Electric Power Systems Research 76 (2005) 106–114

A new power flow method for radial distribution systems including

voltage dependent load models
Ulas Eminoglu, M. Hakan Hocaoglu ∗
Gebze Institute of Technology, Department of Electronics Engineering, Kocaeli 41400, Turkey

Received 27 August 2004; accepted 1 May 2005

Abstract

This paper presents a simple and efficient method to solve the power flow problem in radial distribution systems. The proposed method takes
into account voltage dependency of static loads, and line charging capacitance. The method is based on the forward and backward voltage
updating by using polynomial voltage equation for each branch and backward ladder equation (Kirchoff’s Laws). Convergence ability and
reliability of the method is compared with the Ratio-Flow method, which is based on classical forward–backward ladder method, for different
loading conditions, R/X ratios and different source voltage levels, under the wide range of exponents of loads. Results demonstrate that the
proposed power flow algorithm has a robust convergence ability when compared with the improved version of the classical forward-backward
ladder method, i.e., Ratio-Flow.
© 2005 Elsevier B.V. All rights reserved.

Keywords: Radial distribution systems; Polynomial equation; Power flow; Voltage dependent loads

1. Introduction hand, the second group of methods is utilised by proper

In the last few decades, efficient and reliable load
flow solution techniques, such as: Gauss-Seidel; Newton-
Raphson; and Fast decoupled load flow [1–3], have been
developed and widely used for power system operation,
control and planning. However, it has repeatedly been shown
that these methods may become inefficient in the analysis of
distribution systems with high R/X ratios or special network
structures [4–6]. Accordingly, there are a number of reported
studies in the literature [7–17] specially designed for solu-
tion of power flow problem in radial distribution networks.
Methods developed for the solution of ill-conditioned radial
distribution systems may be divided into two categories.
The first type of methods is based on the forward–backward
sweep process for solution of ladder networks. On the other
∗ Corresponding author. Tel.: +90 262 653 84 97/10 74;
fax: +90 262 653 84 90.
E-mail addresses: u.eminoglu@gyte.edu.tr (U. Eminoglu),
hocaoglu@gyte.edu.tr (M.H. Hocaoglu).
modification of existing methods such as, Newton-Raphson.
Shirmohammadi et al. [7] have presented a compensation-
based power flow method for radial distribution networks
and/or for weakly meshed structure using a multi-port com-
pensation technique and basic formulations of Kirchhoff’s
Laws. The radial part is solved by a straightforward two-step
procedure in which the branch currents are first computed
(backward sweep) and then bus voltages are updated (for-
ward sweep). In the improved version [8], branch power flow
is used instead of branch complex current for weakly meshed
transmission and distribution systems by Luo and Semlyen.
In [9], Baran and Wu propose a methodology for solving
radial power flow for analysing optimal capacitor sizing
problem. In this method, for each branch of the network three
non-linear equations are written in terms of branch power
flows and bus voltages. The number of equations is subse-
quently reduced using terminal conditions associated with
the main feeder and its laterals, and Newton-Raphson method
is applied to this reduced set. Computational efficiency is
improved by making some simplifications in the jacobian.

0378-7796/$ – see front matter © 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.epsr.2005.05.008
 U. Eminoglu, M.H. Hocaoglu / Electric Power Systems Research 76 (2005) 106–114 107

