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

net/publication/256970618

A mixed-integer LP model for the optimal allocation of voltage regulators and

capacitors in radial distribution systems

Article in International Journal of Electrical Power & Energy Systems · June 2013
DOI: 10.1016/j.ijepes.2012.11.027

CITATIONS READS

95 205

4 authors:

John F. Franco Marcos J. Rider

São Paulo State University University of Campinas
125 PUBLICATIONS 3,674 CITATIONS 196 PUBLICATIONS 5,901 CITATIONS

SEE PROFILE SEE PROFILE

Marina Lavorato Rubén Romero

PUC Campinas São Paulo State University
20 PUBLICATIONS 599 CITATIONS 252 PUBLICATIONS 10,074 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by John F. Franco on 12 May 2020.

The user has requested enhancement of the downloaded file.

 A Mixed-Integer LP Model for the Optimal Allocation of Voltage
Regulators and Capacitors in Radial Distribution Systems

John F. Franco, Marcos J. Rider, Marina Lavorato and Rubén Romero

er
John F. Franco, Marcos J. Rider, Marina Lavorato and Rubén Romero are with the UNESP – Universidade

Estadual Paulista, Faculdade de Engenharia de Ilha Solteira, Departamento de Engenharia Elétrica, Ilha

ap
Solteira, São Paulo, Brazil.
Postal address: Prof. Dr. Marcos Julio Rider Flores

UNESP - Universidade Estadual Paulista, Faculdade de Engenharia de Ilha Solteira

Ilha Solteira - SP Brazil

emails: {johnfranco, mjrider, marina, ruben}@dee.feis.unesp.br

Abstract
dP
This paper presents a mixed-integer linear programming model to solve the problem of allocating voltage

regulators and fixed or switched capacitors (VRCs) in radial distribution systems. The use of a mixed-integer

linear model guarantees convergence to optimality using existing optimization software. In the proposed
pte
model, the steady-state operation of the radial distribution system is modeled through linear expressions.

The results of one test system and one real distribution system are presented in order to show the accuracy

as well as the efficiency of the proposed solution technique. An heuristic to obtain the Pareto front for the

multiobjective VRCs allocation problem is also presented.

Keywords: Distribution system optimization, optimal allocation of voltage regulator and capacitor, mixed-
ce

integer linear programming.

1 Introduction

The voltage regulation is an important function of an electrical distribution system (EDS) and is the ability
Ac

of to provide a voltage magnitude within standard ranges for a wide range of load conditions [1]. The voltage

magnitude ranges are imposed by the electricity regulatory agency of each country to guarantee a quality

service delivery to consumers. Thus the utilities are obliged to maintain an adequate voltage profile in EDS,

and this require investments in appropriate devices that must be economically viable. The most common way

1
to regulate the voltage magnitude in EDS is to install voltage regulators and capacitors, evaluating several

aspects like installation cost, equipment utilization rate, quality of service and loss minimization [2].

The installation of capacitors in an EDS is important for providing reactive power support, power factor

er
correction, voltage profile improvement and loss minimization [3]. Therefore, the location, the size and the

number of the equipments to be installed in an EDS must be identified in order to ensure quality service.

Much research about the capacitor allocation (CA) problem can be found in the literature, as shown in [4].

The CA problem is commonly modeled as a mixed integer nonlinear programming (MINLP) problem [5]–[8],

ap
where the objective function is usually to minimize investment costs plus the power losses costs in EDS [6],

[9]. Among the methodologies used to solve the CA problem can be found constructive heuristic algorithms

[5], [8], [10]; metaheuristics like the genetic algorithms [11], [12], tabu search [13], plant growth simulation

[14] and particle swarm [15]; and classical approaches were used to solve the CA problem like the branch and

dP
bound algorithm [16], where the CA problem is modeled like a mixed integer linear programming (MILP)

problem using current injection equations considering the constant current type load.

The installation of voltage regulators in an EDS is also important for controlling the voltage profile mainly

in large feeders and/or with large loads at the feeder’s end, where the greatest problems of voltage drop occur

[17]. Just as in the CA problem, the voltage regulators allocation (VRA) problem can also be modeled as

a MINLP problem, where the objective function is usually to minimize the investment costs plus the active
pte
losses costs in an EDS [17, 18]. In [19] the voltage magnitude deviation is also considered in the objective

function. Among the methodologies used to solve the VRA problem it can be found constructive heuristic

algorithms [17, 18] and metaheuristics like the genetic algorithms [19] where a sweep load flow is used to

calculate the steady-state operation point.

In most cases, the VRA problem is solved separately from the CA problem. For example, in [20]–

[22] the problem of location, sizing and control of capacitors and the problem of location and control of
ce

voltage regulators are considered but are solved separately in two decoupled problems. Only in [23]–[27]

methodologies are presented to jointly solve the problem of allocating voltage regulators and capacitors by

using genetic algorithms [23]–[26] and tabu search [27]. In [23]–[25] the problem of allocation, sizing and type

(fixed or switched) of capacitors and allocation and configuration of voltage regulators, considering different
Ac

load conditions, are modeled as a MINLP problem. In [26] the problem of allocation and coordination of

voltage regulators and capacitors considers the impact on the harmonic distortion of bus voltage in three-

phase electrical distribution system. The problem of allocation and coordination of voltage regulators and

capacitors for the voltage profile control considering the presence of distributed generation in an EDS is

presented in [27].

2
 Although the capacitors can contribute to improve the voltage profile, an EDS can reach a state in which

full reactive support exists and problems of voltage drop still occur [20]. In this cases, the allocation of voltage

regulators, which can provide a better control of the voltage regulation and reduce the operation cost, becomes

er
necessary. Therefore, the joint allocation of voltage regulators and capacitors has the advantage of assessing

which set of equipments is the most appropriate to be installed in an EDS.

In this paper, the voltage regulators and fixed or switched capacitors (VRCs) allocation problem in radial

distribution systems is modeled as a mixed integer linear programming problem. The proposed model allows

ap
the independent or joint solution of the VRCs allocation problems. Linearizations were made to adequately

represent the steady-state operation of the EDS considering the behavior of the constant power type load.

The integer nature of the decision variables represents the allocation, size and type of voltage regulators and

capacitors. The objective is to minimize the total investment and operation costs subject to operation and

dP
physical constraints. The proposed model was tested in systems of 69 and 136 nodes. In order to validate

the linearizations performed, the steady-state operation point was also calculated using a load flow sweep

method. The main contributions of this paper are as follows:

1. A novel model for the steady-state operation of a radial distribution system through the use of linear

expressions;
pte
2. A mixed integer linear programming model for the VRCs allocation problem that presents an efficient

computational behavior with conventional MILP solvers;

3. A heuristic to obtain the Pareto front for the VRCs allocation problem considering two different objec-

tive functions (total cost and maximum voltage deviation).

ce

2 The Problem of Allocating Voltage Regulators and Fixed or Switched

Capacitors in Radial Distribution Systems

2.1 Assumptions

In order to represent the steady-state operation of a radial EDS, the following assumptions are made:
Ac

1. The load is represented as constant real and reactive power;

2. In branch ij the node i is closer to the substation node than node j.

3. The real and reactive power losses on branch ij are concentrated in origin node i.

3
 4. The EDS is balanced and represented by a monophasic equivalent.

