IET Generation, Transmission & Distribution

Research Article

Three-phase power flow calculations using

direct ZBUS method for large-scale unbalanced
distribution networks

ISSN 1751-8687
Received on 13th February 2015
Revised on 1st September 2015
Accepted on 7th October 2015
doi: 10.1049/iet-gtd.2015.0822
www.ietdl.org

Nien-Che Yang
Department of Electrical Engineering, and Innovation Center for Big Data and Digital Convergence, Yuan Ze University,
135 Yuan-Tung Road, Chung-Li, Taoyuan 32003, Taiwan
E-mail: [email protected]

Abstract: In this study, a three-phase power flow solution method using graph theory, injection current, and sparse matrix
techniques for large-scale unbalanced distribution networks is proposed. In the bus frame of reference, a direct iterative
method is adopted. To integrate the electric characteristics of transformers and step voltage regulators into the proposed
method, the existing component models are modified by equivalent injected currents. To validate the performance and
effectiveness of the proposed method, four three-phase IEEE test systems and random test systems are used for
comparison purposes. As the size of the network increases, the proposed direct ZBUS method drastically shows its
superiority over the other methods. The results reveal that the proposed method has good potential for improving the
computational efficiency of optimal planning and design as well as real-time power dispatch applications in large-scale
distribution systems.

Introduction

The obvious recent trend in new energy technologies has been the
application of distributed energy resources (DERs) and plug-in
hybrid electric vehicles (PHEVs), with which it is possible to
realise many advantages such as reducing greenhouse gas
emissions, deferring network upgrades, and increasing system
reliability. However, applications of DERs and PHEVs on
the power grids may pose new power quality safety problems.
The power ow solution algorithm is a basic and important tool
for the analysis of distribution systems in planning and
operational stages.
In power system applications, several common techniques used for
iterative solutions of the non-linear algebraic equations are the direct
iterative method [110], NewtonRaphson method [1117], and
Gauss method [18, 19]. The rst group is based on the direct
iterative methods i.e. the ladder and the forward/backward sweep
methods. Most of the direct iterative methods are designed for radial
distribution systems. In recent years, the direct iterative method has
become the most popularly used power ow solution algorithm in
radial distribution networks owing to its intuitive solution
procedure. However, the ladder solution method uses many
sub-iterations on laterals. Most of the radial power ow techniques
proposed in this literature assign sophisticated procedures for
numbering the radial distribution networks to execute their
algorithms. One of the main disadvantages of this kind of algorithm
is that the relationships among components are built by a direct
observation method. In a practical large radial distribution system,
the data preparation may be cumbersome and prone to errors [20].
That is, most direct iterative methods may confuse program
developers. The second group is based on the Newton algorithm. At
present, the Newton algorithm is preferred to solve power ow
problems because of its quadratic convergence characteristics.
However, the time-consuming building and inversion processes of
the Jacobian matrix cannot be avoided at every iteration.
Furthermore, it often fails to converge for ill-conditioned
distribution systems e.g. small resistance-to-resistance (X/R) ratio
of underground cables and heavy or unbalanced loads. The third
group is based on the Gauss algorithm. The Gauss implicit ZBUS
method is usually adopted to solve distribution power ow solutions

1048

by electric power companies [18, 21]. It has also been widely used
to solve optimal planning and design as well as real-time power
dispatch problems in electric power distribution networks [2226].
However, the full bus admittance matrix must be factorised into the
lower and upper (LU) triangular matrix, which is time consuming,
especially for large-scale distribution networks. If the
time-consuming procedures are avoided, the performance of other
applications based on the Gauss implicit ZBUS method will be
improved immediately.
The objective of the present paper is to develop a methodology
that can improve the solution procedure of the Gauss implicit
ZBUS method. In the proposed method, the LU factorisation,
Gaussian elimination, inverse of the full bus admittance matrix or
Jacobian matrix, and complicated building process of the bus
impedance matrix all can be avoided. By using the corresponding
incidence matrix, the unbalanced three-phase power ow
problems can be solved without the need for other complex
procedures. However, all advantages of using the Gauss implicit
ZBUS method can be achieved in the proposed method. The results
demonstrate the superiority and availability of the proposed
method for solving optimisation problems.

