Journal of King Saud University – Engineering Sciences xxx (xxxx) xxx

Contents lists available at ScienceDirect

Journal of King Saud University – Engineering Sciences

journal homepage: www.sciencedirect.com

Original article

A graph theory application for fast and efficient search of optimal

radialized distribution network topology
Dipu Sarkar a,⇑, Pushpanjalee Konwar a, Abhinandan De b, Sanjay Goswami c
a
Department of Electrical and Electronics Engineering, National Institute of Technology, Nagaland, Chumukedima 797103, Nagaland, India
b
Department of Electrical Engineering, Indian Institute Engineering Science and Technology, Shibpur, Howrah 711103, W.B., India
c
Department of Computer Applications, Narula Institute of Technology, Agarpara, Kolkata 700110, W.B., India

a r t i c l e i n f o a b s t r a c t

Article history: The topological changes and re-configuration can be brought into an existing power network by opera-
Received 25 April 2018 tion of sectionalizing switches on the feeder lines. Distribution networks can be operated either in
Accepted 14 February 2019 meshed or in radial configuration. While primary meshed distribution networks have advantages like
Available online xxxx
lower short circuit current, reduced overall conductor length and increased reliability and fault tolerance,
the radial networks on the other hand stand-out on their merits of simple installation and easier relay
Keywords: coordination. An optimal radial network can operate at higher efficiency owing to its reduced power
Re-configuration
losses and can offer higher voltage stability. Transformation of a primary meshed network to a radially
Graph theory
Edmonds Maximal Spanning Tree algorithm
reconfigured optimal network requires selection of the ‘‘best” set of switches on the feeders that are to
Voltage stability be opened or closed so that each feeder can be operated close to its maximum loadability limit and
Radial distribution network the resulting optimal network operates at its lowest feasible value of power loss and highest voltage sta-
bility index. This theoretical concept of optimal power network radialization is however difficult to
implement in practice due to the heavy computational burden and unacceptably long searching time
as the number of feasible alternative configurations is usually massive for large power networks.
Given the extent of the search space, the heuristic search approaches are usually more practical and com-
mon in power network reconfiguration. This paper presents a novel guided search approach based on
Edmond’s Maximal Spanning Tree Algorithm to achieve the optimal radial configuration for an arbitrary
power network rapidly and efficiently. Efficacy of the proposed method has been tested on 30-node and
57-node mesh distribution networks with encouraging results.
Ó 2019 Production and hosting by Elsevier B.V. on behalf of King Saud University. This is an open access
article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction as shunt capacitors, tap-changing transformers and FACTs devices

at proper locations. Installation of additional hard ware equipment
In the present scenario, problems related with voltage stability involves capital investment, whereas, a small change in the net-
at load points are receiving more attention with the increased work topology, supported by efficient search and optimization
loading in power distribution system. To mitigate the voltage sta- algorithm may give similar or even better system performance
bility problems different approaches have been traditionally used with minimal additional investment and lesser implementation
(Taylor, 1994; Mukhopadhyay and Chakrabarti, 1990; Ramana time.
and Ram, 2011; Sakthivel and Mary, 2011) in power transmission Distribution networks are generally designed in meshed struc-
networks by incorporating additional hardware installations such ture, but often operated radially. The reliability of the meshed net-
work is more as there are two paths for the power to flow to the
load but at the same time they are more expensive as loop net-
⇑ Corresponding author. works require longer length of distribution conductors to connect
E-mail address: [email protected] (D. Sarkar). each and every load to the source in at least two parallel ways.
Peer review under responsibility of King Saud University. On the contrary, the advantages of the radial networks are lower
short circuit current and simpler relay co-ordination. The meshed
networks require more complex scheme and costly protective
devices like directional over current relays to protect loop or paral-
Production and hosting by Elsevier lel feeders.

https://doi.org/10.1016/j.jksues.2019.02.003
1018-3639/Ó 2019 Production and hosting by Elsevier B.V. on behalf of King Saud University.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Please cite this article as: D. Sarkar, P. Konwar, A. De et al., A graph theory application for fast and efficient search of optimal radialized distribution net-
work topology, Journal of King Saud University – Engineering Sciences, https://doi.org/10.1016/j.jksues.2019.02.003
2 D. Sarkar et al. / Journal of King Saud University – Engineering Sciences xxx (xxxx) xxx

