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

net/publication/260497562

Imposing Radiality Constraints in Distribution System Optimization Problems

Article in Power Systems, IEEE Transactions on · February 2012

DOI: 10.1109/TPWRS.2011.2161349

CITATIONS READS
484 1,032

4 authors:

Mario Lavorato John F. Franco

Institute of Scientific and Technical Research for Defense São Paulo State University
45 PUBLICATIONS 1,132 CITATIONS 125 PUBLICATIONS 3,792 CITATIONS

SEE PROFILE SEE PROFILE

Marcos J. Rider Rubén Romero

University of Campinas São Paulo State University
200 PUBLICATIONS 6,034 CITATIONS 252 PUBLICATIONS 10,223 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by John F. Franco on 11 May 2020.

The user has requested enhancement of the downloaded file.

 1

Imposing Radiality Constraints in Distribution

System Optimization Problems

er
Marina Lavorato, Student Member, IEEE, John F. Franco, Student Member, IEEE,
Marcos J. Rider, Member, IEEE, and Rubén Romero, Senior Member, IEEE.

ap
Abstract—Distribution systems commonly operate with a ra- n0ij existent circuit in branch ij.
dial topology, so all models of optimization problems in these lij circuit length of branch ij.
distribution systems should consider radiality in their formula- 0
tion. This work presents a literature review, a critical analysis and Si maximum apparent power limit of existent substation
a proposal for incorporating the radiality constraints in mathe- at node i.
matical models of optimization problems for radial distribution Si maximum apparent power limit of substation rein-
systems. The objective is to show that the radiality constraints forcement or construction at node i.
on such optimization problems can be considered in a simple I ij maximum current flow limit of branch ij.
and efficient way. The reconfiguration and expansion planning
nb number of nodes (nb =|Ωb |).

dP
problems of distribution systems are used to test and verify
the proposed radiality constraints. A generalization of radiality nbs number of bus substation nodes (nbs =|Ωbs |).
constraints is also examined. ndg number of distributed generators (ndg =|Ωbdg |).
Index Terms—Distribution system optimization, distribution PDi active power demand at node i.
system planning, distribution system reconfiguration, mixed bi- QDi reactive power demand at node i.
nary nonlinear programming, radiality constraint of the electrical gij conductance of branch ij.
distribution system. bij susceptance of branch ij.
Functions:
N OTATION
Pi active power calculated at node i.
The notation used throughout this paper is reproduced below
Qi reactive power calculated at node i.
for quick reference.
Pij active power flow that leaves node i toward node j.
pte
Sets: Qij reactive power flow that leaves node i toward node
Ωl sets of branches. j.
Ωb sets of nodes. Irij real current flow component of branch ij.
Ωbs sets of bus substation nodes (Ωbs ⊂ Ωb ). Imij imaginary current flow component of branch ij.
Ωbi sets of connected nodes in the node i (Ωbi ⊂ Ωb ). v total power losses.
Ωbp sets of transfer bus (Ωbp ⊂ Ωb ). f total investment and operation cost.
Ωbdg sets of distributed generation (Ωbdg ⊂ Ωb ).
Variables:
Constants: xij circuit that can be reconfigured on branch ij.
κl capital recovery rate of circuit constructions. nij circuit that can be added on branch ij.
κs capital recovery rate of substation reinforcement or mi substation number that can be added on node i.
ce

construction. Vi voltage magnitude at node i.

cij construction cost of branch ij (US$/km). θij difference of phase angle between nodes i and j.
cfi substation fixed cost at node i (US$). PSi active power provided by substation at node i.
α number of hours in one year (8760 h). QSi reactive power provided by substation at node i.
τl interest rate of the cost of power losses.
τs interest rate of substation operation cost. I. I NTRODUCTION
φl load factor of circuits.
E LECTRICAL distribution systems (EDS) must be ade-
Ac

φs load factor of substations. quately planned to permit an efficient and reliable opera-
cl cost per unit of energy lost (US$/kWh). tion. Although may be found some networked systems in urban
cvi substation operation cost at node i (US$/kVAh2). cores, the majority of EDS operate with a radial topology for
V minimum voltage magnitude. various technical reasons, the two most important follow: a) to
V maximum voltage magnitude. facilitate the coordination and protection; and b) to reduce the
Manuscript sent May 4, 2011. This work was supported by the Brazilian short-circuit current of EDS. Thus the radiality constraint is
institutions CNPq, FEPISA and FAPESP present in almost all of the expansion and operation planning
M. Lavorato, J. F. Franco, M. J. Rider and R. Romero are with the Faculdade problems. The most widely known problems are the distribu-
de Engenharia de Ilha Solteira, UNESP - Universidade Estadual Paulista,
Departamento de Engenharia Elétrica, Ilha Solteira - SP, Brazil. tion system (feeder) reconfiguration problem (DSR) and the
(E-mails: {marina, jfranco, mjrider, ruben}@dee.feis.unesp.br). distribution system expansion planning problem (DSP).
 2

The distribution system reconfiguration problem can be that this is a necessary but not a sufficient condition to
viewed as an operation planning problem of EDS. The main ensure the radiality of an EDS. This study suggests that it
objective of the DSR problem is to find a radial EDS with “would be highly desirable if the radiality constraint could
minimum losses. This problem can be modeled as a mixed be expressed in analytical form.” If this were possible, the
integer nonlinear programming problem (MINLP), in which radiality constraints could be incorporated in a mathematical

er
case working with the radiality of the EDS has always been model and solved by using a precise optimization technique.
considered complicated, as shown in [1], [2]. In [7] an algorithm branch-exchange is presented to solve
The distribution systems expansion planning problem is the DSR problem. The algorithm starts with a feasible radial
another MINLP problem related to the optimization of EDS. topology and then creates new radial topologies successively
In this problem, given an initial topology of the EDS, the by implementing one branch-exchange at a time. Thus the
aim is to obtain its expansion so that it can operate properly, algorithm guarantees the radiality of the EDS.

