Electrical Power and Energy Systems 58 (2014) 291299

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

A load ow method for weakly meshed distribution networks using

powers as ow variables
Hongwei Li , Anan Zhang, Xia Shen, Jin Xu
School of Electrical Engineering and Information, Southwest Petroleum University, Chengdu 610500, PR China

a r t i c l e

i n f o

Article history:
Received 1 April 2013
Received in revised form 7 January 2014
Accepted 18 January 2014

Keywords:
Load ow
Weakly meshed network
PV node
Node-branch incidence matrix
Thevenins equivalent circuit

a b s t r a c t
The relationship between the branch powers and the node-injection powers were developed with the
node-branch incidence matrix. Then based on two accurate formulas to estimate the voltage drop and
angle difference, a new load ow algorithm for weakly meshed distribution systems was presented. By
using active and reactive power rather than complex currents as ow variables, the algorithm reduces
the computational complexity and has higher efciency. Moreover, in order to deal with the PV nodes,
a new solution was introduced based on Thevenins equivalent circuit. The solution deduced an accurate
calculation formula to update the reactive power injections of PV nodes at each iteration and to x
voltage magnitude of PV nodes at specied values. The proposed load ow algorithm is essentially still
belongs to the loop-analysis based method and has a strong ability to deal with meshed network. It
reduces the iteration number and has a faster calculation speed even when network becomes more
meshed and has more PV nodes. The numerical tests proved that the new method is robust and has
excellent convergence characteristics.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
Load ow method as a fundamental tool for distribution management system plays a vital role in evaluation of network condition and optimal operation. Several efcient load ow solution
techniques have been developed during the past decades [14].
The forward/backward sweep based method is one of the most
commonly used load ow solutions in distribution system. It takes
advantages of the distribution network structure and has fast calculation speed for no large matrix calculation, and the programming is relatively simple.
For meshed network, the most commonly used solution is the
compensation based technique [5]. The method has high calculation speed and good convergence, but it has more iteration number
and becomes less efcient when node voltage is lower or when
network become more meshed. In [6], an improved version of
the method in [5] has been presented, which reduced the related
computational effort by using the branch power ows instead of
the branch complex currents as ow variables. But it constructs
the sensitivity matrix to deal with weakly meshed networks and
PV nodes by assuming all bus voltages being close to 1.0 p.u., so
its convergence is affected and becomes bad. Ref. [7] presents a novel forward/backward sweep algorithm for meshed network which
Corresponding author. Tel.: +86 13880768930.
E-mail address: lhwmail@126.com (H. Li).
http://dx.doi.org/10.1016/j.ijepes.2014.01.015
0142-0615/ 2014 Elsevier Ltd. All rights reserved.

is rmly based on loop analysis theorem. The method has excellent

convergence characteristics even when the network becomes more
meshed. Ref. [8] proposed a new load ow method based on the
superposition principle according to the structure feature of the
weakly meshed network. The method, which can be viewed as a
voltage compensation method, is much complicated and not universal. Ref. [9,10] put forward their methods for weakly meshed
distribution load ow solutions. The concise formula to formulate
the relationship between branch currents and node injections is
developed and is used to solve load ow problem directly. The
two methods all have higher calculation efciency and speed, but
the method in [10] has a clear theory foundation and a more universal form.
The emergences of distributed generation (DG), which are most
based on renewable energy sources, into the distribution systems
have great inuence on the voltage quality, load ow and power
loss, etc. [11]. So far, many approaches have been studied to deal
with DG. In [5,6,12], the breakpoint compensation based methods
are used to solve radial and weakly meshed systems with PV nodes
and obtain good results. But the convergence number increases
signicantly and sometimes the load ow does not converge when
network become more meshed and has more PV nodes. In [10,13],
in order to x voltage magnitude of PV nodes at specied values,
two approaches have been presented to cope with PV nodes by
updating the reactive power injections of PV nodes at each
iteration. The two methods all have good convergence. Ref. [14]

292

H. Li et al. / Electrical Power and Energy Systems 58 (2014) 291299

Nomenclature
N
l
b
e
A
aij
A
At
Al
Tt
T ti
Bt
dU
da
dP
dQ

number of the independent nodes in the network

number of link branches (loops) in the network
number of the total branches in the network, b N l
node-branch incidence matrix in the network
e
element for row i and column j in A
reduced node-branch incidence matrix in the network
reduced node-branch incidence matrix corresponding
to the tree branches in A
reduced node-branch incidence matrix corresponding
to the link branches in A
path matrix corresponding to tree branches in the network,
node i corresponding row vectors (path vector) in T t
loop matrix corresponding to tree branches in the network,
branch voltage drop
branch voltage angle difference
branch active power loss
branch reactive power loss