In the past decade a number of research works have been car- challenge of fast adaptability in case of drastic power imbalance
ried out in the area of distribution system reconfiguration using situation. In that context, spanning tree approach takes only one
different techniques. In Haque (1996), by introducing some pass to identify the most optimal power distribution network, that
dummy buses, a meshed network is changed into a radial network too in real time and makes it fit for online applications. Edmond’s
through a reduced order node matrix, inorder to obtain a secured Algorithm, though complex, performs better than other established
system. In Goswami and Basu (1992), a radial distribution network spanning tree algorithms such as Prim’s and Kruskal’s. This is
was determined with the minimum losses through heuristic algo- established in the results obtained in the present work.
rithm. The algorithm is obtained by solving KVL and KCL (Kirchoff’s The feeder load carrying capability is a function of node voltage
voltage and current laws) equations of the network. To obtain a magnitudes, the differences of the voltage angles and feeder reac-
radial network in a single loop, the optimum path for power flow tance between the adjacent nodes. In the proposed technique, the
was obtained by switching operation of the switches. In Wen load carrying capability of each feeder is considered as a cost
et al. (2016) a Radial Structure is obtained from a distribution net- (weight) of the corresponding edge of the graph of the network.
work by changing the sectionalizing switches and tie switches. In The target is then to maximize the cost to search for the optimal
(Rao and Sivanagaraju, 2010) to minimize the losses, a guiding radial network configuration with highest cost. In simple terms,
search over the relevant configuration through network reconfigu- the searching algorithm attempts to open the switches on lower
ration is obtained, and reduction in the losses and improvement in load carrying feeders, so that the power originally flowing through
load balancing were observed. The system operates in a radial con- those feeders are diverted through alternative feeders having
figuration by opening and closing the switches in pairs resulting in higher power carrying capability. The search continues until all
branch exchange, also loop cutting method was incorporated the feeder branches on the network operate optimally at highest
wherein the switches are opened from a completely meshed net- possible loading.
work to obtain a feasible radial configuration. In Swarnkar et al.
(2010), feasible radial topologies were obtained with the help of
graph theory via reconfiguration through evolutionary process. 2. Theory
Nguyen and Truong (2015) presented a method based on a cuckoo
search algorithm (CSA) for minimizing active power loss and 2.1. Voltage stability assessment by L-Indicator
improving voltage magnitudes. Bernardon et al. (2009) proposed
a fuzzy multi-criteria decision making algorithm for network In the present study, the single–feeder equivalence method of
distribution network consisting of multiple feeders (Jasmon and
reconfiguration for reduction of power losses, enhancement of
the voltage profile, and in the increase of reliability. Lee, 1993) has been used for voltage stability assessment. The sin-
gle feeder system, is illustrated in Fig. 1.
Teimourzadeh and Zare (2014) proposed the binary group
search optimization algorithm (BGSO) for solving the optimal dis- Where, P = Injected Real Power, Q = Injected Reactive Power,
r = Resistance of Feeder; V = Sending End Voltage, x = Reactance
tribution network reconfiguration problem. Esmaeilian and
Fadaeinedjad (2015) employed network reconfiguration and of Feeder, PL = Real Load, QL=Reactive Load
From Fig. 1, the real and reactive power equations can be
capacitor placement simultaneously to reduce energy losses and
improve system reliability subjected to operational and power derived as
quality constraints using a fuzzy approach. Hsiao and Chien

(2000) presented a combination of fuzzy-GA method to resolve r P2 þ Q 2
P¼ þ PL ð1Þ
the service restoration problem. Gohokar et al. (2004) proposed a V2
method to solve the reconfiguration problem using network topol-
ogy considering single loop optimization, based upon highest volt-

x P2 þ Q 2
age drop across the open switches. The proposed algorithm was Q¼ þ QL ð2Þ
also utilized to check radiality of the network. Arun and V2
Aravindhababu (2010) proposed a new fuzzy logic based reconfig- Rearranging Eqs. (1) and (2) we obtain
uration algorithm that enhanced voltage stability and improved
the voltage profile besides minimizing losses, without involving xðP  PL Þ ¼ rðQ  Q L Þ ð3Þ
any extra cost of hardware. Kashem et al. (2000) proposed a new
On rearranging Eq. (3), quadratic equations in terms of P and Q
technique for network reconfiguration for enhancement of voltage
are obtained
stability under varying load conditions. The Depth First Search
(DFS) algorithm of graph theory has been utilized by Sarkar et al. 2x2 PL  2rxQ L þ r
(2011) for off line application, such as Power Systems planning. P¼
2ðr2 þ x2 Þ
Kruskal’s algorithm was attempted by Sarkar et al. (2015) in their h   
i12
work to generate radial network configurations from an originally 2x2 P L  2rxQ L þ r2  4 r2 þ x2  x2 P L 2 þ r2 Q L 2  2rxP L Q L þ rP L
meshed network to achieve better performances. 
2ðr2 þ x2 Þ
In the present paper Edmond’s Maximal Spanning Tree Algo- ð4aÞ
rithm has been proposed to generate optimal radial network from
1
a meshed network which may be considered a step forward from 2Q
L 2rxP L þx ½ð2x2 PL 2rxQ L þr2 Þ4ðr2 þx2 Þðx2 PL 2 þr2 Q L 2 2rxPL Q L þrPL Þ2
Q ¼ 2r 2ðr2 þx2 Þ
– 2ðr 2 þx2 Þ
(4b)
the work done in Sarkar et al. (2015) as it consumes less time
and delivers more effective solution to the problem.
Though there are several soft computing and optimization tech-
niques available, spanning tree approach is straight forward and -
works in real time. Optimization techniques such as GA, PSO or
classical approaches take time to converge as they go through sev-
eral randomly generated combinations, evaluating them. Neural
network approach needs the network to be trained offline before
being implemented online to rearrange the network for optimal
performance. This needs time and the network may fall into the Fig. 1. Single Feeder system.