ap
and at the lowest investment cost, for a horizon planning in Meta-heuristics have also been used to solved the DSR
which demand is known. Thus, the DSP problem can add problem. The authors of [8] used an evolutionary algorithm;
new substations and/or repower existing ones or change the [9] and [10] used two specialized genetic algorithms. All
conductors of existing circuits and/or build new circuits in proposals implicitly consider the radial operation constraint
branch candidates, taking into account the radiality of the EDS. of the EDS. In [8] the radial operation constraint of the EDS
As in the DSR problem, the radiality in the DSP problem has is assured using graph theory, while that in [9] and [10] is
also been considered complicated, as shown in [3]. controlled inside genetic operators.
Both the DSP and DSR problems are well known, and sev- An alternative mathematical model that allows the solution

dP
eral papers have proposed contributions to modeling and tech-
niques to solve these problems. In the specialized literature, the
optimization techniques used to solve these problems can be
classified into two main groups: a) exact optimization tech-
niques and b) heuristic algorithms and meta-heuristics. The
first group, which includes the branch and bound algorithm,
has been used in conjunction with relaxed models (linearized
models) of the DSR and DSP problems to obtain a mixed
integer linear programming problem. But when considering
the more accurate models (nonlinear models) of the DSR and
DSP problems, heuristic algorithms and meta-heuristics have
of the DSR problem using commercial software is presented
in [2]. In this model, the radiality of an EDS is represented
by algebraic relations using so-called path-based connectivity
modeling. The authors of this study explicitly recognize that
it is very difficult to find a mathematical model for the DSR
problem and solve it with a conventional technique like the
branch-to-node algorithm.
In [11], [12], [13] and [14], three specialized genetic al-
gorithms and one simulated annealing are proposed to solve
the DSP problem. All four proposals implicitly consider
the radiality of an EDS through the operator application of
pte
been applied with great success in the last decades. each meta-heuristic. In [11] a new codification proposal is
presented, using the concept of prüfer number; this proposal
A. Literature review
allows the genetic operators to generate a radial topology.
In the specialized literature, the radial operation constraint A mathematical model that considers the distributed gener-
of an EDS appears mainly in the DSR and DSP problems. ation in the DSP problem is presented in [3]. In this proposal,
It must be pointed out that, if exact optimization techniques a constraint equivalent of (1) is shown. The authors state that
are employed, the radiality constraints must be explicitly the radiality cannot be guaranteed and that, “to ensure that the
represented in the mathematical modeling. This is not the case, network obtained is always radial, it may be necessary to add
however, when heuristic or meta-heuristic techniques are used, constraints with specific information about the topology of the
where the radiality constraints are controlled implicitly. network under analysis. Considering the diversity of situations
ce

Several works propose only (1) to represent the radial encountered, this task can be relatively complex.”
operation constraint of an EDS in the DSR and DSP problems. The authors of [15] propose the planning of the primary
M = nb − 1 (1) and secondary distribution networks in the same optimization
problem; the radial operation constraint of the EDS is modeled
where M is the number of branches of the solution obtained similarly as in (1). The authors mention that “the radial
in the DSR and DSP problems. characteristic is achieved by allowing a maximum of only
References [1], [4], [5] and [6] present four different one positive power input at each node.” The mathematical
heuristic algorithms to solve the DSR problem. The process model presented is solved with commercial software. The
Ac

begins with a completely meshed network, and in each step same authors present in [16] an ant colony algorithm to solve
of the heuristic algorithm, a branch is removed; the process the DSP problem. In this algorithm, the radial structure of the
ends when a radial topology is encountered. The difference solution is controlled implicitly in the constructive phase of
between each heuristic algorithm is the sensitivity analysis the meta-heuristic.
used to decide which branch should be removed/opened at Two branch-exchange algorithms are presented in [17] and
each step. Thus the radial operation constraint of the EDS is [18] to solve the DSP problem. The radial operation constraint
imposed implicitly by the heuristic algorithms, using (1) as of the EDS is obtained implicitly when the optimization pro-
stop criteria, and not explicitly in the DSR model. However, cess generates an initial radial topology; thus the other topolo-
[1] acknowledges that (1) is often used to represent the radial gies that are found, using the strategy of branch exchanging,
operation constraint in optimization problems of EDS, but should also be radials. In [19] and [20], some topics related
 3

to the DSP problem are presented. These authors mention that - 25 MW

radiality requires that (1) be satisfied and, additionally, that the . .............
........
.
. .. ..
EDS must be connected. A mathematical model for the multi- .. .. ..
f12 .... .. 2 ..
. ... .
stage DSP problem is also shown in [21]. In this proposal, a . . . f23 f25 ....
radial operation constraint similar to (1) is presented. ..
. .  ? .
.
... @ R ...
@

er
1 .. .. .. 3 ..
.. 5
Reference [22] is one of the first papers to propose a ........ f13 - ... f 35 - ....... -
mathematical model for the DSP problem. It recognizes the t-.................................................................
. ...
need to find optimal radial topologies, but the radiality con- g1 ..........
. .. -
... ....... 60 MW
..
..
.. 30 MW . ..
.
straints is not shown in the model. Finally, [23] presents a .. .. ..
..
@
.. f f  . ..
literature review on the DSP problem. This paper recognizes f14 @ .R.. .
. 34 45
..
.
the importance of the radiality constraints but provides no
..
.. ?.... . ..
.. .
. .
.. ... 4 ....

