Energy xxx (2015) 1e15

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

A multi-objective optimization of the active and reactive resource

scheduling at a distribution level in a smart grid context
Tiago Sousa a, *, 1, Hugo Morais b, Zita Vale a, Rui Castro c
a
nio Bernardino de Almeida, 431, 4200-072
GECAD e Knowledge Engineering and Decision Support Research Center e Polytechnic of Porto (IPP), R. Dr. Anto
Porto, Portugal
b
AUTomation and Control Group e Department of Electrical Engineering, Denmark Technical University (DTU), Elektrovej, Bld 326, 2800 Lyngby, Denmark
c
INESC-ID/IST Instituto Superior T

ecnico e University of Lisbon, Lisbon, Portugal

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

Article history: In the traditional paradigm, the large power plants supply the reactive power required at a transmission
Received 6 September 2014 level and the capacitors and transformer tap changer were also used at a distribution level. However, in a
Received in revised form near future will be necessary to schedule both active and reactive power at a distribution level, due to the
23 February 2015
high number of resources connected in distribution levels. This paper proposes a new multi-objective
Accepted 14 March 2015
Available online xxx
methodology to deal with the optimal resource scheduling considering the distributed generation,
electric vehicles and capacitor banks for the joint active and reactive power scheduling. The proposed
methodology considers the minimization of the cost (economic perspective) of all distributed resources,
Keywords:
Energy resource management
and the minimization of the voltage magnitude difference (technical perspective) in all buses. The Pareto
Distributed energy resources front is determined and a fuzzy-based mechanism is applied to present the best compromise solution.
Reactive power scheduling The proposed methodology has been tested in the 33-bus distribution network. The case study shows the
Virtual power player results of three different scenarios for the economic, technical, and multi-objective perspectives, and the
Voltage stability results demonstrated the importance of incorporating the reactive scheduling in the distribution
network using the multi-objective perspective to obtain the best compromise solution for the economic
and technical perspectives.
© 2015 Elsevier Ltd. All rights reserved.

1. Introduction smart grid. For this reason, some authors introduced the VPP (vir-
tual power player) concept as an aggregator of distributed energy
The introduction of DG (distributed generation) units, particu- resources connected to the electric network, mainly at the distri-
larly based on renewable sources, in the distribution networks has bution level [4]. In order, to operate in a complex and competitive
led to a significant change in the operation and planning of these environment, VPPs will need to develop new decision support
kind of networks [1]. In addition to the integration of DG units, systems for helping the management and control of the aggregator
there are other types of distributed energy resources, such as active resources. The complexity of this management is expected to in-
consumers with demand response programs, storage units and EVs crease geometrically with the transition from conventional vehicles
(electric vehicles) [2], that can cause an even more complex oper- to electric ones [5]. However, EVs can be useful in the scheduling as
ation of the distribution networks. flexible load and backup system of renewable sources [6].
Over these years, the smart grid has been presented as a good Reactive power scheduling is a task performed by network op-
concept to handle with this new power system paradigm and new erators to avoid the voltage instability, to maintain the voltage
uncertainties, considering the decentralization of the whole power levels within a proper range, and to minimize the power losses.
system management and control [3]. In this new paradigm, the Conventionally, centralized power plants were responsible to pro-
aggregation of distributed energy resources will be essential to vide a minimum amount of reactive power to maintain the stability
improve the management and control of these resources in the and quality of the system, and for this service the network operator
did not remunerate them. Furthermore, the distribution network
operator participated in the reactive scheduling through capacitors
* Corresponding author. Tel.: þ351 22 8340500; fax: þ351 22 8321159.
E-mail address: [email protected] (T. Sousa). and tap changer transformers in order to improve the voltage
1
WebSite: http://www.gecad.isep.ipp.pt. profile.

http://dx.doi.org/10.1016/j.energy.2015.03.077
0360-5442/© 2015 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077
2 T. Sousa et al. / Energy xxx (2015) 1e15

Nomenclature X Binary variable

Subscript
Parameters 1 Operation cost function
a Objective function F1 weight factor 2 Voltage magnitude difference function
b Objective function F2 weight factor Asyn_DG DG unit with asynchronous generator
l Penalization factor B Bus
hc Grid-to-Vehicle efficiency BatMax Battery energy capacity
hd Vehicle-to-Grid efficiency BatMin Minimum stored energy to be guaranteed at the end of
l Penalization factor used in the objective function F2 period t
m Membership function CAP Shunt capacitor
B Imaginary part in admittance matrix [S] Ch Charge process
cA Fixed component of cost function [m.u./h] Dch Discharge process
cB Linear component of cost function [m.u./kWh] Deg Battery degradation
cC Quadratic component of cost function [m.u./kWh2] DG Distributed generation unit
c Resource cost in period t [m.u./kWh] DGForecast Forecast power of distributed generation unit in
E Stored energy in the battery of vehicle at the end of period t
period t [kWh] EV Electric vehicle
EInitial Energy stored in the battery of vehicle at the beginning GCP Generation curtailment power
of period 1 [kWh] i, j Bus i and Bus j
ETrip Energy consumption in the battery during a trip that L Load
occurs in period t [kWh] Max Upper bound limit
G Real part in admittance matrix [S] Min Lower bound limit
M Total number of non-dominated solutions in the Pareto NSD Non-supplied demand
front o oth objective function
N Total number of resources obj Total number of objectives
NF Normalization factor REF Slack bus
Smax
Lk
Maximum apparent power flow in line k [kVA] SP External supplier
T Total number of periods Step Step of a shunt capacitor with discrete regulation
U Voltage in polar form [V] Stored Stored energy in the battery of the vehicle
y Series admittance of line that connects two buses [S] TFR_HV_MV Transformer that connects from high voltage to
ysh Shunt admittance of line that connects two buses [S] medium voltage
TFR_MV_LV Transformer that connects from medium voltage to
Variables low voltage
q Voltage angle
P Active power [kW] Superscript
Q Reactive power [kVAr] i Bus i
S Apparent power [kVA] s Non-dominated solution s
V Voltage magnitude [V]

The VPP will require the use of adequate optimization tech- one as the minimization of the voltage magnitude differences in all
niques for handling with the active and reactive power scheduling buses of the distribution network. The main goal is helping the
of distribution networks considering a specific objective [7]. Typi- VPP's management of a distribution network with high penetration
cally, this problem is formulated to minimize the operation cost of of several distributed energy resources, such as distributed gener-
the available distributed energy resources [8], or just considering ation units, electric vehicles, and capacitor banks.
the DG units [9], but in a competitive environment, as the smart The proposed methodology will obtain the Pareto front of the
grid, it is also important to consider other issues than just this envisaged optimal resource scheduling problem, which it will help
economic perspective. Therefore, the VPP needs to be aware about the decision maker to have a clear view of all non-dominated so-
the power quality, voltage stability, and active power losses. The lutions of the problem. After determining the Pareto front, a fuzzy-
optimal resource scheduling problem handled by the VPP requires based mechanism [10] is applied to obtain the best compromise
the incorporation of new strategies to deal with these issues. solution for the VPP.
Hence, the incorporation of reactive power control is essential to The results presented in the case study are expected to show
enable the VPP to present an optimal scheduling that considers all that is relevant to take into account the reactive power control in
the issues mentioned above. the optimal resource scheduling for the distribution network's
The optimization methodology proposed in this paper is based operation with the inclusion of the voltage magnitude difference
on a multi-objective approach to handle with day-ahead optimal function (technical perspective). In the case study, it has been used
resource scheduling of a VPP in a distribution network considering three scenarios with different reactive control strategies to evaluate
different reactive power management strategies. The proposed the impact of the reactive scheduling into the distribution network.
methodology will determine an optimal resource scheduling This paper is structured with the following sections: after the
considering two competing objective functions. One objective introductory section, Section 2 presents a literature review related
function is expressed as the minimization of the operation cost of to the reactive optimal resource scheduling. Section 3 focuses on
all distributed energy resources managed by the VPP, and the other the proposed methodology and describes the mathematical

