8 Load flow studies

Ensure:
1. the system is stable in the steady-state, i.e. there is enough transmission capacity;
2. transmission capacity is adequate even with some lines out of service;
3. all busbar voltages are within limits;
4. the flow of reactive power in the system is acceptable.

Nodal Voltage Analysis

Example:
a b

generator, a generator, b
transmission line, c

load, d load, e

Equivalent circuit for one phase:

Zc

Za Zb
Va Vb Ze
Zd

Ea Eb

Convert to current sources and admittances

Yc

Va Yd Vb Ye Yb Yb Eb
Ya Ea Yd

summing currents at the two nodes:

Ya Ea = Va .(Ya + Yd ) + (Va − Vb ).Yc

Yb Eb = Vb .(Yb + Ye ) + (Vb − Va ).Yc

8-1
and re-arranging:
Ya Ea = Va .(Ya + Yc + Yd ) − Vb .Yc

Yb Eb = − Va .Yc + Vb .(Yb + Yc + Ye )
In general, for a system with r nodes, then at node n:
r
I n = Yn1V1 + Yn2 V2 +......+ Ynn Vn +.....+Ynr Vr = ∑ Ynk Vk
k =1
where: Ynn = sum of all admittances connected to node n
Ynk = sum of all admittances connected between nodes n and k
In = current injected at node n

For the complete system of r nodes:

 I1  Y11 .. Y1n .. Y1r   V1 
:  : : :  : 
    
I n  = Yn1 .. Ynn .. Ynr .  Vn  or [I] = [Y][. V]
    
:  : : :  : 
 I r   Yr1 .. Yrn .. Yrr   Vr 

[Y] is the nodal admittance matrix

Nodal elimination
Faster solutions are obtained by minimising the number of nodes. In particular, if
there are e nodes at which the injected current (Ik) is zero and n nodes at which the
injected current is non-zero, the nodal admittance matrix [Y] can be re-arranged and
then partitioned. The column vector of injected currents is partitioned into two
column vectors, one containing the n injected currents [In] and the other [Ie]
containing the e zero injected currents:
[I n ] [I n ] [Ynn ] : [Yne ] [Vn ]
      
 ...  =  ...  =  ... .:. ... .  ... 
[I e ]  [0]  [Yen ] : [Yee ] [Ve ]
the solution of the network may then be limited to the calculation of the nodal
voltages [Vn], using:
[0] = [Yen ][. Vn ] + [Yee ].[Ve ]
−1
∴ [Ve ] = −[Yee ] .[Yen ].[Vn ]
also:
[I n ] = [Ynn ][. Vn ] + [Yne ][. Ve ]

8-2
and substituting for [Ve]:

[ ]
[I n ] = [Ynn ][. Vn ] + [Yne ].{−[Yee ]−1.[Yen ][. Vn ]} = [Ynn ] − [Yne ][Yee ]−1.[Yen ] [Vn ]
so the new (nxn) nodal admittance matrix is:

[Y ] = [[Y
'
nn ] − [Yne ][Yee ]−1.[Yen ] ]
Conventional circuit analysis involves the calculation of nodal voltages from known
injected currents by inverting the nodal admittance matrix.

The Load Flow Problem

Nodal data is in various forms:
1. Load nodes: complex power Sn= Pn +jQn taken from or injected into the network.
2. Generator nodes: the injected power, Pn, and the magnitude of the nodal voltage |
Vn | are specified.
3. Floating bus or slack bus: the nodal voltage magnitude | Vn | and phase angle δn are
specified.
The floating bus is always needed because it is not possible to specify the total power
in and out of the system without knowing the system losses, which depend on the
nodal voltages.
Solution of the load flow problem uses an iterative technique, known as the Newton-
Raphson method, in which estimates of the nodal voltage are gradually improved until
convergence is obtained.
General Principles of the Newton Raphson Method for a Two Variable Problem
Suppose f1 and f2 are functions of the two variables x1 and x2, such that:
K1 = f1 (x1 , x 2 ) and K 2 = f 2 ( x1 , x 2 )
where K1 and K2 are constants.
We want to calculate values of x1 and x2 of which satisfy these two equations. Suppose
initial estimates of these solutions are: x10 , x 02 and that the corrections which must be
made to these initial estimates to reach the solutions are: ∆x10 , ∆x 02 , so:

(
K1 = f1 x10 + ∆x10 , x 02 + ∆x 02 )
K2 = f (x
2
0
1 + ∆x10 , x 02 + ∆x )0
2

and the problem becomes one of calculating ∆x10 , ∆x 02 . Using Taylor’s Theorem:

8-3
 ( ) ( ) ∂f
K1 = f1 x10 + ∆x10 , x 02 + ∆x 02 = f1 x10 , x 02 + ∆x10 . 1
∂x1 x 0 , x 0
∂f
+ ∆x 02 . 1
∂x 2 x 0 , x 0
+...
1 2 1 2

( ) ( ) ∂f
K 2 = f 2 x10 + ∆x10 , x 02 + ∆x 02 = f2 x10 , x 02 + ∆x10 . 2
∂x1 x 0 , x 0
∂f
+ ∆x 02 . 2
∂x 2 x 0 , x 0
+...
1 2 1 2

neglecting higher order terms and simplifying the notation:

( ) ∂f
∂x1 0
∂f
K1 − f1 x10 , x 02 = ∆x10 . 1 + ∆x 02 . 1 = ∆K10
∂x 2 0

( ) ∂f
∂x1 0
∂f
K 2 − f2 x10 , x 02 = ∆x10 . 2 + ∆x 02 . 2 = ∆K 02
∂x 2 0

which can be re-written in matrix form:

( ) ∂f
∂x1 0
∂f
K1 − f1 x10 , x 02 = ∆x10 . 1 + ∆x 02 . 1 = ∆K10
∂x 2 0

( ) ∂f
∂x1 0
∂f
K 2 − f2 x10 , x 02 = ∆x10 . 2 + ∆x 02 . 2 = ∆K 02
∂x 2 0
 ∂f1 ∂f1 
 
[ ]
 ∆K10   ∂x1 0 ∂x 2  0  0
0 . ∆x1 = J 0 . ∆x1
 0 =  ∂f 2  ∆x 0   0
∆K 2  ∂f 2   ∆x 2 
  2
 ∂x1 0 ∂x 2 0 

where the matrix [J] of partial differentials, evaluated using the estimates of x1 and x2,
is called the Jacobian.
The equation above can be solved to find the corrections required to improve the
estimates of the solutions to the set of equations:

[ ]
 ∆x10  0 −1  ∆K1 
0
 0 = J . 0 
∆x 2  ∆K 2 
The solution proceeds by forming new estimates for the solutions:
x11 = x10 + ∆x10
x12 = x 02 + ∆x 02
because the Taylor series was truncated, these new estimates are not exact solutions,
but are closer estimates to the required solutions.

A simple example
Suppose f1 and f2 are functions of the two variables x1 and x2, such that:
f1 = x12 + x22 = K1 = 1 and f 2 = x1 + x2 = K 2 = 0
where K1 and K2 are constants.

8-4
We can show analytically that x1 = 0.707, x2 = -0.707 or x1 = -0.707, x2 = 0.707, but
can we use Newton-Raphson to get the same result? Suppose initial estimates of these
solutions are: x10 = 1, x20 = 0 :
( )
K1 − f1 x10 , x20 = 1 − (12 + 0 2 ) = 0 = ∆K10
K2 − f (x , x ) = 0 − (1 + 0) = −1 = ∆K
2
0
1
0
2
0
2

 ∂f 1 ∂f 1 
 
∆K   ∂x1 ∂x 2 ∆x10  ∆x10 
[ ]
0
0
= = J . 0 
1 0
  ∆x20 
0
.
∆K   ∂f 2 ∂f 2 ∆x2 
0
2

 ∂x1 0
∂x 2 
0

 0  2 x10 2 x20  ∆x10 

− 1 =  . 
   1 1  ∆x20 
 1 − 2 x20   1 0
−1  0   − 1 2 0
∆x  2 x
0 0
2 x20   0  − 1 2 x1   0    .  =  0 
∴ =
1 1
 .  = .  =
∆x   1
0
2 1  − 1 (2 x10 − 2 x20 ) − 1 (2 − 0) − 1 − 1

The solution proceeds by forming new estimates for the solutions:

x11 = x10 + ∆x10 = 1 + 0 = 1
x12 = x20 + ∆x20 = 0 − 1 = −1
so:
 ∆K 1 
calculate:  11 
∆K 2 
 ∂f 1 ∂f 1 
 
 ∆K11   ∂x1 1 ∂x 2  ∆x1   ∆x1 
 1 = 
1
[ ]
. 11  = J 1 . 11 
 ∆x2 
∆K 2   ∂f 2 ∂f 2 ∆x2 

