Electrical Power and Energy Systems 121 (2020) 106134

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

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

A hierarchical energy management system for islanded multi-microgrid T

clusters considering frequency security constraints
Mohammadreza Mazidia, , Navid Rezaeib, Fatemeh Jahanbani Ardakania, Maryam Mohitia,
⁎

Josep M. Guerreroc
a
Department of Electrical Engineering, Yazd University, Yazd, Iran
b
Department of Electrical Engineering, University of Kurdistan, Sanandaj, Iran
c
Department of Energy Technology, Aalborg University, Aalborg, Denmark

ARTICLE INFO ABSTRACT

Keywords: With the widespread development of microgrids (MGs) in future smart distribution networks, a number of
Energy and reserve scheduling neighboring MGs can be connected and form a multi-microgrid (MMG) cluster. In this regard, the energy
Energy management system management of a MMG is challenging due to more complex components and higher degrees of uncertainty in a
Islanded multi-microgrid clusters small region of power system. Likewise, in the islanded MMG (IMMG) clusters, due to the low-inertia and high
Hierarchical control
intermittent energy delivery of renewable resources, the frequency security should be considered in the energy
Two-stage stochastic optimization
management. To address this issue, this paper proposes an energy management system (EMS) in which hier-
archical control structure of IMMG clusters is precisely modeled. The proposed EMS aims to minimize total
operation cost of IMMG cluster while sufficient primary and secondary reserves are scheduled to preserve fre-
quency security in a predefined range. Besides, the proposed EMS provides optimal strategies for MGs to ex-
change energy and reserves during scheduling horizon. To consider operational uncertainties, the proposed EMS
is formulated as a two-stage stochastic mixed-integer linear programming problem that guaranties the global
optimal solution. The obtained results verify that through the proposed EMS, total operation cost of the IMMG
cluster is minimized while the frequency can be cost-effectively preserved within a pre-defined secure range.

1. Introduction mode [6]. Likewise, due to the high investment cost of transmission
lines, microgrids in remote areas are constructed to be standalone and
Recently, electrical power systems in response to different needs operated in islanded mode [7].
such as improving the reliability of power system and operational ef- To reach the economic benefit, reduce the operation cost, and in-
ficiency of energy supply, enhancing the security and quality of power crease the security of the system, it is proposed to connect a number of
system, and decreasing the environmental emissions, have experienced neighboring MGs and form a multi-microgrid (MMG) system [8]. These
many changes. In the meantime, governments try to enhance diversity goals are achieved by implementing an effective energy management
of fuel resources by using distributed energy resources (DERs), as they system (EMS) in the MMG as a centralized or decentralized hierarchical
are utilized by power system operators more ever [1,2]. Individual control scheme [9]. The EMS schedules the references of each MG and
distributed generations (DGs) with intermittent renewable sources will dispatchable DGs. Also, for sharing surplus energy, EMS uses the con-
pose more challenges in the power systems as they may solve [3]. trollable loads to manage the power production and energy consump-
Therefore, the idea of microgrid (MG) is proposed to cope with the tion between MGs [10]. In [11], an EMS is developed for a smart dis-
interconnection problems of these DGs to the power system [4]. A MG is tribution grid including residential MGs to minimize total operation
a controlled small-scale power system which consists of small-scale cost in which power losses and voltage profile have been addressed.
DGs, loads, energy storage elements, and control units, serving within a Authors of [12] have proposed a central energy management system
defined area to facilitate power supplementation and/or improved (CEMS) framework as it applies a hierarchical bi-level control scheme
power quality to the consumer’s premises [5]. MGs usually are operated for the utility and multiple electrically coupled MGs. In this paper,
in the grid-connected mode. However, following a disturbance in the primary and secondary level controllers are used. In the primary level,
grid, the MG may be separated from the grid and operate in the islanded each MG operates separately to calculate the power set-points of

⁎
Corresponding author.
E-mail address: [email protected] (M. Mazidi).

https://doi.org/10.1016/j.ijepes.2020.106134
Received 6 January 2020; Received in revised form 10 April 2020; Accepted 19 April 2020
0142-0615/ © 2020 Elsevier Ltd. All rights reserved.
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