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077
 T. Sousa et al. / Energy xxx (2015) 1e15 3

formulation. A case study is presented in Section 4, and the last did not enable the joint scheduling of the active and reactive power
section presents the conclusions. because the active power is fixed before the reactive scheduling.
Additionally, a particle swarm optimization technique is proposed
2. Related work in Ref. [24] for the reactive scheduling to minimize the active power
losses considering a fixed active power scheduling obtained before.
The optimal resource scheduling problem is an optimization This method did not take into account the joint active and reactive
problem with the purpose of obtaining the best scheduling of the scheduling to improve the voltage stability and active power losses
distributed energy resources managed by a certain player. In the while reducing the operation cost. A four-stage multi-objective
smart grid paradigm, this optimization problem can be undertaken approach is proposed in Ref. [25] to deal with the reactive sched-
by a diversity of players in a decentralized operation, including uling considering the reactive cost, active power losses and voltage
distribution system operator and VPPs [11]. In this decentralized stability, but the active power scheduling is not included in the
operation, each player might need to manage his own distributed multi-objective approach as another aspect to improve the voltage
resources and network area by a strictly economic objective. As a stability and to reduce the active power losses. In Ref. [26], a multi-
possible consequence, the electric network might be often operated objective approach is also proposed to solve a reactive scheduling
under stress and closer to its operating limits [12] resulting in considering renewable sources without including the joint active
voltage instability, congested lines, and reduced power quality. The reactive scheduling for the economic and technical perspectives of
voltage stability is important for the future distribution network the future distribution networks.
due to the large penetration of distributed energy resources [13]. As The contributions of this paper are:
mentioned before, a proper reactive power scheduling is important
to avoid the electric network's operation under a stress scenario. In 1. To propose a model to obtain the combined active and reactive
these circumstances, a single player should consider an optimal optimal resource scheduling of the distributed energy resources.
resource scheduling with technical concerns related to the reactive This aspect allows the joint optimization of both active and
power scheduling, instead of just considering the economic reactive power considering the global technical constraints and
perspective in the optimal resource scheduling. Moreover, a coor- not only the reactive power constraints, such as the methods
dinated reactive power scheduling of all involved resources can also proposed in Refs. [15,23,24]. This means an optimization process
lead to a better quality and efficiency of the smart grid operation, considering less approximation and, consequently, more real-
instead of considering the centralized perspective used in the istic model and constraints.
traditional paradigm [14]. 2. To propose a model with a multi-objective function combining
In order to prevent the voltage instability, the reactive power the minimization of the operation cost and the minimization of
management is divided into preventive measures and corrective the voltage magnitude difference, leading to an optimal sched-
measures [15]. In terms of preventive measures there is: the reac- uling with a proper balance between the economic and the
tive power scheduling, the active power rescheduling, and the load technical perspectives. In this sense, the methodology provides
shedding. In the reactive scheduling, external suppliers, distributed a set of solutions and not only one solution, such as the meth-
generation, and capacitor banks can have different cost as it is odologies in Refs. [20,25,26]. With this multi-objective
proposed in Ref. [16] which allows the operator to achieve the more approach, the VPP can decide based not only in the operation
profitable solution to dispatch the reactive power. Furthermore, a cost (economic perspective) and also regarding the technical
novel scheme price for the reactive markets is proposed in Ref. [17] implication of its decision.
which is based on a multi-objective approach. 3. To implement a new function of controlling the voltage profile,
Over the years, artificial intelligence techniques have been used through the minimization of the voltage magnitude difference
for solving the reactive power scheduling. In Ref. [18], it is proposed of all buses with respect to the voltage in the slack bus. The
a multi-objective approach based on hybrid fuzzy multi-objective reduction of voltage differences means less power flow in the
evolutionary algorithm to obtain the Pareto front of a multi- lines and consequently less power losses. On the other hand,
objective reactive power scheduling considering the total power decreasing the voltage magnitude difference means that the bus
cost related with reactive services, total power losses and voltage voltage will be far from the voltage operation boundaries and,
stability index. A particle swarm optimization technique has been subsequently a voltage slack change will have more impact in all
presented in Ref. [19] to solve the reactive dispatch of a wind farm. buses.
Shaw et al. [20] proposed a gravitational search algorithm for the 4. To integrate the management of distributed energy resources,
reactive power scheduling with the goal of minimizing the active such as DG units and EVs, in the improvement of the voltage
power losses. Another multi-objective approach is also proposed in stability of a distribution network. The participation of all types
Ref. [21] to deal with the reactive power scheduling problem. In of distributed resources in the ancillary services will be impor-
these methodologies, it is determined the optimal setting of the tant in a near future to assure a reliability of hierarchical power
control variables: reactive power of the generators, tap positions of systems. Thus, all hierarchical levels should assure a part of their
the tap changing transformers, and the reactive power of the own ancillary services requirements. In this sense, the inclusion
capacitor banks. In Ref. [22], a multiple-input-multiple-output of EVs and DG units in the reactive power control should be
servo controller is proposed for a synchronous generator that addressed.
keeps the active power at the desired level and adjusts the reactive
power of the generator to a certain desired level.
Beyond a substantial number of methodologies proposed in the 3. Optimal resource scheduling formulation
literature, the joint active and reactive optimal resource scheduling
is not considered by the several authors in this field of research. For The proposed methodology will obtain the active and reactive
instance, in Ref. [15] is presented a reactive power scheduling optimal resource scheduling for the next day of a VPP considering
formulation without considering the active power rescheduling two non-commensurable and competing objective functions. First
and load shedding contribution for the voltage stability of the objective is to obtain the minimum cost of all distributed energy
network. In Ref. [23], it is also proposed a method for the reactive resources, and the minimum voltage magnitude differences in all
scheduling problem to stabilize the voltage profile, but this method buses is the second objective. The proposed methodology will

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077
4 T. Sousa et al. / Energy xxx (2015) 1e15