Consequently, numerical properties and convergence rate of 2. Power flow formulation for radial distribution
this algorithm have been studied using the iterative solution systems
of three fundamental equations representing real power, reac-
tive power and voltage magnitude by Chiang in [10].Renato In conventional load flow studies, it is presumed that active
Cespedes [11] makes use of well-known bi-quadratic equa- and reactive power demands are specified constant values,
tion, which relates the voltage magnitude at the receiving regardless of the amplitude of voltages in the same bus.
end to the voltage at the sending end and branch power flow In actual power systems operation, different categories and
for every branch. In this study, only voltage magnitudes are types of loads such as residential, industrial and commercial
computed, bus phase angles do not appear in his formulation, loads might be present. The nature of these types of loads
which was also used by Das et al. [12]. Jasmon and Lee [13], is such that their active and reactive powers are dependent
have proposed a load flow technique for every branch, which on the voltage and frequency of the system. Moreover, load
leads to a pair of quadratic equations relating power flows characteristics have significant effects on load flow solutions
at both ends with the voltage magnitude at the sending end and convergence ability [14]. Common static load models for
for the voltage stability analysis of radial networks. Haque active and reactive power are expressed in a polynomial or an
[14] have formulated power flow problem of distribution exponential form. The characteristic of the exponential load
systems in terms of three sets of recursive equations and models can be given as:
analysed power flow results for various voltage dependent
np
load models. The effects of various load models on the con- V
P = P0 (1)
vergence pattern of the method are also studied. The effect V0
of voltage-dependency of load on the results of power flow
nq
V
solution are also analysed by Mok et al. [15]. Authors have Q = Q0 (2)
shown a new approach for power flow analysis of distribution V0
systems, which is based on Kirchoff’s Laws and demon- where np and nq stand for load exponents, P0 and Q0 stand
strated the effect of different load models on the convergence for the values of the active and reactive powers at the nominal
ability of their method. In ref. [16] authors have proposed voltages. V and V0 stand for load bus voltage and load nominal
Ratio-Flow method, which is based on forward–backward voltage, respectively. Special values of the load exponents
ladder equation for complex distribution system by using can cause specific load types such as: 0, constant power; 1,
voltage ratio for convergence control. This method was constant current; 2, constant impedance. The polynomial load
applied with the standard Newton-Raphson method for model is a static load model that represents the power–voltage
complex distribution systems, which have multiple sources relationship as a polynomial equation of voltage magnitude.
or relatively strong connected loops with extended long It is usually referred as ZIP model, as it is made up of three
radial feeders including laterals, to solve power flow different exponential load models: constant impedance (Z),
problem. Ranjan and Das [17] have proposed a new method constant current (I) and constant power (P) static load models.
to solve radial distribution networks. They have used simple Common values for exponents of static loads are given in
algebraic recursive expression of voltage magnitude and the Table 1 [19–21]. For practical application, the evaluation of
proposed algorithm uses the basic principle of the circuit coefficients np and nq requires use of parameter estimation
theory. techniques.
The aim of this paper is to propose a new power flow The structure of many distribution systems is like a tree
method for radial distribution networks with improved con- with several laterals and sublaterals. Load voltage for branch
vergence characteristics. It is based on polynomial equation of any distribution system, given in Fig. 1, can be calculated
on the forward process and backward ladder equation for
each branch of radial distribution system. In the proposed Table 1
method, line shunt capacitance and exponents of static load Common values for the exponents for different static load models
are included in power flow solution. The proposed method Load component np nq
is tested on two ill-conditioned radial systems for different
Battery charge 2.59 4.06
voltage-dependent load models, and then it is compared with Fluorescent lamps 2.07 3.21
the results of Ratio-Flow method [16] and evaluated against Constant impedance 2 2
a commercial grade power flow program PFLOW [18]. Fluorescent lighting 1 3
A series of tests are under taken to evaluate the proposed Air conditioner 0.5 2.5
method for different loading conditions, different R/X ratios Constant current 1 1
Resistance space heater 2 0
and different voltage levels, under wide range of exponents Pumps, funs other motors 0.08 1.6
of loads in radial distribution systems. Results show that the Incandescent lamps 1.54 0
proposed method has fast and reliable convergence ability Compact fluorescent lamps 1 0.35
when compared with Ratio-Flow method, which is known Small industrial motors 0.1 0.6
with its faster convergence characteristics amongst various Large industrial motors 0.05 0.5
Constant power 0 0
sweep methods [16].
108 U. Eminoglu, M.H. Hocaoglu / Electric Power Systems Research 76 (2005) 106–114

Fig. 2. The per-phase equivalent circuit of one feeder without sub-branch.

Fig. 1. The ␲-circuit model of a branch.
(lateral), as given in Fig. 2, the backward ladder formula can
using the polynomial equation of the branch relates the be written as:
voltage magnitude at the receiving end to the voltage at the
n n−1
 n 
sending end and transferred active and reactive power.   
Vs = Vn + It Zk − Ik Zi (11)
In Fig. 1, transmitted real and reactive power can be
k=1 k=1 i=k+1
derived in terms of the ABCD parameters of lossy line model
using following notation by assuming δr = 0: where n is the number of nodes in the feeder; Vs and Vn stand
for the sending end, receiving end voltages, respectively;
|Vr ||Vs | |A||Vr | 2
P= cos(θZ − δs ) − cos(θZ − δA ) (3) Zk and Zi are the impedance of the kth and ith section of
|Z| |Z| the feeder; Lk is the load at node k; Ik is the load current
at node k; It is the total current sent from the sending end
|Vr ||Vs | |A||Vr | 2
Q= sin(θZ − δs ) − sin(θZ − δA ) (4) node.
|Z| |Z| This formula is used to calculate bus voltages in the back-
√
γ = zy (5) ward process of the proposed method.

A = cosh(γ) (6)

where Vs and Vr stand for voltages at bus s and bus r; z and
y are the line series impedance and shunt admittance. θ Z , δs
and δA stand for phase angles of the line series impedance,
voltage of bus (s) and A parameter of line given in Eq. (6),
respectively. Eqs. (3) and (4) can be rewritten as:
3. Proposed solution method
Consider the simple network in Fig. 2, one feeder without
any sub-branch, the proposed method can be summarized for
the same exponents of all load as in the following algorithm.

