See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3354589

Method of load flow solution of radial distribution network

Article  in  IET Proceedings - Generation Transmission and Distribution · December 1999

DOI: 10.1049/ip-gtd:19990464 · Source: IEEE Xplore

CITATIONS READS

269 1,777

2 authors, including:

Smarajit Ghosh
Thapar Institute of Engineering and Technology, Patiala, India
43 PUBLICATIONS   548 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Effective Loss Minimization and Allocation of Unbalanced Distribution Network View project

All content following this page was uploaded by Smarajit Ghosh on 07 January 2016.

The user has requested enhancement of the downloaded file.

Method for load-flow solution of radial distribution
networks
S.Ghosh and D.Das

Abstract: A simple and efficient method for solving radial distribution networks is presented. The
proposed method involves only the evaluation of a simple algebraic expression of receiving-end
voltages. Computationally, the proposed method is very efficient. The effectiveness of the proposed
method is demonstrated through three examples.

List of symbols developed the ladder technique from basic ladder-network

theory into a working algorithm, applicable to the solution
NB = total number of nodes of radial load-flow problems. Stevens et al. [8] have shown
LN1 = total number of branches that the ladder technique is found to be fastest but did not
j = branch number,j = 1,2, 3, ..., LN1 converge in five out of 12 cases studied. Shirmohammadi er
PL(9 = real power load at ith node al. [9] have proposed a method for solving radial distribu-
tion networks based on the direct application of Kirch-
QL(z) = reactive power load at ith node
hoff's voltage and current laws. They have developed a
V(z) = voltage of ith node branch-numbering scheme to enhance the numerical per-
R(ii) = resistance o f a h branch formance of the solution method. They have also extended
Xh] = reactance o f a h branch their method for solution of weakly meshed networks.
Zh) = impedance of branchjj Baran and Wu [lo] have obtained the load-flow solution in
= current that flows through branchj a distribution system by the iterative solution of three fun-
Zu) damental equations representing real power, reactive power
IL(9 = load current of node i
and voltage magnitude. They have computed the system
LPh) = real-power loss of branchjj Jacobian matrix using a chain rule. In their method, the
LQ(ii) = reactive-power loss of branchj mismatches and the Jacobian matrix involve only the eval-
IScii) = sending-end node of branchj uation of simple algebraic expressions and no trigonometric
IRW = receiving-end node of branchj functions. They have also proposed decoupled and fast
= charging admittance at node i decoupled distribution load-flow algorithms. Chiang [l 11
yo(z)
has also proposed three different algorithms for solving
ZC(9 = charging current at node i radial distribution networks based on the method proposed
DVMAX = maximum voltage dfference by Baran and Wu [lo]. He has proposed decoupled, fast
decoupled and very fast decoupled distribution load-flow
1 Introduction algorithms. In fact decoupled and fast decoupled dlstribu-
Little attention has been given to load-flow analysis of dis- tion load-flow algorithms proposed by Chiang [I 13 are sim-
tribution systems, unlike load-flow analysis of transmission ilar to that of Baran and Wu [lo]. However, the very fast
systems. However, some work has been carried out on decoupled distribution load flow proposed by Chiang [111
load-flow analysis of distribution networks, but the choice is very attractive because it does not require any Jacobian
matrix construction and factorisation. Renato [12] has pro-
of a solution method for a practical system is often dsi-
posed one method for obtaining a load-flow solution of
cult. Generally, distribution networks are radial and the R/
radial distribution networks. He has calculated the electri-
Xratio is very high. For t h s reason, conventional Newton-
cal equivalent for each node summing all the loads of the
Raphson (NR) [l] and fast decoupled load-flow [2] meth-
ods do not converge. Many researchers have suggested network fed through the node including losses and then,
modfied versions of the conventional load-flow methods starting from the source node, the receiving-end voltages of
