J Control Autom Electr Syst (2014) 25:103–111

DOI 10.1007/s40313-013-0070-x

Optimal Reconfiguration of Electrical Distribution Systems

Using Mathematical Programming
Marlon C. O. Borges · John F. Franco ·
Marcos J. Rider

Received: 1 April 2013 / Revised: 9 July 2013 / Accepted: 6 August 2013 / Published online: 28 August 2013
© Brazilian Society for Automatics–SBA 2013

Abstract This paper presents a novel mixed integer non- Constants

linear programming (MINLP) model for the reconfiguration
problem of electrical distribution systems (EDS), aiming to Ri j Resistance of branch ij ()
reduce the active power losses of the system. Two lineariza- Xi j Reactance of branch ij ()
tion techniques are applied in the MINLP model to obtain an Zi j Impedance of branch ij ()
equivalent mixed integer linear programming (MILP) model. V Minimum voltage magnitude (kV)
The use of an MILP model guarantees convergence to opti- V Maximum voltage magnitude (kV)
mality using existing optimization software. The proposed V nom Nominal voltage magnitude (kV)
MILP model represents, with a suitable degree of accuracy, Iij Maximum current magnitude of branch ij (A)
the steady-state operation of the EDS. Considering the typical PiD Active power demand at node i (kW)
operating conditions of the EDS, additional constraints are Q iD Reactive power demand at node i (kVAr)
included in the proposed MILP model to reduce the compu- S
i j Upper bound of each block of the power flow lin-
tational effort demanded by its solution. The results of four earization of branch ij
test systems (33, 70, 136, and 417 nodes) are presented in 
V
Discretization step of Vi
sqr
order to show the accuracy as well as the efficiency of the m iSj,w Slope of the wth block used in the linearization of
proposed solution technique. the power flow of branch ij
N Number of nodes
Keywords Reconfiguration of electric power distribution S Number of discretizations used for the variable Vi
sqr
systems · Active power losses reduction · Mixed integer W Number of blocks of the piecewise linearization
linear programming

Variables

List of symbols PiG Active power generated at node i

Q iG Reactive power generated at node i
Sets Qi j Reactive power flow of branch ij
Pi j Active power flow in the branch ij
b Set of nodes Vi
sqr
Square of voltage magnitude at node i
l Set of branches sqr
Ii j Square of current flow magnitude of branch ij
c Set of nodes that can be connected only by two
Pi+j Active power flow in the branch ij in the forward
branches
direction
Pi−j Active power flow in the branch ij in the backward
M. C. O. Borges · J. F. Franco · M. J. Rider (B) direction
Q i+j
Departamento de Engenharia Elétrica, Universidade Estadual Paulista
(UNESP), Ilha Solteira, São Paulo, Brazil
Non-negative auxiliary variable to obtain |Q i j |
e-mail: [email protected] Q i−j Non-negative auxiliary variable to obtain |Q i j |