~i,d and I~ij,d are the phasors of the

The four considerations are shown in Fig. 1, where, for load level d, V
2
voltage at node i and the current flow on branch ij, respectively. For load level d, Rij Iij,d 2
and Xij Iij,d are

er
the real and reactive power losses of branch ij respectively.

Figure 1 goes here.

ap
2.2 Steady-State Operation of a Radial Distribution System

In Fig. 1, the voltage drop in a circuit ij is defined by (1).

~i,d − V
V ~j,d = I~ij,d (Rij + jXij ) ∀ij ∈ Ωl , ∀d ∈ Ωd (1)

The phasor I~ij,d can be calculated using (2).

I~ij,d = dP
Pij,d + jQij,d
~j,d
V

where the square current magnitude is calculated using (3).

!∗
∀ij ∈ Ωl , ∀d ∈ Ωd (2)
pte
2 + Q2
Pij,d
2 ij,d
Iij,d = 2 ∀ij ∈ Ωl , ∀d ∈ Ωd (3)
Vj,d

Equation (2) is then replaced in (1) to obtain (4).

~i,d − V
(V ~j,d )V
~ ∗ = (Pij,d − jQij,d )(Rij + jXij ) ∀ij ∈ Ωl , ∀d ∈ Ωd (4)
j,d
ce

~i,d = Vi,d ∠θi,d , V

Considering that V ~j,d = Vj,d ∠θj,d and θij,d = θi,d − θj,d , where θi,d is the phase angle at

bus i in the load level d, the Eq. (4) can be written as shown in (5).

2
Vi,d Vj,d [cos θij,d + j sin θij,d ] − Vj,d = (Pij,d − jQij,d )(Rij + jXij ) ∀ij ∈ Ωl , ∀d ∈ Ωd (5)
Ac

After identifying the real and imaginary parts Eq. (5), we obtain the following:

2 + (R P
Vi,d Vj,d cos θij,d = Vj,d ij ij,d + Xij Qij,d ) ∀ij ∈ Ωl , ∀d ∈ Ωd (6)

Vi,d Vj,d sin θij,d = (Xij Pij,d − Rij Qij,d ) ∀ij ∈ Ωl , ∀d ∈ Ωd (7)

4
 By summing the squares of (6) and (7) and considering (3), we obtain the following:

2 2 2 2
Vi,d − 2(Rij Pij,d + Xij Qij,d ) − Zij Iij,d − Vj,d = 0 ∀ij ∈ Ωl , ∀d ∈ Ωd (8)

er
In (8), the angular difference between voltages is eliminated; it is possible to obtain the voltage magnitude

of the final node (Vj,d ) in terms of the voltage magnitude of the initial node (Vi,d ), the real power flow (Pij,d ),

the reactive power flow (Qij,d ), the current magnitude (Iij,d ) and the electrical parameters of branch ij. The

ap
conventional equations of load balance are shown in (9) and (10) (see Fig. 1). Equations (3) and (8)–(10)

represent the steady-state operation and are frequently used in the load flow sweep method [6, 28] and optimal

load flow [5] of a radial distribution system.

X X
2 S D
Pki,d − (Pij,d + Rij Iij,d ) + Pi,d = Pi,d ∀i ∈ Ωb , ∀d ∈ Ωd (9)

2.3
X
ki∈Ωl

ki∈Ωl
Qki,d −
ij∈Ωl
X

ij∈Ωl

dP 2
(Qij,d + Xij Iij,d

The Model of Voltage Regulators and Capacitors

) + QSi,d = QD
i,d ∀i ∈ Ωb , ∀d ∈ Ωd

As described in [1], a voltage regulator is basically an autotransformer with an automatic changing mechanism

of the tap position (number of turns) of the series winding to maintain a predetermined level of the voltage
(10)
pte
magnitude along a distribution feeder in spite of load level variations. Standard voltage regulators contain

a reversing switch that enables a ±R% regulator range (+ increases and – decreases the voltage magnitude)

respecting the reference voltage magnitude and the maximum number of steps (2nt, which usually are 32

steps). Eqs. (11)–(14) show the mathematical model of the voltage regulator, considering it allocated is node

j of branch ij.
Vj,d = tij,d Vej,d ∀ij ∈ Ωl , ∀d ∈ Ωd (11)
ce

ntij,d
tij,d = 1 + R% ∀ij ∈ Ωl , ∀d ∈ Ωd (12)
nt
−nt ≤ ntij,d ≤ nt ∀ij ∈ Ωl , ∀d ∈ Ωd (13)

ntij,d integer ∀ij ∈ Ωl , ∀d ∈ Ωd (14)

Ac

where Vej,d is the non-regulated voltage magnitude (before the voltage regulator) as shown in Fig. 2. Note

that ntij,d has 2nt steps, and tij,d varies in the range of [(1−R%) , (1+R%)], allowing a regulation of ±R%.

Additionally, each voltage regulator type r has an annualized installation cost (or fixed cost, cvr
r ) and a

maximum current (Ivr ).

Figure 2 goes here.

5
 As shown in [5], the capacitors that are installed on an EDS can be fixed or switched. The fixed capacitors

are formed by one or more standard capacitor units and remain connected in all load levels. The switched

capacitors are formed by one or more standard capacitor units and can be partially or totally switched in

er
each load level. Therefore, for each group of capacitors installed in a bus, different operation states are

possible. Usually the standard capacitor units are treated as reactive power sources; this work presupposes

that each standard capacitor unit has a Qcp value with an annualized unit cost cun . Also, the fixed and

switched capacitors have an installation cost (or fixed cost, cf x ). Additionally, the switched capacitors have

ap
an annualized cost (csw ) associated with the switching equipment of the standard capacitor units.