all the nodes are calculated. Goswami and Basu [13] have
for solving power networks with hgh R/X ratio [3-51.
presented a direct method for solving radial and meshed
Kersting and Mendive [q and Kersting [I have devel-
distribution networks. However, the main h t a t i o n of
oped a load-flow techque for solving radial distribution
their method is that no node in the network is the junction
networks using ladder-network theory. They have
of more than three branches, i.e. one incoming and two
outgoing branches. Jasmon and Lee [14, 151 have proposed
OIEE, 1999 a new load-flow method for obtaining the solution of radial
ZEE Proceedhgs online no. 19990464 distribution networks. They have used the three fundamen-
DO1 l O . l 0 4 9 / i p g t d : l ~ tal equations representing real power, reactive power and
Paper fmt received 5th November 1998 and in revised form 26th March 1999 voltage magnitude derived in [lo]. They have solved the
The authors are with the ElectricalEngineering Department, Indian Institute of radial distribution network using these three equations by
T E ~ o I o ~Kharagpur
Y, - 721302, WB, India reducing the whole network into a single h e equivalent.
IEE Proc.-Gener. Transm. Distrib.. Vol. 146, No. 6, November 1999 641
Das et al. [16] have proposed a load-flow technique for similarly for branch 2,
solving radial distribution networks by calculating the total
real and reactive power fed through any node. They have V ( 3 )= V ( 2 )- 1 ( 2 ) 2 ( 2 ) (2)
proposed a unique node, branch and lateral numbering As the substation voltage V(1) is known, so if Z(l) is
scheme which helps to evaluate exact real- and reactive- known, i.e. current of branch 1, it is easy to calculate V(2)
power loads fed through any node and receiving-end volt- from eqn. 1.
ages. Once V(2) is known, it is easy to calculate V(3) from
In this paper, the main aim of the authors has been to eqn. 2, if the current through branch 2 is known. Similarly,
develop a new load-flow technique for solving radial distri- voltages of nodes 4, 5, ..., NB can easily be calculated if all
bution networks. The proposed method involves only the the branch currents are known. Therefore, a generalised
evaluation of a simple algebraic expression of receiving-end equation of receiving-end voltage, sending-end voltage,
voltages. The proposed method is very efficient. It is also branch current and branch impedance is
observed that the proposed method has good and fast con-
vergence characteristics. Loads in the present formulation V(m2)= V(rn1) - I ( j j ) Z ( j j ) (3)
have been presented as constant power. However, the pro-
posed method can easily include composite load modelling, where j j is the branch number.
if the composition of the loads is known. Several radial dis-
tribution 'feeders have been solved successively by using the m2 = I R ( j j ) (4)
proposed method. The speed requirement of the proposed
ml = IS(jj) (5)
method has also been compared with other existing
methods. Eqn. 3 can be evaluated forjj = 1, 2, ..., LN1 (LN1 = NB
- 1 = number of branches). Current through branch 1 is
2 Assumption equal to the sum of the load currents of all the nodes
beyond branch 1 plus the sum of the charging currents of
It is assumed that the three-phase radial distribution net- all the nodes beyond branch 1, i.e.
works are balanced and can be represented by their equiva- LN1 LN1
lent single-line diagrams.
I(1)= IL(2) + IC(2) (6)
1 2 3 4 5 6 i=2 i=2
The current through branch 2 is equal to the sum of the
load currents of all the nodes beyond branch 2 plus the
sum of the charging currents of all the nodes beyond
branch 2, i.e.
12
I 11
+
I(2)= IL(3) IL(4) IL(5) + + IL(6)
1 + IL(10) + IL(11)
+ I C ( 3 ) + IC(4) + I C ( 5 ) + I C ( 6 )
Fig. 1 Single-line diagram of radial distribution network + IC(10) IC(11)+ (7)
Therefore, if it is possible to identify the nodes beyond all
3 Solution methodology the branches, it is possible to compute all the branch cur-
rents. Identification of the nodes beyond all the branches is
Fig. 1 shows single-line diagram of a distribution feeder. realised through an algorithm as explained in Section 4.
The branch number sending-end and receiving-end node of The load current of node i is
this feeder are given in Table 1. Consider branch 1. The
receiving-end node voltage can be written as
V ( 2 )= V ( l ) - 1(1)2(1) (1)