P|Z| |A||Vr | • Step 1: Initialize the iteration counter, i.e., i = 1 and the bus
cos(θZ − δs ) = + cos(θZ − δA ) (7) voltages of the feeder, i.e., V1 = V2 = V3 = . . . = Vn = Vs and
|Vs ||Vr | |Vs |
Vs is generally chosen as 1 pu in distribution systems.
Q|Z| |A||Vr | • Step 2: Perform the forward process using Eq. (10) to
sin(θZ − δs ) = + sin(θZ − δA ) (8)
|Vs ||Vr | |Vs | obtain bus voltages Vk starting from bus (0). For each load
bus, use total load power summing all load power of net-
using trigonometric identity,
works fed through the bus and use obtained bus voltage as
cos2 (θZ − δs ) + sin2 (θZ − δs ) = 1 (9) a sending end voltage of the next branch of the feeder.
• Step 3: Calculate new active and reactive power of loads
and substituting Eqs. (1), (2), (7) and (8) in to Eq. (9), the using Eqs. (1) and (2), and then calculate loads current
polynomial equation of the load voltage obtained as: profile.
|A|2 Vr4 + 2|A|Vr2 |Z|(P0 Vrnp cos(θZ − δA )
∗
Sk
Ik = (12)
+Q0 Vrnq sin(θZ − δA )) − Vs2 Vr2 Vk

+(P02 Vr2np + Q20 Vr2nq )|Z|2 = 0 (10) • Step 4: Perform the backward process using Eq. (11) to
obtain new bus voltages Vknew and new sending end voltage
Eq. (10) has a straightforward solution and depends on phase Vsnew starting from bus (n) at the end of feeder.
angle, voltage dependency of static load and line parameters. • Step 5: Calculate the ratio of the new sending end voltage
In the proposed algorithm, Vr is numerically obtained by and adjust the Vknew as:
using a Matlab routine, based on least square optimization
techniques, for different static load models. It is noted that Vsnew
Vs r = (13)
from solutions for Vr only the highest positive real root of Vs
this equation is used in the algorithm. The forward process of
the algorithm is based on this polynomial equation for each Adjust Vknew
Vk = (14)
branch of the system. For any feeder without sub-branch Vs r
 Battery charge
Fluorescent lamps
Constant impedance
Air conditioner
Constant current
Pumps, funs other motors
Small industrial motors
Constant power
Tolerance

Load model
Number of iteration required to solution of the power flow problem
Table 2

Fig. 3. Flow chart of the proposed load flow technique.

2
2
2
2
2
2
2
2
10−4 10−5 10−6

Proposed method
3
3
2
2
2
3
3
3

U. Eminoglu, M.H. Hocaoglu / Electric Power Systems Research 76 (2005) 106–114

3
3
3
3
3
3
3
4
3
2
3
3
2
3
3
3
10−4 10−5 10−6

Ratio-Flow method
3
3
3
3
2
3
3
3
4
3
4
4
2
4
4
4

Table 3
Results of power flow solution by two methods for different static load types (pu)
Load type Constant power Small industrial motors Pumps, funs other motors Constant current Air conditioner Constant impedance Fluorescent lamps Battery charge

Node Proposed R.-F. M. PFLOW Proposed R.-F. M. Proposed R.-F. M. Proposed R.-F. M. Proposed R.-F. M. Proposed R.-F. M. Proposed R.-F. M. Proposed R.-F. M.
1 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
2 0.9888 0.9888 0.9887 0.9889 0.9890 0.9891 0.9891 0.9894 0.9894 0.9894 0.9895 0.9900 0.9900 0.9902 0.9902 0.9904 0.9904
3 0.9790 0.9790 0.9789 0.9793 0.9793 0.9796 0.9796 0.9802 0.9802 0.9802 0.9803 0.9812 0.9812 0.9816 0.9816 0.9821 0.9821
4 0.9655 0.9655 0.9654 0.9660 0.9660 0.9665 0.9666 0.9676 0.9676 0.9675 0.9676 0.9693 0.9694 0.9699 0.9699 0.9707 0.9708
5 0.9539 0.9539 0.9538 0.9546 0.9546 0.9552 0.9553 0.9567 0.9567 0.9566 0.9567 0.9591 0.9591 0.9598 0.9599 0.9610 0.9611
6 0.9429 0.9429 0.9429 0.9438 0.9438 0.9446 0.9446 0.9465 0.9465 0.9463 0.9464 0.9494 0.9495 0.9503 0.9505 0.9518 0.9519
7 0.9354 0.9354 0.9354 0.9364 0.9365 0.9373 0.9374 0.9396 0.9396 0.9393 0.9395 0.9430 0.9431 0.9440 0.9442 0.9457 0.9459
8 0.9315 0.9314 0.9314 0.9325 0.9326 0.9335 0.9336 0.9359 0.9360 0.9357 0.9358 0.9397 0.9397 0.9407 0.9409 0.9425 0.9427
9 0.9268 0.9268 0.9268 0.9280 0.9280 0.9290 0.9291 0.9317 0.9317 0.9313 0.9315 0.9357 0.9358 0.9369 0.9370 0.9388 0.9390
10 0.9246 0.9246 0.9246 0.9258 0.9258 0.9268 0.9269 0.9296 0.9297 0.9293 0.9294 0.9338 0.9339 0.9350 0.9352 0.9370 0.9372
11 0.9237 0.9237 0.9237 0.9249 0.9250 0.9260 0.9261 0.9288 0.9289 0.9285 0.9286 0.9331 0.9332 0.9343 0.9344 0.9363 0.9365
12 0.9235 0.9235 0.9234 0.9247 0.9247 0.9257 0.9259 0.9286 0.9287 0.9282 0.9284 0.9329 0.9330 0.9341 0.9342 0.9361 0.9363

