International Journal of Electrical and Computer Engineering (IJECE)

Vol. 6, No. 6, December 2016, pp. 3276 – 3282

ISSN: 2088-8708, DOI: 10.11591/ijece.v6i6.12134 3276

Convergence Evaluation of a Load Flow Method based on

Cespedes’ Approach to Distribution System Analysis
Diego Issicaba1 and Jorge Coelho2
1
Department of Electrical Engineering, Federal University of Technology - Parana (UTFPR), Curitiba-PR, Brazil
2
Department of Electrical Engineering, Federal University of Santa Catarina (UFSC), Florianopolis-SC, Brazil

Article Info ABSTRACT

Article history: This paper evaluates the convergence of a load flow method based on Cespedes’ for-
Received Aug 9, 2016 mulation to distribution system steady-state analysis. The method is described and the
Revised Sep 20, 2016 closed-form of its convergence rate is deduced. Furthermore, convergence dependence
of loading and the consequences of choosing particular initial estimates are verified
Accepted Oct 7, 2016
mathematically. All mathematical results have been tested in numerical simulations,
Keyword: some of them presented in the paper.
Power engineering
Power distribution systems Copyright
c 2016 Institute of Advanced Engineering and Science.
Load flow analysis All rights reserved.
Convergence

Corresponding Author:
Diego Issicaba
Department of Electrical Engineering, Federal University of Technology - Parana (UTFPR)
Av. Sete de Setembro, 3165, Sector D, Rebouças, 80230-910 Curitiba-PR, Brazil
+55 41 3310-4626
[email protected]

1. INTRODUCTION
Load flow methods are fundamental tools to power distribution system analysis [1, 2]. These methods
allow computing steady-state voltages at network nodes as well as the amount of power flowing through power
system devices. Nevertheless, power system literature lacks on formal analysis and comparisons among con-
vergence properties of load flow algorithms. Consistent exceptions can be found in [3, 4, 5, 6, 7]. Particularly,
in [6] the convergence of a forward-backward sweep method is evaluated and its dependence on system loading
verified. In [7], this same method is formally assessed using fixed-point concepts and the contraction mapping
theorem.
In this context, this paper introduces the convergence analysis of a load flow method based on R.
G. Cespedes’ approach to distribution system analysis [8]. For this accomplishment, section 2 introduces the
proposed method focusing on algorithm procedures and main equations. Section 3 presents the deduction of
the convergence rate of the algorithm and a mathematical region where algorithm iterates are confined, as long
as initial estimates are chosen properly. In section 4, numerical results are shown to illustrate the validity of the
mathematical developments. At last, section 5 outlines conclusions and final remarks.

2. A LOAD METHOD INSPIRED ON CESPEDES’ FORMULATION

Consider the radial feeder schematic shown in Fig. 1. In the schematic, zi denotes a series line
impedance, Ei = ei + jfi represents complex node voltages, and SLi denotes complex loads, all refereed to
a general node i. The index ui stands for the node upstream node i. The substation bus is named the 0 (zero)
node with complex voltage denoted by E0 . Furthermore, the index i points out to both a node and its upstream
line.
The complex power flow downstream line i can be computed by summing the loads and losses down-

Journal Homepage: http://iaesjournal.com/online/index.php/IJECE

IJECE ISSN: 2088-8708 3277

Si
-
Eui zi Ei zi+1 Ei+1 EN −1 zn EN
q q q q q
|{z}
? ? ? branch n ?
SLi SLi+1 SLn−1 SLn
Si

Figure 1. Radial distribution network schematic (adapted from [1]).

stream the line as follows  ∗ 2

X X S 
Si = SLi + SLr + zr  r∗  (1)
Er
r∈∆i r∈∆i

where ∆i represents either the set of nodes downstream node i or downstream line i, depending on the variables
involved. As consequence, Si can be expressed as function of variables associated to nodes immediately
downstream line i, as follows
X X  S ∗ 2
Si = SLi + Sd + zd  d∗  (2)
E d
d∈Di d∈Di

where Di denotes either the set of nodes immediately downstream node i or immediately downstream line i.
Furthermore, the difference between node voltages in adjacent nodes can be expressed as
 ∗
Si
Ei = Eui − zi (3)
Ei∗

Separating the real and imaginary parts of (2) and the voltage magnitudes in (3), the recursive equa-
tions utilized in the load flow method proposed by R. G. Cespedes [8] can be deduced. Conversely, by applying
(2) and (3) without further deductions, a similar but also effective load flow method can be designed. Starting
from initial complex voltage estimates, the two steps below can be continually repeated until the convergence
of complex voltages is reached.