Table 1: Branch number (jj, sending-end (ml /S(jj))node, receiving-end

node (m2 = (/R(jll) and nodes beyond branches 1,2,3. ..., 11 of Fig. 1

Total number of
Branch Sending end Receiving end
Nodes beyond branch j j nodes Mjl)
number ( j j rnl = /S(hl m 2 = /R(jj
beyond branch j
1 1 2 2,3,7,8,5,10,9,12,6,11 11
2 2 3 3, 4,5,IO, 6,1 1 6
3 3 4 4,5, IO, 6,1 1 5
4 4 5 5,6 2
5 5 6 6 1
6 2 7 7,8,9,12 4
7 7 8 8,9,12 3
8 a 9 9 1
9 4 10 IO, 1 1 2
10 10 11 11 1
11 8 12 12 1

642 IEE Proc.-Gener. Transm. Disfrib., Vol. 146, No. 6,November 1999
The charging current at node i is First define the variables:
I C ( i )= yo(i)V(i) i = 2 , 3 , . . . ,N B(9) j j = 1, 2, 3, ..., LN1 cij indicates branch of Fig. 1, see also
Load currents and charging currents are computed itera- Table 1);
tively. Initially, a flat voltage of all the nodes is assumed ip is the node count (identifies the number of nodes beyond
and load currents and charging currents of all the loads are a particular branch);
computed using eqns. 8 and 9. A detailed load-flow-calcu- ZK(ip) is the node identifier (helping to identify nodes
lation procedure is described in Section 5. beyond all the branches);
The real and reactive power loss of branchjj are given
by: Mjj] is the total number of nodes beyond branchjj; and
Wjj)= IW)12Wj) (10) IEGJ, ip + 1) is the receiving-end node.
ZE(J, ip + 1) will now be explained. Consider the fxst
L Q ( j j ) = Il(jj)12X(jj) (11) branch in Fig. 1, i.e. j j = 1; the receiving-end node of
4 Identificationof nodes beyond all the branches branch 1 is 2, i.e. ZRb] = ZR(1) = 2. Therefore ZEb, ip + 1)
= ZE(1,ip + 1) will help to identify all the nodes beyond
Before the detailed algorithm is given, the details of the branch 1. This will help to find the exact current flowing
methodology of identifying the nodes beyond all branches through branch 1. S i a r l y , consider branch 2, i.e. jj = 2;
will be discussed. This wdl help in finding the exact current the receiving-end node of branch 2 is 3, i.e. IR(jj] = ZR(2) =
flowing through all the branches. 3. Therefore, E ( J , ip + 1) = ZE(2, ip + 1) will identify all

(7) from A

read sending-end in 5 ip
and receiving-end
to F
nodes and total
number of nodes no

ni
iil
and branches

nc=l
no

1 k=jj+l I
- from B
.(
I II

IE(jj, ip+l)=lR(jj)
from E
iP =i P+1
from C IK(ip)=i
IE(jj, ip+l)=iR(jj) to E

Li nc=O
N(jj) =ip+ 1
L, d
r -

i=i+l

in=i
f IE(LN,I)=IR(LNI)
IE(jj, ip+l)=IR(jj) N(LNl) = 1
N(jj)=ip+1
1
stop

Fig.2

IEE Proc. 643

the nodes beyond branch 2. This will help to compute the seen that the node 7 is connected to node 8. Therefore,
exact current flowing through branch 2. For each node ZE(1, 5 ) = 8.
identitication beyond a particular branch, '@' will be incre- From the above dlscussion, it is again seen that node 3 is
mented by 1. Note here that before identification of nodes connected to node 4 and node 7 is connected to node 8.
beyond a particular branch, '@' has to be reset to zero. Similarly, the proposed logic will check whether nodes 4
For jj = 1 (first branch of Fig. 1 , Table I), ZRG) = ZR(1 ) and 8 are connected to any other nodes. Thls process will
= 2; check whether IR(1)= ZS(i) or not for i = 2, 3, 4, ..., continue unless all nodes are identified beyond branch 1.
LN1. It is seen that ZR(1)= ZS(2) = 2, IR(1)= IS(6) = 2; The nodes beyond branch 1 are also given in Table 1.
the corresponding receiving-end nodes are ZR(2) = 3 and The total current flowing through branch 1 is equal to
ZR(6) = 7. the sum of the load currents of all nodes beyond branch 1
Therefore, IE(1,1) = 2, IE(1, 2) = 3 and ZE(1, 3) = 7 . plus the sum of the charging currents of all the nodes
Note that there should not be any repetition of any node beyond branch 1 .
while identifying nodes beyond a particular branch, and For jj = 2 (second branch in Fig. 1; Table l), ZRG] =
this logic has been incorporated in the proposed algorithm ZR(2)= 3, check whether ZR(2) = ZS(z] or not for i = 3,
and further explained in the flowchart given in Fig. 2. 4, ..., LN1. It is seen that IR(2) = IS(3) = 3. The corre-
From the above discussion, it is seen that node 2 is con- sponding receiving-end node is IR(3) = 4. Therefore, IE(2,
nected to nodes 3 and 7. Smilarly, the proposed logic will 1) = 3 and ZE(2, 2) = 4.
identify the nodes which are connected to nodes 3 and 7. From the above discussion, it is seen that node 3 is con-
First it w d check whether node 3 appears in the left-hand nected to node 4. The proposed logic will identify the nodes
column of Table 1 . It is seen that node 3 is connected to which are connected to node 4. It will check whether node
node 4. Therefore, ZE(1, 4) = 4. Then it will check whether 4 appears in the left-hand column of Table 1. It is seen that
node 7 appears in the left-hand column of Table 1 . It is node 5 and node 10 are connected to node 4. Therefore