109
110 U. Eminoglu, M.H. Hocaoglu / Electric Power Systems Research 76 (2005) 106–114

• Step 6: Return to Step 2 until the convergence tolerance, Table 4

as given in Eq. (15), is reached. Number of iteration required to solution of the power flow problem
Load model Proposed method Ratio-Flow method
j+1 j
||Vk | − |Vk || ≤ ε (15) Constant power 3 4
Small industrial motors 3 4
• Step 7: Repeat Steps 3–5 again to obtain new phase angles Pumps, funs other motors 3 4
of bus voltages and all parameters of system. Constant current 3 3
Air conditioner 2 4
It is noted that the algorithm does not require initial bus Constant impedance 3 4
voltages in case of constant power load models and after Fluorescent lamps 3 4
the first iteration, the forward process have to be performed Battery charge 3 5
by starting from bus (1) to obtain different bus voltages on
the feeder in Step 2. This algorithm can be applied to the proposed method. The algorithm has been coded in Mat-
whole system, which has sub-branches (lateral) by adding lat- lab and results compared with Ratio-Flow method [16]. To
eral sub-iterations. For the lateral sub-iterations, lateral node demonstrate the accuracy, calculated voltages for constant
voltages are used as lateral convergence target. In adjust- power load model are evaluated against a standard power
ing of lateral node voltages, for calculating the ratio Vs r , flow program PFLOW [18], which is actually designed for
lateral sending end voltage on the feeder is used instead of transmission systems, and also used in distribution studies
feeder sending end voltage Vs in Eq. (13). When the lat- [23]. An 11 kV distribution system having 30 buses (includ-
eral reaches convergence tolerance, lateral total current and ing the source bus) is used to observe the reliability of the
power are calculated and added its sending end on the feeder. method with different static load models. The system has a
The obtained total current and power of lateral are used for main feeder and three laterals. Single line diagram of the
the backward and forward processes, respectively. Then the system is given in Fig. 4. The data for the system is obtained
above algorithm is continued until voltages along the feeder from Ref. [22] and given in Appendix A. Initial voltage mag-
reach the convergence tolerance. The algorithm of the method nitude at all buses is considered to be the same as the source
for whole system, which has laterals, is given in the form of bus. A tolerance of 10−4 on voltage magnitude is used for
a flow chart in Fig. 3. the studies.

4.1. Case study 1

4. Application to radial distribution networks
In this study, the convergence characteristic and reliabil-
Two test systems (distribution networks with and with- ity analysis of the proposed method is performed for seven
out laterals) are used to test the convergence ability of the typical exponential load models in a 12-bus radial system.

Fig. 4. Single-phase circuit of a 30-bus distribution system.

Table 5
Results of power flow solution with two methods for different static load types (pu)
Load type Constant power Small industrial motors Pumps, funs other motors Constant current Air conditioner Constant impedance Fluorescent lamps Battery charge
Node Proposed R.-F. M. PFLOW Proposed R.-F. M. Proposed R.-F. M. Proposed R.-F. M. Proposed R.-F. M. Proposed R.-F. M. Proposed R.-F. M. Proposed R.-F. M.