Please cite this article as: D. Sarkar, P. Konwar, A. De et al., A graph theory application for fast and efficient search of optimal radialized distribution net-
work topology, Journal of King Saud University – Engineering Sciences, https://doi.org/10.1016/j.jksues.2019.02.003
 D. Sarkar et al. / Journal of King Saud University – Engineering Sciences xxx (xxxx) xxx 3

The above equations are in quadratic form and, for P and Q to 1, node 2, node 3 and node 4 to be V1\h1, V2\h2, V3\h3, V4\h4
have real roots, the determinant of the equations b2 – 4ac > 0. respectively.
The load carrying capability of feeders 1 to 4 can be formulated
4ðxPL  rQ L Þ2 þ xQ L þ rPL < 1 ð5Þ as:
  
Where L < 1  V 1V 3 
Hence, for the reduced network
LC 1 ¼  sinðh1  h3 Þ ð8Þ
X1
h 2 i
L ¼ 4 xeq P1eq  r eq Q leq þ Xeq Q leq þ req Pleq   
 V 1V 2 
LC 2 ¼  sinðh1  h2 Þ ð9Þ
X 2
Where,
  
 V 3V 4 
req = equivalent resistance for single line LC 3 ¼  sinðh3  h4 Þ ð10Þ
xeq = equivalent reactance for single line X 3

PLeq = total real loads in the distribution network  


QLeq = total reactive loads in the distribution network  V 2V 4 
LC 4 ¼  sinðh2  h4 Þ ð11Þ
X
P 3
Ploss
r eq ¼ n P 2  P 2 o ð6Þ Also consider, the load carrying capabilities of feeder-1 to
P Leq þ P loss þ Q Leq þ Q loss feeder-4 are in decreasing order. Since, the load carrying capability
of the feeder-4 is minimum, while forming radial network as
P shown in Fig. 3, the most desirable option would be opening the
Q loss
xeq ¼ n P 2  P 2 o ð7Þ sectionalizing switch of feeder-4. The power will be fed to node
PLeq þ Ploss þ Q Leq þ Q loss 4 via feeder-3, in that case. Under this condition, the feeder-3 shall
not become overloaded, provided, the load carrying capability of
The stability index (L) ranges between 0 and 1, with the value the feeder-3 is greater than approximately the sum of power flow
close to ‘0’ indicates stable systems and value closer to ‘1’ indicates through feeder-3 and feeder-4 before opening of switch. The new
possibility of voltage collapse. radial configuration ensures that all the feeders have higher load
carrying capability than earlier.
2.2. Reconfiguration strategy In the present study Edmond’s Maximal Spanning Tree algo-
rithm has been successfully utilized to achieve this goal for larger
A 4 node distribution power network is shown in Fig. 2 may be networks with the help of Cost Matrix of the edges of the graph
considered to illustrate the proposed reconfiguration strategy. corresponding to the network. The generalized Cost Matrix
Considering the reactance (p.u.) of Feeder-1, Feeder-2, Feeder-3, (Weight matrix) for a ‘n’ node distribution network is of size
Feeder-4 to be X1, X2, X3, X4 respectively and voltages (p.u.) at node n  n and aij presents a cost- matrix element which refers to the

Fig. 2. A 4-node mesh Distribution System.

Fig. 3. A 4 node radial network developed from mesh distribution system.

Please cite this article as: D. Sarkar, P. Konwar, A. De et al., A graph theory application for fast and efficient search of optimal radialized distribution net-
work topology, Journal of King Saud University – Engineering Sciences, https://doi.org/10.1016/j.jksues.2019.02.003
4 D. Sarkar et al. / Journal of King Saud University – Engineering Sciences xxx (xxxx) xxx

