Sustainable Cities and Society 38 (2018) 714–722

Contents lists available at ScienceDirect

Sustainable Cities and Society

journal homepage: www.elsevier.com/locate/scs

Energy management and control policies of the islanded microgrids T

a,⁎ b
Mohamad Abedini , Moein Abedini
a
Department of Electrical Engineering, Ayatollah Broujerdi University, Borujerd, Lorstan, CT 06516, Iran
b
School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran 1439957131, Iran

A R T I C L E I N F O A B S T R A C T

Keywords: This paper proposes a new algorithm to find droop parameters of a microgrid (MG) in the islanded mode. To do
Microgrid this, a hybrid method is introduced to solve optimization problem based on the non-dominated solution. The
Distributed generation (DG) proposed algorithm is combined of the harmony algorithm (HS) and also genetic algorithm and optimization
Droop controller variables include droop parameters of DGs in the islanded mode. The results show that the proposed algorithm is
capable of significantly reducing islanded MG customer interruptions as well as improving the islanded MGs
stability. The findings also demonstrate that an effective selection of droop parameters settings will facilitate the
successful implementation of the islanded MG concept in distribution systems. The performance of the paper
approach is compared with other optimization and non-optimization methods in MG with 33 bus using MATLAB.

1. Introduction
Distributed generations (DGs), emerging from the utilization of re-
newable energy resources (RES) in distribution networks, can be clas-
sified in different types including: i) controllable, e.g., diesel generators,
fuel cells and micro-turbines, and ii) uncontrollable, e.g., wind turbines
and photovoltaic systems. DGs along with the energy storages, con-
trollers, loads, and communication systems makes a small low-voltage
system, known as Micro-grid (MG), which is undeniably regarded as a
significant and essential part of the development of distribution net-
works (Katiraei, Iravani, Hatziaragyriou, & Dimeas, 2008). MGs operate
in two different modes, namely i) autonomous, also known as islanded,
and ii) connected referred as grid-connected. In the connected mode
typically the MG is connected to a grid via the point of common cou-
pling (PCC).
Among various factors affected on the dynamic behavior of MGs,
optimal sizing, location and operation of DGs are of the most particular
interests in the power grid optimization literature. Although researches
in finding optimal sizing and siting of DGs in a MG is superiorly de-
veloped (Abdi & Afshar, 2013; Fetanat & Khorsaninejad, 2015;
Gnanambal, Marimuthu, & Babulal, 2011; Moradi & Abedini, 2012;
Nekooei, Farsangi, Nezamabai, & Lee, 2013; Soroudi & Afrasiab, 2012),
discussion is still ongoing as to what is the optimal operation of DGs.
To find the optimal operation of the autonomous MG with and
without droop control, the use of a central controller is vital. A number
of researchers have put forward models for the improvement of op-
eration of the autonomous MG with a central controller (MGCC). In
Huang, Huang, Chang, Yeh, and Tsai (2013), some optimum scheduling
techniques for autonomous MGs are discussed focusing on the economic
profits. In Zoka, Sugimoto, Yorino, Kawahara, and Kubokawa (2007)
and Vasiljevska, Peças Lopes, and Matos (2012), MGCC is utilized to
optimize the total fuel cost of the islanded MG where DGs are modeled
based on the droop control and includes a combined heat and power
(CHP) unit. The research of Basu, Bhattacharya, Chowdhury, and
Chowdhury (2012) looks at the power dispatching among DGs based on
an evolutionary algorithm. In Zheng, Meinrenken, and Lackner (2014)
and Basu et al. (2012), different techniques are proposed to reduce the
fuel cost of the islanded MGs. In Khorramdel and Raoofat (2012) and
Dilettoso, Rizzo, and Salerno (2008), an analytical programming
method is introduced for sharing reactive power among DGs by using
DGs droop control and also considering uncertainty of wind plants. In
Mitra, Vallem, and Patra (2006), a dynamic programming method is
employed for a small MG to find the optimal combination of DGs among
photovoltaic, micro-turbine, and energy storage system to supply both
thermal and electrical demands. An optimum combination of the DG is
developed in Pipattanasomporn, Willingham, and Rahman (2005) by
utilizing a linear programming method. The study of Oyarzabal,
Jimeno, RuelaEngler, and Hardt (2005) gives a multi-agent system for
MG management to optimize the system operation where some agents
receive generation scheduling molds from MG information. In
Kuznetsova, Li, Ruiz, and Rio (2014), the operation of DGs is optimized
subject to environmental constraints. In Moradi, Eskandari, and
Hosseinian (2015) and Moradi, Abedini, and Hosseinian (2015), op-
timal sizing and operational strategy of DGs in MGs are represented
based on a hybrid evolutionary algorithm by considering a grid-con-
nected mode for MGs.