U. Eminoglu, M.H. Hocaoglu / Electric Power Systems Research 76 (2005) 106–114

1 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000
2 0.9885 0.9885 0.9884 0.9888 0.9887 0.9890 0.9890 0.9894 0.9894 0.9894 0.9895 0.9902 0.9902 0.9904 0.9904 0.9907 0.9908
3 0.9785 0.9784 0.9784 0.9789 0.9789 0.9793 0.9794 0.9802 0.9802 0.9802 0.9803 0.9816 0.9816 0.9820 0.9821 0.9827 0.9827
4 0.9647 0.9645 0.9645 0.9654 0.9653 0.9660 0.9661 0.9676 0.9676 0.9674 0.9676 0.9699 0.9700 0.9706 0.9707 0.9717 0.9718
5 0.9528 0.9525 0.9525 0.9537 0.9537 0.9545 0.9546 0.9568 0.9567 0.9564 0.9567 0.9599 0.9600 0.9608 0.9610 0.9623 0.9624
6 0.9416 0.9412 0.9412 0.9427 0.9426 0.9436 0.9438 0.9466 0.9465 0.9460 0.9464 0.9505 0.9506 0.9516 0.9518 0.9534 0.9537
7 0.9341 0.9337 0.9337 0.9354 0.9352 0.9364 0.9366 0.9397 0.9396 0.9390 0.9394 0.9441 0.9442 0.9452 0.9455 0.9473 0.9476
8 0.9301 0.9297 0.9297 0.9315 0.9313 0.9325 0.9327 0.9360 0.9360 0.9354 0.9357 0.9407 0.9408 0.9419 0.9422 0.9441 0.9444
9 0.9255 0.9250 0.9251 0.9269 0.9268 0.9280 0.9282 0.9318 0.9317 0.9310 0.9314 0.9368 0.9369 0.9380 0.9383 0.9403 0.9407
10 0.9232 0.9228 0.9228 0.9247 0.9246 0.9259 0.9261 0.9297 0.9297 0.9290 0.9293 0.9349 0.9350 0.9362 0.9365 0.9386 0.9389
11 0.9224 0.9219 0.9220 0.9239 0.9238 0.9250 0.9252 0.9289 0.9289 0.9282 0.9285 0.9342 0.9342 0.9355 0.9358 0.9379 0.9382
12 0.9221 0.9217 0.9217 0.9236 0.9235 0.9248 0.9250 0.9287 0.9287 0.9279 0.9283 0.9339 0.9340 0.9352 0.9355 0.9377 0.9380
13 0.9780 0.9778 0.9778 0.9784 0.9784 0.9788 0.9789 0.9797 0.9797 0.9796 0.9798 0.9811 0.9811 0.9815 0.9816 0.9822 0.9822
14 0.9775 0.9774 0.9773 0.9779 0.9779 0.9783 0.9784 0.9792 0.9792 0.9792 0.9793 0.9806 0.9806 0.9810 0.9811 0.9817 0.9818
15 0.9773 0.9772 0.9772 0.9778 0.9778 0.9782 0.9782 0.9791 0.9791 0.9790 0.9791 0.9805 0.9805 0.9809 0.9810 0.9816 0.9816
16 0.9773 0.9772 0.9772 0.9778 0.9777 0.9782 0.9782 0.9791 0.9791 0.9790 0.9791 0.9805 0.9805 0.9809 0.9810 0.9816 0.9816
17 0.9325 0.9321 0.9321 0.9339 0.9338 0.9350 0.9352 0.9385 0.9384 0.9379 0.9382 0.9432 0.9432 0.9444 0.9446 0.9465 0.9468
18 0.9250 0.9245 0.9246 0.9266 0.9264 0.9279 0.9280 0.9317 0.9317 0.9311 0.9314 0.9370 0.9371 0.9384 0.9387 0.9409 0.9411
19 0.9166 0.9162 0.9162 0.9185 0.9183 0.9199 0.9201 0.9243 0.9243 0.9236 0.9239 0.9304 0.9304 0.9319 0.9322 0.9346 0.9349
20 0.9099 0.9095 0.9095 0.9119 0.9118 0.9135 0.9137 0.9184 0.9183 0.9176 0.9179 0.9250 0.9250 0.9267 0.9269 0.9297 0.9300
21 0.9040 0.9036 0.9037 0.9063 0.9061 0.9080 0.9082 0.9132 0.9132 0.9123 0.9126 0.9204 0.9204 0.9222 0.9224 0.9254 0.9257
22 0.8973 0.8969 0.8970 0.8998 0.8996 0.9016 0.9018 0.9074 0.9073 0.9063 0.9067 0.9151 0.9152 0.9171 0.9173 0.9205 0.9208
23 0.8918 0.8914 0.8915 0.8944 0.8943 0.8964 0.8966 0.9026 0.9025 0.9014 0.9017 0.9108 0.9109 0.9129 0.9131 0.9166 0.9168
24 0.8867 0.8862 0.8863 0.8894 0.8893 0.8915 0.8916 0.8981 0.8980 0.8968 0.8971 0.9068 0.9069 0.9090 0.9092 0.9128 0.9131
25 0.8842 0.8837 0.8838 0.8870 0.8868 0.8891 0.8893 0.8959 0.8958 0.8946 0.8949 0.9049 0.9049 0.9071 0.9073 0.9111 0.9113
26 0.8832 0.8828 0.8829 0.8861 0.8859 0.8882 0.8884 0.8951 0.8950 0.8937 0.8940 0.9041 0.9042 0.9064 0.9066 0.9104 0.9107
27 0.8829 0.8825 0.8826 0.8858 0.8856 0.8879 0.8881 0.8948 0.8947 0.8935 0.8938 0.9039 0.9040 0.9061 0.9064 0.9102 0.9105
28 0.9333 0.9329 0.9329 0.9346 0.9345 0.9356 0.9358 0.9390 0.9389 0.9383 0.9387 0.9434 0.9435 0.9446 0.9448 0.9467 0.9469
29 0.9328 0.9324 0.9324 0.9341 0.9340 0.9351 0.9353 0.9385 0.9384 0.9378 0.9382 0.9430 0.9431 0.9441 0.9444 0.9462 0.9465
30 0.9325 0.9321 0.9321 0.9339 0.9337 0.9349 0.9351 0.9382 0.9382 0.9376 0.9379 0.9428 0.9428 0.9439 0.9442 0.9460 0.9463