2.4 MINLP Model of the VRCs Allocation Problem

The VRCs allocation problem can be modeled as a mixed integer nonlinear programming problem as follows:

subject to

X
min κc

ki∈Ωl
X
X
(cf x qi + csw qisw + cun ncp
i∈Ωb

ki∈Ωl

Qki,d −
Pki,d −
X

ij∈Ωl
X

ij∈Ωl
i

sqr
(Qij,d + Xij Iij,d
) + κ

dP
r

sqr
X X

(Pij,d + Rij Iij,d

cvr
r v

S
ij,r

) + Pi,d
+ τ l
X

ij∈Ωl r∈Ωr

D
= Pi,d

) + QSi,d + Qcp ni,d = QD

cls
d αd
X

d∈Ωd
sqr
Rij Iij,d
ij∈Ωl

∀i ∈ Ωb , ∀d ∈ Ωd

i,d ∀i ∈ Ωb , ∀d ∈ Ωd
(15)

(16)

(17)
pte
sqr 2 sqr sqr
Vi,d − 2(Rij Pij,d + Xij Qij,d ) − Zij Iij,d − Vej,d =0 ∀ij ∈ Ωl , ∀d ∈ Ωd (18)
sqr sqr
Vej,d 2
Iij,d = Pij,d + Q2ij,d ∀ij ∈ Ωl , ∀d ∈ Ωd (19)
sqr
Vj,d = tsqr e sqr
ij,d Vj,d ∀ij ∈ Ωl , ∀d ∈ Ωd (20)
 2
ntij,d ntij,d
tsqr
ij,d = 1 + 2R% + (R%)2 ∀ij ∈ Ωl , ∀d ∈ Ωd (21)
nt nt
X X
−nt vij,r ≤ ntij,d ≤ nt vij,r ∀ij ∈ Ωl , ∀d ∈ Ωd (22)
ce

r∈Ωr r∈Ωr
2 sqr 2
V ≤ ≤V Vi,d ∀i ∈ Ωb , ∀d ∈ Ωd (23)
sqr 2 X X 2
0 ≤ Iij,d ≤ I ij (1− vij,r )+ min{(Ivr )2 , I ij }vij,r ∀ij ∈ Ωl , ∀d ∈ Ωd (24)
r∈Ωr r∈Ωr
0≤ ni,d ≤ ncp
b qi ∀i ∈ Ωb , ∀d ∈ Ωd (25)
Ac

ni,d ≤ ncp
i ∀i ∈ Ωb , ∀d ∈ Ωd (26)

−ncp sw cp sw
b qi ≤ ni,d − ni,d−1 ≤ nb qi ∀i ∈ Ωb , ∀d ∈ Ωd | d > 1 (27)
X
vij,r ≤ 1 ∀ij ∈ Ωl (28)
r∈Ωr
X
qi ≤ ncp (29)
i∈Ωb

6
 X X
vij,r ≤ nvr (30)
ij∈Ωl r∈Ωr
ni,d integer ∀i ∈ Ωb , ∀d ∈ Ωd (31)

ntij,d integer ∀ij ∈ Ωl , ∀d ∈ Ωd (32)

er
qi ∈ {0, 1} ∀i ∈ Ωb (33)

qisw ∈ {0, 1} ∀i ∈ Ωb (34)

vij,r ∈ {0, 1} ∀ij ∈ Ωl , ∀r ∈ Ωr (35)

ap
The parameters κc and κr converts the present investment values into a stream of equal annual payments

over a specified time, at a specified discount rate or interest. The parameter τl is the interest rate of the cost
sqr sqr e sqr
of power losses. Note that in the MINLP model the variables Iij,d , Vi,d , Vj,d and tsqr
ij,d are used to represent
e 2 and t2 respectively. The objective function (15) stands for the annualized investment and
2 , V2 , V
Iij,d i,d j,d ij,d

dP
operation cost. The first and second part represent the investment cost (allocation of capacitors and voltage

regulators) and the third part represents the annual cost of power losses. Equations (16)–(19) represent the

steady-state operation and are a natural extension of (3) and (8)–(10), considering the presence of voltage
sqr
regulators and capacitors. Note that in (18) and (19), Vej,d sqr
appears instead of Vj,d , as shown in Fig. 2.

Equations (20)–(22) are an extension of (11)–(13), considering the regulation of square voltage magnitude

and the allocation of voltage regulators. Note that, if a voltage regulator is added to the system (vij,r = 1), (22)
pte
sqr sqr
is equal to (13), otherwise ntij,d = 0, tij,d = 1 and Vj,d = Vej,d . Equation (23) represents the constraints of

the square voltage magnitude of the nodes. Equation (24) represents the limit of the flows of current in branch

ij considering also the maximum current of the voltage regulator (Ivr ), which may which may be greater or

less than maximum current magnitude of branch ij (I ij ). Note that, if not installed any voltage regulator
sqr 2 sqr 2
in branch ij (vij,r = 0, ∀r ∈ Ωr ), (24) is equal to 0 ≤ Iij,d ≤ I ij , otherwise 0 ≤ Iij,d ≤ min{(Ivr )2 , I ij }. The
ce

maximum number of standard capacitor units that can be installed in a node of the system is represented by

(25). The number of standard capacitor units installed in the bus i is defined in (26). Equation (27) states

that the number of standard capacitor units of switched capacitors (qisw = 1) can be different in each load

level, while that number is equal in all load levels in the fixed capacitors (qisw = 0).
Ac

Eq. (28) assures that duplication of voltage regulators is not allowed. The maximum number of capacitors

and voltage regulators that can be installed in the systems is represented by (29) and (30). Equations (31) and

(32) represent the integer nature of the standard capacitor units of fixed or switched capacitors and steps of

voltage regulators respectively. Equations (33), (34) and (35) represent the binary nature of allocation of fixed

or switched capacitors and voltage regulators, respectively. An element is constructed if the corresponding

7
value is equal to one and is not constructed if it is equal to zero. The binary investment variables qi , qisw and

vij,r are the decision variables, and a feasible operation solution for the distribution system depends on their

value. The remaining variables represent the operating state of a feasible solution. For a feasible investment

er
proposal, defined through specified values of qi , qisw and vij,r , several feasible operation states are possible.

Note that (15)–(18) and (22)–(30) are linear, while (19), (20) and (21) contain square terms or the product

of two variables. With the aim of using a conventional MILP solver, it is desirable to obtain a linear equivalent

for (19), (20) and (21).

ap
2.5 Linearization
sqr
The left member of (19) is linearized by discretization of Vej,d using the binary variables xj,d,s ∀s = 1 . . . S.
sqr V
Where xj,d,s = 1 if Vej,d is greater than V 2 + s∆ , as is shown in Fig. 3. This condition is modeled in (36)

showing how the variables xj,d,s are calculated.

V2+
S
X

s=1
V

xj,d,s ≤ xj,d,s−1

xj,d,s ∈ {0, 1}
sqr
(xj,d,s ∆ ) ≤ Vej,d
V
≤V2+∆ +
S
X V
(xj,d,s ∆ )
dP s=1



∀j ∈ Ωb , ∀d ∈ Ωd (a) 




∀j ∈ Ωb , ∀d ∈ Ωd , s = 2 . . . S (b) 





∀j ∈ Ωb , ∀d ∈ Ωd , s = 1 . . . S (c) 
(36)
pte
Figure 3 goes here.

sqr sqr
The product Vej,d Iij,d is calculated using the middle point of the first interval of the discretization of

the square voltage magnitude multiplied by the square current flow magnitude, plus the successive power
c ) that depend on ∆ , I sqr and x V
corrections (Pj,d,s ij,d j,d,s , according to (37) and (38).