introduces an improved NewtonRaphson algorithm with lower

initial values requirement. But the method is much complicated
to deal with PV nodes and it needs to calculate the Jacobian matrix.
The Jacobian matrix may be ill-conditioned because of the high R/X
ratio in distribution system [15]. Ref. [16] presents an improved
back/forward sweep load ow method for weakly meshed distribution network with the distributed generation based on the modeling of different DGs (wind turbines, photovoltaic system, fuel
cell, CHP, etc.). Ref. [17] proposed a load ow algorithm based on
the back/forward sweep method which can deal with PV nodes.
The load ow calculation is speeded up by using the laterals delaminating technology.
However, a more efcient method is still needed to deal with
weakly meshed networks and PV nodes. This paper presents a
new load ow method using powers as ow variables for weakly
meshed networks with PV nodes. The main works are the
following.
(1) Two accurate formulas to estimate the voltage drop and
angle difference have been deduced across a branch through
knowledge of terminal power (both active and reactive) and
voltage states.
(2) Based on the node-branch incidence matrix, a new load ow
solution has been proposed by using powers as variables.
(3) In order to deal with PV node, a new solution is presented. It can
be easily integrated into the proposed load ow algorithm.
The advantage of the proposed method and solution are veried
by some numerical tests.

2. Formulation of voltage drop and angle difference

P
Q
P It
Q It
P Il
Q Il
dU
da
Un

an
Il
Ul
Yl
Zb
Z bt
dP di
dQ di

node consumption active power vector

node consumption reactive power vector
tree branch input active power vector
tree branch input reactive power vector
link branch input active power vector
link branch input reactive power vector
branch voltage drop vector
branch voltage angle difference vector
node voltage magnitude vector
node voltage angle vector
loop complex current (link branch current) vector
loop complex voltage vector
loop admittance matrix
branch impedance diagonal matrix
tree branch impedance diagonal matrix
output active power increment of the ith PV node
output reactive power increment of the ith PV node

distribution line can be modeled as a pure p element which shown

in Fig. 1 with related electrical parameters.
Dene voltage drop and angle difference as

dU U i  U o
da ai  ao

And there exists (superscript  indicates complex conjugate)


 

Po jQ o
Pi jQ i
I_

U_ o
U_ i

For simplicity, the shunt impedance G  jB is regarded as the

impedance-constant load. Their expended power (both active and
reactive) can be obtained with terminal voltage as PGi GU 2i ;
Q Gi BU 2i and PGo GU 2o ; Q Go BU 2o . During calculating the
load ow, those powers expended by G  jB can be merged into
the loads at the corresponding node. It is possible to deduce the
accurate formula to calculate difference in voltage magnitude
across a branch through knowledge of terminal power (both active
and reactive) and voltage states.
In Fig. 2, a vector diagram is used to illustrate interdependencies between input and output voltage magnitude and angle.
Apparently, here the output voltage, U_ o U o \0 , has been taken
as the reference vector. The following equation can be obtained
when the vector correlations are examined based on Fig. 2,

Po  jQ o
U_ i U o R jXI_ U o R jX
Uo
RPo XQ o
XPo  RQ o
Uo
j
Uo
Uo

Moreover, a simple change of taking the input voltage,

U_ i U i \0 , as the reference vector will lead to a completely new
diagram and a new equation, as shown by Fig. 3 and (4).

The modeling of the serial components in distribution network

is used as a basis of most load ow calculation algorithms. A typical
Pi + jQi
Ui ,i

Po + jQo

U o ,o

R+jX

G-jB

Fig. 1. Equivalent

G-jB

p circuit of distribution lines.

Fig. 2. Input and output voltages vector diagram for equivalent

as the reference.

p-model using U_ o

293

H. Li et al. / Electrical Power and Energy Systems 58 (2014) 291299

Pi  jQ i
U_ o U_ i  R jXI_ U i  R jX
Ui
RPi XQ i
XPi  RQ i
Ui 
j
Ui
Ui

3. Implementation of load ow algorithm

Let

M o RPo XQ o =U o ; No XP o  RQ o =U o ; Mi RPi XQ i =U i
and Ni XPi  RQ i =U i . Substituting them into (3) and (4), the following equations can be obtained with magnitude calculation,

U 2i U o Mo 2 N2o
U 2o U i  Mi 2 N2i

U 2i  U 2o U o M o 2 N2o  U i  M i 2  N 2i
U 2o  U 2i 2U o Mo U i M i M2o  M2i N2o  N2i
Rearranging it gives,



2U i U o U i  U o U o M o U i Mi M 2o  M2i N2o  N2i
That is,

U o M o U i Mi M2o  M2i N 2o  N2i

Ui Uo
2U i U o

For a weakly meshed distribution network with N 1 nodes

and l link branches (loops), there are N independent nodes when
the power supply be referred as the reference node (the rst node).
The number of the total branches b equals N l. Node-branch incie of N 1  b can be used to describe the network.
dence matrix A
e is dened as:
The element in A
8
>
< 1 Node i is the starting point of branch j
aij 1 Node i is the ending point of branch j
>
:
0 Node i is neither the starting point nor the ending point of branch j

And then U 2i  U 2o can be solved as,

dU U i  U o

3.1. Node-branch incidence matrix

There exists (Appendix A describes the deduction procedure of

(6) in detail)

e and each row corresponds to

There are N 1 row vectors in A,
one node and shows which branch connect to this node. There are
e and each column corresponds to one branch.
b column vectors in A
There are only two non-zero elements, 1 and 1, in each column, which give the starting node number and the ending node
number of the branch. Removing the rst row corresponding to
e gets the reduced node-branch incidence
the reference node from A
matrix A of N  b. Designating one tree in the weakly meshed network and numbering the tree branches in the front and the link
branches in the back in A, then A can be expressed as,

A At ; Al 

12

The order of invertible At is N  N and the order of Al is N  l.