Gauss implicit ZBUS method

In the Gauss implicit ZBUS method, the principle of superposition

is used to evaluate system bus voltages. The superposition
principle dictates that only one type of source is considered at
the same time when calculating the bus voltages. When the
voltage source of the swing bus is activated, all injected currents
are disconnected from the system, whereas when all injected
current sources are connected to the system, the voltage source
of the swing bus is short-circuited to the ground. The major
solution procedures of the Gauss implicit ZBUS method are listed
as follows:
(k)
for loads, shunt elements,
Step 1: Compute bus injected currents IBUS
transformers, line chargings, DERs, PHEV charging demands etc.
Step 2: Compute voltage deviation due to loads, shunt elements,
transformers, line chargings, DERs, PHEV charging demands etc.

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 4, pp. 10481055
& The Institution of Engineering and Technology 2016

In this step, the time-consuming LU factorisation, Gaussian

elimination, inverse of the full bus admittance matrix, or formation
of the bus impedance matrix cannot be avoided
(k)
(k)
(k)
= Y BUS DVBUS
= [ L][U ]DVBUS
IBUS

(1)

Step 3: Apply the voltage superposition algorithm. Two different

Slack
contributions i.e. the specied voltage source VBUS
and current
(k)
injection sources DVBUS are
(k)
(k)
Slack
VBUS
= VBUS
+ DVBUS

3
3.1

(2)

Proposed algorithm

Fig. 1 One-line diagram of sample feeder

Basic concepts

To systematically represent the topology of a network, graph theory

is usually used in mathematics and computer sciences. That is, the
incidence matrix can refer to the geometric arrangements of power
system networks. In graph theory, a graph refers to a collection of
vertices and a collection of edges that connect pairs of vertices.
The terms nodes and elements are also used for vertices and
edges, respectively, especially in electrical networks. Any node of
a connected graph can be chosen as the reference node. The
variables of the other nodes, referred to as buses, can then be
measured with respect to the assigned reference. A subgraph called
a tree is a subset of all the nodes of a graph with no closed path.
The elements in a tree are called branches. Elements of a graph
that are not involved in the tree are called links. For a distribution
system with n nodes, e elements, b branches, and links, b = n 1
and e = b + can be obtained. Several incidence matrices of the
oriented graph used in this paper [27] are described as follows.
In an element-node
3.1.1 Element-node incidence matrix A:
the incidence of elements to nodes in a given
incidence matrix A,
is an e n matrix
oriented graph can be described. The matrix A
with elements of different dimensions according to the number of
element phases, where e is the number of elements and n is the
number of nodes in the graph. Entries of element-node incidence
are described as follows:
matrix A



pm , qm = +1 if phase m of the pth element is incident to and
A
oriented
away


 from phase m of the qth node;
pm , qm = 1 if phase m of the pth element is incident to and
A
oriented
toward
phase m of the qth node; and



pm , qm = 0 otherwise.
A
3.1.2 Element-bus incidence matrix A: The incidence of
elements to buses in a given directed graph can be described by an
element-bus incidence matrix A. Matrix A is an e (n1) matrix
with elements of different dimensions according to the number of
element phases, where e is the number of elements and (n1) is
is
the number of buses in the graph. In Fig. 1, substation node
selected as the reference node and the other nodes are
called buses. Matrix A can be obtained from matrix A by deleting
the column corresponding to the reference node, as shown in (3).
Ab is a non-singular square matrix with rank (n1)
buses
 
branches Ab
A=
A
links

(3)

3.1.3 Augmented element-bus incidence matrix : To

transform element-bus incidence matrix A to a square matrix of

dimension e e, the augmented element-bus matrix is dened as

shown in (4)

= Ab
A
A

[0]
U


(4)

3.1.4 Basic loop incidence matrix C: By a basic loop

incidence matrix C, the incidence of elements to basic loops in a
given oriented graph can be described. Matrix C is an e matrix
with elements of different dimensions according to the number of
phases of elements, where e is the number of elements and is the
number of links in the graph. Entries of basic loop incidence
matrix C are described as follows:
C( p m, q m) = 1 if phase m of the pth element is incident to and
assigned the same direction as phase m of the qth basic loop;
C( p m, q m) = 1 if phase m of the pth element is incident to but
assigned the opposite direction as phase m of the qth basic loop; and
C( p m, q m) = 0 otherwise.
An augmented
3.1.5 Augmented loop incidence matrix C:
loop incidence matrix C of a given oriented graph is an e e
matrix with elements of different dimensions according to the
number of element phases. In an oriented graph, the number of
basic loops is equal to the number of links. To obtain a total
number of loops equal to the number of elements, (e) loops
corresponding to the b branches are designated as open loops. An
open loop refers to the incidence of elements to paths in a given
oriented graph. Entries of open-loop incidence matrix Co are
described as follows:
Co( p m, q m) = 1 if phase m of the pth element is incident to and
assigned
the same direction as that of the qth open loop;