S
sqr sqr 1 V sqr X c
ce

e 2
Vj,d Iij,d = (V + ∆ )Iij,d + Pj,d,s ∀ij ∈ Ωl , ∀d ∈ Ωd (37)
2 s=1

V V 2 
0≤∆ sqr
Iij,d − c
Pj,d,s ≤∆ I ij (1 − xj,d,s) ∀ij ∈ Ωl , ∀d ∈ Ωd , s = 1 . . . S (a)
(38)
c V 2 
0 ≤ Pj,d,s ≤ ∆ I ij xj,d,s ∀ij ∈ Ωl , ∀d ∈ Ωd , s = 1 . . . S (b)
Ac

sqr sqr
Constraint (37) is a linear approximation of the product of Vej,d and Iij,d . Constraint (38) defines the
c , ∀ij ∈ Ω , ∀d ∈ Ω , s = 1 . . . S. If x c sqr 2
values of Pj,d,s l d j,d,s = 0, then Pj,d,s = 0 and Iij,d ≤ I ij ; otherwise

c V sqr c V 2 V 2
Pj,d,s = ∆ Iij,d and Pj,d,s ≤ ∆ I ij , where ∆ I ij plays exactly the role of the “Big M” factor and provides
c .
a sufficient degree of freedom to Pj,d,s

8
 The right member of (19) is linearized as described in [29] and defined in (39) and (40).

R
X R
X
2 + Q2
Pij,d ij,d = mSij,r ∆Pij,d,r + mSij,r ∆Q
ij,d,r ∀ij ∈ Ωl , ∀d ∈ Ωd (39)
r=1 r=1

er

+
Pij,d −
− Pij,d = Pij,d ∀ij ∈ Ωl , ∀d ∈ Ωd (a)  





Q+ −
ij,d − Qij,d = Qij,d ∀ij ∈ Ωl , ∀d ∈ Ωd (b)  



XR 


+
Pij,d −
+ Pij,d = ∆Pij,d,r ∀ij ∈ Ωl , ∀d ∈ Ωd (c)  


ap


r=1 

XR 



Q+ −
ij,d + Qij,d = ∆Q
ij,d,r ∀ij ∈ Ωl , ∀d ∈ Ωd (d)  



r=1 
S (40)
0≤ ∆Pij,d,r ≤ ∆ij ∀ij ∈ Ωl , ∀d ∈ Ωd , r = 1 . . . R (e) 



∀ij ∈ Ωl , ∀d ∈ Ωd , r = 1 . . . R (f) 
S
0 ≤ ∆Q
ij,d,r ≤ ∆ij






where
0 ≤ Q+
+
0 ≤ Pij,d
−
0 ≤ Pij,d

ij,d

0 ≤ Q−
ij,d
dP
mSij,r = (2r − 1)∆ij
S
∀ij ∈ Ωl , ∀d ∈ Ωd (g) 

∀ij ∈ Ωl , ∀d ∈ Ωd (h) 

∀ij ∈ Ωl , ∀d ∈ Ωd (i) 

∀ij ∈ Ωl , ∀d ∈ Ωd (j) 

∀ij ∈ Ωl , r = 1 . . . R


















pte
S V I ij
∆ij = ∀ij ∈ Ωl
R

Note that (39) and (40) are each a set of linear expressions of the right member of (19), and mSij,r
S PR S P
PR S Q
and ∆ij are constant parameters. The expressions r=1 mij,r ∆ij,d,r and r=1 mij,r ∆ij,d,r are the linear
2 and Q2 2 + −
approximations of Pij,d ij,d respectively. The linearization of Pij,d is shown in Fig. 4. Pij,d and Pij,d are

non-negative auxiliary variables to obtain |Pij,d | as shown in (40.a), whereas Q+ −

ij,d and Qij,d are non-negative
ce

auxiliary variables to obtain |Qij,d | as is shown in (40.b). Constraints (40.c) and (40.d) state that |Pij,d | and

|Qij,d | are equal to the sum of the values in each block of the discretization. Constraints (40.e) and (40.f) set

the upper and lower limits of the contribution of each block of |Pij,d | and |Qij,d |.
Ac

Figure 4 goes here.

sqr
Using the discretization of Vej,d shown in (36), the right member of (20) is linearized as shown in (41)

and (42).
X S
1 V
tsqr
ij,d
sqr
Vej,d = tsqr
ij,d (V 2
+ ∆ ) + c
Vj,d,s ∀ij ∈ Ωl , ∀d ∈ Ωd (41)
2
s=1

9
 
2 V
(1 − R%) ∆ (1 − xj,d,s ) ≤ tsqr
V c 2 V 

ij,d ∆ − Vj,d,s ≤ (1 + R%) ∆ (1 − xj,d,s ) 


∀ij ∈ Ωl , ∀d ∈ Ωd , s = 1 . . . S (a)  (42)


V V 
(1 − R%)2 ∆ xj,d,s ≤ Vj,d,s
c
≤ (1 + R%)2 ∆ xj,d,s ∀j ∈ Ωb , ∀d ∈ Ωd , s = 1 . . . S (b) 

er
As well as in (37), the product tsqr e sqr
ij,d Vj,d is calculated using the middle point of the first interval of

the discretization of the square voltage magnitude multiplied by the square voltage regulator tap plus the
c ) that depend upon ∆ , tsqr and x V
successive voltage corrections (Vj,d,s j,d,s . Constraint (41) is a linear

ap
ij,d
sqr
approximation of the product of Vej,d and tsqr c
ij,d . Constraint (42) defines the values of Vj,d,s , ∀j ∈ Ωb , ∀d ∈
V
c
Ωd , s = 1 . . . S. If xj,d,s = 0, then Vj,d,s = 0 and (1−R%)2 ≤ tsqr 2 c sqr
ij,d ≤ (1+R%) ; otherwise Vj,d,s = tij,d ∆ and
V V
(1−R%)2 ∆ ≤ Vj,d,s
c ≤ (1+R%)2 ∆ .

The square term in the right member of (21) varies in the range [0 – 1] and can be approximated by the

tsqr
ij,d = 1 + 2R%
dP
constant 13 . Thus (21) is linearized as shown in (43). This linearization has a correlation coefficient of 0.9993

with a maximum error of 0.82% and a mean error of 0.27%. The high value of the correlation coefficient

shows the high level of precision of the linearization.

ntij,d 1
nt
+ (R%)2 ∀ij ∈ Ωl , ∀d ∈ Ωd
3
(43)
pte
2.6 MILP model for the VRCs Allocation Problem

The VRCs allocation problem could be modeled as a mixed integer linear programming problem, as follows:

min (15)

subject to
ce

constraints (16)–(18), (22)–(36), (38), (40), (42) and (43).