ap
detailed analysis of the subject. ........ ...........
- 20 MW
B. Contributions Figure 1. Illustrative problem
From the literature review above, it seems clear that the ex-
plicit representation of the radiality constraints is an issue that
has not yet been appropriately solved. If this representation is
possible, then the DSR and DSP problems can be solved by X
min v = fij2

dP
(2)
using integer programming techniques. The main aim of this
(ij)∈Ωl
paper is to further contribute to this subject. s.t.
The main contributions of this paper are to provide the X X
following: fji − fij + gi = di ∀ i ∈ Ωb (3)
(ji)∈Ωl (ij)∈Ωl
1) The sufficient conditions to ensure a radial topology in
EDS optimization problems; |fij | ≤ f ij xij ∀ (ij) ∈ Ωl (4)
2) A preliminary analysis of the generalization of radiality 0 ≤ gi ≤ g i ∀ i ∈ Ωbs (5)
constraints; xij ∈ {0, 1} ∀ (ij) ∈ Ωl (6)
3) A mathematical model for the DSR and DSP problems X
xij = nb − 1 (7)
in which the radiality constraints are represented explic-
(ij)∈Ωl
itly and solved with an integer programming technique.
pte
where xij is a binary variable and is equal to 1 if the
corresponding circuit is in operation; otherwise it is equal to
II. R ADIALITY C ONSTRAINTS IN THE E LECTRICAL
0. The active power flow between nodes i and j is denoted
D ISTRIBUTION S YSTEM
by fij . The variable gi is the active power supplied by the
The EDS topology can be considered a graph consisting of substation i. The parameter di represents the active power
n arcs and m nodes. From graph theory, it is known that a demand at node i. The variable f ij stands for the maximum
tree is a connected graph without loops; thus it is possible active power flow limit of branch ij. The parameter g i denotes
to compare the radial topology of an EDS with a tree. As the maximum active power limit of the substation i. The
is shown in the Appendix and [24], the tree of a graph is a mathematical model of this illustrative example was written
subgraph connected with (m−1) arcs. Hence one can state that in the mathematical programming language AMPL [25] and
ce

the topology of an EDS with nb nodes is radial if it satisfies solved with the CPLEX 12.1 [26] (called with default options).
the two following conditions: Considering g 1 = 150 MW and f ij = 100 MW for all
• Condition 1 - the solution must have (nb − 1) circuits; branches, the following solution was obtained:
• Condition 2 - the solution must be connected. v = 11525 MW2 g1 = 135 MW
Note that the radiality constraints have to be formed by x12 = x13 = x14 = x45 = 1
Condition 1 and Condition 2, and that Condition 1 alone x23 = x25 = x34 = x35 = 0
does not ensure the radiality of an EDS.
Ac

f12 = 25 MW, f13 = 30 MW, f14 = 80 MW, f45 = 60 MW

f23 = f25 = f34 = f35 = 0 MW
A. Illustrative problem
Note that the solution found is radial and is verifiably the
In an EDS system of 5 nodes, with 4 load nodes and only optimal solution of the illustrative problem.
one substation node, the aim is to minimize the sum of the
squared active power flow and ensure that the final topology B. Proof of the radiality condition
is radial (see Fig. 1). In this problem, only the First Kirchhoff In this subsection, we will prove that the feasible solution
Law (FKL) is considered; therefore there is no need to specify of the illustrative problem satisfies both Condition 1 and
branch reactances and resistances. The mathematical problem Condition 2; namely, a feasible solution must be radial. The
assumes the following form: main features of this illustrative problem are:
 4

(i) Only a single substation exists in the EDS (substation • Suppose that S = nbs − β < nbs , with β > 0. As the
node); minimum number of arcs of i-th graph is ki − 1, the
P Pnbs −β
(ii) All other nodes are load nodes; total number of arcs is ij∈Ωl xij ≥ i=1 P (ki − 1) =
(iii) Kirchhoff’s first law must be followed; and nb − nbs + β > nb − nbs which contradicts ij∈Ωl xij =
(iv) The aim is to find the best radial topology. nb − nbs . As the supposition is not true, therefore S ≥

er
Proof: Condition 1 is satisfied by (7). A solution that nbs ;
satisfies (3) must supply the power demand at every load bus, • As S ≤ nbs and S ≥ nbs , then S = nbs , which means
so that a path between the substation and each other bus exists. that there are nbs subgraphs.
Therefore, every bus is linked with the substation bus, forming • Let Ci be the minimum number of arcs to take off from
a connected graph, which proves Condition 2. i-thPsubgraph toP obtain a tree. The total number
Pof arcs
nbs nbs
Thus when power balance constraints (3) are combined with is ij∈Ωl xij = Pi=1 (ki −1+Ci ) = nb − nbs + i=1 Ci .

ap
constraint (7), each load node is connected by a single path to Thus, to obtain ij∈Ωl xij = nb − nbs , it is necessary
Pnbs
the substation node; that is, the system is connected without that i=1 Ci = 0, which implies that Ci = 0. Then all
meshes. nbs subgraphs are trees.
In the proof only features (i) - (iv) were used, of which • Since FKL must be satisfied, each tree has at least one
hypothesis (iii) is the one that guarantees that the demand in substation. As there are nbs trees and substations, there
each node is fed. The proof is valid for any other optimiza- is one and only one substation for each tree.
tion problem of an EDS, provided the four hypotheses are 2) Distributed generation and/or reactive power sources:
considered. This case also appears frequently in the DSP problem. Note

C. Generalization of Radiality Constraint

Three cases are presented in this section: 1) an EDS

dP
with more than one substation, 2) an EDS with distributed
generation and/or reactive power sources and 3) an EDS with
zero power injection nodes (so-called transfer nodes). Note
that each case is presented individually in this subsection;
however, the three cases can be consider together to form a
general model.
1) More than one substation: This case appears frequently
that hypotheses (i)-(iv) are still valid since reactive power
sources are included in the load nodes. As the distributed
generators can feed some loads independently, a part of the
EDS can operated isolated. In this case, Condition 2 is
not guaranteed by the power balance constraints (3), and
new constraints should be added to ensure that a distributed
generator is not isolated from the substation, as shown in (9)–
(12).
X
kji −
(ji)∈Ωl
X
kij = Ki
(ij)∈Ωl
∀ i ∈ Ωb (9)
in the DSP problem. Thus Condition 1 or constraint (1) must
pte
be modified to: Ki = 1 ∀ i ∈ Ωdg (10)
M = nb − nbs (8) Ki = 0 ∀i ∈
/ Ωdg ∪ Ωbs (11)
where nbs is the total number of substation nodes in the EDS. |kij | ≤ ndg xij ∀ (ij) ∈ Ωl (12)
Constraint (8) plus the power balance constraints guarantee where Ki represented a fictitious load of each distributed
that nb − nbs circuits will be constructed to feed the nb − nbs generator that only can be fed by the substation. Variable
load nodes and nbs radial topologies will be constructed in the kij denotes the fictitious flow associated with branch ij. If
final solution. it is allowed that the distributed generator feeds some loads
Proof: It must be proved that nbs trees (each tree with independently, then (9) – (12) are not considered in the model.
one
P and only one substation) is true if and only if both, 3) Transfer nodes: A transfer node is simply one node with
ce