C o pm , qm = 1 if phase m of the pth element is incident to but
assigned
the opposite direction as that of the qth open loop; and


C o pm , qm = 0 otherwise.
Therefore, augmented loop incidence matrix C can be partitioned
into submatrices as shown in (5). In Fig. 1, the corresponding
matrices A and C are represented as (6) and (7), respectively.
Owing to the sparsity of the incidence matrices, the standard
sparse matrix techniques can be used
open basic
loops loops

 
Co
branches C o
C =
C =
[0]
[0]
links

Cb
C

Co Cb
=
[0] U

(5)

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 4, pp. 10481055

& The Institution of Engineering and Technology 2016

1049

In which
y
, m [ a, b, c
Y3mm =  t
3am bm
y
y
y
I1 = t2 Vpa ; I2 = t2 Vpb ; I3 = t2 Vpc ;
aa
ac
ab


yt
a
yt
b
Vs Vsb ; I5 =
Vs Vsc ;
I4 =
(9)
3ba bb
3bb bc



yt
c
y
V Vsa ; I7 =  t
Vpa Vsb ;
I6 =
3bc ba s
3aa bb




y
y
I8 =  t
Vpb Vsc ; I9 =  t
Vpc Vsa
3ab bc
3ac ba
and yt is the per-unit leakage admittance.
(6)

3.2.2 Step voltage regulators: There are numerous ways to

keep distribution level voltages within permissible limits. SVRs
are designed to automatically maintain a predetermined level of
voltage with time-varying load demands in distribution systems.
As the load increases, the SVR boosts the voltage to compensate
for the increased voltage drop in the primary feeders. The
equivalent circuit based on the equivalent injected currents for a
three-phase Y-connection SVR is shown in Fig. 3. To reect the
effect of the off-nominal tap ratio of SVRs, the articial injected
currents can be used. The tappings on the primary and secondary
windings of SVRs are m and m, respectively.
In which
yt
am bm

yt
bm am m
Vp , m [ a, b, c
Ipm =
am bm
am

yt
am bm m
Vs
Ism =
am bm
bm
Y3mm =

(7)

and yt is the per-unit leakage admittance.

3.3

According to (6) and (7), the relationships between matrices and C

can be found as shown in (8)
 t
Ab C o
A t C =
[0]

3.2

Atb C b + At U
U


=

Ub
[0]

[0]
U


(8)

Component models

To integrate the electric characteristics of transformers and step

voltage regulators (SVRs) into the proposed method, the existing
component models are modied by equivalent injected currents.

(10)

Formulation developments

Formulation developments of the proposed method can be

categorised into two parts: (i) radial network and (ii) non-radial
network.
3.3.1 Radial network: Since there is no link in the network, the
network system is in a radial arrangement. Each branch has two
variables, vb and ib, where vb denotes the voltage across the branch
and ib denotes the current through the branch. In terms of
admittance form, the performance equation of the primitive
network is
 
i b = yb vb

(11)

For radial networks, the performance equation of the primitive

network is pre-multiplied by Atb and yields
3.2.1 Transformers: The component models of three-phase
transformers with common connections have been presented in
[21]. Owing to space limitations, only the three-phase transformer
with the grounded Y connection is presented here. In the
proposed method, the components marked by the dashed block are
retained to form the primitive impedance matrix or primitive
admittance matrix. All other components of the equivalent circuit
for a three-phase grounded Y transformer are represented as
equivalent injected currents. The equivalent circuit based on the
injected currents for a three-phase grounded Y transformer, as
shown in Fig. 2, is adopted. The tappings on the primary and
secondary windings of transformers are and , respectively.