123
104 J Control Autom Electr Syst (2014) 25:103–111

yi+j Binary variable that defines if the active power flow authors solved the model using two different optimization
is in the forward direction techniques: a genetic algorithm and a solver for mixed integer
yi−j Binary variable that defines if the active power flow linear programming (MILP), in which the objective function
is in the backward direction and the electrical constraints were previously transformed
sqr
x j,s Binary variable used in the discretization of V j into linear functions. In 2010, a mixed integer quadratic
iPj,w Value of the wth block associated with active power programming model for the reconfiguration problem was
flow of branch ij developed (Romero-Ramos et al. 2010), with the advan-
Q
i j,w Value of the wth block associated with reactive tage of avoiding nonlinear constraints and additional binary
power flow of branch ij variables.
ai j Auxiliary variable used in the modeling of the state Jabr et al. (2012) presented two models for the EDS recon-
of branch ij figuration problem. One of them used mixed integer conic
Picj,s Power corrections used in the discretization of programming to represent the reconfiguration of the network
sqr sqr
Vi Ii j in a convex model. The second model was an MILP for-
mulation, based on a polyhedral representation of the conic
constraints used in the first model.
The combinatorial nature of the problem led the researc-
1 Introduction hers to develop heuristic methods, such as the one developed
by Baran and Wu (1989), which uses branch exchange and
The reconfiguration of electrical distribution systems (EDS) applies two specific methods for calculating the load flow in
deals with the modification of the system topology by open- radial networks. In 1992, another heuristic was developed by
ing or closing interconnection switches located in strategic Goswami and Basu (1992), which follows the technique of
points of the system in order to isolate faults, improve volt- closing an interconnection switch to form a single loop in the
age profile, or reduce the active power losses of the EDS. system; then, the power flow is calculated to determine the
The optimal reconfiguration of an EDS is currently a relevant branch of the loop with the lowest flow. Lastly, the intercon-
problem, especially when advanced distribution automation nection switch associated with that branch is opened. This
and smart grid technologies become more important and process is performed until the algorithm finds a configura-
more accessible to distribution companies. tion with minimum power losses.
The reconfiguration of an EDS is a combinatorial prob- In 2008, a new algorithm for the reconfiguration prob-
lem, and it can be modeled as a mixed integer nonlinear lem was proposed by Martin and Gil (2008). This algorithm
programming (MINLP) problem (Lavorato et al. 2012) with begins with a radial topology by opening all switches that
the objective of minimizing the active power losses in the form loops. In each step, an interconnection switch is closed,
EDS, which is subject to various constraints, such as voltage and the heuristic procedure establishes a reduced number
magnitude limits on the nodes, current magnitude limits of of candidate interconnection switches to be opened in order
branches, active and reactive power balances, and radial oper- to eliminate the generated loop. Finally, the choice of the
ation of the EDS. The dimension of the problem is directly most effective switch for minimizing the power losses is
related to the number of interconnection switches in the EDS. obtained from a power flow calculation. In the same year,
Therefore, if an EDS has C switches, the number of possible another algorithm was presented (Raju and Bijwe 2008),
topologies is 2C . Most of those topologies are not feasible, which begins with all interconnection switches closed, and
do not satisfy the radiality constraint, or have disconnected in each step, sensitivities are used to determine the intercon-
nodes (Schmidt et al. 2005). nection switch that is open, a step that is repeated until a
One of the first technical solutions applied to the reconfig- radial configuration is obtained. In a second phase, the algo-
uration problem is the constructive heuristic algorithm pro- rithm of branch exchange is applied to improve the quality of
posed by Merlin and Back (1975). This work presented an the solution. Abul’Wafa (2011) developed another heuristic
integer programming model that allows the determination of algorithm, which used the branch exchange technique and
high quality solutions. In 1996, Abur (1996) presented a for- a load flow algorithm based on graph theory. The load flow
mulation for the reconfiguration problem using a model for used makes it possible to reduce the interconnection switch-
minimum active power flow cost; by ignoring the capacity ing combinations investigated.
limits of the branches, the author solved the reconfiguration Metaheuristics are also used to address the problem of
problem using linear programming. reconfiguration, including genetic algorithms (Zhu 2002),
A more complete formulation for the reconfiguration simulated annealing (Chiang and Jean-Jumeau 1990), ant
problem was proposed in 2005 (Ramos et al. 2005), which colony search (Chang 2008), and tabu search algorithms
modeled both topological and electrical constraints, mak- (Guimarães et al. 2004), among others. Other researchers
ing the model more realistic than previous proposals. The have solved the problem of reconfiguration using artificial

123
J Control Autom Electr Syst (2014) 25:103–111 105

intelligence techniques such as fuzzy logic (Kangan and Bar- Vk Vi Vj

ioni 1999) and artificial neural networks (Hayashi et al. 1996; Pki , jQki , I ki Pij , jQij , I ij
Salazar et al. 2006).
Rki , jX ki , Z ki Rij , jX ij , Z ij
If the reconfiguration problem is represented using an
k i j
MINLP model, the solution technique does not guarantee
Rki I ki2 + jX ki I ki2 Rij I ij2 + jX ij I ij2
that the configuration found is actually the optimal solution PkS + jQkS Pi + jQ
D
i
D
P + jQ Dj
j
D

of the reconfiguration problem due to the non-convexity of

Fig. 1 Radial distribution system
the model; the same situation is repeated when heuristics,
metaheuristics, and artificial intelligence are employed. In
With the assumptions described above, the reconfiguration
this way, an interesting alternative is to represent the EDS
problem of electrical distribution systems can be represented
reconfiguration problem as an MILP, considering the effi-
using an MINLP model, as shown in (1)–(15).
ciency of existent solvers for these problems and the fact that  sqr
in this case, it can be ensured that the solution is actually min v = Ri j I i j (1)
the optimal one. With the evolution of optimization software i j∈l
and the increasing use of powerful solvers, some researchers
subject to
(Lavorato et al. 2012; Jabr et al. 2012; Romero-Ramos et   sqr
al. 2010) have developed mathematical models for the EDS Pki − (Pi j + Ri j Ii j ) + PiS = PiD ∀i ∈ b
reconfiguration problem. ki∈l i j∈l
In this work, an MINLP model for the EDS reconfigura- (2)
tion problem is proposed. Linearizations were made to obtain   sqr
Q ki − (Q i j + X i j Ii j ) + Q iS = Q iD ∀i ∈ b
an equivalent MILP model. The use of an MILP model guar- ki∈l i j∈l
antees convergence to optimality using existing optimization
(3)
software. The proposed model was tested in systems of 33, sqr
Vi − 2(Ri j Pi j + X i j Q i j )−
70, 136, and 417 nodes. In order to validate the approxi- sqr sqr ∀i j ∈ l (4)
Z i2j Ii j − V j − ai j = 0
mations performed, the steady-state operation point for the
sqr sqr
solutions was compared to that obtained using the load flow V j Ii j = Pi2j + Q i2j ∀i j ∈ l (5)
sweep method. sqr 2
The main contributions of this paper are as follows: V 2 ≤ Vi ≤V ∀i ∈ b (6)
Ii j ≤ I i j (yi+j + yi−j )
sqr 2
∀i j ∈ l (7)
(1) A novel MINLP model for the reconfiguration problem Pi+j − Pi−j = Pi j ∀i j ∈ l (8)
of EDS.
0 ≤ Pi+j ≤ V I i j yi+j ∀i j ∈ l (9)
(2) An equivalent model of MILP for the EDS reconfigu-
ration problem that presents an efficient computational 0 ≤ Pi−j ≤ V I i j yi−j ∀i j ∈ l (10)
behavior with conventional MILP solvers. −V I i j (yi+j + yi−j ) ≤ Q i j ≤ V I i j (yi+j + yi−j ) ∀i j ∈ l
(3) Additional constraints are included in the proposed
(11)
MILP model to reduce the computational effort dema-
)(1 − (yi+j yi−j ))
2
nded by its solution. |ai j | ≤ (V − V +2
∀i j ∈ l (12)