schedule the active and reactive generation power of DG units, the "
X
T NB 
X  X
NL
active charge power and active discharge power of EVs, and the  
min F2 ¼ VðREF;tÞ  Vði;tÞ  þ l  cNSDðL;tÞ  PNSDðL;tÞ
reactive power of shunt capacitors.
t¼1 i¼1 L¼1
In this section, the mathematical formulation of the proposed !#
multi-objective methodology is described. This section is divided X
NDG
þ cGCPðDG;tÞ  PGCPðDG;tÞ
into five subsections. Subsection 3.1 shows the expression used to
DG¼1
determine the operation cost of the distributed energy resources.
(2)
Subsection 3.2 shows the expression applied to minimize the
voltage magnitude differences in all buses in the distribution
The objective is to approximate the voltage magnitude of all buses
network. Subsection 3.3 presents the multi-objective function used
to the voltage magnitude fixed in the slack bus. The increase of the
in the proposed methodology and Subsection 3.4 shows the related
reactive generation power of the available distributed energy re-
constraints. Subsection 3.5 describes the fuzzy-based mechanism
sources and shunt capacitors will enable the VPP to achieve the
used to determine the best compromise solution.
proposed minimization of the voltage magnitude differences. This
function will reduce the voltage magnitude differences and it will
directly minimize the active power losses in the distribution
3.1. Objective function e F1
network, enabling economic savings in the energy resource man-
agement of the VPP. The inclusion of non-supplied demand and
The minimization of the operation cost F1 used in the proposed
generation curtailment power variables in this function is impor-
methodology can be formulated as
tant to avoid the cut of power demand or to avoid the excess of
" ! generation from DG units with the only purpose of increasing the
X
T X
NDG cAðDG;tÞ XDGðDG;tÞ þcBðDG;tÞ PDGðDG;tÞ þ
minF1 ¼ 2
voltage magnitude of a particular bus. These variables are multi-
t¼1 DG¼1
cCðDG;tÞ PDGðDG;tÞ þcGCPðDG;tÞ PGCPðDG;tÞ plied by a penalization factor (l) to penalize the objective function.
NEV 
X 
þ cDchðEV;tÞ þcDegðEVÞ PDchðEV;tÞ cChðEV;tÞ
3.3. Multi-objective function e F
EV¼1
#
 XNSP X
NL
PChðEV;tÞ þ cSPðSP;tÞ PSPðSP;tÞ þ cNSDðL;tÞ PNSDðL;tÞ The proposed multi-objective function F is formulated as
SP¼1 L¼1
min F ¼ a  F1 þ b  F2  NF
(1) (3)
a þ b ¼ 1 ; a and b 2 ½0; 1
where the VPP's goal is to obtain a minimum F1 for the next day
where the two previous objective functions are considered. In this
while satisfying the consumers' demand and EV users' re-
paper, the weighted sum method is used to deal with the proposed
quirements. Concerning to EVs, the user must indicate to the VPP
multi-objective function F [30]. The weighted sum method (3) will
the amount of energy that allows him to travel with the EV for the
transform the multi-objective function F into a single function by
next day, and consequently, the VPP needs to respect to this user's
summing all objective functions, where each objective is multiplied
requirement. In order to achieve these demand requirements, the
by a different weight factor. The sum of the two weight factors must
VPP can use the DG units (controllable units) connected to its dis-
be equal to 1, and each weight factor is between 0 and 1. Several
tribution network that use a quadratic function (typically genera-
simulations with different weight factor values will be executed in
tors based on fossil fuels) [27]. The DG based on non-controllable
order to determine the Pareto front of the multi-objective problem.
renewable sources do not follow the quadratic polynomial function,
For levelling purposes, the function F2 uses an additional factor NF.
in which it is possible to establish a unitary energy price in the
linear cost (cB) for these renewables and the other coefficients are
zero. Additionally, the generation curtailment power (cGCP) is 3.4. Optimization problem constraints
included to prevent emergency situations of high generation form
non-controllable renewable sources turning the proposed meth- In a smart grid context, the VPP requires an adequate repre-
odology more robust to deal with the optimal resource scheduling sentation of the network with the purpose of obtaining a sched-
problems. Another way to achieve the VPP's requirements is uling result that is also feasible in the distribution network
through the use of the EVs battery for discharging that is commonly (avoiding lines congestion and voltage magnitude or angle viola-
referred to as V2G (vehicle-to-grid) [2]. It is assumed that the EVs tions). In this paper, the use of an AC power flow [27] is essential to
battery charging can be controlled by the VPP (i.e. smart charging). determine the active and reactive power that flows in each line of
In terms of EVs, the costs for the VPP consider the charge and the distribution network, the power losses and the voltage
discharge processes. Moreover, degradation cost associated with magnitude and angle. The AC power flow will be executed during
additional cycling [28] from the charge and discharge processes is the optimal resource scheduling problem, and the power injected
also formulated, as proposed in Ref. [29]. External suppliers located in each bus is obtained from the sum of the power flow through the
outside of the distribution network are other option for the VPP's lines that connect each bus with the rest of the network. The active
requirements for the next day with a respective cost for this energy and reactive power injected in each line depends on a non-linear
acquired (cSP). It is also considered a penalty cost for the VPP related combination between the admittance matrix, on the voltage
with the non-supplied demand (cNSD), preventing critical situations magnitude, and on voltage angle [27]. The equations that compose
with high consumer demands. the problem constraints are exposed in Appendix A.
Eq. (4) establishes an equality constraint between the active
power injected in bus i and the active power generation minus the
3.2. Objective function e F2 active power demand in the same bus. The active power generation
is the sum of the active generation power of DG units of the active
The objective function F2 minimizes the voltage magnitude generation power of external suppliers, and of the active discharge
differences of all buses in the distribution network and is given by power from EVs. On the other hand, the active power demand is

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077
 T. Sousa et al. / Energy xxx (2015) 1e15 5

determined through the sum of the consumers' active load con- several steps (XCAP(CAP,Step,t)) with different amounts of reactive
sumption and the EVs battery charging. Eq. (5) is a reactive power power for each step (QMax(CAP,Step,t)) (Eq. (19)), and the VPP can
equality constraint considering the reactive power injected and choose the adequate step. In this paper, two types of use for the
consumed in the bus i. The reactive power generation can be ob- capacitors in the distribution network are considered. The first one
tained from reactive generation power of DG units, reactive gen- is associated with the asynchronous generator in the DG unit to
eration power of external suppliers, and reactive power from enable the generation of reactive power by that DG unit. The second
capacitor banks. The reactive power demand results from the is the VPP's possibility of having its own capacitor installed in
consumers' reactive demand and reactive consumption corre- strategic buses in order to increase the voltage level.
sponding to the asynchronous generators used by some DG units. The VPP can also establish “take-or-pay” contracts with DG
The asynchronous generators consume reactive power and do not units, mainly based on renewable sources. Therefore, the VPP is
have the capability of generating reactive power [31]. For this forced to dispatch all the active generation power by the DG units
reason, the use of a capacitor with asynchronous generators is with this type of contract (Eq. (20)). Eqs (21) and (22) represent the
essential to suppress the reactive power consumption of these maximum active and reactive power of each external supplier.
generators and to generate reactive power in order to help the In terms of electric vehicle constraints, Eq. (23) expresses the
VPP's reactive power scheduling. amount of energy stored at the end of period t for each EV. In this
The voltage magnitude and angle are important variables to constraint, it is necessary to consider the typical daily travel profile
determine the power injected in each bus, and these variables must of each EV user, as the VPP is forced to guarantee the energy
have maximum and minimum limit. Eqs. (6) and (7) will keep the required for the EV user to travel in the time horizon of the optimal
voltage magnitude and angle between their own limits. A slack bus resource scheduling. This trip consumption can be sent by the EV
is previously selected in the network, and the fixed voltage owner [32] or it can be used a forecast tool [33] to predict the EVs
magnitude and angle are specified for it. owners behaviour. The proposed methodology takes into account
The power flow must be below a maximum power limit in each the behaviour of the users, driving pattern, travel distance, driving
line (line thermal limit). Eqs. (8) and (9) define the inequality route, vehicles’ internal characteristics and batteries performance
constraint to maintain the power flow from bus i to bus j and vice as influences of the trip consumption in the EVs [34]. The charging
versa below the maximum power limit of each line, respectively. (hc) and discharging (hd) efficiency battery are also used in Eq. (23).
The term U represents the voltage in polar form. The efficiency to charge and discharge considers the efficiency of
As mentioned before, the VPP's distribution network is con- the equipment used to plug the EV to the network (ac/dc converters
nected to upstream networks through transformers that change and so on) and the round trip efficiency.
the voltage level from high voltage (HV) to medium voltage (MV). The energy stored on the battery requires a maximum and min-
From these connections comes the energy from the external sup- imum limit of energy in all optimization periods. Eq. (24) defines a
pliers that are located outside of the network. The energy minimum energy stored in the battery for each period of the opti-
exchanged between the VPP and the externals suppliers is limited mization process. The EV users should inform the VPP which is the
by the maximum capacity of the transformer HV/MV that connects minimum amount of energy stored in the battery and in which
the distribution network with the upstream network. Eq. (10) es- period that energy must be guaranteed. The VPP and the EV users
tablishes that the power from external suppliers must be lower or need to use an adequate communication system to exchange infor-
equal to the maximum capacity of the HV/MV power transformer. mation about the minimum energy stored in the battery [35]. Eq. (25)
The same happens with the buses in the distribution network. maintains the energy stored on the EV battery lower than its capacity.
For each bus i, there is a transformer from MV to low voltage (LV) The charge/discharge rates have their own maximum limit (26)
that connects small resources, such as photovoltaic units and EVs, and (27), respectively. In this formulation, it was considered that
to the MV side of the respective bus i. Eqs. (11) and (12) determine the charge/discharge rate limit can change depending on the
the active and reactive power that flows in the transformer MV/LV different parts of the network that the EV user is connected. Each
of bus i. Eq. (13) establishes that the power from distribution EV can be connected in a single phase (e.g. at home), with the rate
network to the LV side must be lower or equal to the maximum limit being lower than when EVs are connected in three phases (e.g.
capacity of the respective MV/LV power transformer in bus i. a parking lot at the work). In this formulation, two binary variables
In this paper, the apparent generation power of the DG units for each EV are used to control the charge/discharge power,
must be kept within a minimum limit (14) and a maximum limit avoiding that the two operations occur in the same period (28).
(15). The VPP can only choose to use the active generation power The mathematical formulation of the proposed multi-objective
(PDG) to minimize the function F1 (1), or it can just use the reactive methodology has been implemented in mixed-integer non-linear
generation power (QDG) to minimize the function F2 (2), or it can programming (MINLP) approach (deterministic technique) using
also combine the active and reactive generation power until it the software GAMS (general algebraic modeling system) [36].
reaches the maximum limit (15). Eqs. (14) and (15) are only applied
for DG units that use a synchronous generator, because this 3.5. Best compromise solution
generator is capable of generating active and/or reactive power. The
binary variable (XDG) is included in both equations with the object After determining the Pareto front, a fuzzy set theory has been
of deciding whether the DG unit is turning on or turning off. used to choose the best candidate solution for the VPP [10]. The
On the other hand, Eqs. (16) And (17) define the maximum and membership function mo related with the oth objective function of
minimum limit for the active generation power of DG units with each solution in the Pareto set Fo is defined as
asynchronous generator, respectively. As referred before, the
asynchronous generator consumes reactive power, instead of gen- 8
erates reactive power and the parameter QAsyn_DG is not considered >
> 1 ; Fo  Fomin
>
>
in these expressions. >
< F max  F
o
The capacitors used in the proposed methodology can have a mo ¼ o
; Fomax < Fo < Fomin (29)
> F max  F min
>
continuous or discrete regulation. In the continuous regulation, the >
> o o
>
:
capacitor can have a reactive power between zero and a maximum 0 ; Fo  Fomax
limit (Eq. (18)). In the discrete regulation, the capacitor can have