1050

 
Atbib = Atb yb vb

(12)

According to Kirchhoffs current law (KCL), the algebraic sum of

the current entering each node is
I BUS = Atbi b

(13)

 
I BUS = Atb yb vb

(14)

Therefore

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 4, pp. 10481055
& The Institution of Engineering and Technology 2016

By inverting the bus admittance matrix, the bus impedance matrix

can directly be obtained, as shown in (23)


1
t   1
 
= Ab yb Ab
= C to zb C o
Z BUS = Y BUS

Fig. 2 Equivalent circuit based on equivalent injected currents for

three-phase grounded Y transformer

Since the power is invariant, the powers in the interconnected and

primitive networks must be equal. That is

t


I BUS V BUS = i b t vb

t


i b Ab V BUS = i b t vb

3.3.2 Non-radial network: Each element has two variables, v

and i, where v denotes the voltage across the element, and
i denotes the current through the element. For non-radial networks,
the performance equation of the primitive network is
t , yielding
pre-multiplied by A
 
t y v
A t i = A

(16)


  t
I
Ab
=
IBUS = BUS
i
[0]

I BUS



i b
i

t i
=A

 
t y v
IBUS = A
(17)

(25)

(26)

Since the power is invariant

Substituting (17) into (14) yields

 
= Atb yb Ab V BUS

At
U

where i b and i are the vectors of currents through the branches and
links of the network, respectively.
Therefore

Since this equation is valid for all values of ib

vb = Ab V BUS

(24)

According to KCL, the algebraic sum of the current entering each

node is

(15)

Substituting (13) into (15) yields

(23)

t

t
IBUS V BUS = i v

(27)

(18)
Substituting (25) into (27) yields

In the bus frame of reference, the performance equation of the

network in admittance form is
I BUS = Y BUS V BUS

(19)

t


V BUS = i t v
i A
Since this equation is valid for all values of i


The bus admittance matrix can then be decomposed into three

non-singular square matrices

v =

 
F BUS = Atb yb Ab

Substituting (29) into (26)

(20)

vb
v

Ab
A

[0]
U



V BUS
[0]

V BUS
=A

 
t y A
V BUS
IBUS = A

Since the network is in a radial arrangement

Atb C o

(28)

= Ub

(21)

1
C o = Atb

(22)

Therefore

(29)

(30)

In the bus frame of reference, the performance equation of the

network in admittance form is
IBUS = Y BUS V BUS

(31)

 
t y A
Y BUS = A

(32)

Then

Since the network is in a non-radial arrangement

t C = U
A

(33)

 t 1
C = A

(34)

1  t   1
yA

= A
= C t [ z]C
Z BUS = Y BUS

(35)

Therefore

Therefore
Fig. 3 Equivalent circuit based on equivalent injected currents for a
three-phase Y-connection SVR

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 4, pp. 10481055
& The Institution of Engineering and Technology 2016

1051

(36)

Z2
Z4



I BUS
i


Vcn




Z1
V BUS

= ZBUS IBUS =
VBUS =
Z3
[0]

(37)

Method 4

In the bus frame of reference, the performance equation of the

network in impedance form is

Therefore, the bus impedance matrix can be obtained by (38)

(38)
Method 3

Z BUS = Z 1 Z 2 Z 1
4 Z3

In impedance form, the performance equations in the bus frame of

reference are
V BUS = Z BUSI BUS

(39)

(k)
(k)
V BUS = Z BUSI BUS

IEEE results

As mentioned previously, all loads and shunt capacitors as well as

mutual couplings between phases and phase to ground for feeder
conductors, DERs, transformers, SVRs, and electric vehicles are
represented as bus current injections. At the kth iteration, the bus
voltages contributed by equivalent current injections are

The bus impedance matrix ZBUS can be found easily as shown

in (38).
Step 5: Calculate the bus injected currents I BUS .
(k)

Method 4

Van

Method 3

Step 4: Calculate C t [ z]C and ZBUS

Node ID

(42)

Table 1 Accuracy comparison for IEEE 13 bus system

IEEE results

Step 1: Input the data.