S R R
1 V sqr X c X X
(V 2 + ∆ )Iij,d + Pj,d,s = mSij,r ∆Pij,d,r + mSij,r ∆Q
ij,d,r ∀ij ∈ Ωl , ∀d ∈ Ωd (44)
2 s=1 r=1 r=1
sqr sqr 2 X
|Vj,d − Vej,d | ≤ (V − V 2 ) vij,r ∀ij ∈ Ωl , ∀d ∈ Ωd (45)
Ac

r∈Ωr
S
X X
sqr 2 1 V 2
|Vj,d − tsqr
ij,d (V + ∆ ) − c
Vj,d,s | ≤ (V − V 2 )(1 − vij,r ) ∀ij ∈ Ωl , ∀d ∈ Ωd (46)
2
s=1 r∈Ωr

Constraints (36), (38), (40) and (44) replace constraint (19). Constraints (36), (42), (45) and (46) replace

constraint (20). Constraint (43) replaces constraint (21). Constraints (45) and (46) improve the accuracy

10
 P sqr sqr sqr
of the results. If r∈Ωrvij,r = 0, then Vj,d = Vej,d . Otherwise Vj,d is equal to the right member of (41).

Note that the number of operation variables has been increased with the linearization, while the number of

investment variables does not change and, as will be illustrated later in Section 3, this kind of optimization

er
problem can be solved with the help of standard commercial solvers.

Note that (16)–(18), (36), (38), (40) and (44) represent the steady-state operation of a radial distribution

system and are linear expressions. Considering the assumptions in Section 2.1, these expressions can be used

to analyze an EDS with distributed generators or to model other optimization problems of radial distribution

ap
systems through the use of linear expressions and solve it with the help of classical optimization techniques.

2.7 Multiobjective Approach for the VRCs Allocation Problem

The VRCs allocation problem can also be modeled as a multiobjective optimization problem with two different

∆V = 100
dP
objectives: one associated with the total cost, described by (15), and another one with the maximum voltage

deviation, according to [30], as shown in (47).

max
∀i∈Ωb ,∀d∈Ωd
{1 −
q
sqr
Vi,d / V }%

To minimize the total cost and to minimize the maximum voltage deviation are conflicting objectives
(47)

because investments made to improve the voltage profile cause an increase in the total cost. The Pareto front
pte
of the multiobjective VRCs allocation problem can be generated by successively solving the MILP model of

the VRCs allocation problem setting appropriately the V value, as described in the following steps:

1. Let it = 1 and define a value for the minimum voltage magnitude V it .

2. Solve the VRCs allocation problem defined by (15)–(18), (22)–(36), (38), (40), (42)–(46) to obtain the
ce

it-th Pareto front solution. If the problem is feasible, go to step 3. Otherwise exit.

3. Let it = it + 1, set V it = V it−1 + 0.0001pu and return to step 2.

The heuristic presented above allows to find the Pareto front because the maximum voltage deviation

is related to the minimum voltage magnitude V it , that is, a solution found at iteration it must satisfy the
Ac

minimum voltage constraint (23). As the minimum voltage magnitude V it is increased over certain value, the

solution of the VRCs allocation problem changes in order to meet (23), implying a reduction of the maximum

voltage deviation and a increase in the costs; that means that the solution is non-dominated in relation to

the maximum voltage deviation but dominated in relation to the costs when is compared with the previous

solution. The process finishes when it is not possible to find a feasible solution that meets the minimum

11
voltage magnitude V it . So, the proposed procedure found a set of non-dominated solutions which form the

Pareto front for the multiobjective optimization problem.

Knowing the Pareto front brings flexibility to the decision process and allows for better adaptation to the

er
policies of each electrical distribution company. Thus, a set of solutions is available that ranges between one

that minimizes total cost and another one that minimizes the maximum voltage deviation each satisfying the

operational constraints.

ap
3 Tests and Results

A test system of 69 nodes and a real system of 136 nodes were used to show the performance and robustness

of the proposed methodology. For all tests, the maximum voltage magnitude and the voltage magnitude of

the substation is 1.00 pu; the energy cost is 0.06 US$/kWh for all load levels; S is equal to 4, and R is equal

dP
to 20. The value of κc and κr is equal to 1.0, and τl is equal to 1.0. Further, cf x is equal to US$1000, cun is

equal to US$900 and Qcp is equal to 300kVAr. Finally, ncp

b is equal to 4. Two types of voltage regulators are

considered, with respective annualized costs of US$10000 and US$20000, and maximum current magnitude

of 200A and 400A. The value of nvr is equal to 2. The VRCs allocation problem model has been implemented

in AMPL [31] and solved with CPLEX [32] (called with default options) using a workstation with an Intel

XEON W3520 processor.

pte
3.1 69-node Distribution System

The 69-node distribution system data are available in [6]. It is a 12.66-kV distribution system that feeds 69

load nodes and supplies 4.66-MVA. The value of csw is equal to US$300, and ncp is equal to 6. Three load

levels are considered, obtained by multiplication of the loads by the factors 1.0, 0.8 and 0.5, with respective
ce

durations of 1000, 6760 and 1000 hours. The minimum voltage magnitude is equal to 0.95pu. For this

system, four tests were performed: 1) CA, 2) VRA, 3) VRCs allocation (VRCA) and 4) VRCA considering

the presence of a distributed generator (VRCA-DG) at bus 41 with a maximum active power generation of

1000kW working with a power factor of 0.95. Table 1 shows a summary of the obtained results, including
Ac

the initial states of the system without (IS) and with distributed generator (IS-DG).

Table 1 goes here.

Note that in all tests the minimum voltage magnitude of system in all load levels is higher than 0.95pu,

and the power losses in all load levels are lower than their initial states. In the CA test, one fixed capacitor

12
and two switched capacitors are allocated; in the VRA test, one voltage regulators of 200A are allocated; and

in the VRCA test one fixed capacitor and one voltage regulator of 200A are allocated. Note that the total

cost of the VRCA test (US$63344) is lower than the costs of the tests CA (US$66995) and VRA (US$80974),

er
showing that the proposed methodology evaluates the most appropriate set of equipment to be installed on

the EDS at minimum cost. In the IS-DG, the distributed generator is working at maximum capacity in

the first two load levels and generating 658.21kW in the third load level. In the VRCA-DG test, one fixed

capacitors and two switched capacitors are allocated, with a total cost of US$52674, and the distributed

ap
generator operates at maximum capacity in the first two load levels and generating 823.28kW in the third

load level.

The operation points for the four tests were compared using a load flow sweep method. The relative

errors for the power losses and the minimum voltage magnitude are indicated in parentheses in the Table 1,

dP
for all load levels. Note that the errors are small, showing the accuracy of the proposed model.

Using the heuristic presented in Section 2.7, the Pareto fronts for the tests CA and VRCA were found as

seen in Fig. 5, in which the solutions are represented by dots connected with lines to facilitate visualization.

Note that using only CA is possible to obtain a minimum ∆V until 4.87%, while that using VRCA is possible

to obtain a minimum ∆V until 2.86%. In the test VRCA, only it is necessary to allocate capacitors to obtain

a minimum ∆V until 5.25%.

pte
Figure 5 goes here.