load carrying capability(p.u.) of the feeder connected between 2.4. Edmond’s algorithm for finding maximal spanning tree
nodes i and j.
Table 1 presents Cost Matrix for the system as shown in Fig. 2. This algorithm (Edmonds, 1967) is applicable for a directed
The cost of non-connected edges between two nodes, such as graph. The algorithm begins by traversing the directed graph
a11 = a14 = a22 = a23 = a32 = a33 = a41 = a44 = 0. For the present analy- (Digraph) and examines vertices and edges. The examined vertices
sis, element of the cost-matrix for a larger system can normally be are placed in a so called Vertex Bucket (BV). Provisionally selected
obtained using Distribution Load Flow Analysis. Once the power edges are placed in an Edge Bucket (BE). Throughout the execution
flow and load carrying capacity of all feeders are obtained by of the algorithm BE always contains an acyclic collection of direc-
Distribution Load Flow Analysis, the Edmond’s Maximal Spanning ted edges with at most one edge being incident on any vertex not
Tree search a radial network having feeders with higher load carry- in BV.
ing capacity, and rejects those with lower load carrying capacity. This algorithm works in two phases, as described below:
For example, Fig. 3 presents new network derived from the
mesh network Fig. 2 having the highest total cost (sum of load car- Edmond’s Algorithm:
rying capability of all the feeders present in the network).
1. BV = BE = U
Thus, in the proposed technique sectionalizing switches on the
2. i = 0
feeders would be opened in such manner that the feeders with load
3. if BV – Vi(Vi ? set of total vertices of the graph Gi)
carrying capabilities in descending order are selectively chosen and
Go to 4.
the lower load carrying capability feeders are kept out of service by
Else
opening sectionalizing switches.
Go to 8.
The present problem can therefore be formulated as a multi-
4. For some vertex v R BV and v 2 Vi,
objective optimization problem to achieve the goals of low over-
Do
all power loss and maximum over all voltage stability under
a. BV = BV [ {v}
given operating condition, subject to the constraints that none
b. Find an edge e = {x,v} such that
of the nodes are left isolated or supply of power to each and
w(e) = max{w(y,v) | (y,v)  Ei}(Ei ? set of all edges of the
every load is ensured, and power flow through none of the feed-
graph Gi)
ers exceeds the maximum power transfer capability limit. The
a.cIf w(e)  0, then go to 3(don’t select negative weight
primary tool that has been used to realize this practice is first
edges)
to radialize an originally meshed distribution network configura-
Done
tion and then to employ an efficient search algorithm like
5. If BE [ {e} contains a circuit
‘‘Edmond’s Maximal Spanning Tree” algorithm to search for an
a. i = i + 1
optimal radial configuration, looking for those trees (Set of feed-
b. Construct Gi by shrinking Ci to new vertex ui.
ers) which delivers the desired optimization goal subject to con-
c. Modify BE, BV and corresponding edge weights
straint satisfaction.
6. BE = BE [ {e}
There are 30 and 57-nodes in the test systems and hence total
7. Go to 3
number of possible network configuration would be 230 and 257
8. While i – 0,
respectively. The best voltage stability condition will result in from
Do
a unique switching configuration out of these huge combinations.
a. Reconstruct Gi-1 and rename some edges in BE
However, it would be tedious task to study each one of this huge
b. If ui was a root of an outbound tree in BE, then
number of possible combinations individually, as each of config-
BE = BE[fejeC i ande–e0i g
ured network Distribution Load Flow Analysis solution would take
unacceptably long time. Thus for on-feeder operation, repetitive Else
Distribution Load Flow Analysis solution is not at all a practical BE = BE[fejeC i ande– ei g
solution to the problem. c. i = i  1
Done
P
9. Maximum Branching Weight = e2BE wðeÞ
2.3. Maximum spanning trees

Consider a directed graph, G = (V,E) in which each edge (i,j) (vi, Working Principle
vjV) has a score s(i, j). Since G is directed, s(i,j) does not necessarily Phase 1:
equal s(j,i). The maximum spanning tree (MST) of G is the tree y Fig. 4(a) shows the original diagraph G0 = (V0, E0). Gi = (Vi, Ei) is
P
that maximizes the value ði;jÞy sði; jÞ; such that ði; jÞE and every the graph obtained by replacing the ith circuit Ci by a single vertex
vertex in V is used in the construction of y. The maximum projec- ui (Fig. 4(b)). With every circuit Ci encountered, a new graph Gi has
tive spanning tree of G is constructed, similarly, except that it can to be constructed with Ci shrunk into a new artificial vertex ui. This
only contain projective edges relative to some Feeder ordering on gives rise to a succession of graphs G1, G2,. . ., Gk and so on, at every
the vertices of G. The MST problem for directed graphs is also pass encountering a circuit. This process goes on until all the ver-
known as the maximum arborescence problem. tices in the current graph Gi has been covered and Vertex Bucket
(BV)is equal to Vi (the set of vertices in the current graph).
At any time, Gi contains every vertex of Gi-1(the previous
graph) except those belonging to Ci. With each circuit Ci encoun-
Table 1 tered, the vertices belonging to Ci are removed from Vi and
Feeder Cost Matrix for the system shown in Fig. 2.
replaced by the artificial vertex ui. So Vi contains all vertices of
Node 1 Node 2 Node 3 Node 4 Vi-1 except those in Ci, plus ui. Similarly, Ei contains all edges of
Node 1 0 LC2 LC1 0 Ei-1 except those having one or more end-points in Ci.
Node 2 LC2 0 0 LC4 New edges in Ei can be added as follows:
Node 3 LC1 0 0 LC3 For every edge (x,y)  Ei-1, with   Ci and y  Ci, Ei contains an
Node 4 0 LC4 LC3 0
edge (x,ui) such that ui  Vi, any edge in Gi has the same weight as