Step 2: Perform the initial voltage estimation, data checking, and
component modelling.
as shown in
Step 3: Form the augmented loop incidence matrix C,
(7) and the primitive impedance matrix [z], as shown in (42)

1.0000120.00
1.0500120.00
1.0419121.68
1.0400121.73
1.0216122.19
1.0327121.87
1.0310121.95
1.0528122.26
1.0528122.26

1.0528122.26
1.0550122.42
1.0000120.00
1.0500120.00
1.0419121.68
1.0400121.73
1.0216122.19
1.0327121.87
1.0310121.95
1.0528122.26
1.0528122.26

1.0528122.26
1.0550122.42

Method 3

To improve the rate of convergence for the iterative process, the

variable relaxation factor can be used [28]. To describe the
solution procedure of the proposed method as described below,
step by step, the distribution system shown in Fig. 1 is used to
explain the solution procedure of the proposed method:

1.00000.00
1.06250.00
1.02102.49
1.01802.56
0.99413.23

0.99005.30
0.99005.30
0.98815.32

0.98185.25
0.99005.31
0.98355.56

(41)

1.00000.00
1.06250.00
1.02112.51
1.01812.57
0.99403.25

0.99035.34
0.99035.34
0.98845.36

0.98215.29
0.99035.34
0.98415.59

Vbn

(k+1)
Slack
(k)
V BUS = V BUS + V BUS

Method 4

(40)

By applying the superposition principle, two different contributions

i.e. the specied voltage source and current injection sources are

1052

1.0000120.00
1.0687120.00
1.0174117.83
1.0148117.82
0.9960117.34
1.0155117.86
1.0134117.90
0.9778116.02
0.9778116.02
0.9758115.92
0.9738115.78

0.9777116.02
0.9758116.03

Z4

1.0000120.00
1.0687120.00
1.0176117.76
1.0150117.75
0.9961117.27
1.0157117.78
1.0137117.83
0.9781115.87
0.9781115.87
0.9761115.77
0.9740115.61

0.9781115.87
0.9759115.88

1.0000120.00
1.0687120.00
1.0176117.76
1.0150117.75
0.9961117.27
1.0157117.78
1.0137117.83
0.9781115.87
0.9781115.87
0.9761115.77
0.9740115.61

0.9781115.87
0.9759115.88

Z2

1.0000120.00
1.0500120.00
1.0420121.72
1.0401121.72
1.0218122.22
1.0329121.90
1.0311121.98
1.0529122.34
1.0529122.34

1.0529122.34
1.0553122.52


 
Z1
C to zb C b
  =
t 
Z3
C b zb C b + z

1.00000.00
1.06250.00
1.02112.51
1.01812.57
0.99403.25

0.99035.34
0.99035.34
0.98845.36

0.98215.29
0.99035.34
0.98415.59

 
C to zb C o
 
C tb zb C o

650
RG60
632
633
634
645
646
671
680
684
611
652
692
675

Z BUS =

IEEE results

Then

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 4, pp. 10481055
& The Institution of Engineering and Technology 2016

(43)

(44)

For the specic distribution network mentioned above, the bus

injected currents I BUS can be expressed as

I BUS = I abc
1

I2a

I3bc

I4b

I5c

I6bc

In this paper, all computer programs were developed using the

MATLAB R2009b software package on a Windows 8.1-based Intel
Core i7-2600 personal computer. To avoid the effect of different
software and hardware development platforms on the execution
times, all execution times are normalised. The convergence tolerance
is 103 pu. In this paper, four methods are used for comparisons:

(k)
Step 6: Calculate the bus voltages V BUS contributed by the
equivalent current injections.

For the specic distribution network mentioned above, the bus

voltages V BUS can be expressed as

V BUS = V1abc

V2a

V3bc

V4b

V5c

V6bc

Test results

Method
Method
Method
Method

Step 7: Check for convergence. If not converged, update the bus

(k+1)
voltages V BUS contributed by the specied voltage source and
equivalent current injections. Go to Step 5; otherwise, go to the
next step.
Step 8: Stop.

4.1

1:
2:
3:
4:

the
the
the
the

Gauss method;
GaussSeidel method;
implicit Gauss ZBUS method [18]; and
proposed direct ZBUS method.