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077
6 T. Sousa et al. / Energy xxx (2015) 1e15

where Fomax and Fomin are the maximum and minimum values of the 4.3 and 4.4 is presented. Subsection 4.6 presents the results of the
oth objective function, respectively. For each non-dominated so- proposed methodology considering the multi-objective F.
lution s, the normalized membership function ms is calculated as
4.1. Case study's characterization
PNobj s
o¼1 mo
ms ¼ P PNobj (30) A 33-bus distribution network [39] is used to evaluate the per-
M
s¼1 o¼1 mso formance of the proposed methodology. Fig. 1 depicts the 33-bus
distribution network with a projection of DG and EVs for the year
2040 [39]. Bus 0 connects the distribution network to the rest of the
where M and Nobj are the total number of non-dominated solutions electric network, enabling the VPP to buy and/or sell energy to the
and the total number of objective functions, respectively. The best rest of the network (external suppliers). In this figure, each DG unit
compromise solution is the one with the maximum ms value. For the has a different colour, which depends on the primary energy source
decision maker, it could be interesting to arrange all Pareto solu- used in each unit. The VPP will manage this 33-bus distribution
tions in descending order according to their membership function network with: 66 DG units, 1000 EVs connected to the network,
and having a priority list of Pareto solutions. This will give a better and the possibility to negotiate with 10 external suppliers. It is
view of the best solutions for the proposed multi-objective problem necessary to satisfy the consumption of 218 consumers spread over
for the next day [10]. Readers who are interested can find more the 32 buses [39]. The case study considers a specific consumption
details of this method in Ref. [37] and fuzzy set theory in Ref. [38]. scenario, however the VPP can use the proposed methodology to
obtain the optimal resource scheduling for a different set of con-
4. Case study sumption scenarios.
In terms of EVs' projection, a simulator tool proposed in Ref. [33]
The results of the proposed methodology to solve the multi- was used to generate the daily travelling profiles for the 1000 EVs.
objective optimal resource scheduling problem will be presented The simulator tool considers 7 EV models: Renault Zoe Z.E., Renault
in this section. The proposed methodology will be applied to a VPP Kangoo Z.E., Renault Fluence Z. E., Nissan Leaf, Toyota Prius, Citroen
that manages distributed energy resources and a distribution C-Zero and Mitsubishi i-MiEV. The technical information of all EV
network. The VPP can handle DG units, external suppliers, capacitor models used in this case study is presented in Table 1. The charge
banks and EVs. The charge and discharge processes of the EVs are rate considers the slow and fast charging modes, and the discharge
controlled by the VPP, in terms of amount of power, time and rate only considers the slow mode in order to prevent more damage
duration. The discharge capability of an EV is called V2G option. of the batteries’ cycle life. The case study uses a charging (hc) and
This section is divided into six subsections. Subsection 4.1 pre- discharging (hd) efficiency battery of 76.5%, respectively.
sents the information and input data used in the case study of this The price and installed capacity of each available resource are
paper. Subsection 4.2 explains the three scenarios adopted in the presented in Table 2. For this case study, the photovoltaic DG units
present case study. Subsection 4.3 shows the results of the simula- have a “take-or-pay” contract with the VPP, which establishes the
tion considering the objective function F1. Subsection 4.4 shows the duty by the aggregated player to dispatch all the active power
results of the simulation considering the objective function F2. In generated by the photovoltaic generators. For the DG units, the
Subsection 4.5 a comparison of the results obtained in Subsections linear cost coefficient (cB) of objective function F1 (Eq. (1)) is the

Fig. 1. 33-bus distribution network and resources projection in 2040 (based on [39]).

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077
 T. Sousa et al. / Energy xxx (2015) 1e15 7

Table 1 Table 3
Electric vehicle technical information. Shunt capacitor information.

EV model Battery Charge rate Discharge rate Resource Bus Type Number Maximum
(kW) (kW) regulation of steps reactive
Capacity Range
power (kVAr)
(kWh) (km)
Wind 7, 14, 20, 24 Discrete 1 50
MiEV 16.0 160 3 3
and 30
C-Zero 16.0 150 2.2 2.2
Small hydro 2 and 18 40
Fluence Z.E. 22.0 185 3 3
VPP 5 Continuous e 50
Leaf 24.0 160 6.6 6.6
17, 21, 24 and 32 30
Kangoo Z.E. 22.0 170 3 3
Zoe 22.0 150 3 3
Prius 4.4 20 3 3
In the first scenario, the VPP can only use the reactive power
from external suppliers located outside the 33-bus distribution
network. In this scenario, the DG units only generate active power
only one used in this case study. In terms of EVs information, the EV
and the capacitor banks are not considered.
charge price is considered equal to zero, because the objective is to In the second scenario, the VPP established a minimum amount
minimize the operation cost, and to minimize the voltage magni-
of reactive generation power by the DG units in the peak periods. In
tude difference, and for this reason, the consumer consumption and these periods, the reactive power consumption is usually high and
EV charge are assumed as a “mandatory” service provided by the
the DG units must contribute to reduce the impact of this reactive
VPP in the proposed methodology. The discharge price for the EVs consumption. DG units must guarantee an amount of reactive po-
was established in 0.08 m.u./kWh and the degradation cost of the
wer equal or higher than 40% of the active power generated. The DG
battery is equal to 0.042 m.u./kWh, as can be seen in Ref. [29]. units with asynchronous generators will have capacitor banks to
Table 3 shows the shunt capacitors used in the case study. In the
inject the required amount of reactive power in the peak periods. In
table, the type of resource that manages the shunt capacitor is the off-peak periods, the DG units will only generate active power.
indicated. The DG units based on wind and hydro require the use of
The VPP can also use the external suppliers to supply reactive po-
capacitors, because they have asynchronous generators to supply wer to the distribution network.
active power. The VPP also has its own capacitor banks placed in
In the third scenario, the DG units will generate active and
buses 5, 17, 21, 24 and 32 of the distribution network. The type of
reactive power without the reactive requirement in the peak pe-
regulation and the number of steps are also indicated in this table.
riods that is used in the second scenario. In this scenario, the VPP
The capacitors used by the DG units have discrete regulation, and
can select properly the amount of active and reactive generation
the VPP uses continuous regulation in the capacitors. The last col-
power for each DG unit without violating the constraints (14)e(19).
umn of Table 3 indicates the maximum reactive power injected by
The DG units with asynchronous generators will also have capacitor
each capacitor bank.
The simulations of the proposed methodology were executed in
a computer with two processors Intel® Xeon® W3520 2.67 GHz,
each one with two cores, 3 GB of RAM (random-access-memory)
and Windows 7 Professional 64 bits operating system.