Please cite this article as: D. Sarkar, P. Konwar, A. De et al., A graph theory application for fast and efficient search of optimal radialized distribution net-
work topology, Journal of King Saud University – Engineering Sciences, https://doi.org/10.1016/j.jksues.2019.02.003
 D. Sarkar et al. / Journal of King Saud University – Engineering Sciences xxx (xxxx) xxx 5

with ui as an end-point are replaced by their equivalent edges in

Gi-1. However, selection of the edges of Ci which are to be added
back to BE depends upon whether BE already contains an edge
incident to a vertex in Ci. In other words, if ui is already a root of
a maximum branching tree passing through Ci, then add an edge
from Ci to BE which is not the minimum weight edge. This can
be expressed as:

BE BE [ fejeC i and e–e0i g

Otherwise, add an edge from Ci to BE which not have a common
vertex as end-point with an outer incident edges, that is:

BE BE [ fejeC i and e– ei g
Fig. 4. Shrinking of a cycle into a new artificial vertex.
Finally, the final set of edges BE defines a maximum branching
for G0. This can be calculated as: Maximum Branching Weight =
P
in Gi-1 except that incident to ui.If (x,ui) is a new edge in Ei, then its e2BE wðeÞ
weight is assigned by the following rule: The process of edge addition is demonstrated in Fig. 5.
 
wðx; uiÞ ¼ wðeÞ  w ei wðe0i Þ 3. Implementation of proposed algorithm and results

Where e=(x,y) is the equivalent edge of (x,ui) in Ei-1, ei ! unique 3.1. Description of the test system
edge in Ci which is incident to y  Ci, e0i !an edge of minimum
weight in Ci.BV is then modified by simply removing the vertices The proposed algorithm was successfully implemented to
falling into Ci. BE is modified by removing edges of Ci and replacing reconfigure 30 and 57 node distribution systems as illustrated in
them with edges (x,ui) having single end-point at Ci (i.e. ui). Figs. 6 and 9 respectively with non-time varying load at specific
Phase 2: nodes. It is assumed that, each of the feeders in test system is con-
This phase may be viewed as the reverse process of what nected between pairs of nodes through sectionalizing switches
described in phase 1.Gi-1 is reconstructed from Gi. Edges in BE (not shown in Fig. 6). The system was originally operating in
meshed configuration when all the sectionalizing switches were
in closed condition. A new network is formed when a set of sec-
tionalizing switches are opened to disengage some of the feeders
from the meshed network. For a given system, the graph of the net-
work is first developed and Edmond’s algorithm is applied on the
present configuration to obtain the optimal radial configuration,
meeting the necessary objectives cited earlier in Section 2.2. (See
Fig. 7)

3.1.1. Test System 1

This 30-node mesh network is shown in Fig. 6, where each fee-
der consists of one series sectionalizing switch and is assigned
Fig. 5. Reverse process of Edge addition for reconstruction. numbers like S1 to S41. The data for the system was taken from

Fig. 6. The 30-node distribution system under study.

Please cite this article as: D. Sarkar, P. Konwar, A. De et al., A graph theory application for fast and efficient search of optimal radialized distribution net-
work topology, Journal of King Saud University – Engineering Sciences, https://doi.org/10.1016/j.jksues.2019.02.003
6 D. Sarkar et al. / Journal of King Saud University – Engineering Sciences xxx (xxxx) xxx

Fig. 7. Biograph of Optimal radial network of 30-node.