ij∈Ωl xij = nb − nbs and FKL, are satisfied. no generation or demand. Transfer nodes are not so frequent
P First we will prove that nbs trees implies that both, in an EDS; they are used to connect a load node to other load
ij∈Ωl ij = nb − nbs and FKL, are satisfied. Let ki be the
x nodes. A transfer node is not a terminal node (this is the main
number of nodes of the i-th subgraph. The number of branches condition about using the transfer nodes); thus there are at
of the i-th tree is ki −1, according
Ptonb(1). Therefore, the number least two more circuits “leaving” the transfer node.
of branches of the nbs trees is i=1 s
(ki − 1), which is equal To model the use of a transfer node, one must define the
to nb − nbs . Moreover, as each tree has a substation, FKL is binary variable yi such that yi is equal to 1 if the transfer node
satisfied. P
Ac

is used; otherwise yi is equal to 0. To consider transfer nodes

Now, it must be proved that if both, ij∈Ωl xij = nb − nbs in the EDS, (8) is replaced by (13).
and FKL, are satisfied implies that there are nbs trees. Let S X X
be the number of subgraphs. It must be proved that S = nbs xij = nb − nbs − (1 − yj ) (13)
and that all the subgraphs are trees, each tree with one and (ij)∈Ωl j∈Ωbp
only one substation. xij ≤ yj ∀ (ij) ∈ Ωl , ∀ j ∈ Ωbp (14)
• Suppose that S = nbs + β > nbs , with β > 0; so there
xji ≤ yj ∀ (ji) ∈ Ωl , ∀ j ∈ Ωbp (15)
is at least one subgraph without substation and its load X X
nodes cannot be supplied, which contradicts that FKL is xij + xji ≥ 2yj ∀ j ∈ Ωbp (16)
satisfied. As the supposition is not true, therefore S ≤ (ij)∈Ωl (ji)∈Ωl
nbs ; yj ∈ {0, 1} ∀ j ∈ Ωbp (17)
 5

Constraints (13)–(17) avoid loop generation due to the pres- Equation (21) represents the constraints on the voltage
ence of transfer nodes in the EDS and also prevent the ap- magnitude of nodes. The elements of current flow in branch
pearance of a terminal transfer node (with only one connected ij of (22) are given by (27) and (28).
circuit). On the other hand, if it is known that all transfer nodes
Irij = gij (Vi cos θi −Vj cos θj )−bij (Vi sin θi −Vj sin θj ) (27)
are part of the radial topology of the EDS, a simple method

er
is to assume a small value of load (for example, 0.001 pu) in Imij = gij (Vi sin θi −Vj sin θj )+bij (Vi cos θi −Vj cos θj ) (28)
all transfer nodes to ensure that all nodes are connected. Equation (23) represents the binary nature of xij . The circuit
Proof: Consider a transfer node at bus j. Two cases must between buses i – j is connected if the corresponding value
be considered. is equal to one and is not connected if it is equal to zero.
Case 1 – The transfer node is not used. In this case, using In the proposed model of the DSR problem, transfer nodes
the main condition of a transfer node, no branches connect to and a single substation are considered. Thus PSi and QSi

ap
bus j, and we have: have non-zero values only at the substation node. Note that
xij = 0, ∀ (ij) ∈ Ωl , ∀ j ∈ Ωbp the existence of this unique substation node is represented by
xji = 0, ∀ (ji) ∈ Ωl , ∀ j ∈ Ωbp (24). The presence of transfer nodes in the DSR problem is
modeled by constraints (14)-(17) and (24), as shown in the
As (16) guarantees that yj = 0, that implies that bus j section II-C.
is disconnected. Then the number of nodes that remains
connected must be nb − 1, requiring nb − nbs − 1 branches,
B. Distribution system planning problem
which agrees with (13).

only one branch connect to bus j, then yj = 1 according to

(14) and (15), which means that the right member of (16)

this possibility unfeasible. Now, assuming that at least two

dP
Case 2 – The transfer node is used. First, considering that

equals 2; however the left member of (16) is one, which turns

branches connect to bus j, (14) and (15) force yj = 1, so (13)

turns into (8), satisfying the radiality conditions.

III. O PTIMIZATION P ROBLEM OF R ADIAL D ISTRIBUTION

S YSTEMS
In this section, the DSR and DSP problems are modeled
The DSP problem is modeled as follows:
min f = κl

s.t.
δl
X

X
(cij nij lij ) + κs

(ij)∈Ωl

δs
X

i∈Ωbs
(ij)∈Ωl

(gij (n0ij
X
(cfi mi ) +

+ nij )

(cvi (PS2i + Q2Si ))

PSi − PDi −
X
i∈Ωbs

(Vi2 +Vj2 −2Vi Vj

((n0ij + nij )Pij ) = 0

cos θij )) +

∀ i ∈ Ωb (30)
(29)
pte
j∈Ωb
as mixed binary nonlinear programming problems using the X
radiality constraints described in Section II. QSi − QDi − ((n0ij + nij )Qij ) = 0 ∀ i ∈ Ωb (31)
j∈Ωb