 ∂x1 1 ∂x 2 1

 1 − 2 x12   1 2
−1   − 1 2 − 1
 ∆x  2 x
1 1
2 x12  − 1 − 1 2 x11  − 1   .  = − 0.25
∴ =
1 1
 .  = .  =
∆x   1
1
2 1   0  (2 x11 − 2 x12 )  0  (4)  0   0.25 
New estimates for the solutions are:
x12 = x11 + ∆x11 = 1 − 0.25 = 0.75
x22 = x12 + ∆x12 = −1 + 0.25 = −0.75
Application to Load Flow
Use the Newton-Raphson method to solve for the node voltages, using polar form of
phasors and complex admittances:
Vk = Vk ∠δ k Vn = Vn ∠δ n Ykn = Ykn ∠θ kn
At node n, we have:

8-5
 r
complex power, S n = Vn I *n and I n = ∑ Ynk Vk
k =1
r r
∴ S n = Vn ∑ * *
Ynk Vk = ∑ Vn Vk Ykn ∠{δ n − δ k − θ kn }
k =1 k =1
r
∴ Pn = ∑ Vn Vk Ykn cos{δ n − δ k − θ kn } (1)
k =1

r
∴Qn = ∑ Vn Vk Ykn sin{δ n − δ k − θ kn } (2)
k =1

If Pn and Qn are specified at a given node (busbar), we need to calculate the magnitude
Vn and angle δn of the voltage at that busbar. The values of Pn and Qn are fixed and
correspond to K1 and K2 in the general approach, while the functions f1 and f2 are
replaced by Equations 1 and 2. The number of variables is equal to 2x number of
nodes (a magnitude and an angle at each node), while the number of equations is also
2x number of nodes (a power and a reactive power at each node).
Following the general approach, the first step in the solution is to make initial
estimates of the variables: Vn0 ,δ 0n and using these estimates to calculate the power
and reactive power from Equations 1 and 2. These calculated values are compared
with the specified values to give a power and reactive power error:
r
∴ ∆Pn0 = Pn − ∑ Vn0 Vk0 Ykn cos{δ 0n − δ 0k − θ kn }
k =1
r
∴ ∆Q 0n = Qn − ∑ Vn0 Vk0 Ykn sin{δ 0n − δ 0k − θ kn }
k =1

Having calculated the errors ∆Pn0 , ∆Q0n , the corrections to Vn ,δ n are calculated,
using the relations:
: : : : : : : : : : :   ∆δ 0n − 1 
 :   ∂Pn ∂Pn ∂Pn ∂Pn ∂Pn ∂Pn  
 ∆P 0  − − : − −   ∆δ 0n 
 n  ∂δ n − 1 ∂δ n ∂δ n + 1 ∂ Vn − 1 ∂ Vn ∂ Vn + 1
 0 
 :  : : : : : : : : : : :   ∆δ n + 1 
  
 − − −  = − −−− −−− −−− − −− −− −−− −−− −−− − .  − − − 
 0 
 :  : : : : : : : : : : :   ∆ Vn − 1 
 0  ∂Q n ∂Q n ∂Q n ∂Q n ∂Q n ∂Q n  0 
 ∆Q n  − − : − −   ∆ Vn 
∂δ n − 1 ∂δ n ∂δ n + 1 ∂ Vn − 1 ∂ Vn ∂ Vn + 1 
 :    0 
 
: : : : : : : : : : :  ∆ Vn + 1 
 

Jacobian, J

8-6
by inverting the J matrix, the corrections to the initial values of voltage magnitude and
angle can be found and the process repeated until the corrections are within the
desired accuracy of solution.
Other Bus Types
The description above applies to a load bus, where the power and reactive power flow
are specified and the voltage magnitude and angle is to be calculated.

For a generator node the voltage magnitude Vn and power Pn are specified, but the
reactive power is not specified. The order of the calculation can be reduced by 1: there
is no need to ensure that the reactive power is at a set value and only the angle of the
bus voltage needs to be calculated, so one row and column are removed from the
Jacobian.
For the floating bus, both voltage magnitude and angle are specified, so there is no
need to calculate these quantities.

Example
1 2
Z=j0.25 (Y=-j4)

V2 = 11
.
. ∠0o
V1 = 10

Z=j0.20 (Y=-j5) Z=j0.10 (Y=-j10) P2s= 10

.