The Pareto front allows selecting a solution according to the needs and policies of the electrical distribution

company. For example, if the limit for investments is US$64000, with the help of Fig. 4 can be determined

that the best solution that satisfies that limit and also the operational constraints has a minimum ∆V of

4.3%. On the other hand, if the electrical distribution company wants to reduce their minimum ∆V under a
ce

goal value of 5.0%, the Pareto front provides a solution with an investment cost of US$63344.

3.2 136-node Distribution System

The 136-node distribution system data are available in [7]. It is a 13.8-kV distribution system that feeds 107
Ac

load nodes and supplies 19.96-MVA. Three load levels are considered, obtained by multiplication of the loads

by the factors 1.8, 1.0 and 0.5, with respective durations of 1000, 6760 and 1000 hours. The value of csw is

equal to zero, and ncp is equal to 4. For this system four tests were performed: 1) CA with V equal to 0.85pu

(CA1), 2) CA with V equal to 0.90pu (CA2), 3) VRA with V equal to 0.90pu and 4) VRCA with V equal

to 0.90pu. Table 2 shows a summary of the obtained results including the initial state of the system.

13
 Table 2 goes here.

Note that in all tests the system’s minimum voltage magnitude in all load levels are higher than their

respective V , and the power losses in all load levels is lower than their initial states. In the CA1 test, one

er
fixed capacitor and two switched capacitors are allocated, with a total cost of US$192044. This solution is

better than the solution presented in [7] at a total cost of US$192074. In the VRA test, one voltage regulator

of 400A is allocated. In the CA2 test, one fixed capacitor and two switched capacitors are allocated. This is

ap
the same solution as found in the VRCA test because there is no need for voltage regulators to satisfy the

minimum voltage magnitude. Thus the proposed methodology found the equipments most appropriate to

be installed. The operation points for the four tests were compared using a load flow sweep method. The

relative errors for the power losses and the minimum voltage magnitude are indicated in parentheses in the

Table 2, for all load levels. As in the previous test, the errors are small, demonstrating the accuracy of the

proposed model.

4 Conclusions dP
A mixed-integer linear programming model to solve the VRCs allocation problem in radial distribution

systems was presented. The use of a MILP model guarantees convergence to optimality using conventional
pte
MILP solvers. The joint allocation of voltage regulators and capacitors evaluates the set of equipments most

appropriate to be installed in the EDS at minimum cost.

In the proposed MILP model, the steady-state operation of the radial distribution system is modeled

through the use of linear expressions. The results show that the power losses and voltage magnitudes are

calculated with great precision in comparison with the load flow sweep method.

The Pareto front for the VRCs allocation problem considering two different objective functions is easily
ce

found using a heuristic, making it possible to obtain the set of non-dominated solutions according to total

costs and maximum voltage deviation.

One test system and one real distribution system were used to test the proposed model. For the real

distribution system, the proposed methodology found a better solution when compared to the methodology
Ac

shown in [7].

14
References

[1] W. H. Kersting, Distribution System Modeling and Analysis. CRC Press, 2001, ch. 7: Regulation of

Voltages, pp. 145-198.

er
[2] T. Gönen, Electric Power Distribution Systems Engineering, New York: McGraw-Hill, 1986.

[3] S. F. Mekhamer, M. E. El-Hawary, S. A. Soliman, M. A. Moustafa and M. M. Mansour, “New heuristic

strategies for reactive power compensation of radial distribution feeders,” IEEE Trans. Power Del., vol.

ap
17, no. 4, pp. 1128-1135, Oct. 2002.

[4] H. N. Ng, M. M. A. Salama and A. Y. Chikhani, “Classification of capacitor allocation techniques”,

IEEE Trans. Power Del., vol. 15, no. 1, pp. 387-392, Jan. 2000.

[5] S. Segura, R. Romero and M. J. Rider, “Efficient heuristic algorithm used for optimal capacitor placement

dP
in distribution systems, ” Int. J. Electr. Power Energy Syst., vol. 32, no. 1, pp. 71-78, Jan. 2010.

[6] M. E. Baran and F. F. Wu, “Optimal capacitor placement on radial distribution systems,” IEEE Trans.

Power Del., vol. 4, no. 1, pp. 725-734, Jan. 1989.

[7] R. A. Gallego, A. J. Monticelli and R. Romero, “Optimal capacitor placement in radial distribution

networks,” IEEE Trans. Power Syst., vol. 16, no. 4, pp. 630-637, Nov. 2001.

[8] I. C. Silva Junior, S. Carneiro Junior, E. J. Oliveira, J. S. Costa, J. L. R. Pereira and P. A. N. Garcia, “A
pte
heuristic constructive algorithm for capacitor placement on distribution system,” IEEE Trans. Power

Syst., vol. 23, no. 4, pp. 1619-1626, Nov. 2008.

[9] J. Y. Park, J. M. Sohn and J. K. Park, “Optimal capacitor allocation in a distribution system considering

operation costs,” IEEE Trans. Power Syst., vol. 24, no. 1, pp. 462-468, Feb. 2009.

[10] M. R. Raju, K. V. S. R. Murthy and K. Ravindra, “Direct search algorithm for capacitive compensation
ce

in radial distribution systems, ” Int. J. Electr. Power Energy Syst., vol. 42, pp. 24-30, Nov. 2012.

[11] G. Boone and H. D. Chiang, “Optimal capacitor placement in distribution systems by genetic algorithm,”

Int. J. Electr. Power Energy Syst., vol. 15, no. 3, pp. 155-161, June 1993.

[12] G. Levitin, A. Kalyuzhny, A. Shenkman and M. Chertkov, “Optimal capacitor allocation in distribution
Ac

systems using a genetic algorithm and a fast energy loss computation technique,” IEEE Trans. Power

Del., vol. 15, no. 2, pp. 623-628, April. 2000.

[13] D. F. Pires, A. G. Martins and C. H. Antunes, “A multiobjective model for VAR planning in radial

distribution networks based on tabu search,” IEEE Trans. Power Syst., vol. 20, no. 2, pp. 1089-1094,

May. 2005.

15
[14] R. S. Rao, S. V. L. Narasimham and M. Ramalingaraju, “Optimal capacitor placement in a radial

distribution system using Plant Growth Simulation Algorithm, ” Int. J. Electr. Power Energy Syst., vol.

33, pp. 1133-1139, Jun. 2011.

er
[15] S. P. Singh and A. R. Rao, “Optimal allocation of capacitors in distribution systems using particle swarm

optimization, ” Int. J. Electr. Power Energy Syst., vol. 43, pp. 1267-1275, 2012.

[16] S. Haffner, L. A. Pereira, L. V. Gasperiny and L. Barreto, “Alocação de bancos de capacitores em redes

ap
de distribuição de energia visando eliminar violações de tensão,” SBA Controle & Automação, vol. 20,

no. 4, pp. 546-563, Oct. Nov. Dec. 2009.