1. In a backward process, the complex power flow at each node is calculated using (2), starting at end-nodes
and stopping at the first node immediately downstream from the substation node.

2. In a forward process, complex voltage are updated, away from the substation node, using (3).

The complex voltage at the substation bus is assumed constant during the procedures. The convergence criterion
refers to the maximum absolute mismatch between subsequent complex voltage iterates.

3. CONVERGENCE ASSESSMENT
This section addresses the converge assessment of the algorithm, focusing on aspects related to the
implications of choosing an initial estimate and the deduction of the convergence rate.

3.1. Initial Estimate

The update rule of the algorithm can be written recursively for iteration k as

(k) zi  ∗ (k)∗

ψi = ψu(k) i
− (k)∗ Si,c + Li,ac (4)
Ei

(k) (k)
where Ei denotes the complex voltage at node i and iteration k, ψi is the complex voltage at node i and
iteration k + 1, Si,c denotes the sum of all complex loads downstream node i (including the one at node i) and
(k)
Li,ac represents the sum of all electrical losses downstream node i and iteration k.

Convergence Evaluation of a Load Flow Method based on Cespedes’ Approach ... (Diego Issicaba)
3278 ISSN: 2088-8708

Let us define ~i as the set of lines in the path between node 0 and i. By (4), we have
 
(k) z1 ∗ (k)∗
ψ1 = E0 − (k)∗ S1,c + L1,ac
E1
.. (k) ..
. = ψ1 − .
.. .. .. (5)
. = . −  . 
(k) (k) zi ∗ (k)∗
ψi = ψui − (k)∗ Si,c + Li,ac
Ei

By summing the equations above, a closed-form for the update rule of the algorithm can be obtained.
(k)
X zm  
∗ (k)∗
ψi = E0 − (k)∗
S m,c + L m,ac (6)
m∈~i Em

For instance, consider a radial network with 5 nodes and connections {0–1, 1–2, 2–3, 1–4}. For this
network, the closed-form for the update rules are the following:
(k) z1  ∗ (k)∗

ψ1 = E0 − (k)∗ S1,c + L1,ac (7)
E1
(k) z1  ∗ (k)∗
 z2  ∗ (k)∗

ψ2 = E0 − (k)∗ S1,c + L1,ac − (k)∗ S2,c + L2,ac (8)
E1 E2
(k) z1  ∗ (k)∗
 z2  ∗ (k)∗
 z3  ∗ (k)∗

ψ3 = E0 − (k)∗ S1,c + L1,ac − (k)∗ S2,c + L2,ac − (k)∗ S3,c + L3,ac (9)
E1 E2 E3
(k) z1  ∗ (k)∗
 z4  ∗ (k)∗

ψ4 = E0 − (k)∗ S1,c + L1,ac − (k)∗ S4,c + L4,ac (10)
E1 E4

and, in a compact matrix notation:

z1
0 0 0
 
(k) (k)∗
S ∗ + L1,ac
  (k)∗

ψ1
 
E0 E1
(k)  z1 z2
0 0   1,c
∗ (k)∗

ψ2
  E0   (k)∗
E1
(k)∗
E2
  S2,c + L2,ac 

 =
  − z1 z2 z3
 
(11)
 (k)  E0   0   S ∗ + L(k)∗ 
 ψ3   (k)∗
E1
(k)∗
E2
(k)∗
E3  3,c 3,ac

(k)
ψ4 E0 z1
0 0 z4 ∗
S4,c
(k)∗
+ L4,ac
(k)∗ (k)∗
E1 E4

This example suggests the update rule can be written in a general compact matrix notation. For this
accomplishment, assuming an indexing where ui < i, let us define the lower triangular path matrix T with size
n × n and entries 
1, if m ∈ ~i
tim = (12)
0, otherwise
Then, the update rule of the algorithm can written in the compact form