-+
from A
start

-1
read substation voltage V(l),
1
line data and load data.
jj 5 LN1
assume a flat voltage starI,i.e. V(i)
=V(l)=I~O'for i=2,3 ,........, NB

set W(i)=V(i) for i=2,3 ,......., NB

set lSS(ij)=lS(jj) and IRR(jj)=
IR(jj) for jj=1,2,3 ,........, LNl
1 no

set iteration count k=O

calculate IL(i) and IC(i)

,-, L7J converged

calculate line

I
using eqns. 8 and 9
for i=2,3, .... NB

calculate branch currents

by using eqn. 12
1
W(m2)=V(m2)
for m2=2,3,.... NB

4
1
losses, line
flows etc. and
print required
dyta 1
to B

set m l =ISS(jj) and rnS=IRR(jj)

compute receiving-end voltage V(m2)

by using eqn. 3
calculate absolute change in voltage at
node m2, DV(m2)=ABS(/V(m2)/-/W(m2)/)

I jj=jj+l

-1
to A
Fig.3 Fibwehartfor load-jhw calculation of r& dimibution network

644 IEE Proc-Gener. Transm. Distrib., Vol. 146, No. 6,November 1999
ZE(2, 3) = 5 and IE(2, 4) = 10. The proposed logic will Table 2: Load-flow solution of example 1: 29-node radial dis-
check whether nodes 5 and 10 are connected to any other tribution network
nodes. This process will continue unless all nodes are iden-
~ ~