A. Distribution system reconfiguration problem V ≤ Vi ≤ V ∀ i ∈ Ωb (32)

0
The DSR problem is modeled as follows: PS2i + Q2Si ≤ (S i + mi S i )2 ∀ i ∈ Ωbs (33)
X 2
min v = (gij xij (Vi2 +Vj2 −2Vi Vj cos θij )) (18) 2
(n0ij + nij )(Ir2ij + Im ij
)≤ I ij ∀ (ij) ∈ Ωl (34)
(ij)∈Ωl n0ij+ nij ≤ 1 ∀ (ij) ∈ Ωl (35)
s.t.
X nij ∈ {0, 1} ∀ (ij) ∈ Ωl (36)
PSi − PDi − (xij Pij ) = 0 ∀ i ∈ Ωb (19)
mi ∈ {0, 1} ∀ i ∈ Ωbs (37)
ce

j∈Ωbi
X Eqs. (9) − (11)
QSi − QDi − (xij Qij ) = 0 ∀ i ∈ Ωb (20)
|kij | ≤ ndg nij ∀ (ij) ∈ Ωl (38)
j∈Ωbi X
V ≤ Vi ≤ V ∀ i ∈ Ωb (21) (n0ij + nij ) = nb − nbs (39)
2 (ij)∈Ωl
2
xij (Ir2ij + Im ij
)≤ I ij ∀ (ij) ∈ Ωl (22)
where δl =ατl φl cl and δs =ατs φs . The objective function (29)
xij ∈ {0, 1} ∀ (ij) ∈ Ωl (23)
is the total investment and operation cost based on [27].
Ac

Eqs. (14) − (17) The first part represents the investment cost (construction of
X X
xij = nb − 1 − (1 − yj ) (24) circuits and construction/reinforcement of substations); the
(ij)∈Ωl j∈Ωbp second and third part represent the annual cost of power
The objective function (18) represents the power losses of losses and substation operation respectively. Equations (30)
the distribution system operation. Equations (19) and (20) rep- and (31) represent the conventional equations of load balance,
resent the conventional equations of load balance; the elements and the elements of Pij and Qij are given by (25) and
of Pij and Qij are given by (25) and (26) respectively. (26) respectively. Equation (32) represents the constraints
on voltage magnitude of nodes, while (33) represents the
Pij = Vi2 gij − Vi Vj (gij cos θij + bij sin θij ) (25) maximum capacity of substation i. Note that, in (33), both
0
Qij = −Vi2 bij − Vi Vj (gij sin θij − bij cos θij ) (26) the reinforcement of the existing substation (S i 6= 0) and the
 6

0 Table I
construction of a new substation (S i =0) are modeled. The R ESULTS S UMMARY OF THE DSR P ROBLEM
elements of current flow in branch ij of (34) are given by (27)
and (28). Equation (35) ensures that duplication of circuits Systems 33-nodes 84-nodes 119-nodes 136-nodes 417-nodes
(existing and proposed) is not allowed. Equations (36) and Losses (kW) 139.55 469.88 853.61 280.19 685.88
(37) represent the binary nature of the circuits and substations

er
Times (s) 19 3030 4007 4473 14256
that can be added to the distribution system respectively. The Binary variables 37 96 133 156 474
element (circuit or substation) is constructed in the EDS if the
Table II
corresponding value is equal to one and is not constructed if O PEN C IRCUITS OF THE DSR P ROBLEM
it is equal to zero.
The presence of distributed generation in the DSP problem Systems
is modeled by constraints (9)-(11) and (38), as shown in the 33-nodes 6 − 7, 8 − 9, 13 − 14, 24 − 28, 31 − 32.

ap
section II-C. In the presented DSP model, consider all load 84-nodes 6 − 7, 11 − 43, 12 − 13, 14 − 18, 16 − 26, 28 − 32,
nodes to have non-zero power injection (no transfer nodes 33 − 34, 38 − 39, 61 − 62, 71 − 72.

present) and nbs substations. This condition is represented by 119-nodes 23 − 24, 25 − 26, 34 − 35, 39 − 40, 42 − 43, 43 − 54,
50 − 51, 58 − 59, 71 − 72, 74 − 75, 83 − 108, 86 − 105,
(39). Note that the existent circuits plus the added circuits 91 − 96, 97 − 98, 109 − 110.
must be equal to nb − nbs . 136-nodes 6 − 7, 9 − 24, 15 − 83, 31 − 35, 48 − 51, 50 − 96,
55 − 98, 66 − 79, 79 − 131, 84 − 135, 89 − 90, 90 − 129,
91 − 104, 92 − 104, 92 − 132, 95 − 96, 105 − 106,
C. Comments 108 − 114, 125 − 126, 128 − 77, 134 − 135.

dP
The binary variables nij and mi , of the DSP problem, and 417-nodes 2 − 6, 2 − 9, 19 − 32, 20 − 71, 45 − 49, 59 − 57, 74 − 80,
83 − 98, 48 − 38, 66 − 67, 24 − 34, 57 − 19, 23 − 28,
xij , of the DSR problem are decision variables, and a feasible 39 − 46, 76 − 77, 48 − 44, 67 − 383, 93 − 112, 56 − 146,
operation solution for the distribution system depends on their 65 − 251, 147 − 56, 220 − 219, 274 − 208, 262 − 263,
382 − 250, 279 − 280, 280 − 281, 234 − 235, 104 − 107,
values. The remaining variables represent the operating state of 106 − 102, 303 − 305, 313 − 317, 128 − 345, 146 − 141,
a feasible solution. For a feasible investment proposal, defined 159 − 180, 207 − 208, 374 − 201, 133 − 138, 175 − 178,
through specified values of nij,a , mi and xij , several feasible 172 − 178, 368 − 369, 322 − 321, 229 − 230, 295 − 294,
134 − 144, 137 − 154, 265 − 266, 287 − 299, 242 − 248,
operation states are possible. 313 − 308, 308 − 312, 316 − 318, 302 − 303, 310 − 303,
The proof that the solutions of problems (18)–(24) and (29)– 311 − 306, 259 − 323, 333 − 335, 380 − 267, 335 − 322.
(39) are radial solutions is a natural consequence of the proof
presented in Section II. Note that the power balance constraints
(19), (20), (30) and (31) guarantees that the solution is a The data of the real 417-node system were adapted to be
pte
connected one and that (1), (24) and (39) are equivalent used in the DSR problem. Table I presents a summary of the
constraints. obtained results as well as the total CPU time used by the
The presence of transfer nodes, nbs substations, distributed nonlinear B&B algorithm and the number of binary variables
generation and/or reactive power sources can be included in for each test system. For the 119-node system, the B&B
the DSP and DSR models shown above; thus a more general algorithm obtained a solution better than the one presented in
formulation of each optimization problem is obtained. [30]; while the results obtained for the 33-, 84- and 136-node
systems are equal to the presented in [10]. Table II shows the
IV. T ESTS AND R ESULTS open circuits in all systems. In the 84- and 136-node systems
all transfer nodes are used. The radial final topology of the
The objective of this section is to present an analysis of the 33-node system is shown in Fig. 2 and the final topologies
radiality constraints shown in section II. Thus, to exemplify obtained for the other four test systems are also verifiably
ce