⁎
Corresponding author.
E-mail addresses: m.abedini@abru.ac.ir (M. Abedini), m.abedini@ut.ac.ir (M. Abedini).

https://doi.org/10.1016/j.scs.2018.01.032
Received 4 June 2017; Received in revised form 5 December 2017; Accepted 16 January 2018
Available online 31 January 2018
2210-6707/ © 2018 Elsevier Ltd. All rights reserved.
M. Abedini, M. Abedini Sustainable Cities and Society 38 (2018) 714–722

Nomenclature Pj Injected active power to bus j

Qj Injected reactive power to bus j
HMS Harmony memory size Cm(PDGi) Fuel consumption of DGi
PAR Pitch adjusting rate Piˆ+ 1, Qiˆ+ 1 Total active and reactive power at bus i + 1
HMCR Harmony memory considering rate Ndr Number of DGs with droop control
BW Distance bandwidth Vi* Voltage magnitude of DGi at no load
t Time index ωi* Angular frequency of DGi at no load
Nobj Number of objective functions Sp, Sq Active and reactive power static droop gain
Nrep Number of solutions in repository Yij (ω) The ijth element of admittance matrix
κi The decision maker preference for all objective functions θij Phase angle
δi Phase angle of voltage at bus i Vmin, Vmax Minim and maximum bus voltage
Ii,Iimax Current and rated current of line i NB Number of buses
Sgimax Maximum apparent power of DGi P i, Q i Demand active and reactive power of bus i
Pgimax Maximum active power of DGi Ri, Xi Resistance and reactance of line i
Qgimax Maximum reactive power of DGi Pgi, Qgi Active and reactive power delivered to bus i

Despite the great contribution of the above studies in the area of
power grid optimization, a review of the literature reveals that few
studies have looked at optimal operation of DGs along with its optimal
location. To overcome the mentioned drawbacks, this paper represents
a novel hybrid optimization method to find optimal operation of an
autonomous MG. The operation is optimized via finding the optimal
droop gain parameters of DGs. To do this, the optimization problem is
formulated in terms of a multi-objective problem to: i) minimize the
fuel consumption of DGs, ii) improve the voltage profile, and iii) sta-
bility of the MG by considering both operation and security constraints.
A hybrid algorithm, referred as HS-GA, is developed to solve the opti-
mization problem.
The rest of the paper is organized as follows. First, a background to
the paper is discussed. Then the problem formulation along with its
corresponding solution is developed. Finally, the obtained results and
discussion are represented.
DGs (based on DG characteristics) (Geem, Kim, & Loganathan, 2001).
To cope with these drawbacks, tuning of DG droop parameters is of
particular usefulness. MGCC might be employed to make such para-
meters tuning possible. In this paper, a combination of the MGCC with a
decentralized control method is developed to solve the optimal power-
sharing problem of DGs. The MGCC is employed to provide optimum
DGs’ droop parameters while a decentralized controller is used to op-
erate with optimal droop parameters.
3. Problem formulation
The problem is to find the optimal operation and also static char-
acteristics of DGs to reduce fuel cost of DG units (f1); improve voltage
stability index (f2); and reduce total voltage variation (f3), subject to
actual constraints. These three objective functions can be mathemati-
cally formulated as a minimization problem as follows:

⎧ f (i, k ) ⎫
2. Control methods of microgrids ⎪ 1 ⎪
Minimize f2 (i, k )
⎨ ⎬
Two modes of operation can be considered for the MGs, referred as ⎪ f3 (i, k ) ⎪ (1)
⎩ ⎭
i) grid-connected mode, and ii) autonomous mode. In the grid-con-
where i is the bus number, representing DGs location; and k represents
nected mode, both voltage and frequency of the MG are dictated by the
static characteristics including references of the frequency, voltage, and
upstream grid where the MG just controls the interchange of electrical
static droop gains as follows:
power with the grid (Lasseter & Piagi, 2004). In the autonomous mode,
where there is no connection between MG and the main grid, the MG k = {ωi*, Vi* , spi, sqi} i ∈ Ndr (2)
control is responsible for both frequency and voltage regulation and
thus the power generated by MG needs to meet its local demand. To
optimality share power between the corresponded DGs, a proper con- 3.1. Fuel cost of DG units
trol method is required to manage the supply of MG’s load. Under such
a condition, there are two control methods: i) centralized, and ii) de- Fuel consumption cost can be represented by,
centralized. A centralized method is too costly due to transferring re- nDG

markable data and requiring a reliable communication link (Green & f1 = ∑ PDGi × Ci (PDGi )
i=1 (3)
Prodanovic, 2007; Li & Shi, 2012). Consequently, centralized methods
sound to be appropriate for small-scale MGs where DG units are suffi-
ciently close together. 3.2. Voltage stability index (VSI)
In a decentralized method, the load demand is shared between DGs
by use of droop controller characteristics and local measurement (fre- This index can be expressed by Moradi, Abedini et al. (2015),
quency and voltage variations). Although decentralized methods pro-
VSI (i+1)=Vi 4−4 [Piˆ+ 1 Xi − Qiˆ+ 1 Ri]2 −4 [Piˆ+ 1 Ri + Qiˆ+ 1 Xi ]2 Vi 2. (4)
vide a proper voltage regulation in PCC, there are some drawbacks
include: voltage drop in some buses, lack of minimizing power gen- Here, VSI is utilized to calculate the stability of the MG transmission
eration costs, and lack of dispatching reactive power generation among line as illustrated in Fig. 1. For the stable operation of the MG, VSI

Fig. 1. One-line diagram of MG line.