Node Voltage Node Voltage

tified beyond branch 2. The nodes beyond branch 2 are number magnitudes (P.u.) number magnitudes (P.u.)
given in Table 1.
Similarly it is necessary to consider the receiving-end 1 1 .ooooo 15 0.88945
node of branch 3, branch 4, ..., branch LN1 in Fig. 1 and, 2 0.95884 16 0.88560
in a similar way to that discussed above, the nodes have to 3 0.91877 17 0.88296
be identified beyond these branches. The nodes beyond all 4 0.89598 18 0.87556
the branches are also given in Table 1. 5 0.83736 19 0.86914
Note that, if the receiving-end node of any branch in
6 0.85068 20 0.86378
Fig. 1 is an end node of a particular lateral, the total cur-
rent of this branch is equal to the load current of this node 7 0.82855 21 0.86240
plus the charging current of this node itself. For example, 8 0.82189 22 0.86120
consider node 6 in Fig. 1 (branch 5, Table 1); this is an end 9 0.81233 23 0.81710
node. Therefore, the branch current 1(5)is equal to the load 10 0.80821 24 0.81079
current of node 6 plus the charging current of node 6 (i.e. 11 25 0.80725
0.94989
4 5 ) = ZL(6) + ZC(6)). Sirmlarly, 9, 11 and 12 are end nodes
of Fig. 1. The proposed computer logic will identify all the 12 0.94655 26 0.80467
end nodes automatically. 13 0.93980 27 0.80884
The concept of identifying the nodes beyond all the 14 0.93684 28 0.80774
branches, whch helps in computing the exact current flow-
ing through all the branches, has been realised using an Table 3: Load-flow solution of example 2:69-node radial dis-
algorithm (Fig. 2) and applied in the load-flow technique tribution network
as shown in the flowchart in Fig. 3.
Node Voltage Node Voltage
5 Load-flow calculation number magnitudes (P.u.) number magnitudes (P.u.)
1 1 .ooooo 36 0.99992
Once all nodes beyond each branch are identified, it is very 2 0.99997 37 0.99975
easy to calculate the current flowing through each branch
3 0.99993 38 0.99959
as described in Section 3. For this purpose, the load current
and charging current of each node are calculated by using 4 0.99984 39 0.99954
eqns. 8 and 9. Once the nodes are idenflied beyond each 5 0.99902 40 0.99954
branch, the expression of branch current is given as 6 0.99009 41 0.99884
N(33) Nb3) 7 0.98079 42 0.99855
I ( j j )= I L { I E ( j j ,i)} + I C { I E ( j j ,i)} 8 0.97858 43 0.99851
2=1 2=1 9 0.97745 44 0.99850
(12) 10 0.97245 45 0.99841
Initially, a constant voltage of all the nodes is assumed and 11 0.97135 46 0.99840
load currents and charging currents are computed using 12 0.96819 47 0.99979
eqns. 8 and 9. After load currents and charging currents
13 0.96526 48 0.99854
have been calculated, branch currents are computed using
eqn. 12. The voltage of each node is then calculated by 14 0.96237 49 0.99470
using eqn. 3 with eqn. 4. Real and reactive power loss of 15 0.95950 50 0.99415
each branch is calculated by using eqns. 10 and 11, respec- 16 0.95897 51 0.97854
tively. Once the new values of the voltages of all the nodes 17 0.95809 52 0.97853
are computed, convergence of the solution is checked. If it 18 0.95808 53 0.97466
does not converge, then the load and charging currents are
19 0.95761 54 0.97142
computed using the most recent values of the voltages and
the whole process is repeated. 20 0.95731 55 0.96694
The convergence criterion of the proposed method is that 21 0.95683 56 0.96257
if, in successive iterations the maximum difference in volt- 22 0.95683 57 0.94010
age magnitude (DVMAX) is less than O.OOOlp.u., the solu- 23 0.95675 58 0.92904
tion has then converged. The convergence analysis of the 24 0.95660 59 0.92476
proposed method is given in Appendjx 1 (Section 9.1).
25 0.95643 60 0.91974
The proposed distribution load-flow algorithm for solv-
ing radial distribution networks is given in the form of a 26 0.95636 61 0.91234
flowchart in Fig. 3. 27 0.95634 62 0.91205
28 0.99993 63 0.91166
6 Example 29 0.99985 64 0.90976
To demonstrate the effectiveness of the proposed method, 30 0.99973 65 0.90919
three examples are selected. The first example is a 29-node 31 0.99971 66 0.97129
11kV rural distribution feeder of India. Data for this 29- 32 0.99961 67 0.97129
node system are given in Appendix 2 (Section 9.2). Table 2 33 0.99935 68 0.96786
gives the load-flow results of a 29-node radial distribution 34 0.99901 69 0.96786
network. The real- and reactive-power loss of this system
35 0.99895
are 303.78kW and 124.74kVAr, respectively.
IEE Proc -Gener Transm Distrcb , Vol 146, No. 6, November 1999 645
 Table 4 Load-flow results of example 3: 33-node radial distribution network

Real-power Reactive-power Minimum

Status of the network
loss (kW) loss (kVAr) voltage (P.u.)
Without charging admittance 202.67 135.14 Vi8 = 0.91309

With charging admittance 196.54 131.03 V78 = 0.91513

Table 5: Comparison of speed of proposed method with other existing methods

Example 1: 29-node Example 2: 69-node Example 3: 33-node

Proposed method and other four distribution network distribution network distribution network
existing methods CPU time Iteration CPU time Iteration CPU time Iteration
(S) number (s) number (S) number
Proposed method 0.07 3 0.16 3 0.09 3
Chiang [ I l l very fast decoupled 0.09 3 0.24 3 0.11 3
distribution load flow
Load flow using single-line 0.1 1 3 0.29 3 0.13 3
equivalent [14, 151
Renato [I21 load flow using 0.12 4 0.33 4 0.14 4
forward-sweeping method
Kersting [71 load flow using ladder 0.14 4 0.37 4 0.16 4
technique