111
112 U. Eminoglu, M.H. Hocaoglu / Electric Power Systems Research 76 (2005) 106–114

The system is composed from 30-bus system omitting later-
als and total power of each laterals are added to its sending
end on the feeder. Tables 2 and 3 show the results of power
flow solution and number of iterations by using three meth-
ods. From Table 2, it is clear that, for different load models,
the proposed method has fast convergence speed when com-
pared with the Ratio-Flow method for the tolerance values
from 10−4 to 10−6 on voltage magnitudes. From the power
flow solution provided in Table 3, it can be concluded that
the proposed method is robust and reliable for the problem in
distribution systems. Both methods are converged in almost
the same voltage magnitudes. Absolute difference between
voltage magnitudes obtained using both methods is in the
range of maximum 0.0002 pu, which occurs for the higher
number of load exponents. The same observation could be
made for the phase angles.
4.2. Case study 2
In this study, the proposed algorithm, given in Fig. 3, is
tested on the 30-bus radial system with laterals to analyse
performance of the method for different voltage-dependent
load models. The single line diagram of the distribution sys-
tem and its data are given in Fig. 4 and in Appendix A,
respectively. The convergence speed of the proposed method
is compared with Ratio-Flow method and evaluated against

Fig. 5. Variation of the number of iteration for different parameters of the system for the proposed method (䊉), Ratio-Flow (
) [(a) different tolerance values;
(b) different source voltage levels; (c) different loading conditions; (d) different R/X ratio].
 U. Eminoglu, M.H. Hocaoglu / Electric Power Systems Research 76 (2005) 106–114
PFLOW results for constant power load model for a tolerance
of 10−4 on voltage magnitude. The effect of different charac-
teristics of loads on the performance of methods and system’s
voltage profile are analysed with seven types of the common
static loads given in Table 1. From Table 4, it can be said
that the proposed method has faster convergence ability than
Ratio-Flow method for different types of loads. Moreover,
from Table 5, different load models cause different voltage
profiles and the most voltage-dependent load model provides
the highest voltage magnitudes as expected. However, in this
case the convergence gets difficult and the required num-
ber of iteration increases in Ratio-Flow method because the
load values are adjusted in each iteration according to newly
calculated voltage magnitudes. Again, the results obtained
using both methods are in close agreement. Maximum dif-
ference between results for all types of loads is in the range
of 0.0004 pu.
The convergence characteristic of methods is also anal-
ysed for various system parameters and loading conditions
for different value of exponents of loads. The value of
np and nq are selected as equal and varied from 0 to 5.
The number of iterations required to solve the power flow
problem for various voltage levels, loading conditions,
R/X ratios and tolerance values in the 30-bus distribution
system are given in Fig. 5. Results show that, the proposed
method has an obvious advantage on the convergence speed
over Ratio-Flow method under different system conditions.
Convergence tolerance for both methods are varied from
10−2 to 10−5 and results depicted in Fig. 5a. For all
tolerance values, the proposed algorithm needs less number
of iteration than Ratio-Flow method. It is also evident from
Fig. 5a that the proposed algorithm is less sensitive to the
tolerances.
Sending end source voltage (Vs ) are varied from 0.9
to 1.1 pu by keeping tolerances at 10−4 , and results are
depicted in Fig. 5b. For different source voltage levels, the
proposed method has fast convergence speed under wide
range of load exponents as shown in the figure. In addition