P3s=-1.5
Q3s =-0.2
3

1: floating bus 2: generator bus 3: load bus

Nodal admittance matrix:
 I 1  − j9 j4 j 5  V1 
     
 I 2  =  j4 − j14 j10  . V2 
 I 3   j 5 j10 − j15 V3 
Values for P2, P3, Q3 are given, so the iterative solution centres on these quantities.
The bus voltages are:
. ∠ 00
V1 = 10
V2 = 11. ∠δ 2 , so the solution variables are δ 2 , V3 , δ 3
V3 = V3 ∠δ 3
Express the set values in terms of the variables:

8-7
 . ∠δ 2 {j4. V1 − j14 V2 + j10V3 }
*
S 2 = V2 . I *2 = 11

{ ( ) ( ) ( )}
*
= 11
. ∠δ 2 4.0∠ 90 o + 15.4∠ δ 2 − 90 o + 10 V3 ∠ δ 3 + 90 o

. ∠δ {4.0∠ (−90 ) + 15.4∠(− δ + 90 ) + 10 V ∠ (− δ − 90 )}

= 11 2
o
2
o
3 3
o

( ) ( )
= 4.4∠ δ 2 − 90 o + 16.9∠ 90 o + 11 V3 ∠ δ 2 − δ 3 − 90 o ( )
( )
P2 = 4.4 cos δ 2 − 90 o + 11 V3 cos δ 2 − δ 3 − 90 o ( )
S 3 = V3 . I*3 = V3 ∠δ 3 {j5.0 + j11 − j15V3 }
*

{ ( ) ( ) ( )}
*
= V3 ∠δ 3 5.0∠ 90 o + 11∠ δ 2 + 90 o + 15 V3 ∠ δ 3 − 90 o

= V ∠δ {5.0∠(−90 ) + 11∠ (− δ − 90 ) + 15 V ∠(− δ + 90 )}

3 3
o
2
o
3 3
o

( ) (
= 5.0 V3 ∠ δ 3 − 90 o + 11 V3 ∠ δ 3 − δ 2 − 90 o + 15 V3 ∠ 90 o ) 2
( )
( ) (
P3 = 5.0 V3 cos δ 3 − 90 o + 11 V3 cos δ 3 − δ 2 − 90 o )
sin(δ − 90 ) + 11 V sin(δ − 90 ) + 15 V
2
Q 3 = 5.0 V3 3
o
3 3 − δ2 o
3

Iterative Solution, starting with initial estimates of δ 2 , V3 , δ 3

1. Calculate power and reactive power errors:
∆P2 = P2S − P2
∆P3 = P3S − P3
∆Q 3 = Q 3S − Q 3
2. Form the Jacobian:
 ∂P2 ∂P2 ∂P2 
 
 ∆P2   ∂δ 2 ∂δ 3 ∂V3   ∆δ 
2
 ∆P  =  ∂P3 ∂P3 ∂P3  
. ∆δ 3 
 3   ∂δ ∂δ 3 ∂V3  
∆Q 3   ∂Q2 ∂Q 3 ∂Q 3  ∆V3 
 3 
 ∂δ 2 ∂δ 3 ∂V3 
where:
∂P2
∂δ 2
( )
= −4.4 sin δ 2 − 90o − 11V3 sin δ 2 − δ 3 − 90o ( )
∂P2
∂δ 3
(
= 11V3 sin δ 2 − δ 3 − 90o )

8-8
∂P2
∂V3
(
= 11cos δ 2 − δ 3 − 90o ) etc.

3. Invert the Jacobian and hence calculate the corrections to the estimates
∆δ 2 , ∆δ 3 , ∆V3
4. Form new estimates δ 2 → δ 2 + ∆δ 2 ; δ 3 → δ 3 + ∆δ 3 ; V3 → V3 + ∆V3 and repeat
from stage 1.

Sample results

δ2 o()δ() 3
o V3

10 -20 0.9
-2.8 4.8 1.24
-0.34 -3.82 1.09
-0.23 -5.21 1.05
-0.22 -5.28 1.05
-0.22 -5.28 1.05
from which the power flow can be calculated:

P1= 0.5pu; P13= 0.483pu; P12= 0.017pu; P23= 1.017pu

Alternative Load Flow Methods

1. Direct Methods

Direct methods involve inverting the nodal admittance matrix [Y] to give the nodal
impedance matrix [Z}:

[V = [Y]]−1 [I] = [ Z][I]

In a power system analysis the matrices can be very large, so inversion of the nodal
admittance matrix is a considerable task. The computational effort required to invert
an rxr matrix is proportional to r3. One other important feature of the [Y] matrix is
that it is sparse: many of the off-diagonal elements are zero, indicating no direct
connection between nodes. There is a range of numerical methods that are well suited
to the inversion of large sparse matrices.

The direct method of solution can be summarised by the following flowchart (note
that an iterative approach is still needed):

8-9
 Enter from routine to
calculate [Z]

For each busbar estimate

voltage and complex power,
unless already specified

Obtain estimates of node

currents: In = Sn*/Vn* At floating bus: correct
voltage to specified
magnitude and phase.
Improve estimate of
Obtain a corrected set of power input.
voltages from [V]=[Z][I]

Are the successive voltage NO

estimates within limits?