the two optimization models previously presented, seven test radial. Furthermore, due to the use of constraints (14)-(17)
systems were used. For the DSR problem tests, the systems and (24) in the DSR problem, the presence of transfer nodes
of 33, 84, 119, 136 and 417 nodes were used, while the 23- does not result in loops in the systems.
and 54-node systems were used for the DSP problem tests. To show the efficiency of the proposed DSR model, a second
The DSR and DSP problems were solved using a nonlinear test is performed in the 33-node system. In this test, the
B&B algorithm based on [28] and coded in AMPL language, nodes 5, 14, 20 and 31 were replaced by transfer nodes. The
while the nonlinear programming (NLP) problems were solved radial final topology obtained by the DSR model is shown
Ac

through the commercial solver KNITRO [29]. The numerical in Fig. 3. Note that the 5, 14 and 20 nodes are used in the
results have been obtained using a PC Intelr CoreTM 2 Duo final topology, while the node 31 is not connected because it
6700, 2GB RAM. became a terminal node. The total losses for this test were
equal to 107.79 kW.
A. Distribution System Reconfiguration
The test systems data can be found in [7], [10], [27], [30] B. Distribution System Planning
and [31]. The 33-, 119- and 417-node systems are 12.66- 1) The 23-node distribution system: The data for this
kV, 11-kV and 10-kV systems respectively without transfer system are available in [14]. This is a 34.5-kV distribution
nodes, while the 84- and 136-node systems are 11.4-kV and system, supplied by a 10-MVA substation, which feeds an
13.8-kV systems with 17 and 28 transfer nodes respectively. oil production area with 22 load nodes, without distributed
 7

22 23 24 25 26 27 28 29 30 31 32

9
8
2

er
SE 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

18 19 20 21

Figure 2. Distribution system reconfiguration for the original 33-node system 4

6
(without transfer nodes) 16

ap
10 20
5 14
1 22
23
22 23 24 25 26 27 28 29 30 31 32 19
17
11
18
21
12
15 13
SE 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Figure 4. Distribution system expansion plan for the 23-node system

18 19 20 21

generation. For this test, the voltage threshold is 3%, the

average power factor is equal to 0.9, the cost of energy losses
is 0.05 US$/kWh, the loss factor equals 0.35, the interest rate
is 10% and the planning period extends to 20 years.
dP
Figure 3. Distribution system reconfiguration for the 33-node system with
four transfer nodes (nodes 5, 14, 20 and 31)
Circuit
cost
39452
Loss
cost
2923
Table IV
R ESULTS S UMMARY FOR THE 54- NODE SYSTEM (US$)

Substation
cost
540000

substations is 0.001 US$/kVAh2 and the planning period is 20

years.
Operation
cost
2640891
Total
cost
3223266
pte
The radial distribution system expansion plan is depicted in Table IV presents a summary of the obtained results; the
Fig. 4, and Table III shows a summary of the obtained results. solution is presented in Fig. 5. Four radial systems were built
In Fig. 1, the dashed lines represent the proposed routes, and (one for each substation), and two new substations (S3 and
continuous lines show the circuits constructed. In this case, S4) were necessary to feed the load nodes, while substation S1
the investment and operation costs of substations are equal to was expanded. Note that (39) guarantees that only 50 circuits
zero. Note that the constraint of (39) guarantees that only 22 are constructed, and (30) and (31) guarantee that all 50 load
circuits can be constructed and (30) and (31) guarantee that nodes are fed in four radial systems. Additionally, constraints
all 22 load nodes are fed by the existent substation (node 1) (9)-(11) and (38) prevents that the distributed generators can
in one radial system. The total CPU time is equal to 1706s. feed some loads independently. The total CPU time is equal
The topology obtained by the proposed methodology is equal to 6872s.
ce

to the one found by [14].

2) The 54-node distribution system: This is a 15-kV system V. C ONCLUSIONS
with 107.8-MVA to feed 50 load nodes. The objective of this
test is planning the distribution system considering four substa- This paper presented a literature review, a critical analysis
tions, two existing substations and two candidate substations and a proposal to incorporate the radiality constraints into
(see [13]). There are 17 existent circuits and 44 paths to new distribution system optimization problems more simply and
circuits (candidates). Three distributed generators are installed efficiently. It also presented a preliminary analysis of the
in the buses 10, 16 and 33. The maximum voltage deviation generalization of radiality constraints.
Ac

considered is 5%, the average power factor is equal to 0.9, the The DSR and DSP problems were modeled as mixed binary
cost of energy losses is 0.05 US$/kWh, the loss factor is equal nonlinear programming problems with radial constraints, and
to 0.35, the interest rate is 10%, the cost of operation of these the solution was obtained through the use of a nonlinear
branch and bound algorithm. The radiality constraints model
presented can be extended to other distribution system opti-
Table III
R ESULTS S UMMARY FOR THE 23- NODE SYSTEM (US$) mization problems.
The analysis and the assumptions made in this paper are
Circuit Losses Total important for the understanding and appropriate use of the
cost cost cost
radiality constraints, which can be included explicitly in the
151892 20227 172119 optimization model of radial distribution systems.
 8