(yi+j + yi−j ) = N − 1 ∀i j ∈ l (13)
i j∈l

yi+j + yi−j ≤ 1 ∀i j ∈ l (14)

2 A Novel MINLP Model for EDS Reconfiguration
Problem yi+j and yi−j ∈ {0, 1} ∀i j ∈ l . (15)
In the proposed model, it is assumed that all branches have an
Aiming to represent the steady-state operation of an EDS,
interconnection switch that can be opened or closed in order
the following assumptions are made, commonly used in load
to obtain the best topology for the EDS operation. This is
flow formulations (Shirmohammadi et al. 1988), and are
realized in order to demonstrate the efficiency and robustness
shown in Fig. 1.
of the proposed model using test systems as shown in Sect. 4.
The objective function (1) represents the active power
• The load on the EDS is represented as constant active and losses of the EDS. Constraints (2) and (3) correspond to the
reactive power. balance of active and reactive power, respectively. The volt-
• The EDS is balanced and represented by its single phase age magnitude drop in the branches is represented by (4), and
equivalent. the current flow magnitude calculation is expressed by (5).

123
106 J Control Autom Electr Syst (2014) 25:103–111

Equation (4) is similar to the related one shown in Ces- xi ,1 = 1

xi ,2 = 1
pedes (1990), but an additional variable (ai j ) was added. This
xi , s = 1
additional variable depends on the state of the interconnec- xi , S = 1
tion switches. This equation allows to obtain the voltage mag-
nitude of the final node (V j ) in terms of the voltage magnitude V2 V 2 + ΔV V 2 + 2ΔV V 2 + sΔV V 2 + S ΔV V2
of the initial node (Vi ), the real power flow (Pi j ), the reactive V −V2 2

power flow (Q i j ), the current magnitude (Ii j ), and the elec- ΔV =

S +1
trical parameters of branch ij. The voltage magnitude limits
and the current flow capacity are represented by (6) and (7), Fig. 2 Discretization of square of voltage magnitude
respectively. The active power flow of the EDS is represented
using two variables in (8), to identify the active power flow in the EDS reconfiguration problem with an MILP model, it is
the forward direction (ij) and the backward direction (ji) with necessary to replace the nonlinear Eq. (5) by linear equivalent
the variables Pi+j and Pi−j , respectively. These variables are constraints (16)–(17) and (19)–(20).
bounded by (9) and (10) and can be used to include additional The left member of (5) is linearized through the discretiza-
sqr
constraints in the model as shown in Sect. 3.2. The reactive tion of V j using the binary variables x j,s ∀s = 0. . .S, in
sqr sqr
power flow is limited by (11). The auxiliary variable ai j is which x j,s = 1 if V j is greater than V j , as shown in
zero if the branch is active ij according to (12); otherwise, ai j Fig. 2 and explained in Franco et al. (2013b). The state of the
2
is free within the limits defined by (12) in order to maintain V (V −V 2 )
variables x j,s is modeled in (16), with  = (S+1) .
V
and satisfy (4). S
sqr
Constraint (13) is necessary (although not sufficient) to V2 +  x j,s ≤ V j
ensure radial configurations. Constraint (13), combined with s=1
(2) and (3), which represents the Kirchhoff s First Law V V 
S
(KFL), is used to obtain a radial topology for the reconfigura- ≤ V2 +  +  x j,s ∀i ∈ b (16a)
tion problem, like shown in (Lavorato et al. 2012). Constraint s=1
(14) defines that only one binary variable related to the power x j,s ≤ x j,s−1 ∀ j ∈ b , s = 1, . . ., S (16b)
flow direction is active. x j,s ∈ {0, 1} ∀ j ∈ b , s = 0, . . ., S. (16c)
Equation (15) represents the binary nature of the opera- sqr sqr
The product V j Ii j is approximated using the middle point
tion state of the interconnection switch of branch ij. In the
of the first interval of the discretization of the square voltage
model’s Eqs. (1)–(15), two binary variables (yi+j and yi−j )
magnitude multiplied by the square current flow magnitude,
are used to represent the state of a branch due to its com-
plus the successive power corrections (Picj,s ), which depend
putational efficiency compared to the solution of a model
V sqr
that employs only one binary variable; this comparison was on  , V , Ii j , and x j,s according to (17)
achieved experimentally. Additionally, these variables can be V 2 2
used to include additional constraints in the model as shown − V I i j (1 − x j,s ) ≤ Picj,s −
∀i j ∈ l , s = 0, . . ., S
in Sect. 3.2. The binary variables yi+j and yi−j are associated V sqr V 2 2
 Ii j ≤  V I i j (1 − x j,s )
with the direction of active power flow. yi+j is related to the (17a)
ij direction, while yi−j is related to the ji direction. If both 0≤ Picj,s ≤
V 2 2
V I i j x j,s ∀i j ∈ l , s = 0, . . ., S.
variables are zero, we obtain that the interconnection switch
of the branch is open. If one of the variables has a value of (17b)
1, it means that the branch is active, which means the inter- Using (16) and (17), we can write (18), which approximates
connection switch is closed. the left-hand side of (5).
  
