solution
• Examples of “Elementary Operation”:
– Multiply rows (equations) with a constant
– Add rows (equations) to each other
• This is the Gauss elimination method
 Solutions to Linear Equations in Power Systems. A
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 25

Special Case (triangular form)

Assume for a moment that a matrix has the
special triangular form case -- The matrix, A is
triangular with all zeros below the diagonal:

⎡a1,1 a1, 2 a1,n ⎤ ⎡ x1 ⎤ ⎡ b1 ⎤

⎢0 a2 , 2 ⎥
a2 , n ⎥ ⎢ x ⎥ ⎢b ⎥
⎢ ⋅⎢ 2⎥ = ⎢ 2⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣0 0 an , n ⎦ ⎣ xn ⎦ ⎣bn ⎦

Solution method: Back substitution

Lecture 20 Power Engineering - Egill Benedikt Hreinsson 26

Back Substitution
⎡ a1,1 a1,2 a1, n ⎤ ⎡ x1 ⎤ ⎡ b1 ⎤
⎢0 a2, n ⎥⎥ ⎢ x2 ⎥ ⎢b2 ⎥
⎢ a2,2 an − 2,n − 2 xn − 2 + an − 2,n −1 xn −1 + an − 2,n xn = bn − 2
⋅⎢ ⎥ = ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ bn − 2 − an − 2,n xn − an − 2,n −1 xn −1
⎢⎣ 0 0 an , n ⎥⎦ ⎢⎣ xn ⎥⎦ ⎢⎣bn ⎥⎦ xn − 2 =
an − 2 , n − 2
The 3rd line from
an ,n xn = bn The last line n the end

xn =
bn bk − ∑a
j = k +1
k, j xj
an , n The second last line xk =
ak , k
an −1,n −1 xn −1 + an −1,n xn = bn −1
b1 − a1, 2 x2 − a1,3 x3 − … − a1,n xn
bn −1 − an −1,n xn x1 =
xn −1 = a1,1
an −1,n −1
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 27

Solution method
• How do we get any matrix to a triangular form?
• Transform the matrix by Gauss elimination to a lower
triangular form matrix
• Solve such a matrix by back substitution

• We have already checked out back

substitution. Let us look at Gauss
elimination!!
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 28

st
Gaussian Elimination (1 step)
The original matrix: ⎡ a1,1 a1,2 a1,3 a1, n ⎤
⎢a a2,2 a2,3 a2, n ⎥⎥
⎢ 2,1
⎢ a3,1 a3,2 a3,3 ⎥
⎢ ⎥
Multiply the 1st row a2,1 ⎢ ⎥
with a2,1/a1,1 and ⎢ an ,1 an , n ⎥⎥⎦
a1,1 ⎢⎣ an ,2 an ,3
subtract from the 2nd row
⎡ a1,1 a1,2 a1,3 a1, n ⎤
⎢ a2,1 a2,1 a2,1 ⎥
Multiply the 1st row ⎢0 a2,2 − ⋅ a1,2 a2,3 − ⋅ a1,3 a2, n − ⋅ a1, n ⎥
⎢ a1,1 a1,1 a1,1 ⎥
with a3,1/a1,1 and ⎢ ⎥
⎢ a3,1 a3,1 a ⎥
subtract from the 3rd row ⎢0 a3,2 −
a1,1
⋅ a1,2 a3,3 −
a1,1
⋅ a1,3 a3, n − 3,1 ⋅ a1, n ⎥
a1,1
⎢ ⎥
⎢ ⎥
⎢ ⎥
Multiply the 1st row ⎢0 ⋅ a1, n ⎥⎥
an ,1 an ,1 an ,1
⎢ an ,2 − ⋅ a1,2 an ,3 − ⋅ a1,3 an , n −
with an,1/a1,1 and ⎣⎢
a1,1 a1,1 a1,1 ⎥⎦
subtract from the last row
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 29

Gaussian Elimination (2nd step)