20 17 S1
[12] I. J. Ramirez-Rosado and J. L. Bernal-Agustı́n, “Genetic algorithm
9 1 3
19 applied to the design of large power distribution systems,” IEEE Trans.
on Power Syst., vol. 13, no. 2, pp. 696-703, May 1998.
2 7 4
18 [13] V. Miranda, J. V. Ranito and L. M. Proença,“Genetic algorithm in
5
22 26 optimal multistage distribution network planning,” IEEE Trans. on
Power Syst., vol. 9, no. 4, pp. 1927-1933, Nov. 1994.

er
21 23 24 25 8 6
[14] J. M. Nahman and D. M. Peric, “Optimal planning of radial distribution
27
S4 32 networks by simulated annealing technique,” IEEE Trans. on Power
10 28 Syst., vol. 23, no. 2, pp. 790-795, May 2008.
33
DG
[15] P. C. Paiva, H. M. Khodr, J. A. Domı́ngues-Navarro, J. M. Yusta and
30 31 34 A. J. Urdaneta, “Integral planning of primary-secondary distribution
DG
38 35 36
29 37 39 S3 systems using mixed interger linear programming,” IEEE Trans. on
Power Syst., vol. 20, no. 2, pp. 1134-1143, May 2005.
44 15 40 41 [16] J. F. Gómez, H. M. Khodr, P. M. de Oliveira, L. Ocque, J. M. Yusta,
43

ap
45 16 R. Villasana and A. J. Urdaneta, “Ant colony system algorithm for the
13 42
DG
planning of primary distribution circuits,” IEEE Trans. on Power Syst.,
vol. 19, no. 2, pp. 996-1004, May 2004.
12 [17] E. Miguez, J. Cidras, E. Diaz-Dorado and J. L. Garcia-Dornelas,
11 14 46 47 “An improved branch-exchange algorithm for large-scale distribution
48 network planning,” IEEE Trans. on Power Syst., vol. 17, no. 4, pp. 931
50 - 936, Nov. 2002.
S2 49 [18] S. K. Goswami, “Distribution system planning using branch exchange
technique,” IEEE Trans. on Power Syst., vol. 12, no. 2, pp. 718-723,
Figure 5. Distribution system expansion plan for the 54-node system May 1997.
[19] K. Nara, T. Satoh, H. Kuwabara, K. Aoki, M. Kitagawa and T. Ishi-

VI. ACKNOWLEDGMENT
The authors would like to thank Prof. Ariovaldo V. Garcia
from the University of Campinas, São Paulo, Brazil for his
valuable comments and suggestions.

R EFERENCES
[1] H. P. Schmidt, N. Ida, N. Kagan and J. C. Guaraldo, “Fast reconfig-
uration of distribution systems considering loss minimization,” IEEE
Trans. on Power Syst., vol. 20, no. 3, pp. 1311-1319, Aug. 2005.
dP hara, “Distribution systems expansion planning by multi-stage branch
exchange,” IEEE Trans. on Power Syst., vol. 7, no. 1, pp. 208-214,
Feb. 1992.
[20] K. Aoki, K. Nara, T. Satoh, M. Kitagawa and K. Yamanaka, “New
approximate optimization method for distribution system planning,”
IEEE Trans. on Power Syst., vol. 5, no. 1, pp. 126-132, Feb. 1990.
[21] T. Gonen and I. J. Ramirez-Rosado, “Optimal multi-stage planning of
power distribution systems,” IEEE Trans. on Power Del., vol. 2, no. 2,
pp. 512-519, Apr. 1987.
[22] D. L. Wall, G. L. Thompson and J. E. D. Northcote-Green, “An
optimization model for planning radial distribution networks,” IEEE
Trans. on Power App. and Syst., vol. PAS-98, no. 3, pp. 1061-1068,
May 1979.
[23] S. K. Kathor and L. C. Leung, “Power distribution planning: A review
[2] E. R. Ramos, A. G. Exposito, J. R. Santos and F. L. Iborra, “Path-based
pte
of models and issues,” IEEE Trans. on Power Syst., vol. 12, no. 3, pp.
distribution network modeling: application to reconfiguration for loss 1151-1159, Aug. 1997.
reduction,” IEEE Trans. on Power Syst., vol. 20, no. 2, pp. 556-564, [24] M. S. Bazaraa, J. J. Jarvis and H. D. Sherali, Linear Programming and
May 2005. Network Flows. New York: John Wiley & Sons, 1990.
[3] S. Haffner, L. F. A. Pereira, L. A. Pereira and L. S. Barreto, “Multistage [25] R. Fourer, D. M. Gay and B. W. Kernighan, AMPL: A Modeling
model for distribution expansion planning with distributed generation- Language for Mathematical Programming, 2nd ed., Pacific Grove, CA:
part I: Problem formulation,” IEEE Trans. on Power Del., vol. 23, no. Brooks/Cole-Thomson Learning, 2003.
2, pp. 915-923, Apr. 2008. [26] CPLEX Optimization Subroutine Library Guide and Reference, Version
[4] F. V. Gomes, S. Carneiro, J. L. R. Pereira, M. P. Vinagre, P. A. N. Garcia 12.1.0. Incline Village, NV: CPLEX Division, ILOG, Inc., 2009.
and L. R. Araujo, “A new heuristic reconfiguration algorithm for large [27] J. L. Bernal-Agustı́n, Aplicación de Algoritmos Genéticos al Diseño
distribution systems,” IEEE Trans. on Power Syst., vol. 20, no. 3, pp. Optimo de Sistemas de Distrubuición de Energı́a Eléctrica, Ph.D.
1373-1378, Aug. 2005. dissertation, Universidad de Zaragoza, España, 1998.
[5] F. V. Gomes, S. Carneiro, J. L. R. Pereira, M. P. Vinagre, P. A. N. Garcia [28] C. R. Nugoli, M. J. Rider, E. Amorim and J. R. S. Mantovani, “Reactive
and L. R. Araujo, “A new distribution system reconfiguration approach Power Dispatch and Planning Using a Non-Linear Branch-and-Bound
ce