Accuracy comparison

The accuracy of the proposed method is a key factor in the

technological requirements. The IEEE 13 bus test feeder is
adopted for comparison purposes. The nal converged voltages of

Table 2 MAE and MAPE for bus voltages between the calculated solutions and IEEE results
Test case

Method 3

Method 4

MAE

IEEE-13
IEEE-34
IEEE-37
IEEE-123

MAPE

MAE

MAPE

Mag., pu

angle, deg

Mag., %

angle, %

Mag., pu

angle, deg

Mag., %

angle, %

0.00015
0.00210
0.00268
0.00047

0.048
0.251
0.322
0.025

0.015
0.206
0.272
0.046

0.177
0.479
0.634
0.333

0.00015
0.00210
0.00268
0.00047

0.048
0.251
0.322
0.025

0.015
0.206
0.272
0.046

0.177
0.479
0.634
0.333

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 4, pp. 10481055
& The Institution of Engineering and Technology 2016

1053

Table 3 NIs and NET for specific case tests

Case

Method 1
NI

IEEE-13
IEEE-34
IEEE-37
IEEE-123

18
25
350
26

Method 2

Method 3

Method 4

NET

NI

NET

NI

NET

NI

NET

6.89
37.59
400.75
100.11

9
12
243
17

3.83
21.59
299.79
75.89

3
16
3
8

1.25
22.19
4.95
22.24

3
16
3
8

1.00a
17.22
3.25
18.48

Method 3 and Method 4 are listed in Table 1. The mean absolute

error (MAE) and mean absolute percentage error (MAPE) for the
magnitudes and angles of the bus voltages for the nal converged
voltage solutions obtained by the calculated solutions and IEEE
results are listed in Table 2. The simulation results show that the
nal converged results obtained by the proposed method are
similar to those of Method 3. In terms of accuracy and precision,
the proposed method can be identied with the implicit Gauss
ZBUS method.

Normalised by the execution time of IEEE-13 for Method 4

4.2

Table 4 Feasible ranges of system parameters in TPC distribution

Performance test

To verify the performance of the proposed method, specic and

random test cases were performed as follows.

systems
Factor

Feasible ranges of system

parameters

system short-circuit capacity

X/R ratios of equivalent system
impedance of high-voltage
transmission systems
voltage levels of primary feeders
rated capacities of substation
transformers
per cent impedances of substation
transformers
X/R ratios of substation
transformers
size of feeder conductors
length of primary feeders
total load demands of primary
feeders
power factors of feeder loads

4008000 MVA
6.06.5

11.4 kV, 22.8 kV

12.5, 25, 30, 60 MVA
515%
1020
0.131 + j0.364 /km, 0.1469 +
j0.1325 /km
520 km
600 kW12 MW for 11.4 kV test
systems and 1.230 MW for 22.8
kV test systems
0.8 lagging0.8 leading

Table 5 Data for test systems

Case

1
2
3
4

Number of buses

Total length, km

50
100
500
5000

10
30
200
2000

Maximum distance, km
Mean

Standard deviation

1.55
2.8
5.46
16.15

0.32
0.43
0.76
1.75

4.2.1 Specic cases: Four IEEE test cases were used to test
specic feature purposes. Table 3 lists the number of iterations
(NIs) and the normalised execution time (NET) for specic test
cases. Compared with the implicit Gauss ZBUS method, the
percentage reductions in the execution times would range between
20 and 52% by using the proposed method.

4.2.2 Random cases: To demonstrate the validity of the

proposed method for a broad range of operating conditions in
distribution systems, the random test cases are implemented.
Sample cases are randomly produced by the practical Taiwan
Power Company (TPC) system parameters. The feasible ranges of
system parameters in practical TPC distribution systems are listed
as Table 4.
Table 5 lists the data for the tests. There are four kinds of test
cases. For each kind of case, 100 distribution systems are
generated randomly. That is, 100 combinations of system
topologies, parameters, and load conditions are generated
randomly. Tables 6 and 7 show the NI and the NET for random
test cases after 100 runs, respectively. The results show that the NI
of the proposed method is stable. In other words, the performance
of the proposed method is less sensitive to the number of buses.
As the size of networks increases, the proposed method drastically
demonstrates its superiority over the other methods. The NET of
the proposed method is up to 75 times faster than that of the
implicit Gauss ZBUS method. Dramatically, the proposed method is
960 times faster than the conventional GaussSeidel method and is
1882 times faster than the conventional Gauss method.