4.2. Case study's scenarios

The case study has been divided into three different scenarios,
considering different reactive control perspectives by the VPP. The
objective is to evaluate the impact of the different reactive control
perspectives in the optimal resource scheduling. Fig. 1 presents four
areas in the 33-bus distribution network regarding the voltage
profile. These areas will contribute to a better evaluation of the
different reactive control scenarios established in this case study.

Table 2
Distributed energy resources information.

Resource Units # Total installed Linear cost -cB

capacity (kW) (m.u./kWh)

Max. Mean Min.

Photovoltaic 32 1320 0.254 0.187 0.11

Wind 5 505 0.136 0.091 0.06
Small hydro 2 80 0.145 0.117 0.089
CHP 15 725 0.105 0.075 0.057
Biomass 3 350 0.226 0.201 0.186
WTE 1 10 0.056 0.056 0.056
Fuel cell 8 440 0.2 0.055 0.01
External suppliers 10 5800 0.15 0.105 0.06
EVs charge 1000 e 0
EVs discharge 0.08
Fig. 2. Scheduled energy for the three scenarios with objective function F1: a) Active
EVs battery degradation cost 0.042
energy; b) Reactive energy.

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077
8 T. Sousa et al. / Energy xxx (2015) 1e15

Table 4 determine the optimal resource scheduling minimizing the oper-

Performance of the proposed methodology with objective function F1. ation cost (objective function F1). In this table, the voltage magni-
Scenario Operation Voltage Active power tude difference and the active power losses are also presented.
cost e F1 (m.u.) difference e F2 (pu.) losses (kW) Scenario 3 presents the best result in terms of operation cost,
1 6963.14 14.77 823.10 voltage difference, and active power losses. On the other hand,
2 6948.89 13.63 684.29 scenario 1 presents the worse results in these three indicators. The
3 6898.10 12.44 603.05 operation cost's variation between the best scenario (scenario 3)
and scenarios 1 and 2 is around 0.94%, 0.74%, respectively. The
voltage magnitude's variation is of approximately 18.92%, 9.74%
banks to inject the reactive power required by the VPP, but these
between scenario 3 and scenarios 1 and 2, respectively. For the
capacitor banks can only be used if the respective DG unit is turning
active power losses, the difference is around 36.65%, 13.60% be-
on. Otherwise, the capacitor cannot be used in the reactive man-
tween scenario 3 and the scenarios 1 and 2, respectively. The
agement, because it does not belong to the VPP. The VPP owns itself
voltage profile and active power losses present a much higher
several capacitor banks strategically located in the 33-bus distri-
difference than the operation cost due to the different reactive
bution network that will be used in this scenario to control the
control perspectives implemented in each scenario. The different
voltage magnitude.
reactive control perspectives do not affect much the operation cost
result of the optimal scheduling, yet they affect directly the voltage
4.3. Simulation 1 e objective function F1 (a ¼ 1; b ¼ 0) profile and active power losses.
The total active and reactive energy scheduled obtained by the
In simulation 1, it is only considered the objective function F1, three scenarios in simulation 1 is presented in Fig. 2a and b)
and the three scenarios presented before are used in this simula- respectively. The three scenarios obtained solutions very similar.
tion. Fig. 2 and Table 4 show the results of the three scenarios to For this reason, the operation cost results are very close, as it is

Fig. 3. Bus voltage profile for all scenarios with objective function F1: a) In period 4; b) In period 20.

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077
 T. Sousa et al. / Energy xxx (2015) 1e15 9

shown in Table 4. The different reactive control strategies inter- Table 5

fere directly on the voltage profile and active power losses, but Performance of the proposed methodology with objective function F2.

not in the optimal scheduling of the active power. Regarding to Scenario Voltage Operation Active power
the reactive energy scheduled, in scenario 1, the external sup- difference e F2 (pu.) cost e F1 (m.u.) osses (kW)
pliers generate all the reactive power resulting in the highest 1 7.91 9485.59 474.30
active power losses value. In scenario 2, the external suppliers 2 6.73 9102.16 330.32
and DG units generate similar amounts of reactive power. In 3 5.66 8719.84 274.84

scenario 3, the distributed resources, such as DG units and ca-

pacitors, generate more reactive power than the external sup-
pliers, which enables the VPP to obtain the lowest active power
In terms of voltage magnitude difference, scenario 3 presents the
losses value.
best result with 5.66 pu (sum of the differences of the voltage
The voltage magnitude of the slack bus was fixed at 1 pu. The
magnitude in all buses during the 24 periods); scenario 2 achieved
voltage profile of the three scenarios will be shown for one off-peak
the second best result with 6.73 pu; and scenario 1 obtained the
period (period 4) in Fig 3a) and for one peak period (period 20) in
worse result with 7.91 pu. These values represent a variation of
Fig 3b). In period 4, scenarios 1 and 2 present a worse voltage
39.75% and 18.90% comparing the scenario 3 with the scenarios 1
profile than scenario 3. Scenario 1 uses the external suppliers to
and 2 respectively. The active power losses present a similar
supply the reactive power requirement, which leads to a reduction
behaviour as it is possible verify in Table 5. In terms of operation cost,
of the voltage magnitude in the buses further of the slack bus, such
scenarios 1, 2 and 3 obtained a result of 9485.59 m.u., 9102.16 m.u.
as buses 16 and 17 (zone 3). Scenario 2 had the same behaviour as
and 8719.84 m.u., respectively. The cost variation between scenario
scenario 1, because in this scenario the endogenous resources (DG
3 and scenarios 1 and 2 is around 8.78% and 4.38%, respectively.
units and capacitors) can only be used in the peak periods. In period
Again, the voltage magnitude difference and the active power losses
4, the voltage magnitude achieves a low value around 0.97 pu in the
presented variations between scenarios much higher than the ones
further buses of the distribution network (buses 16, 17, 31 and 32).
obtained in the operation cost's results. This fact proves the
As expected, scenario 3 presents better results than the other two
importance of the reactive scheduling to the network operation.
scenarios, even considering that in simulation 1 the objective is just
Scenario 3 presents again the best results in terms of voltage
to minimize the operation cost. This scenario has more flexibility to
magnitude difference, active power losses and operation cost.
schedule the active and reactive resources, where the VPP can use
Fig. 4 depicts the active and reactive energy scheduled obtained
the external suppliers and the endogenous resources to increase
by the three scenarios with objective function F2. Again, in this
the voltage profile.
simulation the different scenarios present similar results in terms
Concerning the period 20, scenario 1 presents the worse voltage
of active energy scheduling. In terms of reactive scheduling (see
profile of the three scenarios, because the external suppliers are the
Fig. 4b), the scenario 3 uses DG units and capacitor banks to supply
only ones to supply reactive power. One way to increase this voltage
the reactive requirement in order to increase the voltage profile of
profile comprises the use of distributed energy resources to supply
reactive power. The voltage magnitude achieves the lowest value of
0.96 pu. in buses 16, 17, 31 and 32. Scenario 2 presents a better
result than scenario 1 due to the requirement by the DG units to
supply a minimum of reactive power equal to 40% of the active
power generated in the same period. With this requirement, the
VPP was able to increase the voltage magnitude of 0.005 pu. in
buses 16, 17, 31 and 32 (further buses). With the reactive control
strategy, adopted in scenario 3, the VPP was able to achieve the best
voltage profile in this period. There is more flexibility to schedule
the available resources in terms of active and reactive power, which
enables the VPP to mitigate the voltage drop in the further buses of
the network. For instance, the minimum voltage magnitude is
around 0.97 pu. In both periods, it is possible to observe that the
network operates near by the boundaries regarding the minimum
voltage limits (0.95 pu).
The performance presented by the three scenarios in simu-
lation 1 can be problematic in the future smart grid paradigm
with a higher resources penetration. The VPP must consider more
technical indicators that simulate the impact of the resources in
the network performance, avoiding the operation of the net-
works under stress. In the next simulation, it will be presented
the results using the objective function F2 that is a technical
indicator.