YES

EXIT

2. Modification of the nodal admittance matrix and its inverse

Load flow studies are often used to investigate the effects of changes to a power
system, e.g. a transmission line being added. For each modification a new nodal
admittance matrix, and therefore a new nodal impedance matrix, is generated. If the
nodal impedance matrix had to be re-calculated by matrix inversion on each occasion
that the system is modified, then considerable effort would be required. Fortunately
the process can be streamlined.

Suppose the original network has r nodes, giving an rxr admittance matrix [Y].The
nodal impedance matrix [Z] = [Y]-1 is obtained in the first load flow study and is
therefore available for subsequent calculations. When the system is modified a new
nodal admittance matrix [Ya] is established. If no new nodes are involved, then [Ya]
is also an rxr matrix and can be expressed in terms of the original matrix [Y] and a
modification matrix [Ym]:

[Y ] = [Y] + [Y ]
a m (2.1)

8-10
The form of the modification matrix depends on the particular system change. For
example, suppose a new transmission line of impedance Z is connected between
nodes d and f. The self admittances at nodes d and f are increased by 1/Z, while the
mutual admittances between nodes d and f are reduced by –1/Z:

 : : 
 1 1 
d L
 Z L − Z L
[Y ]
m = 

: : 

(2.2)
1
f L − Z L 1
Z L
 : : 

Any modification matrix can be expressed in the standard form:

[Y ] = α[G][ H]
m t (2.3)

where α is a scalar and [G} and [H] are column vectors of order r. In the example
above:
 : 
d 1 
 
α = 1Z; [G ] = [ H ] =  :  (2.4)
 
f − 1
 : 

[Y ] = [Y] + α[G][ H]
a t (2.5)

and we need to find [Ya]-1 in terms of [Y]-1, α, [G] and [H]. The following formula
can be used (it does not need to be memorised!):

([Y] −1
[ G ])([ H ] t [ Y] −1 )
[Y ] −1
= [ Y]
−1
− (2.6)
( )
a
1
+ [ H ] t [ Y] [ G ]
−1

3. Another Iterative Method: The Gauss-Seidel Method

Load flow studies involve solution of the equations:

r
S n = Vn I *n and I n = ∑ Ynk Vk (3.1)
k =1

for each node n= 1…r.

Eliminating In between the equations 3.1:

8-11
 Sn r n −1 r
= ∑ Ynk* Vk* = ∑ Ynk* Vk* + Ynn* Vn* + ∑ Ynk* Vk* (3.2)
Vn k =1 k =1 k = n +1

which gives the following expression for Vn* :

Ynk* Vk* n −1
r Ynk* Vk* Sn
V =−∑
*
n * − ∑ * + (3.3)
k =1 Ynn k = n +1 Ynn Vn Ynn*

The above equation is used as the basis for an iterative routine. If Vnp and Vnp+1
denote the values of the voltage at node n after p and p+1 iteration cycles:

p +1 p
* p +1
Y* V*
n −1 Y* V*
r S
V n = − ∑ nk *k − ∑ nk * k + p n * (3.4)
k =1 Ynn k = n +1 Ynn Vn Ynn

Note that in evaluating the nth node voltage, the latest estimates of the other node
voltages are used. In the p+1th iteration cycle when performing the calculation for
node n, the voltages at the nodes k=1…n-1 are available, but for the other node
voltages the values from the previous cycle have to be used.

Types of node:
a) Load nodes where Sn is specified and Vn unknown. At these nodes Eqn 3.4 can be
used directly.
b) At the floating bus Vn is known. The node voltage does not need to be re-
calculated.
c) Generator nodes, where power Pn and voltage magnitude Vn are specified. If:
p +1 p

( ) Ynk* Vk* Ynk* Vk*

# rn −1 P
* p +1
V n =−∑ * − ∑ * + pn * (3.5)
k =1 Ynn k = n +1 Ynn Vn Ynn
then substituting into Eqn. 3.4:

Vn*
p +1
= Vn* ( p +1 #
) +
jQ n
Vn p Ynn
* (3.6)

At the p+1th iteration cycle Eqn 3.5 can be evaluated. Qn can be calculated from Eqn.
3.6 by working with the voltage magnitudes:

  p +1 jQ   p +1
2
jQ 
2

( ) ( )
# # 2
Re Vn* + p n *  + Im Vn* + p n*   = Vn (3.7)
  Vn Ynn   Vn Ynn  

Eqn 3.7 is solved for Qn and this value is substituted into Eqn. 3.6 to give the
improved estimate for the generator node voltage.

8-12