to that, required iteration number for the proposed method
is not affected by the magnitude of the source voltage. On
the other hand, convergence speed of Ratio-Flow method
is more sensitive to load exponents when source voltage is
increased.
Different loading conditions are considered by multi-
plying each nodes power by a load factor (λ) as S = λ × S
and results depicted in Fig. 5c. From the figure, for heavy
loading conditions of the system, maximum number of
iteration for the proposed method is only equal to four. On
the other hand, Ratio-Flow converges in the ninth iteration
for the same load. Moreover, the proposed method is not
sensitive to the loading levels. On the contrary, convergence
speed of Ratio-Flow method is heavily under influence of
load exponents when system’s loads are increased. Finally,
branch impedances R/X ratios are varied multiplying each
branch resistance by a coefficient (k) as Z = k × R + jX
and results depicted in Fig. 5d. Likewise, it is shown
113
that convergence speed of the proposed method is less
effected with the increase of R/X ratios of the system when
compared with the Ratio-Flow method. It is known that the
Ratio-Flow method is based on current summation, thus,
when loads are modelled as a constant current (np = nq = 1),
the required iteration is in the minimum level as shown
in Fig. 5.
The number of iteration needed to perform power flow
study is given in Fig. 5 for wide range of load exponents. It
can be observed that the proposed method converged very
quickly and less sensitive to load exponents and parameters
of the system. It takes four iterations for the maximum load
exponents when the system is heavily loaded, R/X ratio and
source voltage level are increased for the proposed method.
On the other hand, the Ratio-Flow method needs 10 iterations
to solve power flow study at the same system as can be seen
in Fig. 5.
5. Conclusions
In this paper, a new power flow technique, which is simple,
efficient and has fast convergence ability, has been proposed
for the power flow analysis in ill-conditioned radial distri-
bution networks. The method is based on the forward and
backward voltage updating by using polynomial equation for
each branch and backward ladder formula. The advantages of
the proposed method are that, it has fast convergence ability
and gives an ability to use of the line shunt capacitance and
exponents of static loads in solution of power flow problems.
Although a comprehensive comparison on the computation
time is not under taken, the initial results suggest that
computation time of the proposed algorithm is somehow
slower than Ratio-Flow method. This partly due to the
usage of a general built in Matlab routine for the solution of
Eq. (10).
The new technique has been applied to 12-bus distri-
bution system (without laterals) and 30-bus distribution
system (with three laterals). The power flow problem
has been successfully solved for two distribution systems
with different static load types and obtained results by two
methods have been compared. They have also been evaluated
under different loading conditions, different R/X ratios and
different voltage levels, under wide range of load exponents
in radial distribution systems. Results show that the proposed
method is robust, simple and has fast convergence ability.
Moreover, it needs less number of iteration than Ratio-
Flow method and less sensitive to the distribution system
parameters.
The study will continue to improve the algorithm by
implementing an optimized numerical solver for Eq. (10).
Further parametrical analysis, specifically aimed to deter-
mine the computational burden, complexity and overall
speed of the proposed algorithm, will be under taken as a
future work. In the second phase of the study, the algorithm
will be extended for power flow solution of balanced and
114 U. Eminoglu, M.H. Hocaoglu / Electric Power Systems Research 76 (2005) 106–114