Ψ(k) = E0 − TZp S∗c K(k)∗ − TZp L(k)∗

ac K
(k)∗
(13)
(k)
where Ψ(k) is a n × 1 vector with entries given by ψi , E0 is a n × 1 vector with entries equal to E0 , Zp is
(k)
the n × n primitive impedance matrix, Sc denotes a n × n diagonal matrix with elements equal to Si,c , Lac
(k)
represents a n × n diagonal matrix with elements given by Li,ac and K(k) is a n × 1 vector with reciprocals of
(k)
Ei .
Let now R be a closed region in the complex space C n defined by R = {E ∈ C n , ||Ei || ≥ (E0 − α),
∀i = 1, . . . , n}, for some α ∈ R such that

E0 √
≤ α < E 0 − ρα (14)
2
where
ρα = ||TZp S∗c || + ||TZp || L∗ac,α 
 
(15)

IJECE Vol. 6, No. 6, December 2016: 3276 – 3282

IJECE ISSN: 2088-8708 3279

and Lac,α is a n × n diagonal with entries given by the maximum accumulated losses downstream each node,
computed using a backward process with voltage magnitudes equal to (E0 − α).
Given an iterate E(k) ∈ R, by (13) we have
   
Ψ − E0  = TZp S∗c K(k)∗ + TZp L(k)∗
 (k) (k)∗ 
ac K
 

     
≤ ||TZp S∗c || K(k)∗  + ||TZp || L(k)∗
     (k)∗ 
ac    K 
 ∗ 
||TZp S∗c || ||TZp || Lac,α 
≤ + (16)
(E0 − α) (E0 − α)
and, by using (14) we have

 (k)
 ρα
Ψ − E0  ≤ < (E0 − α) ≤ α (17)

(E0 − α)

Therefore, assuming the existence of R, for a given α if E(k) ∈ R, then Ψ(k) belongs to an open ball
(in C ) centered in E0 and radius equal to α. Particularly, Ψ(k) ∈ R for all E(0) ∈ R, ∀k, then if the initial
n

estimate belongs R, all other iterates also belong to R.

3.2. Convergence Rate

Once a region where iterates are confided through the iterative process have been deduced, let us
examine the convergence rate of the algorithm. Notice that two subsequent iterates of the algorithm can be
written as
(k+1)
X zm  ∗ (k+1)∗

ψi = E0 − (k+1)∗
Sm,c + L m,ac (18)
m∈~i Em
(k)
X zm  
∗ (k)∗
ψi = E0 − (k)∗
Sm,c + L m,ac (19)
m∈~i Em

and, subtracting (19) from (18) we have

(k+1)
X  zm   zm  ∗ 
∗ (k)∗ (k+1)∗
∆ψi = (k)∗
Sm,c + Lm,ac − (k+1)∗ Sm,c + Lm,ac (20)
m∈~i Em Em

Hence, by manipulating the terms of the equation above, we have

(k+1)∗
" #
∗ (k+1)∗
(k+1)
X Sm,c + Lm,ac ∆Lm,ac (k+1)∗
∆ψi = zm (k+1)∗ (k)∗
− (k+1)∗ (k+1)∗
∆Em (21)
m∈~i E m E m E m ∆E m

(k+1) (k+1) (k) (k+1) (k+1) (k)

where ∆Em = Em − Em and ∆Lm,ac = Lm,ac − Lm,ac .
In a compact form, this expression can be rewritten as
(k+1)
X (k+1)
(k+1)∗
∆ψi = Dim ∆Em (22)
m∈~i

in which
(k+1)∗
" #
∗ (k+1)∗
(k+1)
X Sm,c + Lm,ac ∆Lm,ac
Dim = zm (k+1)∗ (k)∗
− (k+1)∗ (k+1)∗
(23)
m∈~i Em Em Em ∆Em
Equation (22) can also be written in matrix notation as