(Sarkar et al., 2015). The real and reactive power demands of the ples of Voltage stability indicator, active and reactive power loss
system are 283.40 MW and 126.20 MVAR respectively. Table 2 pre- data obtained with different switch combinations of radial net-
sents the connectivity of sectionalizing switches among different works have been presented in Table 4. In Table 4, the first three
nodes. Distribution Load Flow analysis was carried out to deter- configurations correspond to arbitrary (and manually chosen) net-
mine the cost of the edges (feeder) of the test system. For a 30 node work configurations. The active power losses for these three radial
system, the cost matrix size is 30  30. Newton Raphson Load Flow configurations are 0.30584p.u., 0.41121p.u. and 1.22419p.u.
analysis was conducted in each individual case on the test system respectively. For the same configurations, the reactive power losses
for evaluation of Initial voltage magnitudes and voltage angles of obtained are 1.25630p.u., 1.30056p.u. and 4.40431p.u. respec-
all buses. These information were then used for the computation tively. The optimum voltage stability index obtained out of these
of load carrying capability of the feeders. The cost values of the net- three radial configurations is 0.75904 (configuration 1). The 4th
work edges were utilized to find the maximal directed spanning configuration was obtained using Kruskal’s Maximal Spanning Tree
tree to form the optimal radial network. Edmond’s Maximal Span- algorithm (Sarkar et al., 2015) and the active and reactive power
ning Tree algorithm described in Section 2.4 was used to search for losses obtained are 1.099p.u. and 2.244p.u. respectively. The volt-
the optimal configuration. The optimal graph as obtained is age stability index obtained using this algorithm was 0.2821 while
depicted in Fig. 8. the voltage stability index obtained with the Edmond’s algorithm
To show the effectiveness of the proposed algorithm, alternative is 0.1685.
radial configurations were obtained both manually and also using The results are clearly indicative of the fact that the present
Kruskal’s Spanning Tree algorithm (Sarkar et al., 2015) and the algorithm is far superior than the algorithm used in (Sarkar et al.,
results have been compared with Edmond’s algorithm. A few sam- 2015), as it succeeded in finding even better (Optimal) switching

Table 2
Sectionalizing Switch placement data for the 30 bus distribution system.

Start Bus End Bus Switch Start Bus End Bus Switch Start Bus End Bus Switch Start Bus End Bus Switch
1 2 S1 6 10 S12 15 18 S22 23 24 S32
1 3 S2 9 11 S13 18 19 S23 24 25 S33
2 4 S3 9 10 S14 19 20 S24 25 26 S34
3 4 S4 4 12 S15 10 20 S25 25 27 S35
2 5 S5 12 13 S16 10 17 S26 28 27 S36
4 6 S7 12 14 S17 10 21 S27 27 29 S37
5 7 S8 12 15 S18 10 22 S28 27 30 S38
6 7 S9 12 16 S19 21 23 S29 29 30 S39
6 8 S10 14 15 S20 15 23 S30 8 28 S40
6 9 S11 16 17 S21 22 24 S31 6 28 S41

Please cite this article as: D. Sarkar, P. Konwar, A. De et al., A graph theory application for fast and efficient search of optimal radialized distribution net-
work topology, Journal of King Saud University – Engineering Sciences, https://doi.org/10.1016/j.jksues.2019.02.003
 D. Sarkar et al. / Journal of King Saud University – Engineering Sciences xxx (xxxx) xxx 7

Fig. 8. Optimal configuration obtained from Edmond’s algorithm (For 30 node System).

Fig. 9. The 57-node distribution system under study.

combinations, as revealed by the lowest computation time and the system under study is taken from (Bhuyan et al., 2014). The real
most voltage stable network condition among all the tested 600 and reactive power demands on the system are 1195.8 MW and
plus configurations. 319.40 MVAR respectively. Table 3 presents the sectionalizing
switch data for the system. (See Fig. 10)
3.1.2. Test System 2 Table 5 presents the results obtained from the case study. The
The 2nd case study was conducted on a 57-node mesh distribu- optimal network as obtained using the Edmond’s algorithm is
tion system as shown in Fig. 7. The data for 57-node distribution shown in Fig. 11.

Please cite this article as: D. Sarkar, P. Konwar, A. De et al., A graph theory application for fast and efficient search of optimal radialized distribution net-
work topology, Journal of King Saud University – Engineering Sciences, https://doi.org/10.1016/j.jksues.2019.02.003
8 D. Sarkar et al. / Journal of King Saud University – Engineering Sciences xxx (xxxx) xxx

Table 3
Sectionalizing Switch placement data for the 57 nodes distribution system.

Start Bus End Bus Switch Start Bus End Bus Switch Start Bus End Bus Switch Start Bus End Bus Switch
1 2 S1 5 6 S21 7 29 S41 47 48 S61
2 3 S2 7 8 S22 25 30 S42 48 49 S62
3 4 S3 10 12 S23 30 31 S43 49 50 S63
4 5 S4 11 13 S24 31 32 S44 50 51 S64
4 6 S5 12 13 S25 32 33 S45 10 51 S65
6 7 S6 12 16 S26 34 32 S46 13 49 S66
6 8 S7 12 17 S27 34 35 S47 29 52 S67
8 9 S8 14 15 S28 35 36 S48 52 53 S68
9 10 S9 18 19 S29 36 37 S49 53 54 S69
9 11 S10 19 20 S30 37 38 S50 54 55 S70
9 12 S11 21 20 S31 37 39 S51 11 43 S71
9 13 S12 21 22 S32 36 40 S52 44 45 S72
13 14 S13 22 23 S33 22 38 S53 40 56 S73
13 15 S14 23 24 S34 11 41 S54 56 41 S74
1 15 S15 24 25 S35 41 42 S55 56 42 S75
1 16 S16 24 25 S36 41 43 S56 39 57 S76
1 17 S17 24 26 S37 38 44 S57 57 56 S77
3 15 S18 26 27 S38 15 45 S58 38 49 S78
4 18 S19 27 28 S39 14 46 S59 38 48 S79
4 18 S20 28 29 S40 46 47 S60 9 55 S80

Table 4
Comparison of different radial configuration obtained by manual operations (configuration 1–3), Kruskal’s algorithm (configuration 4), and Edmond’s algorithm (configuration 5)
for 30 bus system.