S
sqr sqr 1 V sqr
V j Ii j = V 2 +  Ii j + Picj,s ∀i j ∈ l . (18)
2
3 Novel MILP Model for EDS Reconfiguration Problem s=1

The right member (5) is linearized as described in Franco et

3.1 Linearization al. (2013a) and defined in Eqs. (19)–(20).
 
The model (1)–(15) is nonlinear because (5) presents the V S
V 2 + 2 Ii j +
sqr
sqr sqr Picj,s =
product of two variables (V j Ii j ) and square terms (Pi2j and s=1
Q i2j ). Two linearization techniques are applied in (5) to obtain 
W 
W
Q
m iSj,w iPj,w + m iSj,w i j,w ∀i j ∈ l
an equivalent MILP model; more information about these lin- w=1 w=1
earizations is shown in Franco et al. (2013a,b). Thus, to solve (19)

123
J Control Autom Electr Syst (2014) 25:103–111 107

w =W tion problem. Some constraints may be added to the proposed

model in order to reduce the computational effort required to
Pij2
solve it. These constraints, which may be called cuts, restrict
W 2 ⎡⎣ Δ ijS ⎤⎦ = (V nom I ij )
2 2

the feasible region of the binary variables, but do not elimi-

w=3 nate the optimal solution.
mijS, w = (2 w − 1)Δ ijS
3.2 Additional Constraints
w=2
mijS,3 = 5Δ ijS For an EDS, without distributed generation and considering
the radiality constraint, we obtain by definition that each node
mijS,2 = 3Δ ijS
in the final solution is fed by one and only one branch. This
2 w =1 condition is represented by constraint (23) for each node j,
9 ⎣⎡ Δ ijS ⎦⎤

and it ensures that the sum of the binary variables that limit the
active power flow in the forward direction (yi+j ) and the binary
mijS,1
2
4 ⎡⎣ Δ ijS ⎤⎦

⎡⎣ Δ ijS ⎤⎦
2
variables that limit the active power flow in the backward
Δ ijP,1 Δ ijP,2 Δ ijP,3 Δ ijP,W
Pij
direction (yi−j ) is equal to one.
For example, assume that, in the case shown in Fig. 4, we
V nom I ij = W Δ ijS have found that the branch hj feeds the node j; therefore, yh+j
Fig. 3 Illustration of piecewise linear modeling of Pi2j
is equal to 1. Consequently, other variables associated with
flows that could feed node j, namely, y − +
ji and yk j , should be
equal to zero.
Q i+j − Q i−j = Q i j ∀i j ∈ l (20a)  
yi+j + y−
ji = 1 ∀ j ∈ b . (23)

W
i j∈l ji∈l
Pi+j + Pi−j = iPj,w ∀i j ∈ l (20b)
w=1 So, for this case, (23) will look as follows:
W
yh+j + yk+j + y −
Q i+j + Q i−j = i j,w
Q
∀i j ∈ l (20c) ji = 1.
w=1 Another constraint is defined for nodes that can be connected
0≤ iPj,w ≤
S
i j ∀i j ∈ l , w = 1, . . ., W (20d) only via two branches, which are represented by the set c ,
S as shown in Fig. 5 (node i). Assuming that there is a branch
0 ≤ iQj,w ≤ i j ∀i j ∈ l , w = 1, . . ., W (20e) ki with end node i, if the active power flow in the forward
0 ≤ Q i+j ∀i j ∈ l (20f) direction of branch ij is greater than zero, then the active
power flow in the forward direction of branch ki is necessarily
0≤ Q i−j ∀i j ∈ l , (20g)
S
in which the constant parameters m iSj,w and i j are defined
by h
S
m iSj,w = (2w − 1)i j ∀i j ∈ l , w = 1, . . ., W (21) yh+ j
S V Iij
i j = ∀i j ∈ l . (22) j
W
W W Q y +ji yk− j
The expressions w=1 m iSj,w iPj,w and w=1 m iSj,w i j,w
are the linear approximations of Pi2j and Q i2j , respectively.
The linearization of Pi2j is shown in Fig. 3. Q i+j and Q i−j are i k
non-negative auxiliary variables to obtain |Q i j | as shown in
Fig. 4 Representation of partial EDS
(20a). Constraints (20b) and (20c) state that |Pi j | and |Q i j |
are equal to the sum of the values in each block of the dis- yki+ yij+
cretization. Constraints (20d) and (20e) set the upper and
lower limits of the contribution of each block of |Pi j | and yki− yij−
|Q i j |. k i j
Thus, the model defined by (1)–(4), (6)–(17), and (19)–
(20) is an MILP model that represents the EDS reconfigura- Fig. 5 Control of circuit state of EDS