4.4. Simulation 2 e objective function F2 (a ¼ 0; b ¼ 1)

In simulation 2, the objective function F2 is used in the three

considered scenarios. The proposed methodology will obtain an
optimal scheduling that minimizes the voltage magnitude differ-
ence (see equation (3)) of all buses regarding the voltage in the Fig. 4. Scheduled energy for the three scenarios with objective function F2: a) Active
slack bus. The obtained results are presented in Table 5. energy; b) Reactive energy.

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077
10 T. Sousa et al. / Energy xxx (2015) 1e15

the buses, scenario 2 uses the external suppliers and the DG in in the management of the future networks. The use of function F2
equal percentage and in scenario 1 only considers the external helped the VPP achieving a good voltage profile, and very close to
suppliers. 1 pu. This technical indicator was able to avoid the instability point
The voltage profile in periods 4 and 20 obtained for each sce- presented in simulation 1. However, the operation cost obtained in
nario is illustrated in Fig. 5. The three scenarios present similar this simulation is higher than the one achieved in simulation 1.
voltage profile. However, scenario 1 presents the worse result of the These two objectives are in conflict and it requires the use of a
three scenarios. In scenario 3, the proposed methodology was able multi-objective approach to deal with this kind of behaviour.
to manage the reactive power in order to obtain a voltage profile in
all buses near to the voltage fixed in the slack bus (1 pu). Buses 31 4.5. Simulation 3 e multi-objective function F (a2[0,1]; b2[0,1])
and 32 (the further buses) achieved the lowest voltage magnitude
of around 0.991 pu. The voltage drop of these buses was of In simulation 3, the results of the multi-objective optimal
approximately 0.7%, and in simulation 1 the voltage drop achieved resource scheduling approach are presented. As referred before, the
3%. In period 20, the three scenarios present more distinct results in weighted sum method has been implemented to determine the
terms of voltage magnitude, because of the peak demand obtained Pareto front for the proposed multi-objective function F in the three
in this period. In scenario 3, buses 31 and 32 obtained the lowest scenarios. For each scenario, the proposed methodology had run
voltage magnitude of approximately 0.977 pu. In simulation 1, for 501 times and in each run the weighted factors changed from 0 to 1
the same scenario, the methodology achieved a voltage profile in steps of 0.002. For each run, the sum of weighted factors (a,b)
below 0.97 pu. With this objective function F2, the VPP was able to must be equal to 1. Fig. 6 shows the obtained Pareto.
increase the voltage profile in 0.07 pu. The Pareto front presents 393, 385 and 382 non-dominated
The results obtained in this simulation proved the importance of solutions for the scenarios 1, 2 and 3, respectively. The Pareto
technical indicators, such as the one used in function F2 (see Eq. (3)) front solutions are well spread in the three scenarios. As it

Fig. 5. Bus voltage profile for all scenarios with objective function F2: a) In period 4; b) In period 20.

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077
 T. Sousa et al. / Energy xxx (2015) 1e15 11

Fig. 6. Pareto front for the scenarios 1, 2 and 3 of simulation 3.

happened in the previous simulations, the best Pareto front was

obtained for scenario 3, and the worst result was determined for
scenario 1. This happen for the same reasons pointed in the pre-
vious simulations.
After determining the Pareto front, the fuzzy set approach has
been used to identify the best compromise solution for each sce-
nario. This method helps the VPP to choose the most suitable so-
lution to be applied for the optimal resource scheduling problem.
Table 6 shows the membership function ms value, the operation
cost, the voltage magnitude difference and the active power losses Fig. 7. Scheduled energy for the three scenarios for the best compromise solution: a)
of the best compromise solution. Active energy; b) Reactive energy.

Scenario 3 presents the best compromise solution with the

lowest value of the three scenarios for the operation cost, voltage
difference, and active power losses. On the other hand, scenario 1
In comparison with simulation 1 (see Fig. 3), these voltage
presents the worst result in these three indicators. In terms of
profiles are higher in all scenarios. On the other hand, the voltage
operation cost's variation, it is seen a low difference of 0.59% and
profiles of this simulation are lower than the ones obtained by
0.35% between scenario 3 and scenarios 1 and 2, respectively. On
simulation 2 (see Fig. 5). The multi-objective approach was able to
the other hand, the voltage difference achieved a higher variation of
find a middle-term solution between the individual optimal solu-
around 32.42% and 20.11% between scenario 3 and scenarios 1 and
tions obtained by the operation cost (simulation 1) and the voltage
2, respectively. Moreover, the active power losses variation be-
magnitude difference (simulation 2).
tween the best one and scenarios 1 and 2 is approximately 48.00%
The use of a multi-objective approach is essential to obtain all
and 31.29%, respectively. Fig. 7 presents the active and reactive
possible solutions of the active and reactive optimal resource
energy scheduling obtained by the best compromise solutions in
scheduling problem (non-dominated solutions of the Pareto front)
the three scenarios. As expected, the results are between the ones
with the purpose to help the VPP to select the best solution for its
obtained in the previous simulations. The reactive energy sched-
own interests. The multi-objective approach is more useful to the
uling is very close to the one obtained with the objective function
VPP than to optimize individually each objective function such as in
F2. This happens due to the cost of the reactive power is considered
simulation 1 and 2. The proposed methodology was able to obtain
equal to zero in the simulations.
the Pareto front and the fuzzy set approach presented the best
Fig. 8 illustrates the voltage magnitude in all buses in periods 4
compromise solution for the three scenarios.
and 20. As expected, scenario 3 obtains the best voltage profile than
the other two scenarios in both periods, presenting a solution that
4.6. Comparison of simulations 1, 2 and 3
minimizes operation cost and voltage magnitude at the same time.
This scenario uses more DG units and capacitor banks to increase
The comparison of the operation cost of simulations 1, 2 and 3
voltage in the further buses.
(considering the best solution) is presented in Fig. 9. As expected,
simulation 1 achieved a better operation cost for the three sce-
narios. Simulation 3 presents the second best results with very
close results to the ones presented by simulation 1. The operation
Table 6 cost variation between simulation 1 and 3 is around 1.13%, 1.09%
Best compromise solution of the proposed methodology with multi-objective
and 1.48% for scenarios 1, 2 and 3, respectively. Moreover,
function F.
the operation cost increased from simulation 1 to simulation 2
Scenario ms Operation Voltage Active power in 36.23%, 30.99% and 26.41% for scenarios 1, 2 and 3,
cost e F1 (m.u.) difference e F2 (pu.) losses (kW)
respectively.
1 0.003260 7042.53 12.58 670.39 Fig. 10 depicts the voltage magnitude difference of simulations 1,
2 0.003282 7025.75 11.41 528.64 2 and 3 in the three scenarios. As expected, simulation 2 achieved a
3 0.003582 7001.41 9.50 452.96
best voltage magnitude difference in the three scenarios.

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077
12 T. Sousa et al. / Energy xxx (2015) 1e15

Fig. 8. Bus voltage profile for all scenarios for the best compromise solution: a) In period 4; b) In period 20.

Simulation 3 presents the middle-term voltage profile result and simulation 2 in approximately 86.73%, 102.53% and 119.79% for the
simulation 1 obtains the worst result of the three simulations. The scenarios 1, 2 and 3, respectively. These variations are higher than
voltage profile variation between simulation 2 and 3 is around the ones presented in the previous comparison of the operation
59.04%, 69.54% and 67.84% for scenarios 1, 2 and 3, respectively. The cost (see Fig. 9), proving once again the importance of using this
voltage magnitude difference decreased from simulation 1 to technical indicator in order to improve the operation of the future

Fig. 9. Operation cost comparison between simulation 1, 2 and 3. Fig. 10. Voltage magnitude difference comparison between simulation 1, 2 and 3.