715
M. Abedini, M. Abedini Sustainable Cities and Society 38 (2018) 714–722

should be positive in all microgrid buses. To improve VSI, expression 4.1. Three fuzzy-subordination functions
(4) should be maximized or equivalently the following should be
minimized, In this paper, a fuzzy logic is employed to find the best solution in
1 the two separate parts: one is finding the best solution of the optimal
f2 = i = 1,2,..., NB. objective functions from their set of Pareto front solutions and another
VSI (i + 1) (5)
is finding the best HM from its two vectors provided by GA and HS. For
a set of Pareto front solutions, fuzzy membership functions can be re-
3.3. Total voltage variation (TVV) presented by Papageorgiou (2011),

To uniform voltage in MG, TVV index can be introduced by, ⎧ 1 fi < fi min
⎪ fimax − fi
NB μf = max min fi min ≤ fi ≤ fi max
∑i = 1 1 − Vi i
⎨ fi − fi
f3 = ⎪
NB (6) ⎩ 0 fi ≥ fi max (13)
where μ fi is the fuzzy subordination of fi; fi min is the best solution of fi,
3.4. Constraints and fi max is the worst solution. Linear-partition of fall half trapezoid is
used for the fuzzy membership functions. To solve multi-objective
There are two types of limitations in the optimization problem: functions, in each iteration, there is a set of non-dominated answers
equality and inequality. Equality limitations are nonlinear load flow which should be kept in a repository. Answer x1 dominates x2 if the
equations which can be obtained by Moradi and Abedini (2012), following equations are warranted,
NB
fn (x1) ≤ fn (x2) ∀ n ∈ {1,2,..., τ } (14)
Pgi = Pi + Vi ∑ Vj Yij cos(δi − δj − θij )
j=1 (7)
fm (x1) < fm (x2) ∃ m ∈ {1,2,..., τ } (15)
NB
Qgi = Qi + Vi ∑ Vj Yij sin(δi − δj − θij ) where x represents solutions includes droop parameters and sitting of
j=1 (8) DGs. To evaluate each individual in repository, the following normal-
ized function can be used as follows:
1) Voltage Limitation: The limitation of the voltage magnitudes in the
MG buses can mathematically be represented by Moradi and Abedini N
∑i =obj1 κi × μ fi (x n )
(2012), λμ (n) = N N
∑n =rep1 ∑i =obj1 κi × μ fi (x n ) (16)
Vmin ≤ Vi ≤ Vmax (9)
Eq. (16) is used to sort the non-dominated solutions, according to
2) Thermal Limitation: The thermal limitation can be obtained based
the decision making preference, for all objective functions.
on the current flow in the network lines as follows:
Ii < Ii max i = 1,..., NBr . (10) 4.2. Proposed hybrid optimization algorithm
3) Power Generation Limitation of DGs: For an islanded MG with a
droop controller, the power generation limits can be obtained based on The proposed hybrid optimization algorithm, HS-GA, is a combi-
active, reactive and nominal powers (Schmidt, Steele, Dillon, & Chang, nation of the both harmony algorithm (HS) (Rezaei Mozafara, Moradi,
2007) as follows: & Amini, 2016) and the genetic algorithm (GA). Although HS is re-
garded as a popular optimization approach, it has some deficiency due
max 1
0 ≤ Sgni − Pgni (S ) ⊥ *(ωi* − ω) − Pgni (S ) ≥ 0 to its low speed divergence and sticking to local optimal solutions. To
kpi (11) cope with such a deficiency, HS is combined with the GA to provide a
1 new optimization algorithm. The combination of HS and GA is based on
max
0 ≤ Qgni − Qgni (S ) ⊥ *( Vi* − Vi ) − Qgni (S ) ≥ 0 adding two GA operators, namely, mutation and crossover to HS. Mu-
kqi (12)
tation is exploitative and it generates random small diversions. Thus,
The notation ⊥ in Eqs. (11) and (12) denotes a complement index. In the new solution stays near the parent and its promising region.
the cases where the generated active and reactive powers reach to their Crossover is explorative and generates random big diversions. Thus, the
max max
maximum values, Pgni and Qgni are set to Pgni and Qgni , respectively. new solution jumps to a region between two parents. These two op-
Under such a condition, the droop controller would be converted to a erators are called HS-GA that compared with HS, it can effectively
PQ bus controller. search the solution region and reduce convergence calculation time in
the local minima region.The scheme of HS-GA can be discussed in terms
4. Proposed solution of three harmony memories, referred as HM1, HM2 and HM3. Firstly,
HM1 is generated similar to the conventional HS methods. Then, har-
A novel solution is proposed to solve the mentioned multi-objective mony vectors in HM1 are used as parents of GA in a way that both
problem. The solution is developed in three parts: crossover and mutation operators of the GA algorithm are applied to
HM1 to provide new harmony vectors, saved in harmony memory,
i) Proposing a hybrid optimization algorithm, referred as HS-GA, to called HM2. Finally, harmony vectors in HM1 and HM2 are compared
solve the multi-objective problem; together according to the objective function fitness to obtain the best
ii) Developing a novel load flow formulation to run HS-GA to optimize solution, saved in harmony memory, called HM3. It should be noted
the following variables: a) the steady state frequency of system, b) that in the HS approach, HM1 is considered as a next generation in
reference frequency, c) reference voltage, and d) droop coefficients contrast in the HS-GA approach, HM3 is considered as a next genera-
of DGs; tion.
iii) Adopting a fuzzy method to obtain the best solution of the non- In summary, the steps of the HS-GA algorithm can be outlined as
dominated results. It should be noted that HM is a memory space follows:
that includes the set of decision parameters (vector of DG control Step 1: Input MG parameters (e.g., MG topology, load and branch
parameters). information).
Step 2: (initialization of HS): Set t = 0 and generate answer vectors