Nomenclature s Probability of scenarios

VLLi, t Value of loss of load of control level i at hourt
Acronyms rampjup Ramp-up limit of DG j
rampjdown Ramp-down limit of DG j
BTB Back-To-Back converter UTj Minimum up-time of DG j
CEMS Central Energy Management System DTj Minimum down-time of DG j
EMS Energy Management System ¯
PwWT Maximum active power generation of WTw
GAMS General Algebraic Modelling System ¯
P PV Maximum active power generation of PV p
IMMG Islanded Multi Microgrid
p
fm, i Minimum allowable frequency excursion limit of MG m
MG Microgrid _
DG Distributed Generation during control leveli
DER Distributed Energy Resource f¯m, i Maximum allowable frequency excursion limit of MG m
VLL Value of Loss of Load during control leveli
PV Photovoltaic system
WT Wind Turbine Variables
MGCC Microgrid Central Controller
u j, t Binary variable indicating commitment state of DG j at
Indices hourt
u j, s , i , t Commitment of DG j in scenario s and control level i at
j DG(m) Index of DGs belong to MGm hourt
w WT(m) Index of WTs belong to MGm uON
j, t Binary variable indicating turn-on state of DG j at hourt
l Tie(m : m') Index of tie-lines between MG m andm' uOFF
j, t Binary variable indicating turn off state of DG j at hourt
p PV(m) Index of PVs belong to MGm P jDG
,t Active output power of DG j at hourt
t NT Index of energy management hours P jDG
, s, i, t Active output power of DG j in scenario s and control level
m MG Index of MGs i at hourt
i CI Index of hierarchical control level could be equal to pri RUjDG
, i, t Deployed upward reserve of DG j in control level i at hourt
(primary), sec (secondary), and ter (tertiary) RDjDG
, i, t Deployed downward reserve of DG j in control level i at
s NS Index of scenarios hourt
P jDG
, s, i, t Active power deviation of DG j in scenario s and control
Parameters level i at hourt
PlTie
, s, i, t Active power deviation of tie-line l in scenario s and
Dm, s, t Frequency elasticity of load in microgrid m in scenario s at control level i at hourt
hourt PwWT
, s , i, t Active power deviation of WT w in scenario s and control
Rj Droop gain of DG level i at hourt
f0m Reference frequency in MGm PpPV
, s , i, t Active power deviation of PV p in scenario s and control
level i at hourt
PmL, t Load demand of MG m at hourt
fm, s, i, t Frequency deviation of MG m in scenario s and control
Fixed operation cost of DG j
j
level i at hourt
First order operation cost of DG j
j
LSm, s, i, t Load shedding amount of MG m in scenario s and control
ON
Start-up cost of DG j
j level i at hourt
OFF
j Shut down cost of DG j P jDG , ref
Reference power deviation of DG j in scenario s and
, s, i, t
j, i Cost of providing reserve by DG j in control intervali control level i at hourt
WT
w Cost of power generation by WTw f mref, s, i, t Reference frequency deviation of MG m in scenario s and
PV
p Cost of power generation by PV p control level i at hourt
µl Cost of exchanging energy throughout tie-linel PmL, s, i, t Active power deviation of MG m in scenario s and control
l, i Cost of exchanging reserve throughout tie-line l in control level i at hourt
intervali