The second example is a 69-node radial distribution net- TRIPATHY, S.C., DURGAPRASAD, G., MALIK, O.P., and
HOPE, G.S.: ‘Load flow solutions for ill-conditioned power system by
work. Data for this system are available in [lo]. Load-flow a Newton like method’, ZEEE Tram., 1982, PAS-101, pp. 36843657
results of this system are given in Table 3. Real- and KERSTING, W.H., and MENDIVE, D.L.: ‘An application of ladder
reactive-power losses of this system are 224.96kW and network theory to the solution of three phase radial load flow prob-
lem’. IEEE PES winter meeting, New York, January 1976
114.15kVAr, respectively. KERSTING, W.H.: ‘A method to the design and operation of a dis-
The third example is a 33-node radial distribution net- tribution system’, ZEEE Trans., 1984, PAS-103, pp. 1945-1952
work. Data for this system are available in [17, 151. T h ~ s STEVENS, R.A., RIZY, D.T., and PURUCKER, S.L.: ‘Performance
of conventional power flow routines for real-time distribution automa-
33-node radial network is solved with and without consid- tion application’. Proceedings of 18th southeastern symposium on
ering charging admittance. Table 4 gives real- and reactive- System theory, 1986, (IEEE Computer Society), pp. 196200
SHIRMOHAMMADI. D.. HONG. H.W.. SEMLYEN. A.. and
power losses and the minimum voltage with and without LUO, G.X.: ‘A compensa6on-based power ’flow method for weakly
charging admittance. meshed distribution and transmission networks’, IEEE Trans., 1988,
The proposed method is also compared with four other PWRS-3, pp. 753-743
10 BARAN, M.E., and WU, F.F.: ‘Optimal sling of capacitors placed on
existing methods. Table 5 shows the CPU time and a radial distribution system’, IEEE Trans., 1989, P-2, pp. 735-
number of iterations of all three examples. All these three 743
11 CHIANG, H.D.: ‘A decoupled load flow method for distribution
examples were simulated on a Meteor 400VT with a power network algorithms, analysis and convergence study’, Electr.
66MHz clock. From Table 5, it is seen that the proposed Power Energy Syst., 1991, 13, (3), pp. 13C138
method is better than the other four existing methods. 12 RENATO, C.G.: ‘New method for the analysis of distribution net-
works’, IEEE Trans., 1990, PWRD-5, (l), pp. 9-13
It is worth mentioning here that the authors have tried to 13 GOSWAMI, S.K., and BASU, S.K.: ‘Direct solution of distribution
solve the above three examples using the Newton-Raphson systems’, IEE Proc. C. , 1991, 188, (I), pp. 78-88
(NR) and Gauss-Seidel (GS) methods. However, for all 14 JASMON, G.B., and LEE, L.H.C.C.: ‘Distributionnetwork reduction
for voltage stability analysis and load flow calculations’, Electr. Power
these three examples, the NR and GS methods did not Energy Syst., 1991, 13, ( I ) , pp. 9-13
converge. 15 JASMON. G.B.. and LEE. L.H.C.C.: ‘Stabilitv of load flow tech-
niques for’distribution system voltage stability ahysis’, IEE Proc. C,
1991, 138, (6), pp. 479484
7 Conclusions 16 DAS, D., NAGI, H.S., and KOTHARI, D.P.: ‘Novel method for
solving radial distribution networks’, IEE Proc. C, 1994, 141, (4), pp.
391-3911
A simple and efficient load-flow technique has been pro- - ,I I,-