123
108 J Control Autom Electr Syst (2014) 25:103–111

active. Similarly, for the active power flow in the backward In the initial configuration of the system, the lower voltage
direction of branch ki to be greater than zero, it is required magnitude is at node 17 with a value of 0.91309 pu. After the
that the active power flow in the backward direction of branch reconfiguration, the lower voltage magnitude is found at node
ij must be greater than zero; thus, yi−j has to be equal to one. 31 with a value of 0.93779 pu; that voltage drop does not vio-
So, for nodes in c (24) can be written as late the minimum voltage magnitude, which is 7 %. The com-
 putational time spent by the proposed methodology to find the
yi+j ≤ yki
+
optimal solution for the system was 4 s; this time is required
∀i ∈ c . (24)
yki ≤ yi−j
−
for the B&B algorithm to explore the entire search tree.

Additionally, an initial superior quota is added in the pro- 4.2 70-Node Distribution System
posed methodology, which limits the objective function of the
reconfiguration problem with a superior bound. This limit This system has a 69 load nodes, one substation, and 74
corresponds to the active power losses associated with the branches. The nominal voltage magnitude is 12.66 kV, sup-
initial topology of the EDS. This initial value allows the plying a total load of 1117.86 kW and 902.25 kVAr. This
Branch-and-Bound (B&B) algorithm to remove the nodes system has 5 interconnection switches that are initially open.
of the B&B tree with an objective function greater than the Table 2 presents the results found by the proposed method-
initial active power losses. Consequently, it is possible to ology and the best results found in the literature. Note that
reduce the evaluated solutions during the search process of the proposed methodology found the same solution as that
the B&B algorithm. reported in the specialized literature.
In the initial configuration of the system, the lower volt-
age magnitude is found at node 66 with a value of 0.971963
4 Results pu. After the reconfiguration, the lower voltage magnitude is
found at node 62 with a value of 0.982399 pu. The compu-
The proposed methodology was implemented in the mathe- tational time spent by the proposed methodology to find the
matical programming language AMPL (Fourer et al. 2002) optimal solution for the system was 8 s; this time is required
and was solved using the commercial software CPLEX for the B&B algorithm to explore the entire search tree.
(CPLEX 2008) with default options. All simulations were
done using a computer with an Intel processor i7PC of 1.87 4.3 136-Node Distribution System
GHz. The test systems of 33 nodes (Chiou et al. 2005), 70
nodes (Chiang and Jean-Jumeau 1990), 136 nodes (Carreño This system has 136 nodes and 156 branches. The nomi-
et al. 2008), and 417 nodes (Ramírez and Bernal 1998) were nal voltage magnitude is 13.8 kV, supplying a total load of
used to show the performance and robustness of the proposed 18313.809 kW and 9384.827 kVAr. This system has 21 inter-
methodology. The operation point for the solution of the connection switches initially open. The results found by the
reconfiguration problem was compared with the one obtained proposed methodology and the best results found in the lit-
using the load flow sweep method (AC-LF). erature are shown in Table 3.
Note that the result found by the proposed method, even
4.1 33-Node Distribution System utilizing distinct values for the constant S, are the same as
that reported in the specialized literature, as can be seen in
This system has 33 nodes and 37 branches, with a nomi- Carreño et al. (2008). The computational time spent by the
nal voltage magnitude of 12.66 kV, supplying a total load proposed methodology to find the optimal solution for the
of 3715 kW and 2315 kVAr. This EDS has 32 interconnec- system utilizing S = 0 was 13 s and utilizing S = 1 was
tion switches, normally closed, and 5 open interconnection 1,236 s; this time is required for the B&B algorithm to explore
switches. Table 1 presents the results found by the proposed the entire search tree, but the solution was found in 363 s.
methodology and the best results found in the literature. Note After the reconfiguration, the lowest voltage magnitude in the
that the proposed methodology found the same solution as system is 0.9589 pu at node 106, whose voltage drop does not
that reported in the specialized literature. violate the minimum voltage magnitude of 5%. In the initial

Table 1 Results obtained

for 33-node system Methodology Open interconnection Active power Active power losses
switches losses (kW) using the AC-LF (kW)