generation resources, power generation shortage or surplus in the MGs, using the robust optimization. A hybrid robust distributed model for
and the constraints. The calculated data in the primary level is used in operation of MMG systems is proposed in [16]. To model the renewable
the secondary level to decide about the surplus or shortage of the MGs. generation and load uncertainties, the risk-averse information gap de-
The optimal operation of MMGs by a cooperative energy and reserve cision theory approach is applied. Also, to model the day-ahead market
scheduling model based on game theory is proposed in [13]. In that price uncertainty, a robust optimization approach is used.
study, the energy and reserve can be utilized among MMGs. Moreover, The energy management of a MMG system is challenging due to
wind turbines (WTs) and photovoltaics (PVs) are considered as un- more complex components and higher degrees of uncertainty in a small
certain parameters and the popular robust optimization approach is region of power system. Large load increases/renewable generation
used for solving the problem. A decentralized model is presented in decreases followed by an inadequate system response in the absence of
[14] which the operator of distribution network interacts with con- high inertia, may lead to frequency excursions and endangers the secure
trollers of MMG system to minimize total operation cost. However, the operation of islanded MMG (IMMG) clusters [17,18]. In grid connected
uncertainty has been ignored in the proposed model. The authors of operation mode unlike the islanded mode, the variations in loads and
[15] have proposed a decentralized robust model for optimal operation renewable generations can be compensated by exchanging power with
of MMG system within a distribution network. The proposed model the main grid and therefore, frequency of MMG system is controlled
independently coordinates operation of MMG system with the dis- within the secure range. However, in IMMG clusters frequency devia-
tribution network to minimize total operation cost. Meanwhile, the tions should be taken into consideration in the EMS to preserve fre-
uncertainty of renewable generation and load demand are considered quency within secure range. The frequency security of MMG clusters

2
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

can be further improved by exchanging power with neighboring mi- Authors of [29] have proposed a droop based strategy for frequency
crogrids to balance power and supply. In [19], a model predictive control of IMMGs. In the proposed model adjacent MGs are connected
control strategy is proposed to regulate voltage, frequency, and power with the use of AC/DC converter. Accordingly, the MGs power to im-
of the inverters in the IMMG clusters. The goal of that study is defined prove frequency security of system. In this paper, the tertiary control
as increasing in the reliability and security of the system. In [20] a level is not adapted to optimize the operational cost of the IMMGs.
decentralized energy management system is proposed for an IMMG Likewise, reserve scheduling which is used to guarantee the frequency
cluster which each MG has different frequency requirement according security of system has not been considered.
to load configurations to improve reliability from the economical Careful review of previous works indicates that most of them mainly
viewpoint. In this paper the hierarchical control structure of the system have focused only on energy scheduling of IMMG clusters while reserve
and uncertainty are not considered. A similar model based on dis- scheduling which is used to guarantee the frequency security of system
tributed cooperative control and adaptive neural networks is proposed has not considered. Moreover, in the most of studies, for the sake of
in [21] for optimal voltage and frequency regulation in an IMMG which simplicity, the frequency security is not considered. It is notable that
the stability of the system is guaranteed. In [22], a CEMS is proposed the security of the IMMG cluster is reflected by the frequency whose
for optimal operation of MMG system in grid connected and islanded proper control can cause of achieving the economic targets. Also, the
modes. The objective of the proposed model is to minimize operation hierarchical control structure and the operational uncertainties is not
cost of the system in grid conceded mode and minimize total load modeled by the studies, simultaneously. To fulfill these gaps, this paper
shedding in islanded mode. In [23], a multi-agent based distributed proposes an EMS in which hierarchical control structure of IMMG
optimal tie-line power flow control strategy is proposed to coordinate cluster is precisely modeled. The aim of the proposed EMS is to mini-
the MGs in an IMMG cluster load fluctuation while guaranteeing op- mize total operation cost of IMMG cluster while sufficient primary and
erational stability. In [24], a daily energy management strategy is secondary reserves are scheduled to preserve frequency security in a
proposed to determine optimal energy supply in IMMG clusters with predefined range. Besides, the proposed EMS provides optimal strate-
minimum cost. In this paper, the frequency security and operational gies for the MGs to exchange energy and reserves during scheduling
uncertainties have not been considered. Authors of [25] have proposed horizon. To consider operational uncertainties, the proposed EMS is
a bi-level EMS for a networked MMG system in which the outer-level formulated as a two-stage stochastic mixed-integer linear programming
EMS is aimed to exchange the required information and power between problem that guaranties achieving the global optimal solution. Table 1,
the interconnected MGs, and the inner-level EMS is intended for energy presents the comparison between the proposed model and other ex-
scheduling of each on-fault MG in case of separation from other MGs. In isting methodologies.
[26], a robust optimization framework is proposed to determine the The main contributions of this study are summarized as following:
day-ahead energy and reserves between the constituent MGs of the
MMG system. Authors of [27] have proposed a pricing mechanism in • The joint energy and reserve scheduling problem are modeled in an
MMG system to minimize local cost of MGs via energy trading with islanded multi-microgrids clusters using two-stage optimization
other MGs/main grid. In [28], a hierarchical frequency control scheme method.
is proposed for IMMGs. The proposed control scheme includes three • The hierarchical control structure of IMMG cluster is model pre-
levels including distributed management system, central autonomous cisely to guarantee the frequency security of MGs in a cost-effective
management controller, and microgrid central controller which aims to manner.
preserve the primary frequency of the system within secure range by
adjusting the set points of DGs and controllable loads of each microgrid. The rest of the paper is organized as follows: system configuration
In this paper, the secondary and tertiary control levels of microgrids are and the formulation of the IMMG is presented in Section 2. Section 3
ignored. Moreover, the reserve scheduling has not been considered. introduces the model formulation for the objective function and