716
M. Abedini, M. Abedini
in the harmony memory (HM1) to calculate control parameters,
j based on (14) and (15).
xi = x imin + rand (0,1)*(x imax − x imin) j = 1, ..., HMS
Step 3: (load flow): Run load flow for each of answer vectors to
calculate voltage stability, generated power cost and voltage deviation.
Step 4: (fuzzy): Determine the fitness of each objective function
using fuzzy method (13) and calculate μ f1, μ f2 and μ f3 based on f1, f2 and
f3.
Step 5: (fitness evaluation): Use (16) to sort the objective functions.
Step 6: (time updating): Update the time counter, t = t + 1.
Step 7: Generate a new harmony (xnew) in HM1 according to the
following steps:
- Generate a random number (M) between 0 and 1.
- If M < HMCR, then generate variable x newi of the memory con-
sideration (HM1).
With a probability of PAR, modify x newi by a small amount according
to
x newi = x newi + M × BW
- If M > HMCR, then generate variable xnewi by a random choice of
admissible region.
- If the fitness function of the new harmony vector xnew is better than
the worst harmony vector in HM1, then change the vector position in
HM1.
- Save fitness of each vector in HM1
Step 8: (initialization of GA): Generate harmonies for GA by using
HM1 vectors that they used as parents in GA.
Step 9: Make new harmonies by the crossover and mutation op-
erators of GA and then save them in HM2.
Step 10: (Elitism): Generate a new harmony in HM3 according to the
following steps:
- Merge HM1 vectors with the generated population in HM2 of GA.
- Run a load flow algorithm for answer vectors of HM2.
- Determine the fitness of each objective function using the fuzzy
method (14).
- Sort answer vectors in the descending order of the fitness values and
chose the best answer vectors for HM3 based on the fitness of HM1
and HM2 vectors to continue the optimization process.
Step 11: Update HM1 and repository according to the following
steps:
- Swap the position of the harmony vectors in HM1 with vectors found
in HM3
Fig. 2. 33-bus-islanded microgrid system.
Sustainable Cities and Society 38 (2018) 714–722
- Identify all the non-dominated answers and saving in the repository
(17)
Step 12: (End): If the ending criterion is not met go to step 6.
It should be noted that Steps 9–11 are added to the conventional HS
in this paper. To calculate the objective functions, a new load flow al-
gorithm is developed as discussed below.
4.3. Proposed load flow algorithm in autonomous MGs
Load flow algorithms, e.g., Newton-Raphson and Fast-Decoupled
are of the most frequent approach employed to solve the load flow
problem. However, such algorithms may not be effective in distribution
systems and MGs because of divergence of results due to the low value
of X/R in MGs. To cope with such divergence problem, several algo-
rithms have been proposed in the literature (Abdi & Afshar, 2013;
Gnanambal et al., 2011; Huang et al., 2013; Moradi & Abedini, 2012;
Soroudi & Afrasiab, 2012). Unfortunately, the use of such algorithms
faces two issues, i) the slack bus consideration issue, and ii) lack of
(18) droop bus consideration issue. Therefore, it is required to reformulate
the load flow problem in a way that the droop bus is considered rather
than a slack bus. The Appendix A establishes a new algorithm for the
load flow analysis problem in an islanded microgrid to model different
control modes of DGs, including: droop and PQ.
5. Obtained results and discussion
One case study, based on autonomous operation, is studied and si-
mulated in MATLAB environment to evaluate the performance of the
proposed method.
5.1. 33-Bus test system microgrid
The system, a 33-bus MG, shown in Fig. 2, is used (Moradi, Abedini
et al., 2015) with total active and reactive loads of 3.715 MW and
2.30 MVar, respectively. Four DGs are set on buses 26, 22, 25 and 9.
The data of DG units along with the fuel prices are shown in Table 1 and
the parameters of HS-GA are provided in Table 2. The performance of
HS-GA is compared with HS, GA and a case with pre-determined (PD)
DGs parameters where any optimization process is not conducted. For
each method, the proposed load flow algorithm is used. Also for each
method, the Pareto front solution is employed and then the fuzzy ap-
proach for finding the best solution is used. Fig. 3 shows the solution
provided by HS-GA.
Table 3 shows droop control parameters of DGs obtained by HS, GA
and HS-GA, respectively. This table also shows pre-determined values
of such parameters for the PD case. The permissible frequency and
voltage deviation for PD are considered to be 0.5% and 5% of the
717
M. Abedini, M. Abedini Sustainable Cities and Society 38 (2018) 714–722