∆Ψ(k+1) = D(k+1) ∆E(k+1)∗ (24)

(k+1)
where ∆Ψ(k+1) = Ψ(k+1) − Ψ(k) , D(k+1) is a n × n matrix with entries given by Dim and ∆E(k+1) =
E(k+1) − E(k) . Matrix D(k+1) indicates the convergence rate of the algorithm in each iteration. It also proves
convergence direct dependence on network loading, losses iterates and convergence of losses.

Convergence Evaluation of a Load Flow Method based on Cespedes’ Approach ... (Diego Issicaba)
3280 ISSN: 2088-8708

Figure 2. Schematic of the actual distribution network utilized in the evaluations.

4. NUMERICAL RESULTS
Numerical load flow analysis are presented in this section to verify the provided mathematical results.
Fig. 2 shows an actual 13.80 kV distribution network with 490 nodes utilized to this purpose. Without loss
of
 generality, the value of 8.28 kV (0.6000 pu) has been chosen as α. This implies in a feasible region R ,
E ∈ C 489 , ||Ei || ≥ E0 − α, ∀i = 1, . . . , 489}, which meets the interval

E0 √
0.5000 pu = ≤ α < E0 − ρα = 0.8160 pu (25)
2

where ρα equals 0.0339 pu.

Complex voltages have been obtained using the load flow method. Results have been validated using
the approach proposed in [1]. Table 1(a) shows the real and imaginary parts of voltage iterates as well as error
values for node 300, assuming the uncommon initial estimate of 2.00∠63.03o pu, ∀i. On the other hand, Table
1(b) shows the same variables, but for a case considering a loading increased by sevenfold. The maximum
absolute mismatch between voltage iterates has been chosen as convergence criterion. The last iterate has been
set as solution for the sake of error computation. Tolerance has been specified to 10−6 .

(0)
Table 1. Numerical results for bus 300 assuming the initial estimate Ei = 2.00∠63.03o pu, ∀i

(a) Normal loading (b) Increased loading

   
 (k)   (k) 
k e300 (pu) f300 (pu) D  Error k e300 (pu) f300 (pu) D  Error
0 0.90719 -1.78241 - - 0 0.90719 -1.78241 - -
1 0.99995 -0.01043 0.00013 1.79E−00 1 0.99915 -0.07408 0.00677 1.86E−00
2 0.98115 -0.00927 0.00073 2.28E−02 2 0.85611 -0.06065 0.05374 1.94E−01
3 0.98096 -0.00936 0.00069 3.16E−04 3 0.84061 -0.06560 0.07107 2.48E−02
4 0.98096 -0.00936 0.00079 2.91E−06 4 0.83888 -0.06560 0.07926 2.89E−03
5 0.98096 -0.00936 0.00074 2.18E−08 5 0.83870 -0.06563 0.07756 3.13E−04
6 0.98096 -0.00936 0 0 6 0.83868 -0.06563 0.07852 3.20E−05
7 0.83868 -0.06563 0.07816 3.23E−06
8 0.83868 -0.06563 0.07826 2.96E−07
9 0.83868 -0.06563 0 0

Fig. 3 shows the first two voltage iterates for the first case in a level curve of the convergence region.
As expected, one can notice that even by choosing a nonrealistic solution as initial estimate, the first
iterate belongs to an open ball centered in E0 and limited by radius α, followed that the convergence of the
algorithm is reached. Convergence rate is shown to be, in these cases, lower to the unit. Furthermore, as
deduced, the increase of loading caused an increase of convergence rate, impacting on the algorithm efficiency.

IJECE Vol. 6, No. 6, December 2016: 3276 – 3282

IJECE ISSN: 2088-8708 3281

1.5
Solution
1 Voltage iterate

0.5

f300 [pu]
1
0

-0.5
E0 − α
-1

-1.5
0
-2
-2 -1 0 1 2
e300 [pu]
(0)
Figure 3. Iterates in a level curve of the region R, for the bus 300. Initial estimate: Ei = 2.00∠63.03o pu.