Table 1
Comparison of proposed model with existing approaches.
Reference Configuration of MMG Hierarchical control structure Uncertain parameters Operational constraints
system

Connected Isolated Primary Secondary Tertiary Wind Solar Load Energy Reserve Frequency Exchange MILP
mode mode power power demand scheduling scheduling security between format
MGs

[9] * * * * *
[10] * * * * * *
[11] * * * * *
[12] * * * * * * * *
[13] * * * * * * *
[14] * * * * * *
[15] * * * * * *
[16] * * * * * *
[19] * * * * *
[20] * * * * *
[21] * * * * *
[22] * * * * *
[23] * * * * * *
[24] * * *
[25] * * * * * * *
[26] * * * * * * * *
[27] * * * *
[28] * * *
[29] * * * * * *
This paper * * * * * * * * * * * *

3
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

constraints of the scheduling energy and reserve problem. Numerical as the hourly frequency. Accordingly, an hourly frequency excursion in
results and simulations are discussed in Section 4. Finally, Section 5 is the primary and secondary control intervals can be extracted for the
devoted to some concluding remarks. MGs and included in to the day-ahead energy and reserve scheduling
problem. Therefore, to schedule the hourly primary and secondary re-
serves the steady-state hourly values of the system frequency should be
2. System configuration and formulation
used. In other words, since average hourly load variations can be
modeled, corresponding average hourly frequency excursions can be
Fig. 1 shows the configuration of an IMMG cluster including two
also defined after any active power deviations. Indeed, to schedule the
MGs which are connected through a back-to-back (BTB) converter. In
primary and secondary reserves it can simply assume that at ‘t0′, the
order to control and management the IMMG cluster the use of a hier-
maximum power imbalance has been occurred and after that there is no
archical control scheme is necessary. Similar to the conventional power
greater deviation during that hour. Accordingly, the hourly average
systems, secure operation of the IMMG cluster can be accomplished
value of the system frequency excursion can be extracted and assigned
through a hierarchical three control levels namely primary, secondary,
to whole of that hour and consequently utilized for scheduling the
and tertiary. The aim of primary control level is to share the active
primary and secondary reserves. Notably, the scheduled primary re-
power among DG units by drooping system frequency and to provide a
serves are deployed instantaneously within some seconds after t0. The
stable balance between the generation and demand in the MG. Second
assumption is that primary reserve can be deployed at a very fast rate,
control level tries to optimistically restore the frequency of the MG to its
sufficiently so as to stabilize the system frequency within 5–10 s, and
nominal value, which is performed by the microgrid central controller
that it can maintain the system power in balance for several minutes
(MGCC). Meanwhile, the tertiary control level is in the charge of
until the slower secondary reserves can take over.
managing exchanged powers between the MGs using BTB converters.
In the following of this section, each control level is explained in
The CEMS has the most important role in coordination of the hier-
details with corresponding mathematical formulations. In the derived
archical control levels. Indeed, the CEMS provides a synergy in a way
formulations it is assumed that all the transient and oscillatory modes
that not only the operation cost of IMMG cluster is minimized but also
are damped and the frequency of the MGs are settled to the steady state
the frequency security is preserved in case of WTs, PVs, and load
which is considered as hourly frequency. Then, the proposed model for
fluctuations. To this end, the scheduled energy and regulation reserves
energy and reserve scheduling of the IMMG cluster considering fre-
as well as the exchanged energy and reserves between MGs are opti-
quency security constraints is presented in the next section.
mized by the CEMS and sent to the MGs and BTB converters. Note that
the use of BTB converters for interconnecting MGs provides the ability
to control the frequency of MGs independently [30]. 2.1. Primary control level
To obtain the proposed model, it is assumed that all the transients
and oscillatory modes are damped. Indeed, the frequency excursions in The frequency excursion of MGs may occur due to the power var-
the MGs are settled into a steady-state equal point which is considered iations of WTs, PVs, and load. To compensate these variations, the