using optimum power flow and sensitivity analysis for loss reduction,” Algorithm,” IET Generation, Transmission & Distribution, vol. 4, no.
IEEE Trans. on Power Syst., vol. 21, no. 4, pp. 1616-1623, Nov. 2006. 8, pp. 963-973, Aug. 2010.
[6] D. Shirmohammadi and H. W. Hong, “Reconfiguration of electric [29] R. H. Byrd, J. Nocedal and R. A. Waltz, “KNITRO: An integrated
distribution networks for resistive line losses reduction,” IEEE Trans. package for nonlinear optimization,” in Large-Scale Nonlinear Opti-
on Power Del., vol. 4, no. 2, pp. 1492-1498, Apr. 1989. mization, New York: Springer Verlag, pp. 35-59, 2006.
[7] M. E. Baran and F. F. Wu, “Network reconfiguration indistribution [30] D. Zhang, Z. Fu and L. Zhang, “An improved tabu search algorithm
systems for loss reduction and load balancing,” IEEE Trans. on Power for loss minimum reconfiguration in large-scale distribution systems,”
Del., vol. 4, no. 2, pp. 1401-1407, Apr. 1989. Electrical Power & Energy Systems, vol. 77, no. 5-6, pp. 685-694, Apr.
[8] A. C. B. Delbem, A. C. de Carvalho and N. G. Bretas, “Main 2007.
chain representation for evolutionary algorithms applied to distribution
[31] J. P. Chiou, C. F. Chang and C. T. Su, “Variable scaling hybrid
Ac

system reconfiguration,” IEEE Trans. on Power Syst., vol. 20, no. 1,

differential evolution for solving network reconfiguration of distribution
pp. 425-436, Feb. 2005.
systems,” IEEE Trans. on Power Syst., vol. 20, no. 2, pp. 668-674, May.
[9] J. Mendoza, R. Lopez, D. Morales, E. Lopez, P. Dessante and
2005.
R. Moraga, “Minimal loss reconfiguration using genetic algorithms
with restricted population and addressed operators: real application,”
IEEE Trans. on Power Syst., vol. 21, no. 2, pp. 948-954, May 2006.
[10] E. M. Carreño, R. Romero and A. Padilha-Feltrin, “An efficient codifi- A PPENDIX
cation to solve distribution network reconfiguration for loss reduction
problem,” IEEE Trans. on Power Syst., vol. 23, no. 4, pp. 1542-1551, The following properties were obtained from [24].
Nov. 2008. Property 1: Let T be a proper tree graph with m(≥ 2)
[11] Y. Y. Hong and S. Y. Ho, “Determination of network configuration nodes, and let (i, j) ∈ T . Then disconnecting (i, j) from T ,
considering multiobjective in distribution systems using genetic algo-
rihms”, IEEE Trans. on Power Syst., vol. 20, no. 2, pp. 1062-1069, that is, removing the arc (i, j) from T but leaving the nodes
May 2005. i and j in T , decomposes T into two trees T1 and T2 .
 9

Property 2: A proper tree graph T has at least two end

nodes.
Property 3: A tree with m nodes has (m − 1) arcs. This
is clearly true for m = 1 or m = 2. By induction, assume
that this property holds for a tree with (m − 1) nodes and

er
consider a tree with m nodes, m ≥ 3. By Property 2, an end
node i exists. Disconnect the (unique) arc incident at the end
node and obtain two trees T1 and T2 (by Property 1), where
T1 = {i}. Hence T1 has zero arcs, and T2 has (m − 1) nodes.
By the induction hypothesis, it has (m − 2) arcs. Therefore T
has (m − 2) + 1 = (m − 1) arcs.

ap
Marina Lavorato (S’07) received the B.Sc. and M.Sc. degrees in 2002 and
2004 respectively, from the Federal University of Juiz de Fora, Brazil, and the
Ph.D. degree in 2010 from the University of Campinas, Brazil, all in electrical
engineering.

power systems.

dP
Currently she is carrying out postdoctorate research at the Universidade
Estadual Paulista, Ilha Solteira, Brazil. Her areas of research are the develop-
ment of methodologies for the optimization, planning and control of electrical

John F. Franco received the B.Sc. and M.Sc. degrees in 2004 and 2006 re-
spectively from the Universidad Tecnológica de Pereira, Risaralda, Colombia.
Currently, he is pursuing the Ph.D. degree in electrical engineering at the
Universidade Estadual Paulista, Ilha Solteira, Brazil. His area of research
pte
is the development of methodologies for the optimization and planning of
distribution systems.

Marcos J. Rider (S’97, M’06) received the B.Sc. (Hons.) and P.E. degrees
in 1999 and 2000 respectively from the National University of Engineer-
ing, Lima, Perú; the M.Sc. degree in 2002 from the Federal University
of Maranhão, Maranhão, Brazil; and the Ph.D. degree in 2006 from the
ce

University of Campinas, Brazil, all in electrical engineering. Currently he

is a professor in the Electrical Engineering Department at São Paulo State
University (FEIS-UNESP) in Ilha Solteira, Brazil.
His areas of research are the development of methodologies for the opti-
mization, planning and control of electrical power systems and applications
of artificial intelligence in power systems.
Ac

Rubén Romero (M’93, SM’08) received the B.Sc. and P.E. degrees in
1978 and 1984 respectively from the National University of Engineering,
in Lima, Perú, and his M.Sc. and Ph.D. degrees from the University of
Campinas, Brazil, in 1990 and 1993 respectively. Currently he is a professor
in the Electrical Engineering Department at São Paulo State University (FEIS-
UNESP) in Ilha Solteira, Brazil.
His general research interests lie in the area of electrical power systems
planning.

View publication stats