Radial Configuration No. Switches which are open Active Power loss Reactive Power loss Voltage Stability Time for
(p.u.) (p.u.) Index computation
1 S3,S6,S8,S12,S20,S24,S26,S29,S31,S35,S39,S40 0.30584 1.25630 0.75904 Not Applicable
2 S3,S4,S9,S10,S14,S17,S21,S23,S32,S33,S38 0.41121 1.30056 0.7938 do
3 S3,S4,S8,S10,S14,S20,S22,S26,S28,S32,S33,S39 1.22419 4.40431 0.9461 do
4 (Kruskal’s) S3,S8,S12,S20,S21,S23,S29, S32, S33, S39,S40 1.099 2.244 0.2821 0.3665
5 (Edmond’s) S3,S6,S8,S12,S20,S24,S26,S29,S31,S35,S39,S40 1.0484 2.0065 0.1685 0.1618

Fig. 10. Biograph of Optimal radial network of 57-node system.

Please cite this article as: D. Sarkar, P. Konwar, A. De et al., A graph theory application for fast and efficient search of optimal radialized distribution net-
work topology, Journal of King Saud University – Engineering Sciences, https://doi.org/10.1016/j.jksues.2019.02.003
 D. Sarkar et al. / Journal of King Saud University – Engineering Sciences xxx (xxxx) xxx 9

Table 5
Comparison of different radial configuration obtained by manual operations (configuration 1–3), Kruskal’s algorithm (configuration 4), and Edmond’s algorithm (configuration 5)
for 57 bus system.

Radial Switches which are open Active Reactive Voltage Time for
Configuration Power loss Power loss Stability computation
No. (p.u.) (p.u.) Index
1 S6,S8,S10,S11,S12,S13,S18,S21,S25,S26,S30,S34,S44S54,S59,S62,S64,S69,S73,S74,S75,S78 1.16982 5.53721 0.9348 Not Applicable
2 S6,S8,S10,S11,S12,S18,S21,S25,S26,S30,S34,S4,S54,S59,S62,S64,S69,S73,S74,S75,S78,S28 1.17350 5.54935 0.9356 do
3 S8,S10,S11,S12,S13,S18,S21,S25,S26,S30,S34,S44,S54,S59,S62,S64,S69,S73,S74,S75,S78,S7 1.25835 5.92732 0.9558 do
4 (Kruskal’s) S6,S8,S10,S11,S12,S13,S18,S21,S25,S26,S30,S34,S44,S54,S59,S62,S64,S69,S73,S74,S75,S78 0.8904 5.59068 0.7478 1.5213
5 (Edmond’s) S6,S8,S10,S11,S12,S13,S18,S21,S25,S26,S30,S34,S54,S59,S62,S64,S69,S73,S74,S75,S78 0.8213 5.4625 0.7468 0.8237

Fig. 11. Optimal configuration of 57-node system obtained from Edmond’s algorithm.