M 2o  M2i N2o  N2i

0
2U i U o

6
3.2. Load ow algorithm deduction

Thus,

U o M o U i Mi RPo XQ o RPi XQ i

Ui Uo
Ui Uo
RP o Pi XQ o Q i R2Pi  dP X2Q i  dQ

Ui Uo
Ui Uo

dU

where dP is the branch active power loss and dQ is the branch reactive power loss.
Eq. (7) is an accurate formula to estimate the voltage drop.
Based on Fig. 3, the angle difference can also be deduced, that is

sinda

Ni XP i  RQ i

Uo
Uo Ui

And then
1

da sin



XPi  RQ i
XPi  RQ i

Ui Uo
Ui Uo

The dP and dQ in (7) can be determined by

dP P i -Po RI2

R P2i Q 2i

dQ Q i  Q o XI

U 2i


X P2i Q 2i
U 2i

10

11

Fig. 3. Input and output voltages vector diagram for equivalent p-model using U_ i as
the reference.

Considering a simple network as shown in Fig. 4, there are

N 6 independent nodes and l 2 link branches (branch 7 and
branch 8).
The p-models are used to model transmission lines as Fig. 1. Let
superscript T indicate matrix transpose. Dene Q I Q i1 ; Q i2 ; Q i3 ;
Q i4 ; Q i5 ; Q i6 ; Q i7 ; Q i8 T and P I Pi1 ; Pi2 ; P i3 ; P i4 ; Pi5 ; Pi6 ; P i7 ; P i8 T be the
input reactive and active powers of branches, and then the power
balance equations can be obtained from Fig. 4 as below,

8
>
P  Pi2  Pi3 PN1 G1 U 21 dP1 P1
>
> i1
>
2
>
>
>
> Pi2  Pi4  Pi5 PN2 G2 U 2 dP2 P2
>
< P  P P G U 2 dP P
N3
3 3
3
3
i3
i7
> Pi4  Pi6 Pi7 PN4 G4 U 24 dP4 dP7 P 4
>
>
>
>
> Pi5  Pi8 PN5 G5 U 2 dP5 P5
>
5
>
>
:
Pi6 Pi8 PN6 G6 U 26 dP6 dP8 P6

13

8
>
Q i1  Q i2  Q i3 Q N1  B1 U 21 dQ 1 Q 1
>
>
>
>
>
Q i2  Q i4  Q i5 Q N2  B2 U 22 dQ 2 Q 2
>
>
>
>
< Q  Q Q  B U 2 dQ Q
3 3
i3
i7
N3
3
3
2
>
Q

Q

Q

B
U

dQ
> i4
4 4
i6
i7
N4
4 dQ 7 Q 4
>
>
>
2
>
> Q i5  Q i8 Q N5  B5 U 5 dQ 5 Q 5
>
>
>
:
Q i6 Q i8 Q N6  B6 U 26 dQ 6 dQ 8 Q 6

14

Fig. 4. A simple weakly meshed distribution network.

294

H. Li et al. / Electrical Power and Energy Systems 58 (2014) 291299

where Gi and Bi are the summation of shunt conductance and shunt

susceptance corresponding to node i respectively. Dene P P 1 ; P2 ;
P3 ; P 4 ; P5 ; P6 T and Q Q 1 ; Q 2 ; Q 3 ; Q 4 ; Q 5 ; Q 6 T be the node consumption active and reactive powers which include the loads, the
serial branch power loss and the shunt branch power loss corresponding to the node.
Eq. (13) can be expressed in general form with the reduced
node-branch incidence matrix A, that is,

P AP I At ; Al 

P It
P Il

At P It Al P Il

1
T
T
P It A1
t P  At Al P Il Tt P Bt P Il

where

branches

A1

t
and Bt BTt

Now, in order to calculate load plow of weakly meshed

network, the key is to calculate P Il and Q Il . For a weakly meshed
network, the loop complex current (link branch current) vector
I l l  1, the loop complex voltage vector U l l  1 and the loop
admittance matrix Yl l  l satisfy the following equation [7,10],

I l Yl U l

15

is the path matrix corresponding to tree


T
is the loop matrix corre A1
t Al Tt Al

sponding to tree branches [10]. P It is the tree branch input active

power vector and P Il is the link branch input active power vector.
Similarly, Eq. (14) can also be expressed in general form as (Q It
is the tree branch input reactive power vector and Q Il is the link
branch input reactive power vector),