⎛ a1,1 a1,2 a1,3 a1, n ⎞

⎜ (1) (1) (1) ⎟
⎜ 0 a 2,2 a 2,3 a2, n ⎟
The matrix after the ⎜ ⎟
1st step: ⎜ ⎟
⎜ 0 an −1, n ⎟
(1) (1) (1)
a n −1,2 a n −1,3
The 1st column ⎜⎜ 0 a (1)
a (1) (1) ⎟
an , n ⎟⎠
is with zeros ⎝ n ,2 n ,3

except at the top row

Lecture 20 Power Engineering - Egill Benedikt Hreinsson 30

nd
Gaussian Elimination (2 step)
The matrix after the 1st step:
⎡ a1,2 a1,2 a1,3 a1, n ⎤
Now the 2nd row plays the same role⎢ ⎢ 0
(1)
a2,2 (1)
a2,3 a2,(1)n ⎥
⎥
⎢ ⎥
as the 1st row before: ⎢
⎢ 0 an(1)−1,2 an(1)−1,3 an(1)−1, n ⎥
⎥
⎢ 0 an(1),2 an(1),3 an(1), n ⎥⎥⎦
⎢⎣
Multiply the 2nd row
with a3(1, 2)
(1) ⎛ a1,1 a1,2 a1,3 a1, n ⎞
a 2, 2 ⎜ (1) (1) ⎟
⎜ 0 a2,2 a2,3 a2,(1)n ⎟
and subtract from ⎜ (1) (1) ⎟
a3,2 a3,2
the 3rd row ⎜ 0 0 a (1)
− ⋅ a2,3
(1) (1) ⎟
a3, n − (1) ⋅ a2, n
(1)

⎜ ⎟
3,3 (1)
a 2,2 a2,2
⎜ ⎟
Multiply(1the 2 nd row ⎜ ⎟
⎜ an(1),2 an(1),2 (1) ⎟
an , )2
with (1) ⎜ 0 0 a (1)
− ⋅ a2,3
(1)
an , n − (1) ⋅ a2, n ⎟
(1)

⎜ ⎟
n ,3 (1)
a a2,2
a2 , 2 ⎝ 2,2
⎠
and subtract from
the last row
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 31

rd
Gaussian Elimination (3 step)

The matrix after the

2nd step:
The 1st column with ⎛ a1,2 a1,2 a1,3 a1, n ⎞
⎜ (1) ⎟
zeros except at the top ⎜ 0
(1)
a2,2 (1)
a2,3 a2, n ⎟
⎜ ⎟
⎜ ⎟
⎜ 0 an(2)−1,3 an −1, n ⎟
(2)
0
The 2nd column with ⎜⎜ 0 0 an(2),3 an(2), n ⎟⎟⎠
⎝
zeros, except in the
2 top rows
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 32
Gauss-Elimination Transforms the Matrix to the
Following Form :
⎛ 1 a12′ a1′n ⎞ ⎡ x1 ⎤ ⎡b1′ ⎤
⎜ ⎟⎢ ⎥ ⎢ ′⎥
U⋅x = G ⎜0 1 ′
a2 n ⎟ ⎢ x2 ⎥ ⎢b2 ⎥
=
⎜ ⎟⎢ ⎥ ⎢ ⎥
⎜⎜ ⎟⎟ ⎢ ⎥ ⎢ ⎥
⎝0 0 1 ⎠ ⎢⎣ xn ⎥⎦ ⎢⎣bn′ ⎥⎦
This matrix is U or an “upper” or “unit upper”
triangular matrix. The elements are calculated with
iteration, where each step is an “elementary operation”. It
is possible to solve the equation with a simple “back
substitution”. It is found by dividing each row with the
diagonal element
Lecture 20 Power Engineering - Egill Benedikt Hreinsson 33

Gaussian Elimination

• After n-1 steps the matrix has

become triangular!
• Computer time grows fast with the matrix size
(increasing n)
• Computer time grows in proportion to n3, if the
matrix is not sparse
• We can obtain a “1” on the diagonal by dividing by
the diagonal element in each step