4. Conclusion Voltage stability conditions were also found to improve when

using the proposed search algorithm.
Radialization of primary mesh distribution networks by optimal
operation of the sectionalizing switches on feeders was attempted
References
in this paper with the objective to achieve better operational per-
formance in terms of power loss reduction and voltage stability Arun, M., Aravindhababu, P., 2010. Fuzzy based reconfiguration algorithm for
improvement. While radial networks are arguably more efficient voltage stability enhancement of distribution systems. Expert Syst. Appl. 37,
in respect of power loss and voltage stability, transformation of 6974–6978.
Bernardon, D.P., Garcia, V.J., Ferreira, A.S.Q., Canha, L.N., 2009. Electric distribution
an originally mesh network to an optimal radially reconfigured network reconfiguration based on a fuzzy multi-criteria decision making
network demands fast and efficient search of the optimal switch algorithm. Electr. Power Syst. Res. 79, 1400–1407.
combinations, as the number of feasible alternative configurations Bhuyan, S., Hazarika, S., Bardalai, A., 2014. Power flow analysis on IEEE 57 bus
system using MATLAB. Int. J. Eng. Res. Technol. 3, 1161–1171.
is usually huge and most traditional search algorithms are unac- Edmonds, J., 1967. Optimum branchings. J. Res. Natl. Bur. Stand. 71, 233–240.
ceptably slow for real-time application. The paper demonstrated Esmaeilian, H.R., Fadaeinedjad, R., 2015. Distribution system efficiency
a novel application of Edmond’s Maximal Directed Spanning Tree improvement using network reconfiguration and capacitor allocation. Int. J.
Electr. Power Energy Syst. 64, 457–468.
based search algorithm to render fast and efficient search for opti- Gohokar, V.N., Khedkar, M.K., Dhole, G.M., 2004. Formulation of distribution
mal network configuration under a given set of operating condi- reconfiguration problem using network topology: a generalized approach.
tions. Efficiency of the proposed algorithm was tested on 30- Electr. Power Syst. Res. 69, 304–310.
Goswami, S.K., Basu, S.K., 1992. A new algorithm for the reconfiguration of
node and 57-node mesh distribution networks and the results distribution feeders for loss minimization. IEEE Trans. Power Delivery 7, 1484–
depict that the proposed algorithm consistently converged to bet- 1491.
ter solutions compared to other efficient search algorithms like the Haque, M.H., 1996. Efficient load flow method for distribution systems with radial
or mesh configuration. IEE Proceedings-Generation, Transmission and
Kruskal’s Maximal Spanning Tree Algorithm. The proposed algo-
Distribution, vol. 143, pp. 33–38.
rithm achieved 55.85% shorter convergence time compared to Hsiao, Y.T., Chien, C.Y., 2000. Enhancement of restoration service in distribution
Kruskal’s Algorithm when applied to 30-node distribution network systems using a combination fuzzy-GA method. IEEE Trans. Power Syst. 15,
and 45.85% shorter convergence time when applied to 57-node 1394–1400.
Jasmon, G.B., Lee, L.H.C.C., 1993. New contingency ranking technique incorporating
distribution network. The proposed algorithm also achieved a voltage stability criterion. In: IEE Proceedings C- Generation, Transmission
greater loss reduction in 30-node system and in 57-node system. and Distribution, pp. 87–90.

Please cite this article as: D. Sarkar, P. Konwar, A. De et al., A graph theory application for fast and efficient search of optimal radialized distribution net-
work topology, Journal of King Saud University – Engineering Sciences, https://doi.org/10.1016/j.jksues.2019.02.003
10 D. Sarkar et al. / Journal of King Saud University – Engineering Sciences xxx (xxxx) xxx

Kashem, M.A., Ganapathy, V., Jasmon, G.B., 2000. Network reconfiguration for Sarkar, D., De, A., Chanda, C.K., Goswami, S., 2015. Kruskal’s Maximal Spanning Tree
enhancement of voltage stability in distribution networks. IEE Proceedings – Algorithm for Optimizing Distribution Network Topology to Improve Voltage
Generation, Transmission and Distribution, vol. 147, pp. 171–175. Stability. Electr. Power Compon. Syst. 43, 921–1930.
Mukhopadhyay, A.K., Chakrabarti, A., 1990. A microprocessor based industrial load Sarkar, D., Goswami, S., De, A., Chanda, C.K., Mukhopadhyay, K., 2011. Improvement
bus static var controller. Electr. Mach. Power Syst. 18, 449–460. of voltage stability margin in a reconfigured radial power network using graph
Nguyen, T.T., Truong, A.V., 2015. Distribution network reconfiguration for power theory. Can. J. Electr. Electron. Eng. 2, 454–462.
loss minimization and voltage profile improvement using cuckoo search Swarnkar, A., Gupta, N., Niazi, K.R., 2010. Minimal loss configuration for large scale
algorithm. Int. J. Electr. Power Energy Syst. 68, 233–242. radial distribution systems using adaptive genetic algorithms. In: 16th National
Ramana, G., Ram, B.S., 2011. Power system stability Improvement using FACTS with Power Systems Conference, pp. 647–652.
expert systems. Int. J. Adv. Eng. Technol. 1, 395–404. Taylor, C.W., 1994. Power System Voltage Stability. McGraw-Hill.
Rao, P.R., Sivanagaraju, S., 2010. Radial distribution network reconfiguration for loss Teimourzadeh, S., Zare, K., 2014. Application of binary group search optimization to
reduction and load balancing using plant growth simulation algorithm. Int. J. distribution network reconfiguration. Int. J. Electr. Power Energy Syst. 62, 461–
Electr. Eng. Inf. 2, 266. 468.
Sakthivel, S., Mary, D., 2011. Voltage stability limit improvement incorporating SSSC Wen, J., Tan, Y., Zhang, J., 2016. An effective encoding scheme of obtaining radial
and SVC under line outage contingency condition by loss minimization. Eur. J. topology structure in distribution networks. SpringerPlus 5, 1463.
Sci. Res. 59, 44–54.

Please cite this article as: D. Sarkar, P. Konwar, A. De et al., A graph theory application for fast and efficient search of optimal radialized distribution net-
work topology, Journal of King Saud University – Engineering Sciences, https://doi.org/10.1016/j.jksues.2019.02.003