Q It TTt Q BTt Q Il

16

It can be seen from (15) and (16), the tree branch input powers
consist of two parts, P I1 TTt P Q I1 T Tt Q and


T
P I2 Bt P Il Q I2 BTt Q Il . P I1 Q I1 can be viewed as the contribution

P It Q It

of the node consumption powers P (Q), which is equivalent to do a

backward sweep calculation for a radial network by breaking the
link. And P I2 Q I2 can be viewed as the contribution of the link
branch input powers P Il Q Il .
Supposing P Il and Q Il be known. First P It and Q It can be calculated with (15) and (16), then the branch power losses dP and
dQ can be calculated with (10) and (11), and nally the branch
voltage drop dU and the angle difference da can be calculated with
(7) and (9). Let U 0 be the power supply (reference node) voltage
magnitude and U n be the node voltage magnitude vector of N1.
T
Dene dU dU 1 ; dU 2 ; dU 3 ; dU 4 ; dU 5 ; dU 6  be the tree branch voltage drop vector. So the voltage magnitude difference between
power supply and any other node i equals the summation of the
tree branch voltage drop along the path of node i, that is [10],

U ni U 0  Tti dU i 1; 2; . . . ; 6

17

where Tti is the row vector corresponding to node i in Tt .

Let the angle of power supply (reference node) be zero and an
be the node voltage angle vector of N  1. Dene
T

da da1 ; da2 ; da3 ; da4 ; da5 ; da6  . Similarly there exists,

ani Tti da i 1; 2; . . . ; 6

19

where Yl Z 1
Gl  jBl and Z l BZ b BT ; Z b is the branch impedl
ance diagonal matrix of b  b.
Dene I l I lR jI lI ; U l U lR jU lI and substituting them into
(19) and rearranging it gives,

And,

Tt TTt

4. Calculation of link branches powers

18

The calculation steps of the proposed load ow algorithm can

be described as following,
Step 1, data preparation, initial values setting.
Step 2, calculating P and Q, P Il and Q Il . Then P It and Q It can be
calculated by (15) and (16).
Step 3, calculating dP and dQ for each branch with (10) and
(11), then calculating dU and da for each tree branch with (7)
and (9).
Step 4, calculating U n with (17).
Steps 14 will be repeated until convergence is achieved (the
maximum voltage magnitude mismatch between the results
of two times is small enough).
Step 5, calculating an with (18).

I lR Gl U lR Bl U lI
I lI Gl U lI  Bl U lR

20

U l is actually the complex voltage difference vector between the

starting points and the ending points of link branches when only
considering the contribution of P and Q. Referring to Fig. 4, the
starting point and the ending point of link branch 7 in loop
1(12431) are node 3 and node 4. The voltage magnitude and
angle of those two nodes, U n4 ; an4 and U n3 ; an3 , can be calculated
with (17) and (18) just considering the contribution of the node
consumption powers P and Q. Then the complex voltages at node
3 and node 4 can be expressed as,

U_ n3 U n3 cos an3 jU n3 sin an3

U_ n4 U n4 cos an4 jU n4 sin an4

So the real parts and imaginary parts of voltage for loop 1 can be
calculated by,

U l1R U n4 cos an4  U n3 cos an3

U l1I U n4 sin an4  U n3 sin an3

21

As described above, all the loops voltages can be calculated with

(21), then the loop current vector I lR and I lI can be solved with
(20). Now the input powers in link branches can be determined.
For instance, the input powers in link 7 (in loop 1) can be calculated
with,

PI7 U l1R Il1R U l1I Il1I

Q I7 U l1R Il1I U l1I Il1R

22

Then the second step (step 2) described in Section 3 should be

modied as below,
Step 2.1, calculating P and Q, and P I1 TTt P; Q I1 TTt Q .
Step 2.2, calculating dP and dQ with (10) and (11), and calculating dU and da with (7) and (9).
Step 2.3, calculating nodes voltages at the starting points and
the ending points of link branches with (17) and (18), the loop
voltages can be determined by (21).
Step 2.4, solving loop current vector I lR and I lI with (20), and
then calculating input powers P Il and Q Il in link branches with
(22).
Step 2.5, then P I2 BTt P Il and Q I2 BTt Q Il , then P It P I1 P I2
and Q It Q I1 Q I2 . Go step 3 in Section 3.
5. Handling of PV nodes
Some DGs, which can be well represented as PQ nodes, require
little special treatment in the load ow algorithm. But some must
be modeled as PV nodes, have xed values of active power into the
system and require the necessary reactive power to sustain the
voltage at the scheduled voltage magnitude. At each iteration,
the reactive and active powers of each PV node are regarded as