Proposed 7, 9, 14, 32, 37 139.74 139.55

Mantovani et al. (2000) 7, 9, 14, 32, 37 139.55 139.55

123
J Control Autom Electr Syst (2014) 25:103–111 109

Table 2 Results obtained

for 70-node system Methodology Open interconnection Active power losses Active power losses
switches (kW) using the AC-LF (kW)

Proposed 15, 59, 62, 70, 71 9.95 9.34

Mantovani et al. (2000) 15, 59, 62, 70, 71 9.34 9.34

Table 3 Results obtained for 136-node system

Methodology Open interconnection S Time (s) Active power losses Active power losses
switches (kW) using the AC-LF (kW)

Proposed 7, 35, 51, 90, 96, 106, 118, 126, 135, 137, 0 13 284.35 280.14
138, 141, 142, 144, 145, 146, 147, 148, 150,
151, 155
Proposed 7, 35, 51, 90, 96, 106, 118, 126, 135, 137, 1 1236 280.51 280.14
138, 141, 142, 144, 145, 146, 147, 148, 150,
151, 155
Carreño et al. (2008) 7, 35, 51, 90, 96, 106, 118, 126, 135, 137, – 0.4 280.14 280.14
138, 141, 142, 144, 145, 146, 147, 148, 150,
151, 155

Table 4 Results obtained for the 417-node system

Methodology Open interconnection S Time (s) Active power losses Active power losses
switches (kW) using the AC-LF (kW)

Proposed 5, 13, 15, 16, 21, 26, 31, 53, 54, 59, 60, 73, 82, 0 7,445 591.79 581.80
87, 94, 96, 97, 110, 111, 115, 136, 142, 150,
155, 156, 163, 165, 168, 169, 178, 179, 191,
195, 199, 214, 221, 254, 256, 266, 282, 317,
322, 325, 358, 362, 369, 392, 395, 403, 404,
416, 423, 426, 431, 436, 437, 446, 449, 466
Proposed 5, 13, 15, 16, 21, 26, 31, 54, 57, 59, 60, 73, 86, 1 171,425 585.65 581.57
87, 94, 96, 97, 110, 111, 115, 136, 142, 149,
150, 155, 156, 163, 168, 169, 178, 179, 191,
195, 199, 214, 221, 254, 256, 266, 282, 317,
322, 325, 358, 359, 362, 369, 392, 395, 403,
404, 416, 423, 426, 431, 436, 437, 446, 449
Lavorato et al. (2012) 5, 13, 15, 16, 21, 26, 31, 54, 57, 59, 60, 73, 86, – – 685.88 685.88
87, 94, 96, 97, 110, 111, 115, 136, 142, 149,
150, 155, 156, 163, 168, 169, 178, 179, 191,
195, 199, 214, 221, 254, 256, 266, 282, 317,
322, 325, 358, 359, 362, 369, 392, 395, 403,
404, 416, 423, 426, 431, 436, 437, 446, 449

configuration of the system, the lower voltage magnitude than the results reported in the specialized literature, as can be
is 0.9326 pu at node 106. Thus, the reconfiguration of the seen in Lavorato et al. (2012). The computational time spent
system improved the voltage magnitude profile of the system. by the proposed methodology to find the optimal solution for
this system, which is the time needed by the B&B algorithm
4.4 417-Node Distribution System to explore the entire search tree, was 7,445 s for S = 0, but
the optimal solution was found in 1,958 s. For S = 1, the pro-
This system has 415 nodes and 473 branches and is a large- posed methodology found the optimal solution in 171,425 s.
scale EDS. The nominal voltage magnitude is 10.0 kV, sup- After the reconfiguration, the lowest voltage magnitude in the
plying a total load of 27958.01 kW and 13736.81 kVAr. system is 0.954233 pu at node 48, whose voltage drop does
This system has 59 interconnection switches initially open. not violate the minimum voltage magnitude of 5 %. When the
The results found by the proposed methodology and the best parameter S is increased, the proposed methodology finds an
results found in the literature are shown in Table 4. optimal solution of better quality using more computational
The result found by the proposed methodology, utilizing time, because the constant S increases the number of binary
different values for the constant S for this system, is better variables of the model. This difference between the solutions

123
110 J Control Autom Electr Syst (2014) 25:103–111

found by the proposed method is irrelevant when compared x j ,1 = 1

with the computational effort employed to find the optimal x j ,2 = 1
x j ,3 = 1
solution using S = 1.
x j ,0 = 0