unbalanced three-phase distribution systems including local [5] S. Iwamoto, Y. Tamura, A load flow-calculation method for ill-
generation. conditioned power systems, IEEE Trans. Power Apparatus Syst.
PAS-100 (4) (1981) 1736–1743.
[6] G.W. Stagg, A.H. El-Abiad, Computer Methods in Power System
Appendix A. Data for the 30-bus distribution system Analysis, McGraw-Hill, 1968.
[7] D. Shirmohammadi, H.W. Hong, A. Semlyen, G.X. Luo, A
compensation-based power-flow method for weakly meshed distri-
bution and transmission networks, IEEE Trans. Power Syst. 3 (2)
Feeder section Load R + jX (pu) (1988) 753–762.
P (pu) Q (pu) [8] G.X. Luo, A. Semlyen, Efficient load flow for large weakly meshed
networks, IEEE Trans. Power Syst. 5 (4) (1990) 1309–1316.
1–2 0.0042 0.0026 0.0967 + 0.0397i [9] M.E. Baran, F.F. Wu, Optimal sizing of capacitors placed on a
2–3 0 0 0.0886 + 0.0364i radial distribution system, IEEE Trans. Power Delivery 4 (1) (1989)
3–4 0.0042 0.0026 0.1359 + 0.0377i 735–743.
4–5 0.0042 0.0026 0.1236 + 0.0343i [10] H.D. Chiang, A decoupled load flow-method for distribution power
5–6 0 0 0.1236 + 0.0343i networks: algorithms, analysis and convergency study, Elec. Power
6–7 0 0 0.2598 + 0.0446i Energy Syst. 13 (3) (1991) 130–138.
7–8 0.0042 0.0026 0.1732 + 0.0298i [11] G. Renato Cespedes, New method for the analysis of distribution
8–9 0.0042 0.0026 0.2598 + 0.0446i networks, IEEE Trans. Power Delivery 5 (1) (1990) 391–396.
9–10 0.0041 0.0025 0.1732 + 0.0298i [12] D. Das, D.P. Kothari, A. Kalam, Simple and efficient method for load
10–11 0.0042 0.0026 0.1083 + 0.0186i flow solution of radial distribution networks, Elec. Power Energy
11–12 0.0025 0.0015 0.0866 + 0.0149i Syst. 17 (5) (1995) 335–346.
3–13 0.0011 0.0007 0.1299 + 0.0223i [13] G.B. Jasmon, L.H.C. Lee, Distribution network reduction for voltage
13–14 0.0011 0.0007 0.1732 + 0.0298i stability analysis and load flow calculations, Elec. Power Energy
14–15 0.0011 0.0007 0.0866 + 0.0149i Syst. 13 (1) (1991) 9–13.
15–16 0.0002 0.0001 0.0433 + 0.0074i [14] M.H. Haque, Load flow-solution of distribution systems with voltage
6–17 0.0044 0.0027 0.1483 + 0.0412i dependent load models, Elec. Power Syst. Res. (36) (1996) 151–156.
17–18 0.0044 0.0027 0.1359 + 0.0377i [15] S. Mok, S. Elangovan, C. Longjian, M. Salama, A new approach for
18–19 0.0044 0.0027 0.1718 + 0.0391i power-flow analysis of balanced radial distribution systems, Electric
19–20 0.0044 0.0027 0.1562 + 0.0355i Machines Power Syst. (28) (2000) 325–340.
20–21 0.0044 0.0027 0.1562 + 0.0355i [16] J. Liu, M.M.A. Salama, R.R. Mansour, An efficient power-flow algo-
21–22 0.0044 0.0027 0.2165 + 0.0372i rithm for distribution systems with polynomial load, Int. J. Elec. Eng.
22–23 0.0044 0.0027 0.2165 + 0.0372i Edu. 39 (4) (2002) 372–386.
23–24 0.0044 0.0027 0.2598 + 0.0446i [17] R. Ranjan, D. Das, Simple, Efficient computer algorithm to solve
24–25 0.0044 0.0027 0.1732 + 0.0298i radial distribution networks, Electric Power Comp. Syst. (31) (2002)
25–26 0.0044 0.0027 0.1083 + 0.0186i 95–107.
26–27 0.0026 0.0016 0.0866 + 0.0149i [18] C.A. Canizares et al., PFLOW: continuation and direct meth-
7–28 0.0017 0.0011 0.1299 + 0.0223i ods to locate fold bifurcations in AC/DC/FACTS power sys-
28–29 0.0017 0.0011 0.1299 + 0.0223i tems, University of Waterloo, August 1998, available at
29–30 0.0017 0.0011 0.1299 + 0.0223i http://www.power.uwaterloo.ca.
[19] T.V. Cutsem, C. Vournas, Voltage Stability of Electric Power Sys-
tems, Power Electronics and Power System Series, Kluwer, 1998.
[20] C.W. Taylor, Power System Voltage Stability, Electric Power
References Research Institute, McGraw-Hill, 1994.
[21] G. Le Dous, Voltage Stability in Power Systems. Load Modeling
[1] W.D. Stevenson, Elements of Power System Analysis, McGraw-Hill, Based on 130 kV Field Measurements, Technical Report No. 324L.
1982. Department of Electrical Power Engineering, Chalmers University of
[2] W.G. Tinney, C.E. Hart, Power flow solutions by Newton’s method, Technology, Goteborg, Sweden 1999.
IEEE Trans. Power Apparatus Syst. PAS-86 (1967) 1449–1457. [22] M. Salama, A.Y. Chikhani, A simplified network approach to the
[3] B. Stott, O. Alsac, Fast decoupled load flow, IEEE Trans. Power VAr control problem for radial distribution systems, IEEE Trans.
Apparatus Syst. PAS-93 (3) (1974) 859–869. Power Syst. 8 (3) (1993) 1529–1535.
[4] S.C. Tripathy, G.D. Prasad, O.P. Malik, G.S. Hope, Load flow solu- [23] N. Mithulananthan, M.M.A. Salama, C.A. Canizares, J. Reeve, Dis-
tions for ill-conditioned power systems by a Newton-like method, tribution system voltage regulation and VAr compensation for differ-
IEEE Trans. Power Apparatus Syst. PAS-101 (10) (1982) 3648–3657. ent static load models, Int. J. Elec. Eng. Edu. 37 (4) (2000) 384–395.