295

H. Li et al. / Electrical Power and Energy Systems 58 (2014) 291299

the load (output powers) at the connecting node. So the key is to

calculate and update the output reactive power increment of each
PV node.
Breakpoint is selected and opened at the connecting node of PV
node and the Thevenin equivalent circuit from the breakpoint of
each PV node is shown as Fig. 5,
In Fig. 5, the open-circuit voltage U_ di is the calculated voltage
during the load ow calculation at the connecting node i of PV
node. U_ dsi is the voltage of the PV node, whose magnitude is a predened value but angle is uncertain. Z di is the Thevenin equivalent
impedance from the breakpoint, which equals the summation of
the branches impedances along the path of connecting node i in
Tt . Let Z bt be the tree-branch impedance diagonal matrix of
N  N, then [10]

Z di Rdi jX di Tti Z bt TTti

23

Let d \U_ dsi  \U_ di and h \Z di , then the apparent power increment of PV node can be determined by,

U_ dsi  U_ di
dSdi U_ dsi
Z di

!

U 2dsi
U dsi U di
\h 
\d h
jZ di j
jZ di j

24

Expanding (24), the output active and reactive power increment

of the PV node at each iteration, dPdi and dQ di , can be expressed as,

h
i
dPdi U 2dsi cos h  U dsi U di cosd h =jZ di j
h
i
dQ di U 2dsi sin h  U dsi U di sind h =jZ di j

25

2

dQ di  2

28

jZ di j sin h
U 2di

dQ di

U 2dsi
U 2di

29

Here, sin h X di =jZ di j, substituting sin h X di =jZ di j into (29) and

rearranging it has,
2

dQ di  2c1 dQ di c2 0
X di U 2dsi =jZ di j2 ;

enter

U 2di =jZ di j2 .

q
c21  c2

31

Because the less change is the better for the output reactive
power, the minus is here. That is

dQ di c1 

First, a 33-bus, ve loops distribution system [18] and a 69-bus,

ve loops distribution system [19] are used to test the proposed
method (see Figs. 7 and 8). The test results of the proposed algorithm using powers as ow variables are compared with those of
the backward/forward sweep with compensation method [5] and
the loop theory based method [7]. As listed in Table 1, six cases
with different loops are used to test the ability of three methods
to deal with the weakly-meshed networks.
The initial values of the voltage magnitudes at all nodes are set
to be U 0 (power supply voltage magnitude) and the initial values of

30
U 2dsi U 2dsi

where c1
c2

Then, dQ di can be gotten by solving (30) as

dQ di c1 

7. Numerical tests

27

Squaring the both sides of (27) and (28), and then adding them
together to eliminate d h gives

jZ di j
U dsi U di

The load ow solution should consist of two nested iteration

loops including the meshed load ow calculation (inter-loop) and
PV node voltage magnitude compensation (outer-loop) respectively. The inter-loop solves the meshed load ow by regarding
PV node as PQ node, and the maximum node voltage magnitude
mismatch is used as the termination condition. The outer-loop
solves the output reactive power increment at each PV node and
then re-enter the inter-loop until the maximum voltage mismatch
at each PV node is small enough.
The reactive power increments during each iteration are generally small and have very little impact on the load ow of the whole
system, so the outer-loop can be cancelled. The iterative and computational ow chart of the proposed load ow algorithm is shown
in Fig. 6. It can obtain a higher efciency. First the meshed load
ow is solved at each iteration, then the reactive power increments
are calculated and the next iteration starts. But in some cases it
need more iteration numbers and even may not converge.

7.1. Efciency analysis of load ow algorithm

U dsi cos h
U di

jZ di j
sind h U dsi sin h=U di  dQ di
U dsi U di

6. Iterative and computational ow chart

26

The output active power of PV node is xed values, so dPdi 0.

Substituting it into (25) and rearranging (25) and (26) give,

cosd h

Before the beginning of the rst iteration, the initial value of

output reactive power at each PV node could be set to 0 or other
value. But note that the output reactive power Q d is limited at each
PV node. If Q d violates the upper or lower limit, it will be set to
upper or lower limit and then the PV node will be converted to
PQ node during next iteration.

q
c21  c2

32

Read data and generate related calculation

matrix
Set iteration counter k=1
Solve load flow with the proposed algorithm in
section 3 and 4 (regard PV node as PQ nodes)
Calculating and update the output reactive power
Of each PV node as described in section 5
Maximum voltage
Magnitude Mismatch
is small enough?

k=k+1

No

Yes
stop
Fig. 5. The Thevenin equivalent circuit for PV node.

Fig. 6. A ow chart of the proposed load ow solution.

296

H. Li et al. / Electrical Power and Energy Systems 58 (2014) 291299

Fig. 7. A 33-bus test system with ve loops.

Fig. 9. The iteration curves for 33 bus test system.

Fig. 8. A 69-bus test system with ve loops and six PV nodes.

Table 1
System conguration with different loops.

Fig. 10. The calculation time for 33 bus test system.

No.

Closed loops

No.