81 84.8 88.6 92.4 96.2 100

ΔV = 3.8 V jsqr = 94.09

5 Conclusions
Fig. 7 Illustrative example of the voltage discretization
In this paper, the reconfiguration problem of electrical distri-
bution systems has been modeled as an MINLP model, with  
the objective of reducing the active power losses of the EDS. Q 12 − Q 23 + X 23 I23
2
+ Q S2 − Q S2 = 0 (26)
Two linearization techniques were used to obtain an equiva-
lent MILP model. Additional constraints are included in the
6.2 Illustrative Example of the Discretization of the Square
proposed model to reduce the computational effort demanded
of the Voltage Magnitude
by its solution. The use of an MILP model guarantees con-
vergence to optimality using conventional MILP solvers. The sqr sqr
Consider the following data: V j = 94.09kV 2 ; Ii j =
solutions obtained using the proposed methodology for sys-
tems of 33, 70, 136, and 417 nodes are equal to or better than 62500 A2 ; V = 9kV ; V = 10kV ; and S = 4, the con-
V
the solutions found in the specialized literature. The results stant  = 3.8 kV 2 can be obtained. The Fig. 7 shows an
show that the power losses are calculated with great precision illustrative example of the discretization of the square of the
in comparison with the load flow sweep method. voltage magnitude. Using Eq. (16), the values of the binary
variables x j,S can be established. So, x j,1 = x j,2 = x j,3 = 1
and x j,4 = 0.
From Eq. (17), one can find the values of Picj,s .
6 Appendix
V sqr
Picj,1 = Picj,2 = Picj,3 =  Ii j = 237500 and Picj,4 = 0
6.1 Illustrative Example of the Power Balance Formulation
Therefore, from (18), it is concluded that
In order to clarify the formulation for the balance of active
and reactive power used in this study, a test system of 3 nodes 
V
sqr

4
V2 + Ii j + Picj,s = 5893750 V A2 .
will be used (Fig. 6). 2
s=1
The active power losses of the circuit 2–3 represented by
2 are used in the balance of active power of the node 2
R23 I23 While the exact value is
as follows: sqr sqr
V j Ii j = (94.09)62500 = 5880625 V A2 .
 
P12 − P23 + R23 I23 2
+ P2S − P2D = 0 (25) Note that the result of the linear approximation for this exam-
ple has an error of 0.22 % compared with the exact value.
Thus, we can proceed to represent the active power balance This error can be reduced increasing the value of S.
in node 2, considering the node power injections: supplied
(P2S ) and demanded (P2D ) powers, incoming power flows 6.3 Illustrative Example of the Piecewise Linear Modeling
from previous branches (P12 ), and power flows from pos- of Pi2j
2 ). The
terior branches (P23 ) plus their power losses (R23 I23
same representation is made for the reactive power balance The Fig. 8 shows an illustrative example of the modeling of
2 .
in which the reactive power losses are represented by X 23 I23 Pi2j as a piecewise linear function. Consider the following
The reactive power balance is described below:
data: Pi j = 60kW, V nom I i j = 90 kV A, and W = 5. Thus,
S
one can obtain i j = 18 kV A from (22), and the values of
V1 V2 V3 m iSj,w can be calculated from (21)
P12 , Q12 , I12 , R12 , X 12 P23 , Q23 , I 23 , R23 , X 23

m iSj,1 = 18, m iSj,2 = 54, m iSj,3 = 90,

1 2 3 m iSj,4 = 126 e m iSj,5 = 162.
P1D + jQ1D P2S + jQ2S P2 + jQ2 P3S + jQ3S P3 + jQ3
D D D D

P + jQ
1
S
1
S

From Eq. (20b), one can find the values of the variables
Fig. 6 Illustrative test system iPj,w :