Table 6 Number of iterations for random case tests (100 runs)

Case

1
2
3
4

Method 1

Method 2

Method 3

Method 4

Mean

Standard deviation

Mean

Standard deviation

Mean

Standard deviation

Mean

Standard deviation

370.38
696.33
986.46
999.66

272.75
541.98
987.62
1037.93

212.02
397.06
575.97
516.21

141.60
279.70
546.52
541.97

2.96
3.18
3.32
3.86

0.38
0.47
0.55
0.77

2.96
3.18
3.32
3.86

0.38
0.47
0.55
0.77

Table 7 NET for random case tests (100 runs)

Case

Method 1
Mean

1
2
3
4

108.67
577.15
13,036.43
1,114,143.56

Method 2

Method 3

Method 4

Standard deviation

Mean

Standard deviation

Mean

Standard deviation

Mean

Standard deviation

79.90
440.18
13,059.15
1,155,963.56

65.30
334.77
7685.85
568,268.41

43.92
239.49
7382.77
600,774.12

1.74
5.28
125.81
44,612.03

0.97
1.15
25.30
8967.49

1.00a
2.17
14.32
591.82

0.35
0.59
2.98
83.85

Normalised by the execution time of Case 1 for Method 4

1054

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 4, pp. 10481055
& The Institution of Engineering and Technology 2016

Conclusions

In this paper, a three-phase power ow solution using graph theory,

injection current, sparse matrix techniques, and the direct ZBus
method was proposed to improve the performance of the
conventional Gauss implicit ZBUS method. In the proposed
method, the time-consuming LU factorisation, Gaussian
elimination, inverse of the full bus admittance matrix and the
Jacobian matrix, and complex building process of the bus
impedance matrix can be avoided. The bus impedance matrix can
be built directly by the corresponding C t [ z]C matrix. By using the
corresponding incidence matrix, the unbalanced three-phase power
ow problems can be solved without the need for other
complicated procedures.
To validate the performance and effectiveness of the proposed
method, four three-phase IEEE test systems and random test
systems were used for comparisons. As the size of networks
increased, the proposed direct ZBUS method drastically showed its
superiority over the other methods. The Gauss implicit ZBUS
method can be completely replaced by the proposed direct ZBUS
method. All advantages of using the Gauss implicit ZBUS method
are achieved in the proposed method. That is, the proposed
method can be extended to other applications by using the
conventional Gauss implicit ZBUS method. Therefore, the proposed
method is suitable as a solution engine for optimisation problems.
The results reveal that the proposed method has good potential for
improving the computational efciency of optimal planning and
design, and real-time power dispatch applications in large-scale
distribution systems.

Acknowledgment

This paper has been sponsored by the Ministry of Science and

Technology, Taiwan (MOST 104-2221-E-155-059).

References

1 Cheng, C.S., Shirmohammadi, D.: A three-phase power ow method for real-time

distribution system analysis, IEEE Trans. Power Syst., 1995, 10, (2), pp. 671679
2 Thukaram, D., Banda, H.M.W., Jerome, J.: A robust three phase power ow
algorithm for radial distribution systems, Electr. Power Syst. Res., 1999, 50, (3),
pp. 227236
3 Ciric, R.M., Feltrin, A.P., Ochoa, L.F.: Power ow in four-wire distribution
networks general approach, IEEE Trans. Power Syst., 2003, 18, (4),
pp. 12831290
4 Eminoglu, U., Hocaoglu, M.H.: A new power ow method for radial distribution
systems including voltage dependent load models, Electr. Power Syst. Res., 2005,
76, (13), pp. 106114
5 Moghaddas-Tafreshi, S.M., Mashhour, E.: Distributed generation modeling for
power ow studies and a three-phase unbalanced power ow solution for radial
distribution systems considering distributed generation, Electr. Power Syst. Res.,
2009, 79, (4), pp. 680686
6 Teng, J.H.: Modelling distributed generations in three-phase distribution load
ow, IET Gener. Transm. Distrib., 2008, 2, (3), pp. 330340

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 4, pp. 10481055
& The Institution of Engineering and Technology 2016