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077
 T. Sousa et al. / Energy xxx (2015) 1e15 13

networks. For instance, in scenario 3 the operation cost increases in The main conclusion of this work is the importance of using
1.48% with the objective of reducing the voltage magnitude dif- distributed resources in the reactive power scheduling in order to
ference in 67.84% between simulation 1 and 3. The reduction in the guarantee a safe operation without any kind of stressful situations.
voltage magnitude differences has more impact than an increase in The proposed methodology was able to integrate the active and
the operation cost. For the simulation 3, the differences between reactive power scheduling with the objective of obtaining a solu-
scenarios 1 and 3 is for the operation cost of 0.6% and for the voltage tion that manages the network for the economic and technical
difference of 32%. perspectives. The aggregated players must also be aware of
The active power losses comparison between simulations 1, 2 the technical indicators of the network, and not just in the eco-
and 3 is shown in Fig. 11. The active power losses results have the nomic perspective of cost minimization. The multi-objective
same behaviour as the voltage magnitude difference (see Fig. 10), approach was able to obtain the Pareto front and the respective
and simulation 2 presents better results than simulation 1 and 3. non-dominated solution in order to select the most suitable solu-
Simulation 3 presents the second best result and simulation 1 ob- tion based on the VPP's interest. In the future, the authors intend to
tains the worst result for the power losses. use multi-objective evolutionary algorithms to deal with the pro-
The results obtained in this paper have been technically vali- posed optimization problem.
dated using an adequate network transient simulator tool pre-
sented in Ref. [39]. This simulator provides a large set of realistic Acknowledgements
models of the energy resources and network equipment allowing
the validation of the proposed methodology. This work is supported by FEDER Funds through the ‘‘Programa
Operacional Factores de Competitividade e COMPETE00 program
5. Conclusions and by National Funds through FCT ‘‘Fundaç~ ^ncia e a
ao para a Cie
Tecnologia’’ under the projects FCOMP-01-0124-FEDER: UID/EEA/
This paper presents a methodology that handles with the active 00760/2013, and SFRH/BD/81848/2011 (Tiago Sousa PhD), and by
and reactive optimal resource scheduling in a distribution network the SASGER-MeC, project n NORTE-07-0162-FEDER-000101, co-
with a large number of distributed resources. The proposed funded by COMPETE under FEDER Programme. Hugo Morais is
methodology is helpful to avoid drawbacks introduced by the supported by the SOSPO project has received funding from the
distributed resources, such as bidirectional power flow, instability Danish Council for Strategic Research under grant agreement no.
in the voltage levels. Three objective functions are used in the re- 11-116794. Rui Castro is supported by national funds through FCT
sources scheduling, namely: the operation cost minimization, the “Fundaça~o para a Cie
^ncia e a Tecnologia” with reference UID/CEC/
minimization of the voltage magnitude difference and the mini- 50021/2013.
mization of a multi-objective function resulting in a Pareto front. A
fuzzy set approach has been used to select the best compromise Appendix A
solution. The proposed function to minimize the voltage magnitude
difference in all buses improves the voltage stability of the network NDG 
X
i
 i
NSP
X
i
NEV
X
while reducing consequently the power losses. i i i i
PDGðDG;tÞ  PGCPðDG;tÞ þ PSPðSP;tÞ þ PDchðEV;tÞ
The paper presents a case study for a 33-bus distribution DG¼1 SP¼1 EV¼1
network managed by a VPP with a penetration of 66 DG units and NLi 
X  i
NEV
X
1000 EVs showing clearly the effectiveness of the proposed i i i 2
 PLoadðL;tÞ  PNSDðL;tÞ  PChðEV;tÞ ¼ Gii  ViðtÞ
approach to significantly increase the voltage profile in all buses of L¼1 EV¼1
network with a marginal increase in the costs. With a Pareto front, X  
the VPP can choose for different solutions according its profile and þ ViðtÞ  VjðtÞ  Gij cos qijðtÞ þ Bij sin qijðtÞ
j 2 Li
the operation scenario. Additionally, the fuzzy set approach gives a
good compromised solution between the both considered objec- (4)
tives (costs and voltage profile). Several scenarios have been tested
with the three proposed objective functions, considering the gen- N i
X
i
NSP N
X
i
PDG
i i i
CAP
eration of reactive power by the DG and/or by the capacitor banks. QDGðDG;tÞ þ QSPðSP;tÞ þ
QCAPðCAP;tÞ
The best results are obtained when the DG units generate active DG¼1 SP¼1 CAP¼1
i
and reactive power and when the capacitor banks are used (i.e. NLi 
P  NAsyn
XDG
i i i
scenario 3). In this case, a reduction of 32% of voltage differences is  QLoadðL;tÞ  QNSDðL;tÞ  QAsyn DGðAsyn DG;tÞ
obtained with a cost increasing of only 0.6%. L¼1 Asyn DG¼1
X  
2
¼ ViðtÞ  VjðtÞ  Gij sin qijðtÞ  Bij cos qijðtÞ  Bii  ViðtÞ
j 2 Li
ct2f1; …; Tg; ci2f1; …; NB g; qijðtÞ ¼ qiðtÞ  qjðtÞ
(5)

i i
VMin  ViðtÞ  VMax (6)

qiMin  qiðtÞ  qiMax (7)

 h   i* 
 
UiðtÞ  yij  UiðtÞ  UjðtÞ þ ysh i  UiðtÞ   SMax
Lk (8)

Fig. 11. Comparison of active power losses between simulation 1, 2 and 3.

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077
14 T. Sousa et al. / Energy xxx (2015) 1e15

 h   i*  QSPðSP;tÞ  QMaxðSP;tÞ
 
UjðtÞ  yij  UjðtÞ  UiðtÞ þ ysh j  UjðtÞ   SMax
Lk (22)
(9) ct2f1; …; Tg; cSP2f1; …; NSP g
ct2f1; …; Tg; ci; j2f1; …; NB g; isj; ck2f1; …; NK g
EStoredðEV;tÞ ¼ EStoredðEV;t1Þ  ETripðEV;tÞ þ hcðEVÞ  PChðEV;tÞ
0 12 0 12
i
NSP
X
i
NSP
X  2 1
@ i
PSPðSP;tÞ A þ@ i
QSPðSP;tÞ A  SMax   PDchðEV;tÞ
TFR HV MVðiÞ hdðEVÞ
SP¼1 SP¼1
ct2f1; …; Tg; cEV2f1; …; NEV g; Dt ¼ 1
ct2f1; …; Tg; ci2f1; …; NB g
t ¼ 1/EStoredðEV;t1Þ ¼ EInitialðEVÞ
(10)
(23)