Table 1 6. Discussion
Data of DG Units Based on Droop Controller For a 33-Bus System.
The main idea is to consider the issues of the conventional methods
DGs Smax (MVA) qMAX (Mvar) Cost ($/kwh)
which are unable to handle isolated MGs analysis. The significance of
1 2 1.20 7.705 the proposed approach contrary the conventional methods can be dis-
2 1 0.60 11.731 cussed as follows.
3 1.5 0.90 10.363
4 0.75 0.45 8.475
6.1. Load flow algorithm (DG operating mode)

Table 2
HS-GA Parameters.
F HMCR PAR BW CR Iteration
0.55 0.8 0.45 0.01 0.33 30
nominal values, respectively. The values of ω* and V* for PD are con-
sidered in the basis of meeting the power demands and security con-
ditions of MG in the load flow algorithm. Table 4 represents the values
of steady state frequency calculated by each method. This table de-
monstrates that all frequency values satisfy the constraints. Table 5
compares the results of the generated power of DGs provided by each
method. Table 6 compares the values of the objective functions calcu-
lated by the proposed method and other methods as well. This table
shows that the HS-GA algorithm provides the lowest (best) values of the
objective functions and PD provides the highest (worst) values as well.
This obtained results guaranty the effectiveness of the proposed
method. It should be noted that if solely cost is considered as the op-
timization objective, then DG1 should be nominated as the most gen-
erated power due to fuel price, shown in Table 1. However, due to other
objectives include voltage profile and stability index, the generated
power from high to low belongs to DG1, DG3, DG4, and DG2, as shown in
Table 5. This demonstrates that the necessity of considering multi-ob-
jective functions in finding optimum operation of DGs. Figs. 4 and 5
show the voltage profile and VSI of the MG for different methods, re-
spectively. Based on Fig. 4, HS-GA algorithm provides the best voltage
profile; however PD represents the worst case condition. This figure
also demonstrates that using optimization methods enhances voltage
profile in stressed buses, especially in bus 18.
In addition, Fig. 5 shows that HS-GA gives the best VSI; however PD
provides the worst condition. This figure also shows that using optimal
parameters for DGs helps in improving VSI.
The first significance of the proposed approach is that the proposed
approach presents a new formula for the load flow problem. The de-
HMS veloped approach is based on assuming the steady state frequency as an
unknown variable (this frequency is firstly calculated by the proposed
20 method and then it becomes known and constant parameter) and in-
troducing droop buses along with PQ buses. Despite the conventional
methods, droop control equations are considered in the corresponded
equations. Consequently, the method is able to deal with the case of
islanded MGs as well. In addition, the power mismatch values for two
cases (Moradi, Abedini et al., 2015) (69-bus and 33-bus) were calcu-
lated as the corresponding results are shown in Fig. 6. This figure de-
monstrates a low power mismatch in the initial iterations and it also
shows after several iterations the power mismatch values converge to
zero. According the method is fast in providing convergence results.
6.2. Droop control parameters setting
The values of ω* and V* for PD were considered in the basis of
meeting the power demands and security conditions of MG in the load
flow algorithm. Table 4 gives the values of the steady state frequency
calculated by each method. This table demonstrates that all frequency
values satisfy the constraints and maintain the system frequency within
the allowable deviation tolerance. Table 5 compares the results of the
generated power of DGs provided by each method. In this case, the
lower generation cost of the islanded MG can be achieved on the ex-
pense of smaller stability. These results demonstrate that the trade-off
between the minimization of the system generation cost and the max-
imization of the system stability. As the DGs reaches their maximum
apparent power capacity, the reactive power production switches from
being governed by the droop equation to being governed by the
available capacity. For instance, Table 5 shows that DG5 produces its
maximum apparent power. Hence, further increase in the active power
results in a decrease in the reactive power production of DG5 to

Fig. 3. Pareto front solution of HS-GA (p.u.).

718
M. Abedini, M. Abedini Sustainable Cities and Society 38 (2018) 714–722

Table 3
Droop control parameters of DGs in 33-bus mg.