123
J Control Autom Electr Syst (2014) 25:103–111 111

Pij2 Franco, J. F., Rider, M. J., Lavorato, M., & Romero, R. (2013b). A
mijS,5 = 162 mixed-integer LP model for the optimal allocation of voltage reg-
8100 ulators and capacitors in radial distribution systems. International
Journal of Electrical Power & Energy Systems, 48, 123–130.
Goswami, S. K., & Basu, S. K. (1992). A new algorithm for the recon-
mijS,4 = 126 figuration of distribution feeders for loss minimization. IEEE Trans-
mijS,3 = 90
5184
actions on Power Delivery, 7(3), 1484–1491.
mijS,2 = 54 Guimarães, M. A. N., Lorenzeti, J. F. C., & Castro C. A. (2004).
3672
Reconfiguration of distribution system for voltage stability margin
2916 enhancement using tabu search.In POWERCON 2004: International
mijS,1 = 18 Conference on Power System Technology (pp. 1556–1561). Singa-
1296 60 pore.
324 Hayashi, Y., Iwamoto, S., Furuya, S., & Liu, C. C. (1996). Efficient
Pij determination of optimal radial power system structure using hop-
Δ ijP,1 = 18 Δ ijP,2 = 18 Δ ijP,3 = 18 Δ ijP,4 = 6 Δ ijP,5 = 0 field neural network with constrained noise. IEEE Transactions on
Power Delivery, 11(3), 1529–1535.
90 kVA Jabr, R. A., Singh, R., & Pal, B. C. (2012). Minimum loss network recon-
figuration using mixed-integer convex programming. IEEE Transac-
Fig. 8 Illustrative example of the linearization tions on Power Systems, 27(2), 1106–1115.
Kangan, N., & Barioni, C. C. O. (1999). Fuzzy decision model for the
reconfiguration of distribution networks using genetic algorithms. In
iPj,1 = iPj,2 = iPj,3 = 18, iPj,4 = 6 e iPj,5 = 0. Proceedings of the 13th Power System Computation Conference (pp.
929–935). Trondheim: Norway..
W
It is concluded that w=1 m iSj,w iPj,w = 3672kW 2 . Lavorato, M., Franco, J. F., Rider, M. J., & Romero, R. (2012). Imposing
radiality constraints in distribution system optimization problems.
The exact value for Pi2j is (60)60 = 3600 kW 2 . IEEE Transactions on Power Systems, 11(3), 172–180.
Note that the result of modeling Pi2j as a piecewise linear Mantovani, J. R. S., Casari, F., & Romero, R. A. (2000). Reconfiguração
function has an error of 2 % when compared to the exact de sistemas de distribuição radiais utilizando o critério de queda
de tensão (Reconfiguration of radial distribution systems using the
value. This error can be reduced by increasing the value of W. criterion of voltage drop). SBA Controle & Automação, 11, 150–159.
Martin, J. A., & Gil, A. J. (2008). A new heuristic approach for dis-
tribution systems loss reduction. Electric Power Systems Research,
References 78(11), 1953–1958.
Merlin, A., & Back, H. (1975). Search for a minimal-loss operating
Abul’Wafa, A. R. (2011). A new heuristic approach for optimal recon- spinning tree configuration in an urban power distribution system.
figuration in distribution systems. Electric Power Systems Research, In Power System Computation Conference (pp. 1–18).
81(2), 282–289. Raju, G., & Bijwe, P. R. (2008). An efficient algorithm for minimum
Abur, A. (1996). A modified linear programming method for dis- loss reconfiguration of distribution system based on sensitivity and
tribution system reconfiguration. Electric Power Energy Systems heuristics. IEEE Transactions on Power Systems, 23(3), 1280–1287.
Research, 18(7), 469–474. Ramírez, I. J. R., & Bernal, J. L. A. (1998). Genetic algorithms applied
Baran, M. E., & Wu, F. F. (1989). Network reconfiguration in distrib- to the design of large power distribution systems. IEEE Transactions
ution systems for loss reduction and load balancing. IEEE Transac- on Power Systems, 13(2), 696–703.
tions on Power Delivery, 4(2), 1401–1407. Ramos, E. R., Exposito, A. G., Santos, J. R., & Iborra, F. L. (2005).
Carreño, E., Romero, R., & Feltrin, A. P. (2008). An efficient codifica- Path-based distribution network modelling: Application to reconfig-
tion to solve distribution network reconfiguration for loss reduction uration for loss reduction. IEEE Transactions Power Systems, 20(2),
problem. IEEE Transactions on Power Systems, 23(4), 1542–1551. 556–564.
Cespedes, R. (1990). New method for the analysis of distribution net- Romero-Ramos, E., Riquelme-Santos, J., & Reyes, J. (2010). A simpler
works. IEEE Transactions on Power Delivery, 5(1), 391–396. and exact mathematical model for the computation of the minimal
Chang, C. F. (2008). Reconfiguration and capacitor placement for loss power losses tree. Electric Power Systems Research, 80(5), 562–571.
reduction of distribution systems by ant colony search algorithm. Salazar, H., Gallero, R., & Romero, R. (2006). Artificial neural net-
IEEE Transactions on Power Systems, 23(4), 1747–1755. works and clustering techniques applied in the reconfiguration of
Chiang, H. D., & Jean-Jumeau, R. (1990). Optimal network reconfigura- distribution systems. IEEE Transactions on Power Delivery, 21(3),
tions in distribution systems: Part 2: Solution algorithms and numer- 1735–1742.
ical results. IEEE Transactions on Power Delivery, 5(3), 1568–1574. Schmidt, H. P., Ida, N., Kagan, N., & Guaraldo, J. C. (2005). Fast
Chiou, J. P., Chang, C. F., & Su, C. T. (2005). Variable scaling hybrid reconfiguration of distribution systems considering loss minimiza-
differential evolution for solving network reconfiguration of distribu- tion. IEEE Transactions on Power Systems, 20(3), 1311–1319.
tion systems. IEEE Transactions on Power Systems, 20(2), 668–674. Shirmohammadi, D., Hong, H. W., Semlyen, A., & Luo, G. X. (1988).
CPLEX (2008). Optimization subroutine library guide and reference, A compensation based power flow method for weakly meshed dis-
version 11.0, CPLEX Division, ILOG Inc., Incline Village, NV, USA. tribution and transmission networks. IEEE Transactions on Power
Fourer, R., Gay, D. M., & Kernighan, B. W. (2002). AMPL: A modeling Systems, 3(2), 753–762.
language for mathematical programming. Pacific Grove: Duxbury. Zhu, J. Z. (2002). Optimal reconfiguration of electrical distribution net-
Franco, J. F., Rider, M. J., Lavorato, M., & Romero, R. (2013a). Optimal work using the refined genetic algorithm. Electric Power Systems
conductor size selection and reconductoring in radial distribution Research, 62(1), 37–42.
systems using a mixed-integer LP approach. IEEE Transactions on
Power Systems, 28(1), 10–20.

123