7 Chen, T.H., Yang, N.C.: Three-phase power-ow by direct Z(BR) method for
unbalanced radial distribution systems, IET Gener. Transm. Distrib., 2009, 3,
(10), pp. 903910
8 Chen, T.H., Yang, N.C.: Loop frame of reference based three-phase power ow
for unbalanced radial distribution systems, Electr. Power Syst. Res., 2010, 80,
(7), pp. 799806
9 Teng, J.H.: A direct approach for distribution system load ow solutions, IEEE
Trans. Power Deliv., 2003, 18, (3), pp. 882887
10 Lisboa, A.C., Guedes, L.S.M., Vieira, D.A.G., et al.: A fast power ow method for
radial networks with linear storage and no matrix inversions, Int. J. Electr. Power
Energy Syst., 2014, 63, pp. 901907
11 Lo, K.L., Zhang, C.: Decomposed three-phase power ow solution using the
sequence component frame, IEE Proc. Gener. Transm. Distrib., 1993, 140, (3),
pp. 181188
12 de Moura, A.P., de Moura, A.A.F.: NewtonRaphson power ow with constant
matrices: a comparison with decoupled power ow methods, Int. J. Electr.
Power Energy Syst., 2013, 46, pp. 108114
13 Kamh, M.Z., Iravani, R.: Unbalanced model and power-ow analysis of
microgrids and active distribution systems, IEEE Trans. Power Deliv., 2010, 25,
(4), pp. 28512858
14 Kamh, M.Z., Iravani, R.: A unied three-phase power-ow analysis model for
electronically coupled distributed energy resources, IEEE Trans. Power Deliv.,
2011, 26, (2), pp. 899909
15 Kamh, M.Z., Iravani, R.: Steady-state model and power-ow analysis of
single-phase electronically coupled distributed energy resources, IEEE Trans.
Power Deliv., 2012, 27, (1), pp. 131139
16 Li, H.W., Zhang, A.A., Shen, X., et al.: A load ow method for weakly meshed
distribution networks using powers as ow variables, Int. J. Electr. Power Energy
Syst., 2014, 58, pp. 291299
17 Peralta, J.A., de Leon, F., Mahseredjian, J.: Unbalanced multiphase load-ow
using a positive-sequence load-ow program, IEEE Trans. Power Syst., 2008,
23, (2), pp. 469476
18 Chen, T.H., Chen, M.S., Hwang, K.J., et al.: Distribution system power ow
analysis a rigid approach, IEEE Trans. Power Deliv., 1991, 6, (3),
pp. 11461152
19 Vieira, J.C.M. Jr., Freitas, W., Morelato, A.: Phase-decoupled method for
three-phase power-ow analysis of unbalanced distribution systems, IEE Proc.
Gener. Transm. Distrib., 2004, 151, (5), pp. 568574
20 AlHajri, M.F., El-Hawary, M.E.: Exploiting the radial distribution structure in
developing a fast and exible radial power ow for unbalanced three-phase
networks, IEEE Trans. Power Deliv., 2010, 25, (1), pp. 378389
21 Chen, T.H., Chen, M.S., Inoue, T., et al.: Three-phase cogenerator and transformer
models for distribution system analysis, IEEE Trans. Power Deliv., 1991, 6, (4),
pp. 16711681
22 Dzac, I., Jabr, R.A., Halilovic, E., et al.: A sensitivity approach to model local
voltage controllers in distribution networks, IEEE Trans. Power Syst., 2014, 29,
(3), pp. 14191428
23 Chen, T.H., Cherng, J.T.: Optimal phase arrangement of distribution transformers
connected to a primary feeder for system unbalance improvement and loss reduction
using a genetic algorithm, IEEE Trans. Power Syst., 2000, 15, (3), pp. 9941000
24 Huang, W.T., Chen, S.T., Yao, K.C., et al.: Three-phase line ows calculation for
distribution automation by ZBUS distribution factor. 2008 ICICIC 08 Third Int.
Conf. on Innovative Computing Information and Control; 2008, 1820 June 2008
25 Yang, N.C., Chen, T.H.: Dual genetic algorithm-based approach to fast screening
process for distributed-generation interconnections, IEEE Trans. Power Deliv.,
2011, 26, (2), pp. 850858
26 Yang, N.C., Chen, T.H.: Evaluation of maximum allowable capacity of distributed
generations connected to a distribution grid by dual genetic algorithm, Energy
Build., 2011, 43, (11), pp. 30443052
27 Stagg, G.W., El-Abiad, A.H.: Computer methods in power system analysis
(McGraw-Hill Book Company, New York, 1980, 2nd edn.)
28 Bijwe, P.R., Kelapure, S.M.: Nondivergent fast power ow methods, IEEE Trans.
Power Syst., 2003, 18, (2), pp. 633638

1055