DGs HS-GA HS

Sp Sq V* ω* Sp Sq V* ω*

1 5.9894E-4 9.591E-4 1.0040 0.99967 7.4410E-4 1.2116E-3 1.0035 0.99974

2 3.3519E-4 0.01228 1.00831 0.99877 3.0488E-4 0.01795 1.0045 0.99861
3 1.4342E-3 1.5384E-4 1.0082 1.0001 1.3613E-3 1.3326E-4 1.0117 1.00010
4 2.2694E-3 5.8228E-3 1.0014 1.0002 2.4138E-3 5.3435E-3 0.9991 0.99998

DGs GA PD

Sp Sq V* ω* Sp Sq V* ω*

1 9.43396E-4 9.4869E-3 1.0120 1.0002 1.1151E-3 0.025 1.015 1.000

2 6.4189E-4 2.8986E-3 1.0027 0.99889 2.2284E-3 0.050
3 1.1064E-3 1.9608E-3 1.0134 1.0000 1.4863E-3 0.0333
4 2.5877E-3 5.03778E-3 1.0007 1.0005 2.9712E-3 0.0667

Table 4 preserve its active power sharing capability up to the maximum loading
Values of Steady State Frequency. point. The rate of decrease in the reactive power production as the DG
unit reaches its maximum apparent power capacity is governed by the
Methods HS-GA HS GA PD
active power static droop coefficient, i.e., the rate of active power
Steady state frequency 0.99865 0.99851 0.99870 0.99840 production increase.This demonstrates the necessity of considering
multi-objective functions in finding optimum operation of DGs. Figs. 4
and 5 show the voltage profile and VSI of MG for different methods,
Table 5 respectively. Fig. 4 demonstrates that HS-GA gives the best voltage
Generated Power (PDG (MW), QDG (Mvar)) of DGs for different methods.
profile while PD provides the worst. Also, Fig. 4 shows the minimum
Method PDG1 QDG1 PDG2 QDG2 PDG3 QDG3 PDG4 QDG4 and maximum voltages that can occur at the different system buses in
different methods. As can be seen in Fig. 4, the droop settings obtained
HS-GA 1.71 1.04 0.36 0.24 1.01 0.65. 0.68. 0.39. in the optimization methods are capable of satisfying the operational
HS 1.65 1.07 0.33 0.12 1.17 0.75 0.61 0.40
voltage constraints.The results has been shown that proposed method
GA 1.59 1.03 0.29 0.14 1.18 0.77 0.69 0.39
PD 1.44 0.58 0.72 0.49 1.077 0.82 0.54 0.45
can improve operation of the islanded MG as Table 6 reveals that vol-
tage profile and stability of MG have been improved. In PD method,
droop parameters have been not optimized and those values have been
Table 6 set based on supplying the demand. Because, the values of objective
Comparison among objective functions of proposed algorithm and other methods. functions in PD method are higher in comparison with HS-GA.
Objective function HS-GA HS GA PD
6.3. Optimization method (HS-GA)
f1 ($/h) 33586.73 33857.87 33795.06 35250. 32
f2 (p.u.) 1.1103 1.1139 1.1141 1.1533
f3 (p.u.) 0.0401 0.0455 0.0422 0.0724 The third significance is expanding/enhancing the HS algorithm
literature by adding mutation and crossover operators of GA, enabling
HS to explore and exploit the solutions which are more effectively. In

Fig. 4. Voltage profile of MG buses for different methods.

719
M. Abedini, M. Abedini Sustainable Cities and Society 38 (2018) 714–722

Fig. 5. VSI of MG buses for different methods.

Fig. 6. Convergence graph of power mismatch in two case studies.

addition, a fuzzy approach for finding the best solution from a set of
Pareto front answers is employed. See Table 7 to compare different
optimization methods based on three indexes, SD, BM and WV. Table 7
shows the lowest SD index for HS-GA implying the uniformity of results.
This table also shows that the lowest BM and WV indexes for HS-GA
implying HS-GA ability in effective handling interdependence re-
lationships among decision variables. Furthermore, see Table 6, which
demonstrates that HS-GA is capable to provide the lowest values for the
objective functions compared with HS and GA.
Box-plot is a common tool to show distribution of statistical results.
Fig. 7 shows a compression between results calculated of the c metric
for HS-GA and NSGA-II algorithms. The right box shows how many
answers of NSGA-II are dominated by answers of HS-GA and the left box
shows how many answers of HS-GA are dominated by answers of
NSGA-II. The c (NSGA-II, HS-GA) has much lower distribution than the
c (HS-GA, NSGA-II); namely, the answer set of HS-GA has much better
non-dominated performance compared with NSGA-II. It shows that the
proposed method having better performance in finding the approximate
Pareto-optimal front of feasible answers.
7. Conclusion
This paper presented a novel hybrid optimization method to find the
optimal operation of an autonomous MG. The optimization problem is
formulated in terms of a multi-objective problem to minimize the fuel
consumption of DGs and to improve the voltage profile and stability of

Table 7
Statistical analysis results for different methods.