Closed loops

0
1
2

No loop
Loop 1
Loop 12

3
4
5

Loop 13
Loop 14
Loop 15

the branch power losses are set to be zero. The power ows for
33-bus system and the 69-bus system were calculated by the three
load ow algorithms under different loops respectively. The total
iteration numbers are shown in Figs. 9 and 11 and the calculation
time are shown in Figs. 10 and 12 for two tests (convergence accuracy is 106 p.u.).
As can be seen from Figs. 911, the forward/backward sweep
with compensation method needs more iteration number and
the calculation time has a signicant increase when the loops
increase. But the iteration number decreases and the calculation
time increases less and remains almost constant with the loop
theory based method and the proposed method for more loops.
The reason is that the electrical distances between loads and
source become shorter with more loops for any distribution system, and therefore the deviations of node voltage become smaller
at each iteration. So the iteration number decreases and the calculation time increases not much with more loops for the loop-analysis theory methods. It can also be seen from the converged results,
the node voltage level of the whole system increase although the
loops do not generate any extra active and reactive powers. The
reason is also that the loops shorten the electrical distances between loads and source and distribution of the load ow has been
improved.
The proposed algorithm in this paper is essentially still belongs
to the loop-analysis based method and has a strong ability to deal

Fig. 11. The iteration curves for 69 bus test system.

with meshed network. But the proposed method uses real reactive
and active power rather than complex currents as ow variables
and real operation is in place of complex operation. Thus the proposed method has better convergence and faster calculation speed
even when network become more meshed.
But it need note that the two parts in the right side of (15) and
(16) cannot be completely decoupled because of the quadratic
function relation between the branch power and node voltage.
And thus the link branch powers cannot be solved directly. The
proposed method need solve the loop currents rst before calculating P Il and Q Il , and this will bring errors to the load ow results.

H. Li et al. / Electrical Power and Energy Systems 58 (2014) 291299

Fig. 12. The calculation time for 69 bus test system.

Table 2
Some test results with the three algorithms in 69-bus system under ve loops (loop
15)/p.u.
Bus
no.

The backward/forward
sweep with
compensation

The loop
theory based
method

The proposed method

using powers as ow
variables

19
20
21
22
23
24
25
26

0.977
0.977
0.977
0.976
0.975
0.972
0.971
0.970

0.977
0.977
0.977
0.976
0.975
0.972
0.971
0.970

0.977
0.976
0.976
0.976
0.974
0.972
0.971
0.970

7
1
0
4
1
5
4
8

7
1
0
4
1
5
4
8

4
8
8
2
9
3
2
6

Table 3
The converged maximum voltage magnitude error under different loops with the
proposed method (103, p.u.).
Number of loops

1 2

Maximum voltage error

33-bus system
0
69-bus system
0
69-bus system with 2.5 times the original loads 0

0.6
0.2
1.7

0.8 1.4 1.2 1.3

0.3 0.5 0.4 0.3
2.1 3.9 3.7 1.9

297

theory based method are used as the benchmark in the 33-bus

and 69-bus test systems under different loops. Comparing them
with the results from the proposed method, the maximum voltage
magnitude errors (p.u.) for different loops are listed in Table 3. The
error terms are dened as absolute values of the differences
between calculation results from two algorithms (the same below).
From Table 3, it shows that the maximum voltage magnitude error
(p.u.) is less than 0.0014 for the original loads. And when the load
goes up 2.5 times the original load for 69-bus system, the maximum voltage magnitude error (p.u.) is less than 0.0039 but with
a bad voltage conditions.
In order to examine the algorithms robustness and accuracy, a
local 110-bus oil-eld electric power system was tested (the grid
diagram is omitted here). The system has the characteristics of
the light loads and long distribution lines with nine link branches.
The loop theory based method and the proposed method were
used to calculate the load ow. The converged results for this
light-loaded grid are shown in Table 4. The two methods have
same convergence characteristic but the proposed method has a
faster calculation speed. The results from the loop theory based
method are used as the benchmark too, and the results show that
the maximum voltage magnitude error (p.u.) is less than 0.0001 for
four cases listed in Table 4.
From Tables 3 and 4, it can be seen that the results should be able
to meet the requirement of actual precision. Since the loop theory
based method has been generally acknowledged to be one of the best
performing algorithm for distribution networks in good condition,
the proposed method can be seen as being capable of offering an
enough accuracy level for networks under good voltage conditions.
To further illustrate the algorithms ability to deal with ill-conditioned network, the results of convergence properties under heavy
load for six different network congurations in 69-bus system are
listed in Table 5. In Table 5, the load values are the ratio between
the corresponding load values and their original values, respectively. With even a very small increase (by 0.1 times) in the load
value, the load ow will diverge. From Table 5, it can be seen that
the proposed algorithm has similar convergence characteristic but
has a faster calculation speed compared with the loop theory based
method. The method has a good ability to deal with ill-conditioned
network, such as under heavy load. And from Table 5, it can be also
seen that the maximum voltage errors between two methods are
acceptable because of the ill-conditioned network.
7.2. Efciency analysis considering PV nodes