X 
i
NDG
i i EStoredðEV;tÞ  EBatMinðEV;tÞ (24)
PTFR MV LVði;tÞ ¼ PDGðDG;tÞ  PGCPðDG;tÞ
DG¼1
NEV 
i
 EStoredðEV;tÞ  EBatMaxðEV;tÞ
X (25)
i i
þ PDchðEV;tÞ  PChðEV;tÞ (11)
ct2f1; …; Tg; cEV2f1; …; NEV g
EV¼1
NLi 
X 
i i PChðEV;tÞ  PMaxðEV;tÞ  XChðEV;tÞ (26)
 PLoadðL;tÞ  PNSDðL;tÞ
L¼1
PDchðEV;tÞ  PMaxðEV;tÞ  XDchðEV;tÞ (27)
i i
NDG NCAP
X X
i i
QTFR MV LVði;tÞ ¼ QDGðDG;tÞ þ QCAPðCAP;tÞ XChðEV;tÞ þ XDchðEV;tÞ  1
DG¼1 CAP¼1 (28)
NLi   XChðEV;tÞ and XDchðEV;tÞ 2f0; 1g
X
i i
 QLoadðL;tÞ  QNSDðL;tÞ (12)
L¼1
i
NAsyn
XDG
 i
QAsyn References
DGðAsyn DG;tÞ
Asyn DG¼1
[1] Tafarte P, Das S, Eichhorn M, Thr€ an D. Small adaptations, big impacts: options
for an optimized mix of variable renewable energy sources. Energy
 2  2  2 2014;72(0):80e92.
PTFR MV LV þ QTFR MV LV  SMax
TFR MV LVðiÞ (13) [2] Kempton W, Tomic J. Vehicle-to-grid power fundamentals: calculating ca-
pacity and net revenue. J Power Sources 2005;144(1):268e79.
ct2f1; …; Tg; ci2f1; …; NB g [3] Vale Z, Morais H, Faria P, Ramos C. Distribution system operation supported by
contextual energy resource management based on intelligent SCADA. Renew
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Energy 2013;52(0):143e53.
P 2DGðDG;tÞ þ Q 2DGðDG;tÞ  SMinðDG;tÞ  XDGðDG;tÞ (14) [4] Morais H, Pinto T, Vale Z, Praca I. Multilevel negotiation in smart grids for VPP
management of distributed resources. IEEE Intell Syst 2012;27(6):8e16.
[5] Sousa T, Vale Z, Carvalho JP, Pinto T, Morais H. A hybrid simulated annealing
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi approach to handle energy resource management considering an intensive
P 2DGðDG;tÞ þ Q 2DGðDG;tÞ  SMaxðDG;tÞ  XDGðDG;tÞ (15) use of electric vehicles. Energy 2014;67(0):81e96.
[6] Madzharov D, Delarue E, D'Haeseleer W. Integrating electric vehicles as
flexible load in unit commitment modeling. Energy 2014;65(0):285e94.
[7] Ji B, Yuan X, Chen Z, Tian H. Improved gravitational search algorithm for unit
PDGðDG;tÞ  PMinðDG;tÞ  XDGðDG;tÞ (16) commitment considering uncertainty of wind power. Energy 2014;67(0):
52e62.
[8] Li Y, Wu Q, Li M, Zhan J. Mean-variance model for power system economic
dispatch with wind power integrated. Energy 2014;72(0):510e20.
PDGðDG;tÞ  PMaxðDG;tÞ  XDGðDG;tÞ (17) [9] Aman M, Jasmon G, Bakar A, Mokhlis H. A new approach for optimum
simultaneous multi-DG distributed generation units placement and sizing
based on maximization of system loadability using HPSO (hybrid particle
QCAPðCAP;tÞ  QMaxðCAP;tÞ (18) swarm optimization) algorithm. Energy 2014;66(0):202e15.
[10] Abido M. Multiobjective evolutionary algorithms for electric power dispatch
problem. IEEE Trans Evol Comput 2006;10(3):315e29.
[11] Chun-Hao L, Ansari N. Decentralized controls and communications for
X
NStep autonomous distribution networks in smart grid. IEEE Trans Smart Grid
QCAPðCAP;tÞ ¼ QMaxðCAP;Step;tÞ  XCAPðCAP;Step;tÞ 2013;4(1):66e77.
Step¼1 (19) [12] Mousavi O, Cherkaoui R. Literature survey on fundamental issues of voltage
and reactive power control. Deliverable of the MARS Project financially sup-
ct2f1; …; Tg; cCAP2f1; …; NCAP g ported by “swisselectric research”. 2011.
[13] van Cutsem T, Vournas C. Voltage stability of electric power systems. Springer;
1998.
[14] Zhang A, Li H, Liu F, Yang H. A coordinated voltage/reactive power control
method for multi-TSO power systems. Int J Elec Power 2012;43(1):20e8.
PDGðDG;tÞ þ PGCPðDG;tÞ ¼ PDGForecastðDG;tÞ [15] Mousavi O, Bozorg M, Cherkaoui R. Preventive reactive power management
(20) for improving voltage stability margin. Electr Pow Syst Res 2013;96(0):36e46.
[16] Caldon R, Rossetto F, Scala A. Reactive power control in distribution networks
ct2f1; …; Tg; cDG2f1; …; NDG g
PSPðSP;tÞ  PMaxðSP;tÞ (21) with dispersed generators: a cost based method. Electr Pow Syst Res
2003;64(3):209e17.
[17] Reddy S, Abhyankar A, Bijwe P. Reactive power price clearing using multi-
objective optimization. Energy 2011;36(5):3579e89.

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077
 T. Sousa et al. / Energy xxx (2015) 1e15 15

[18] Saraswat A, Saini A, Saxena AK. A novel multi-zone reactive power market [28] Peterson S, Whitacre J, Apt J. The economics of using plug-in hybrid electric
settlement model: a pareto-optimization approach. Energy 2013;51(0): vehicle battery packs for grid storage. J Power Sources 2010;195(8):2377e84.
85e100. [29] Peterson S, Apt J, Whitacre J. Lithium-ion battery cell degradation resulting
[19] Martinez-Rojas M, Sumper A, Gomis-Bellmunt O, Sudria -Andreu A. Reactive from realistic vehicle and vehicle-to-grid utilization. J Power Sources
power dispatch in wind farms using particle swarm optimization technique 2010;195(8):2385e92.
and feasible solutions search. Appl Energ 2011;88(12):4678e86. [30] Deb K. Multi-objective optimization using evolutionary algorithms. Wiley;
[20] Shaw B, Mukherjee V, Ghoshal S. Solution of reactive power dispatch of power 2001.
systems by an opposition-based gravitational search algorithm. Int J Elec [31] Gomez-Exposito A, Conejo A, Canizares C. Electric energy systems: analysis
Power 2014;55(0):29e40. and operation. Taylor & Francis; 2008.
[21] Zare M, Niknam T, Azizipanah-Abarghooee R, Amiri B. Multi-objective prob- [32] Ferreira J, Monteiro V, Afonso J. Vehicle-to-Anything application (V2Anything
abilistic reactive power and voltage control with wind site correlations. En- app) for electric vehicles. IEEE Trans Industrial Informatics 2013:99. 1-.
ergy 2014;66(0):810e22. [33] Soares J, Canizes B, Lobo C, Vale Z, Morais H. Electric vehicle scenario simu-
[22] Fodor A, Magyar A, Hangos K. Multiple-InputeMultiple-Output linear- lator tool for smart grid operators. Energies 2012;5(6):1881e99.
quadratic control of the energy production of a synchronous generator in a [34] Garcia-Valle R, Lopes J. Electric vehicle integration into modern power net-
nuclear power plant. Electr Pow Compo Sys 2014;42(15):1673e82. works. Springer; 2012.
[23] Lin X, Yu C, David A, Chung C, Wu H, Xu N. A novel market-based reactive [35] Hong L, Huansheng N, Yan Z, Guizani M. Battery status-aware authentication
power management scheme. Int J Elec Power 2006;28(2):127e32. scheme for V2G networks in smart grid. IEEE Trans Smart Grid 2013;4(1):99e110.
[24] Badar A, Umre B, Junghare A. Reactive power control using dynamic particle [36] GAMS. GAMS home page.
swarm optimization for real power loss minimization. Int J Elec Power [37] Sakawa M, Yano H, Yumine T. An interactive fuzzy satisficing method for
2012;41(1):133e6. multiobjective linear-programming problems and its application. IEEE trans-
[25] Kargarian A, Raoofat M, Mohammadi M. Reactive power market management actions on systems. Man Cybern 1987;17(4):654e61.
considering voltage control area reserve and system security. Appl Energ [38] Farina M, Amato P. A fuzzy definition of “optimality” for many-criteria opti-
2011;88(11):3832e40. mization problems. IEEE Trans Syst Man Cybern Part A Syst Humans
[26] Zare M, Niknam T. A new multi-objective for environmental and economic 2004;34(3):315e26.
management of Volt/Var control considering renewable energy resources. [39] Silva M, Morais H, Vale Z. An integrated approach for distributed energy
Energy 2013;55(0):236e52. resource short-term scheduling in smart grids considering realistic power
[27] Grainger J, Stevenson W. Power system analysis. McGraw-Hill; 1994. system simulation. Energ Convers Manage 2012;64(0):273e88.

Please cite this article in press as: Sousa T, et al., A multi-objective optimization of the active and reactive resource scheduling at a distribution
level in a smart grid context, Energy (2015), http://dx.doi.org/10.1016/j.energy.2015.03.077