HS-GA HS GA

SD (%) BM WV SD (%) BM WV SD (%) BM WV

f1 ($) 1.41 2447E5 2449E5 2.03 2463E5 2482E5 2.59 2451E5 2474E5
f2 (p.u.) 1.01 1.117 1.204 1.64 1.091 1.286 1.54 1.215 1.375
f3 (p.u.) 0.55 0.0379 0.0394 0.87 0.0861 0.0914 0.94 0.0683 0.0731

720
M. Abedini, M. Abedini Sustainable Cities and Society 38 (2018) 714–722

was developed to solve the optimization problem.The Pareto front of

non-dominated results was provided and the best solution of non-
dominated results was obtained by employing a fuzzy method. This
paper introduced an improvement to the optimization methods con-
cerning the optimal operation of DGs in an autonomous MG by in-
creasing the divergence speed and also solving the problem of sticking
in local optimal solutions. The following subjects are suggested for fu-
ture studies:

1. Improving the proposed islanded power flow algorithm to enable

harmonic power flow consideration in islanded MGs.
2. Expanding a new algorithm to enable for the optimum re-
configuration of islanded MG systems.

Fig. 7. Box-plot of c metric.

MG subject to operation and security constraints. A hybrid algorithm

Appendix A. The Proposed Load Flow Algorithm

(A1) The Proposed Load Flow Algorithm

(A1) Calculation for PQ Buses

In this paper, the model of Venkatesh and Ranjan (2003) is adopted and modified to deal with PQ buses. Following Venkatesh and Ranjan (2003),
the value of the bus voltage magnitude and phase angle can be obtained, respectively as follows:
2 2
V j2 = ⎛⎜ [rP + x (ω ). Q − Vi ]2 − [r 2 + x 2]. [P 2 + Q 2] ⎟⎞ − [rP + x (ω). Q − Vi ]
j j j j j j
⎝ 2 ⎠ 2 (A1)

x (ω) Pj − rQj
δj = δi − sin−1 ( )
Vi Vj (A2)

(A3) Calculation for droop Buses

Unknown variables of a droop bus, as shown in Fig. A1, might be classified as active and reactive powers. The values of power injection of bus j,
shown in Fig. A1, can be given by,
Pj = −PDG + Pload + PT (A3)

Qj = −QDG + Qload + QT (A4)

Active and reactive powers sharing depend on the frequency and local voltage of each DG unit, and can be given by,
1
PDGi = (ωi* − ω)
sp i (A5)
1
QDGi = ( Vi * − Vi )
sqi (A6)
To calculate the receiving end bus phase angle, (A2) is used where Pj and Qj are given by (A3) and (A4). This is valid for the case where Pj and Qj
are within the permitted limit of DG units. Otherwise, Pj and Qj need to be treated as known variables in which their values are equal to their
corresponding limit values.

(A7) Proposed Load Flow Solution in Autonomous MG

The goal of the optimization problem here is to minimize the sum of absolute mismatch values of active and reactive power,

Fig. A1. Injected power of DG to bus j based on droop.