[17] C. A. N. Pereira and C. A. Castro, “Optimal placement of voltage regulators in distribution systems,”

2009 IEEE Bucharest Power Tech Conference, pp. 1-5, Bucharest, Romania, 2009.

dP
[18] A. S. Safigianni and G. J. Salis, “Optimum voltage regulator placement in a radial power distribution

network,” IEEE Trans. Power Syst., vol. 15, no. 2, pp. 879-886, May 2000.

[19] J. E. Mendoza, D. A. Morales, R. A. López, E. A. López, J. C. Vannier and C. A. C. Coello, “Multi-

objective location of automatic voltage regulators in a radial distribution network using a micro genetic

algorithm,” IEEE Trans. Power Syst., vol. 22, no. 1, pp. 404-412, Feb. 2007.

[20] J. J. Grainger and S. Civanlar, “Volt/Var control on distribution systems with lateral branches using
pte
shunt capacitors and voltage regulators part i: the overall problem,” IEEE Trans. Power App. Syst.,

vol. PAS-104, no. 11, pp. 3278-3283, Nov. 1985.

[21] J. J. Grainger and S. Civanlar, “Volt/Var control on distribution systems with lateral branches using

shunt capacitors and voltage regulators part ii: the solution method,” IEEE Trans. Power App. Syst.,

vol. PAS-104, no. 11, pp. 3284-3290, Nov. 1985.

ce

[22] J. J. Grainger and S. Civanlar, “Volt/Var control on distribution systems with lateral branches using

shunt capacitors and voltage regulators part iii: the numerical results,” IEEE Trans. Power App. Syst.,

vol. PAS-104, no. 11, pp. 3291-3297, Nov. 1985.

[23] E. P. Madruga and L. N. Canha, “Allocation and integrated configuration of capacitor banks and voltage
Ac

regulators considering multi-objective variables in smart grid distribution system,” 9th International

Conference on Industry Applications, pp. 1-6, São Paulo, Brazil, Nov. 2010.

[24] B. A. de Souza, and A. M. F. de Almeida, “Multiobjective optimization and fuzzy logic applied to

planning of the volt/var problem in distributions systems” IEEE Trans. Power Syst.,vol. 25, no. 3, pp.

1274-1281, Aug. 2010.

16
[25] I. Szuvovivski, T. S. P. Fernandes and A. R. Aoki, “Simultaneous allocation of capacitors and voltage

regulators at distribution networks using genetic algorithms and optimal power flow, ” Int. J. Electr.

Power Energy Syst., vol. 40, pp. 62-69, 2012.

er
[26] G. Carpinelli, C. Noce, D. Proto and P. Varilone, “Voltage Regulators and Capacitor Placement in

Three-phase Distribution Systems with Non-linear and Unbalanced Loads,” International Journal of

Emerging Electric Power Systems , vol. 7, no. 4, pp. 1-17, Nov. 2006.

ap
[27] J. Sugimoto, R. Yokoyama, Y. Fukuyama, V. V. R. Silva and H. Sasaki, “Coordinated allocation and

control of voltage regulators based on reactive tabu search,” 2005 IEEE Rusian Power Tech Conference,

pp. 1-6, St. Petersburg, Russia, 2005.

[28] R. Cespedes, “New method for the analysis of distribution networks,” IEEE Trans. Power Del., vol. 5,

no. 1, pp. 391-396, Jan. 1990.

dP
[29] N. Alguacil, A. L. Motto and A. J. Conejo, “Transmission expansion planning: a mixed-integer LP

approach,” IEEE Trans. Power Syst., vol. 18, no. 3, pp. 1070-1077, Aug. 2003.

[30] B. Milosevic and M. Begovic, “Capacitor placement for conservative voltage reduction on distribution

feeders,” IEEE Trans. Power Del., vol.19, no.3, pp. 1360-1367, Jul. 2004.

[31] R. Fourer, D. M. Gay and B. W. Kernighan, AMPL: A modeling language for mathematical programming.
pte
CA: Brooks/Cole-Thomson Learning, Pacific Grove, 2nd ed., 2003.

[32] “CPLEX Optimization subroutine library guide and reference, version 11.0,” CPLEX Division, ILOG

Inc., Incline Village, NV, USA, 2008.

Appendix A
ce

The notation used throughout this paper is reproduced below for quick reference.

Sets:

Ωb sets of nodes.

Ωl sets of branches.
Ac

Ωd sets of load levels.

Ωr sets of voltage regulator types.

Constants:

κc capital recovery rate of capacitor constructions.

17
κr capital recovery rate of voltage regulator constructions.

cls
d energy cost in load level d (US$/kWh).

cvr
r annualized installation cost of the voltage regulator type r (US$).

er
cun annualized unit cost of each standard capacitor unit (US$).

csw annualized switch equipment cost of the switched capacitors (US$).

cf x annualized installation cost of the fixed and switched capacitors (US$).

ap
αd number of hours of load level d in one year (h).

τl interest rate of the cost of power losses.

ncp maximum number of fixed and switched capacitors that can be added in the system.

ncp
b maximum number of standard capacitor units that can be installed in a node of the system.

V minimum voltage magnitude (kV).

V maximum voltage magnitude (kV).

dP
nvr maximum number of voltage regulators that can be added in the system.

Ivr maximum current magnitude of voltage regulator type r (A).

I ij maximum current magnitude of branch ij (A).

pte
D real power demand at node i in load level d (kW).
Pi,d

QD
i,d reactive power demand at node i in load level d (kVAr).

Rij resistance of branch ij (Ω).

Xij reactance of branch ij (Ω).

ce

Zij impedance of branch ij (Ω).

R number of blocks of the piecewise linearization.

sqr
S number of discretizations of the variable Vei,d .

2nt maximum step number of the voltage regulator tap.

Ac

R% regulator range of voltage regulators.

Qcp reactive power of each standard capacitor unit (kVAr).

V sqr
∆ discretization step of Vei,d .

mSij,r slope of the rth block of the power flow of branch ij.

18
 S
∆ij upper bound of each block of the power flow of branch ij.
Variables:

qi binary variable for allocation of a fixed or switched capacitor at node i.

er
qisw binary variable for allocation of a switch equipment for the capacitor units at node i.

vij,r binary variable for allocation of a voltage regulator on branch ij of type r.

ntij,d integer step number of the tap of the voltage regulator on branch ij in load level d.

ap
ni,d integer number of standard capacitor units operating at node i in load level d.

ncp
i number of standard capacitor units installed at node i.

tij,d voltage regulator tap of branch ij in load level d.

tsqr
ij,d square of tij,d .

dP
Pij,d real power flow of branch ij in load level d.

Qij,d reactive power flow of branch ij in load level d.

S real power provided by substation at node i in load level d.

Pi,d

QSi,d reactive power provided by substation at node i in load level d.