Table 4
The converged results in the 110-bus light-loaded grid.
Load ow method

The loop theory based

method

The proposed method

using powers as ow
variables

Item

System conguration with

different loops
No
loop

3
closed
loops

6
closed
loops

9
closed
loops

Iteration
number
Calculation
time (ms)

0.80

1.5

2.0

2.5

Iteration
number
Calculation
time (ms)

0.52

1.2

1.6

1.8

Table 2 shows part of the converged voltage magnitudes for 69bus test system under ve loops solved with the three load ow
method. The backward/forward sweep with compensation method
and the loop theory based method have the same results. But
compared with them, the results from the proposed method have
certain errors. The converged voltages that solved with the loop

Based on the 69-bus test system, six PV nodes are added at node
88, 46, 34, 52, 14 and 23 respectively (see Fig. 8). Their rated active
powers are 200 kW, 300 kW, 250 kW, 300 kW, 200 kW and 250 kW
in turn. The lower and upper limits of output reactive power at
each PV node are set to zero and the value of the rated active
power. The convergence accuracy is 106 p.u. and the threshold
of PV node voltage mismatch is 0.01 p.u.
The proposed method in this paper (new approach) and the
approaches in [13] (approach I) and [10] (approach II) are selected
to deal with PV nodes. The proposed load ow method using
powers as ow variables in this paper is used to solve the load
ow. Six cases in Table 6 will be analyzed and discussed.
The total iteration numbers and calculation time with the three
methods are shown in Figs. 13 and 14 respectively. It can be seen
from Figs. 13 and 14, compared with approach I and approach II,
new approach proposed in this paper has better convergence and
faster calculation speed, and the calculation time of new approach
does not increase much with more loops and more PV nodes. It
shows that the new approach is robust and effective.
The converged output reactive powers at PV nodes solved with
the three approaches are shown in Table 7 for Case1 and Case4.

298

H. Li et al. / Electrical Power and Energy Systems 58 (2014) 291299

Table 5
The algorithms convergence properties under heavy load for six cases (see Table 1) in 69-bus system.
Case

1
2
3
4
5
6

Load

3.2
3.3
3.3
3.4
5.7
6
7.4
7.5

The loop theory based method

The proposed method using powers as

ow variables

Iteration
number

Calculation time
(ms)

Iteration
number

Calculation time
(ms)

73
32
32
31
40
Diverge
34
49

8.1
6.1
6.8
7.7
11

51
33
33
32
17
21
51
Diverge

2.7
4.3
5.1
5.8
3.5
4.2
12

11
15.8

Case1
Case2
Case3
Case4
Case5
Case6

Closed loops
No loop
No loop
Loop 1&4
Loop 1&4
Loop 15
Loop 15

Minimum voltage/maximum
voltage

0
0.0072
0.0070
0.0091
0.0954

0.5020
0.5531
0.5538
0.5576
0.6112
0.5710
0.5117
0.5112

0.0423

Table 7
Part of output reactive power of PV nodes calculated with three approaches/kVar.

Table 6
System operation conguration in 69-bus system.
Case no.

Maximum voltage error

(p.u.)

PV nodes
PV13
PV16
PV13
PV16
PV13
PV16

Case no.

PV nodes

Approach I

Approach II

New approach

Case1

PV1
PV2
PV3

5.1
300
0

5.1
300
0

4.6
300
0

Case4

PV1
PV2
PV3
PV4
PV5
PV6

200
300
250
300
0
150.9

200
300
250
300
0
150.9

200
300
250
300
0
150.5

8. Conclusions
This paper presents a fast and efcient method for the load ow
solution of weakly meshed networks including PV nodes. Results of
some test systems are given to illustrate the performances of the
proposed method. The conclusions are as below.

Fig. 13. The iteration curves under six cases for 69-bus test system with PV nodes.

Fig. 14. The calculation time under six cases for 69-bus test system with PV nodes.

Because the reactive power is limited at each PV node, some PV

nodes that violated the limits have been already converted to PQ
node during the iteration. From Table 7, it can be seen that the
results with three approaches are approximately the same which
proves the feasibility and accuracy of new approach.

(1) The relationship between the branch powers and the nodeinjection powers has been developed with the node-branch
incidence matrix (path matrix and loop matrix). And based
on the accurate formulas to estimate the voltage drop and
angle difference, a load ow method was presented by using
the active and reactive power rather than the complex currents as ow variables. The real operation is in place of complex operation leads to a fasted calculation speed.
(2) The proposed load ow method takes full advantage of the
weakly meshed structure of distribution systems and has
good efciency and convergence behavior. The algorithm is
essentially still belongs to the loop-analysis based method
and has a strong ability to deal with meshed network. It
reduces the iteration number and has a faster calculation
speed even when network become more meshed.
(3) An approach based on Thevenins equivalent circuit to deal
with PV node is employed in order to x voltage magnitude
of PV nodes at specied values. An accurate calculation formula was deduced to update the reactive power injections of
each PV node at each iteration. Numerical tests veried the
correctness and effectiveness of the approach.
(4) And with appropriate modications and extensions, the proposed load ow method is potentially extensible to unbalanced load ow problems.
Acknowledgements
This work is supported nancially by the National Nature
Science Foundation of China with item #51107107 and the Chinese