5. CONCLUSIONS
This paper evaluates the convergence of a load flow method inspired on R. G. Cespedes’ recursive
equations. Numerical results and formal deduction of convergence rate show that the efficiency of the method
depends on the network loading, losses and convergence of losses. Also, a region where algorithm iterates are
confined is deduced as long as initial estimates are chosen properly. Future works will extend these develop-
ments to three-phase load flow approaches.

ACKNOWLEDGMENTS
The authors would like to acknowledge the financial, technical and human support of the CNPq,
CAPES, INESC Porto and INESC P&D Brasil.

REFERENCES
[1] D. Issicaba and J. Coelho, “Rotational load flow method for radial distribution systems,” International
Journal of Electrical and Computer Engineering (IJEPE), vol. 6, no. 3, 2016.
[2] H. Marefatjou and M. Sarvi, “Distributed generation allocation to improve steady state voltage stability
of distribution networks using imperialist competitive algorithm,” International Journal of Applied Power
Engineering (IJAPE), vol. 2, no. 1, pp. 15–26, 2013.
[3] J. F. Chen and W. M. Wang, “Uniqueness of the feasible voltage solutions for radial power networks,” in
IEEE Region 10 International Conference on Microeletronics and VLSI, TENCON’s 95, Novemeer 1995,
pp. 351–354.
[4] H. D. Chiang and M. E. Baran, “On the existence and uniqueness of load flow solution for radial distribution
networks,” IEEE Transactions on Circuits and Systems, vol. 37, no. 3, pp. 410–416, March 1990.
[5] K. N. Miu and H. D. Chiang, “Existence, uniqueness, and monotonic properties of the feasible power
flow solution for radial three-phase distribution networks,” IEEE Transactions on Circuits and Systems –
I:Fundamental Theory and Applications, vol. 47, no. 10, pp. 1502–1514, October 2000.
[6] E. Bompard, E. Carpaneto, G. Chicco, and R. Napoli, “Convergence of the backward-forward sweep
method for the load flow analysis of radial distribution systems,” Electrical Power & Energy Systems,
vol. 22, no. 7, pp. 521–530, October 2000.
[7] D. Issicaba, “Ladder load flow methods to radial and weakly meshed distribution systems,” Master’s thesis,
Federal University of Santa Catarina, 2008.
[8] R. G. Cespedes, “New method for the analysis of distribution networks,” IEEE Transaction on Power
Delivery, vol. 5, no. 1, pp. 391–396, January 1990.

Convergence Evaluation of a Load Flow Method based on Cespedes’ Approach ... (Diego Issicaba)
3282 ISSN: 2088-8708

BIOGRAPHIES OF AUTHORS

Diego Issicaba received the B.S. and M.S. degrees in Electrical Engineering from the Federal Uni-
versity of Santa Catarina (UFSC), Santa Catarina, Brazil, in 2006 and 2008, respectively. Fur-
thermore, he received the Ph.D. degree on Sustainable Energy Systems, under the MIT Doctoral
Program, from the Faculty of Engineering of the University of Porto, Portugal. His research inter-
ests involve smart grids, mutiagent systems, distributed generation and distribution systems. He is
currently a full Professor at Federal University of Technology – Parana (UTFPR), Associate Re-
searcher and Coordinator of the Research Area on Energy and Management of INESC P&D Brasil.

Jorge Coelho received the B.S. and M.S. degrees in electrical engineering from the Federal Univer-
sity of Santa Catarina, Brazil, in 1977 and 1980, respectively. In 1990, he received the Ph.D. degree
in electrical engineering from the Catholic University of Rio de Janeiro, Brazil. He is a Professor of
the Department of Electrical Engineering at the Federal University of Santa Catarina, Brazil, since
March 1978. His research interests include distribution systems expansion and operation planning,
power systems reliability, probabilistic methods applied to power systems, and power quality.

IJECE Vol. 6, No. 6, December 2016: 3276 – 3282