Vi,d voltage magnitude at node i in load level d.

pte
sqr
Vi,d square of Vi,d .

Vei,d non-regulated voltage magnitude at node i in load level d.

sqr
Vei,d square of Vei,d .

Iij,d current flow magnitude of branch ij in load level d.

sqr
ce

Iij,d square of Iij,d .

sqr
xj,d,s binary variable used in the discretization of Vei,d .
sqr sqr
c
Pj,d,s power corrections used in the discretization of Vei,d Iij,d .

c
Vj,d,s voltage magnitude corrections used in the discretization of tsqr e sqr
ij,d Vi,d .
Ac

∆Pij,d,r value of the rth block of |Pij,d |.

∆Q
ij,d,r value of the rth block of |Qij,d |.

19
 List of Captions for Tables

Table 1 Results Summary of the 69-node System.

er
Table 2 Results Summary of the 136-node System.

Set of Tables

ap
Table 1: Results Summary of the 69-node System
Losses Investment Capacitors (kVAr) Voltage Regulators Tap Power Losses (kW) Min. Voltage Magnitude (pu)
Time
Test cost cost Load level Load level Load level Load level
Bus Branch (s)
(US$) (US$) 1 2 3 1 2 3 1 2 3 1 2 3
226.88 139.14 51.55 0.9092 0.9288 0.9567
IS 73140 − − − − − − − − − −
(+0.83%) (+0.16%) (-0.12%) (+0.00%) (+0.00%) (+0.00%)
47 600 600 600
212.66 99.29 37.68 0.9507 0.9508 0.9681
CA 55295 11700 50 1200 600 300 − − − − 6847
(-0.02%) (+0.33%) (+0.53%) (+0.00%) (+0.00%) (+0.00%)
53 900 300 0
VRA 70974

VRCA 49644

IS-DG 67213

VRCA-
DG
41874
10000

13700

10800
−

50

−
−

900

−
16 300 300
50 1200 900
54 900 300
−

900

−
−

900

−
300
600
0
45-46

44-45

−
10

−
dP
7

−
4

−
214.65
(-1.02%)
155.12
(+0.45%)
207.06
(-0.19%)
150.60
(+0.24%)
133.99
(-1.09%)
93.91
(-0.04%)
128.03
(+0.36%)
76.70
(+0.01%)

Table 2: Results Summary of the 136-node System

50.71
(-0.37%)
37.43
(+0.32%)
47.69
(+0.25%)
28.81
(-0.29%)
0.9616
(+0.54%)
0.9531
(-0.30%)
0.9142
(+0.00%)
0.9502
(+0.00%)
0.9707
(+0.54%)
0.9621
(+0.01%)
0.9336
(+0.00%)
0.9541
(+0.00%)
0.9834
(+0.20%)
0.9816
(-0.09%)
0.9598
(+0.00%)
0.9704
(+0.00%)
20

3570

11274
pte
Losses Investment Capacitors (kVAr) Voltage Regulators Tap Power Losses (kW) Min. Voltage Magnitude (pu)
Time
Test cost cost Load level Load level Load level Load level
Bus Branch (s)
(US$) (US$) 1 2 3 1 2 3 1 2 3 1 2 3
1117.96 320.36 77.00 0.8636 0.9307 0.9669
IS 201637 − − − − − − − − − −
(+0.36%) (-0.13%) (+1.14%) (-0.03%) (-0.01%) (+0.00%)
17 600 600 300
1017.16 287.53 69.86 0.8977 0.9607 0.9813
CA1 181844 10200 39 600 600 600 − − − − 4672
(+0.04%) (+0.16%) (+1.43%) (+0.01%) (+0.00%) (+0.00%)
155 1200 1200 600
17 600 600 300
1015.77 288.55 70.10 0.9016 0.9642 0.9830
CA2 182187 10200 39 600 600 600 − − − − 35391
(+0.00%) (-0.02%) (+0.51%) (+0.00%) (+0.00%) (+0.00%)
157 1200 1200 600
1087.32 316.65 76.79 0.9170 0.9563 0.9732
VRA 198279 20000 − − − − 150-154 8 4 1 28
(+0.23%) (+0.19%) (+0.04%) (+0.11%) (-0.30%) (+0.42%)
ce

17 600 600 300

1015.77 288.55 70.10 0.9016 0.9642 0.9830
VRCA 182187 10200 39 600 600 600 − − − − 65457
(+0.00%) (-0.02%) (+0.51%) (+0.00%) (+0.00%) (+0.00%)
157 1200 1200 600

List of Captions for Illustrations

Ac

Fig. 1 Illustrative example.

Fig. 2 Voltage regulator model.

Fig. 3 Discretization of the square voltage magnitude.

Fig. 4 Modeling the piecewise linear Pij2 function.

Fig. 5 Pareto front for the 69-node system.

20
 Set of Illustrations

~k,d
V ~i,d
V ~j,d
V
Pki,d , Qki,d , I~ki,d Pij,d , Qij,d , I~ij,d

er
(Rki , Xki , Zki , I ki ) (Rij , Xij , Zij , I ij )

S
k 2
Rki Iki,d + D i 2
Rij Iij,d + D
j
Pk,d + Pi,d + Pj,d +
2 2
jQSk,d jXki Iki,d jQD jXij Iij,d jQD
i,d j,d

ap
Figure 1: Illustrative example.

Vi,d Vej,d Vj,d

Pij,d , Qij,d , Iij,d tij,d

(Rij , Xij , Zij , I ij )

(Iv r )
D
Pi,d
D
jQi,d
+
i

dP
Figure 2: Voltage regulator model.

xj,d,1 = 1
xj,d,2 = 1
xj,d,s = 1
D
Pj,d
D
jQj,d
+

xj,d,S = 1
j
pte
V V V V
V2 V 2 +∆ V 2 +2∆ V 2 +s∆ V 2 +S∆ V
2

2
V V −V 2
∆ =
D+1

Figure 3: Discretization of the square voltage magnitude.

ce
Ac

21
 2
Pij,d r=R
S
R2 [∆ij ]2 = (V I ij )2

er
r=3
S
mSij,r = (2r−1)∆ij

r=2

ap
S
mSij,3 = 5∆ij
S
mSij,2 = 3∆ij
S
9[∆ij ]2
S r=1
mSij,1 = ∆ij
S
4[∆ij ]2
S
[∆ij ]2
0
∆P
ij,d,1 ∆P
ij,d,2

dP ∆P
ij,d,3

S
R∆ij = V I ij

Figure 4: Modeling the piecewise linear Pij2 function.

∆P
ij,d,r
|Pij,d |
pte
78000
76000
74000
VRCA
72000
Total cost (US$)

70000
CA
ce

68000
66000
64000
62000

60000
58000
Ac

56000
54000
3 4 5 6 7 8
Maximum voltage deviation, ∆V (%)

Figure 5: Pareto front for the 69-node system

22
View publication stats