17 BARAN, M.E., and WU, F.F.: ‘Network reconfiguration in distribu-

posed for solving radial distribution networks. It has been tion systems for loss reduction and load balancing’, ZEEE Tram.
found from the cases with which the method was tested Power Deliv., 1989, PWRD-4, pp. 1401-1407
that the method has good and fast convergence characteris-
tics compared with some other existing methods. The pro- 9 Appendices
posed method has been implemented on a Meteor 400VT
with a 66MHz clock. Several radial distribution networks
9.1 Convergence analysis
It has been explained in Section 3 that the proposed load-
have been solved successfully using the proposed method. flow algorithm starts with an initial set of node voltages
and load currents and charging currents are then computed
8 References
using eqns. 8 and 9.
1 TINNY, W.F., and HART, C.E.: ‘Power flow solution by Newton’s Now, for the convergence analysis of the proposed
method’, IEEE Tram., 1967, PAS86, pp. 1449-1456 method, consider the electrical equivalent of Fig. 1. Fig. 4
2 SCOlT, B., and ALSAC, 0.: ‘Fast decoupled load flow’, ZEEE shows the electrical equivalent of Fig. 1.
Trans., 1974, PAS-93, pp. 859-869
3 RAJICIC, D., and TAMURA, Y.: ‘A modification to fast decoupled In Fig. 4,
power flow for networks with hgh R/X ratios’, IEEE Tram., 1988, j j = branch number
PWRS-3, pp. 743-746
4 IWAMOTO, S., and TAMURA, Y.: ‘A load flow calculation method m l = ISfij) = sending-end node
for illconditioned power systems’, IEEE Trans., 1981, PAS-100, pp.
17361713 m2 = IRfij) = receiving-end node
646 IEE Proc-Gener. Transm. Distrib., Vol. 146, No. 6, November 1999
P(m2) = total real power load fed through node m2 fore, if e(m2) for m2 = 2, 3, ..., NB converges then, in
Q(m2) = total reactive power load fed through node m2 eqn. 3, V(m2),for m2 = 2, 3, ..., NB, will also converge.
Also note that, when one is solving the receiving-end volt-
From Fig. 4, the following equations can be written:
age, the sending-end voltage is known.
V(m2)= V(m1) - W ) { W j )+ j X ( j j ) } (13)
P(m2)- jQ(m2)= V * ( m 2 ) l ( j j ) (14)
In fact, eqn. 13 is similar to eqn. 3:
V(m1)
sending end
- RW+iXUi)
Mi)
V(r”2)
receiving end

I 1P(mZ)+jQ(mZ)
mi =IS(ij) rnP=IR(jj)
Fig. 4 Electrical equivalent of Fig. I

P(m2) = sum of the real-power loads of all the nodes

beyond node m2 plus the real-power load of the node m2
itself plus real-power losses of all the branches beyond node
m2.
Q(m2) = sum of the reactive-power loads of all the nodes To study the convergence analysis of the proposed
beyond node m2 plus the reactive-power load of the node method, assume that
m2 itself plus reactive-power losses of all the branches
beyond node m2.
From eqns. 13 and 14 one obtains
From eqns. 23 and 22 one obtains
V(m2)= V(m1)
e(m2)= g{e(m2)} (24)
Now consider the graph (Fig. 5) of g(e(m2)). In Fig. 5,
(15) e0(m2)is the initial approximation of e(m2), 5(m2) is the
final solution of e(m2), and ~,(rn2) and .en+,(m2)are the
Now the voltages at nodes m2 and ml are expressed as errors in the nth and (n+l)th iterations. From Fig. 5,
+
V(m2)= e(m2) j f ( m 2 ) (16) +
en(m2)= t(m2) ~ , ( m 2 ) (25)
V(m1)= e(m1)+ j f ( m 1 ) (17) +
en+l (m2)= ((ma) ~ ~ + l ( m 2 ) (26)
Substituting eqns. 16 and 17 in eqn. 15 and separating real The solution of e(m2) at various iterative stages can be
and imamary parts gives written using eqn. 24:
+
e2(m2) f’(m2) = {e(ml)e(m2) f ( m l ) f ( m 2 ) ) + el (m2)= g(eo(m2))
- { W ’ j ) P ( m 2+
) X(jj)Q(m2)1 e2 (ma)= d e l (4)
(18)

en+l(m2)= g{en(m2)) (27)

From eqns. 27 and 25 one obtains
en+l(m2)= g(r(m2)+ En(m2)) (28)
From eqns. 28 and 26 one obtains
First let us consider eqn. 18. In fact in a distribution sys-
tem the voltage angle is extremely small. Hence, the imagi- +
[(ma) Enfl(m2)= g(E(m2) En(m2)) (29) +
nary parts of the voltages are extremely small. Therefore, Expanding the right-hand term of eqn. 29 by Taylor series
the termsf2(m2) andjfmlMm2) in eqn. 18 can be dropped. and considering the first three terms, one obtains
Therefore, eqn. 18 can be written as
t(m2)+ En+l(m2)= t(m2)+ &n(m2)g’{t(m2)}