H. Li et al. / Electrical Power and Energy Systems 58 (2014) 291299

National Key Basic Research and Development Program (973) with

item #2013CB228203.
Appendix A
The deduction procedure of (6) is described as following,

M 2o  M2i N2o  N2i

RPo XQ o 2 XPo  RQ o 2

2U i U o
2U i U o U 2o


RPi XQ i 2 XPi  RQ i 2
2U i U o U 2i
2

RPo XQ o 2 XPo 2 RQ o 2
2U i U o U 2o
2

RPi XQ i 2 XPi 2 RQ i 2

2U i U o U 2i




R2 X2 P2o Q 2o
R2 X2 P 2i Q 2i


2U i U o U 2o
2U i U o U 2i

 
3
2 2
2
P2i Q 2i
R2 X2 4 Po Q o
5


2U i U o
U 2o
U 2i
From (2), the following equations are true,


 
 
Po jQ o
P2 Q 2
_ I
_  Po jQ o
I2 I
o 2 o
_U o
_U o
Uo

 
 
2
2
Pi jQ i
P i jQ i
P Q

i 2 i
Ui
U_ i
U_ i
So,

 
3
2 2
2
P2 Q 2
M 2o  M2i N2o  N2i
R2 X2 4 Po Q o
5


2U i U o
2U i U o
U 2o
U 2i

R2 X2 2
I  I2 0
2U i U o

299

References
[1] Chen TH, Chen MS, Hwang KJ, et al. Distribution system power ow analysis: a
rigid approach. IEEE Trans Power Deliv 1991;6(3):114652.
[2] Thukaram D, Banda HMW, Jerome J. A robust three-phase power ow
algorithm for radial distribution systems. Electr Power Syst Res
1999;50(3):22736.
[3] Zimmerman RD, Chiang HD. Fast decoupled power ow for unbalanced radial
distribution systems. IEEE Trans Power Syst 1995;10(4):204551.
[4] Chang GW, Chu SY, Wang HL. An improved backward/forward sweep load ow
algorithm for radial distribution systems. IEEE Trans Power Syst
2007;22(2):8824.
[5] Shirmohammadi D, Hong HW, Semlyen A, et al. A compensation-based power
ow method for weakly meshed distribution and transmission networks. IEEE
Trans Power Syst 1988;3(2):75361.
[6] Luo GX, Semlyen A. Efcient load ow for large weakly meshed networks. IEEE
Trans Power Syst 1990;5(4):130916.
[7] Wu WC, Zhang BM. A three-phase power ow algorithm for distribution
system power ow based on loop-analysis method. Int J Electr Power Energy
Syst 2008;30(1):815.
[8] Che Renfei, Li Renjun. A new three-phase power ow method for weakly
meshed distribution systems. Proc CSEE 2003;23(1):749.
[9] Teng J H. A direct approach for distribution system load ow solutions. IEEE
Trans Power Deliv 2003;18(3):8827.
[10] Li Hongwei, Zhang Anan. Three-phase power ow solution for weakly meshed
distribution system including PV type distributed generation. Proc CSEE
2012;32(4):12835.
[11] Pecas Lopes JA, Hatziargyriou N, Mutale J, et al. Integrating distributed
generation into electric power systems: a review of drivers, challenges and
opportunities. Electr Power Syst Res 2007;77(9):1189203.
[12] Cheng CS, Shirmohammadi D. A three-phase power ow method for real-time
distribution system analysis. IEEE Trans Power Syst 1995;10(2):6719.
[13] Li Hongwei, Sun Hongbin, Zhang Anan, et al. Positive-sequence component
based three-phase unbalanced power ow solution for distribution system
with PV nodes. Proc CSEE 2012;32(1):11521.
[14] Zhu Y, Tomsovic K. Adaptive power ow method for distribution systems with
dispersed generation. IEEE Trans Power Deliv 2002;17(7):8227.
[15] Dukpa A, Venkatesha B, El-Hawary M. Application of continuation power ow
method in radial distribution systems. Electr Power Syst Res
2009;79(4):150310.
[16] Ding M, Guo XF. Three-phase power ow for the weakly meshed distribution
network with the distributed generation. Proc CSEE 2009;29(13):3540.
[17] Zhao Jingjing, Li Xin, Xu Zhong. Calculation of three-phase unbalanced power
ow in distribution network containing distributed generation. Power Syst
Technol 2009;33(3):948.
[18] Wang Shouxiang, Wang Chengshan. Modern distribution system
analysis. Beijing: Higher Education Press; 2007. p. 199204.
[19] Civanlar S, Grainger JJ, Yin H, et al. Distribution feeder reconguration for loss
reduction. IEEE Trans Power Deliv 1988;3:121723.