721
M. Abedini, M. Abedini
nbus nbus
f (x , u) = k1 ∑ ΔPi + k2 ∑ ΔQi
i=1 i=1
where x and u are the vectors of state and control variables, and they can be calculated, respectively, by,
x = [δ V PQ L P DR DG Q DR DG]
u = [ω V1 DR.... VN DR]
Here δ and VLPQ
are the voltage angle and magnitude of the PQ bus; are the generated active and reactive power of the droop bus; V DR is
DR
PDG DR
and QDG
the voltage of the droop bus; ω is the system frequency; N is the number of buses with a droop controller; nbus is the number of buses; and k1 and k2
are penalty coefficients for the objective function. In (A7), ΔPi and ΔQi are power mismatches and they can be calculated, respectively, by,
ΔPi = PDGi − Pload i − ∑ Vi Vj Yij (ω)cos(δi − δj − θij )
j (A10)
ΔQi = QDGi − Qloadi − ∑ Vi Vj Yij (ω)sin(δi − δj − θij )
j (A11)
References
Abdi, S. H., & Afshar, K. (2013). Application of IPSO-Monte Carlo for optimal distributed
generation allocation and sizing. International Journal of Electrical Power & Energy
Systems, 44(February (1)), 786–797.
Basu, A., Bhattacharya, A., Chowdhury, S., & Chowdhury, S. P. (2012). Planned sche-
duling for economic power sharing in a CHP-based microgrid. IEEE Transactions on
Power System, 27(February (1)), 30–38.
Dilettoso, E., Rizzo, S. A., & Salerno, N. (2008). SALHE-EA: A new evolutionary algorithm
for multi-objective optimization of electromagnetic devices. Intelligent Computer
Techniques in Applied Electromagnetics, 119, 37–45.
Fetanat, A., & Khorsaninejad, E. (2015). Size optimization for hybrid photovoltaic-wind
energy system usingant colony optimization for continuous domains based in-
tegerprogramming. Applied Soft Computing, 31, 196–209.
Geem, Z. W., Kim, J. H., & Loganathan, G. V. (2001). A new heuristic optimization al-
gorithm: Harmony search. Simulation, 76(February (2)), 60–68.
Gnanambal, K., Marimuthu, N. S., & Babulal, C. K. (2011). Three-phase power flow
analysis in sequence component frame using Hybrid Particle Swarm Optimization.
Applied Soft Computing Journal, 11(2), 1727–1734.
Green, T., & Prodanovic, M. (2007). Control of inverter-based microgrids. Electric Power
Systems Research, 77, 1204–1213.
Huang, C. L., Huang, W. L., Chang, H. Y., Yeh, Y. C., & Tsai, C. Y. (2013). Hybridization
strate-gies for continuous ant colony optimization and particle swarm optimizatio-
napplied to data clustering. Applied Soft Computing, 13, 3864–3872.
Katiraei, F., Iravani, R., Hatziaragyriou, N., & Dimeas, A. (2008). Microgrids manage-
ment. IEEE Power and Energy Magazine, 3(May-June), 54–65.
Khorramdel, B., & Raoofat, M. (2012). Optimal stochastic reactive power scheduling in a
microgrid considering voltage droop scheme of DGs and uncertainty of wind farms.
Energy, 45, 994–1006.
Kuznetsova, E., Li, Y., Ruiz, C., & Rio, E. (2014). An integrated framework of agent-based
modelling and robust optimization for microgrid energy management. Applied Energy,
129, 70–88.
Lasseter, R. H., & Piagi, P. (2004). Microgrid: A conceptual solution. Proc. of 35th annual
IEEE power electronics specialists conference.
Li, G., & Shi, J. (2012). Agent-based modeling for trading wind power with uncertainty in
the day-ahead wholesale electricity markets of single-sided auctions. Applied Energy,
99, 13–22.
Mitra, J., Vallem, M. R., & Patra, S. B. (2006). A probabilistic search method for optimal
resource deployment in a microgrid. Proc. 9th int. conf. probabilistic methods applied to
power systems (pp. 1–6).
Moradi, M. H., & Abedini, M. (2012). A combination of genetic algorithm and particle
722
Sustainable Cities and Society 38 (2018) 714–722
(A7)
(A8)
(A9)
swarm optimization for optimal DG location and sizing in distribution systems.
International Journal of Electrical Power & Energy Systems, 34(January (1)), 66–74.
Moradi, M. H., Abedini, M., & Hosseinian, S. M. H. (2015). A combination of evolutionary
algorithm and game theory for optimal location and operation of DG from DG owner
standpoints. IEEE Transaction on Smart Grid, (March (7)), 608–616.
Moradi, M. H., Eskandari, M., & Hosseinian, S. M. H. (2015). Operational strategy opti-
mization in an optimal sized smart microgrid. IEEE Transaction on Smart Grid, 6(May
(36–38)), 1087–1095.
Nekooei, K., Farsangi, M. M., Nezamabai, H., & Lee, K. Y. (2013). An improved multi-
objective harmony search for optimal placement of DGs in distribution systems. IEEE
Transactions on Smart Grid, 4(March (1)), 557–567.
Oyarzabal, J., Jimeno, J., RuelaEngler, J., & Hardt, C. (2005). Agent-based-micro-grid-
management-system. International conference on future power systems.
Papageorgiou, E. I. (2011). A new methodology for decisions in medical informatics using
fuzzy cognitive maps based on fuzzy rule-extraction. Applied Soft Computing, 11(1),
500–513.
Pipattanasomporn, M., Willingham, M., & Rahman, S. (2005). Implications of on-site
distributed generation for commercial/industrial facilities. IEEE Transactions on Power
Systems, 20(February (1)), 206–212.
Rezaei Mozafara, M., Moradi, M. H., & Amini, M. H. (2016). A simultaneous approach for
optimal allocation of renewable energy sources and electric vehicle charging stations
in smart grids based on improved GA-PSO algorithm. Sustainable Cities and Society,
32, 627–637.
Schmidt, S., Steele, R., Dillon, T. S., & Chang, E. (2007). Fuzzy trust evaluation and
credibility development in multi-agent systems. Applied Soft Computing, 7, 492–505.
Soroudi, A., & Afrasiab, M. (2012). Binary PSO-based dynamic multi-objective model for
distributed generation planning under uncertainty. IET Renewable Power Generation,
6(June (2)), 67–78.
Vasiljevska, J., Peças Lopes, J. A., & Matos, M. A. (2012). Evaluating the impacts of the
multi-microgrid concept using multi criteria decision aid. Electric Power Systems
Research, 91(7), 44–51.
Venkatesh, B., & Ranjan, R. (2003). Data structure for radial distribution power flow
analysis. IEE Proceedings - Generation, Transmission and Distribution, 150(January),
101–106.
Zheng, M., Meinrenken, C. J., & Lackner, K. S. (2014). Agent-based model for electricity
consumption and storage to evaluate economic viability of tariff arbitrage for re-
sidential sector demand response. Applied Energy, 126, 297–306.
Zoka, Y., Sugimoto, A., Yorino, N., Kawahara, K., & Kubokawa, J. (2007). An economic
evalua-tion for an autonomous independent network of distributed energy resources.
Electric Power Systems Research, 77(7), 831–838.