+ *g”{t(m2)}
2

Assuming that en+1(m2)= E n ( m 2 ) g ’ { t ( m 2 ) } + ~ g ’ r { F ( m 2 ) }

2
(30)
From eqns. 20 and 21 one obtains Now the first and second derivatives of eqn. 23 are:
4 j j )
g’(e(m2))= -
ez(m2)
Now consider eqn. 19. In eqn. 19, the terms on the right-
hand side are independent of jfm2), and jfm2) can be -2Au(jj)
g”(e(m2))=
obtained directly after computing e(m2) iteratively. There- e3 (m2)

IEE Proc.-Gener. Transm. Distrib., Vol. 146, No. 6, November 1999 641
 when e(m2) = gm2) negligible and can be dropped from eqn. 36.
.‘. E n S l ( m 2 ) = &n(m2)g’{l(m2)) (37)
(33)
From eqn. 37 it is seen that the error at the (n + 1)th itera-
and tion is proportional to the error at the nth iteration. Hence
convergence of the proposed method is linear. When r, x, P
(34) and Q are expressed in P.u., the term Aubj) (eqn. 21) is
extremely small and hence g’(Krn2)) = { A u ( ~ j ] / ~ ( mis2)}
From eqns. 30, 33 and 34, one obtains extremely small, i.e. g’(gm2)) << 1.0. Therefore, the con-
vergence of the proposed method is always guaranteed. It
has already been mentioned that solutions of the node volt-
ages are obtained iteratively by using eqn. 3.
From eqns. 35 and 33, one obtains In fact several problems were solved using the proposed
method and an interesting point is that, for all the examples
tested, the proposed method has taken three iterations to
converge. Therefore, it may be concluded that the number
of iterations is independent of the size of the problem.
During the iterative process, it has been observed that, for
all the examples, (~,(m2)/5(m2)}for m2 = 2, 3, ..., NB is 9.2 Data for29-node system

Table 6: Line data and load data of example 1: 29-node radial distribution network

PL of QL of
Branch Sending Receiving
R (8) X(Q) receiving-end receiving-end
number end end
node (kW) node (kVAr)
1 1 2 1.8216 0.7580 140.00 90.00
2 2 3 2.2270 0.9475 80.00 50.00
3 3 4 1.3662 0.5685 80.00 60.00
4 4 5 0.9 180 0.3790 100.00 60.00
5 5 6 3.6432 1.5160 80.00 50.00
6 6 7 2.7324 1.1370 90.00 40.00
7 7 8 1.4573 0.6064 90.00 40.00
8 8 9 2.7324 1.1370 80.00 50.00
9 9 10 3.6432 1.5160 90.00 50.00
10 10 11 2.7520 0.7780 80.00 50.00
11 11 12 1.3760 0.3890 80.00 40.00
12 12 13 4.1280 1.1670 90.00 50.00
13 13 14 4.1280 0.8558 70.00 40.00
14 14 15 3.0272 0.7780 70.00 40.00
15 15 16 2.7520 1.1670 70.00 40.00
16 16 17 4.1280 0.7780 60.00 30.00
17 17 18 2.7520 0.7780 60.00 30.00
18 2 19 3.4400 0.9725 70.00 40.00
19 19 20 1.3760 0.3890 50.00 30.00
20 20 21 2.7520 0.7780 50.00 30.00
21 21 22 4.9536 1.4004 40.00 20.00
22 3 23 3.5776 1.0114 50.00 30.00
23 23 24 3.0272 0.8558 50.00 20.00
24 24 25 5.5040 1.5560 60.00 30.00
25 6 26 2.7520 0.7780 40.00 20.00
26 26 27 1.3760 0.3890 40.00 20.00
27 27 28 1.3760 0.3890 40.00 20.00
Base voltage = 1l.OkV, base MVA = 100.00

648 IEE Proc.-Gener. Transm. Distrib., Vol. 146, No. 6, November 1999

View publication stats