MG MG
f

DG2
p
Converter 1 Converter 2

v abc 1 v abc 2
f f

DG1
v dc
DG1

p p

f
SVPWM SVPWM
DG2

p
ud 1 uq 1 ud 2 uq 2
Current controller 1 Current controller 2
i dref 1 i qref 1 i dref 1 i qref 1
dc-voltage Tie-line power Q2
Q1 control
control control control
*
v dc Q1* Ptit* Q2*
MGCC MGCC
Back to Back
converter

CEMS

Fig. 1. Configuration of an IMMG cluster.

4
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

committed converter based DGs are automatically adjusted and ap- preserving the frequency in a secure range. To this end, the MGCC re-
propriate amount of power is exchanged with neighboring MGs. adjusts the reference power of converter based DGs to restore frequency
Accordingly, the supply-demand balance in each MG of IMMG cluster of the MG to the nominal value or control it in a more secure range
during primary control level can be described by the following equa- which is shown in Fig. 3. Likewise, the MG can exchange power with
tion: neighboring MGs to improve frequency restoration process during
secondary control level. According to Fig. 3, the static frequency control
( P jDG L
, s, i, t . uj, s, i, t )|i = pri = Pm, s, i, t |i = pri PlTie
j DG(m) l Tie
, s, i, t
function of converter based DGs in the secondary control level can be
(m : m')
written as follows:
|i = pri PwWT
, s, i, t |i = pri PpPV
, s, i, t
w WT(m) p PV(m) 1
P jDG
, s, i, t |i = sec = uj, s, i, t |i = sec P jDG , ref
, s, i, t |i = sec f |i = sec ;
|i = pri + Dm, s, t . fm, s, i,t |i = pri LSm, s,i, t Rj m, s, i, t
|i = pri ; m MG, j DG(m) , s NS , t NT (1) m MG, j DG(m) , s NS , t NT (10)
Note that the natural inertia of converter based DGs is much smaller
Where,
than a synchronous generator. Thus, to mimic the governor of a syn-
chronous generator, virtual inertia is added to the converter-based DGs P jDG
, s, i, t |i = sec = P jDG
, s, i, t |i = sec P jDG
, s, i, t |i = pri ; j DG(m) , s NS , t NT
by using a power-frequency (P/f) droop controller which can be de- (11)
scribed as follows:
P jDG , ref
, s, i, t |i = sec = P jDG , ref
, s, i, t |i = sec P jDG , ref
, s, i, t |i = pri ; j DG(m) , s NS, t NT
Rj ( P jDG
, s , i, t P jDG , ref
, s , i, t )
(12)
= uj, s, i, t ( f mref, s, i, t fm, s, i, t ); j DG(m) , s NS, i CI , t NT (2)
fm, s, i, t |i = sec = fm, s, i, t |i = sec f m0 ; m MG, s NS, t NT (13)
It should be mentioned that the reference values of converter based
According to Eqs. (6), and (10)–(13), the frequency excursion in the
DGs and frequency of MG can be only changed by the MGCC. Since the
secondary control level can be retrieved as follows:
primary control level has a fast response time, the primary control
functions are performed within a few seconds and therefore, the MGCC fm, , i, t |i = sec
has not enough time to alter reference values of DGs. Meanwhile, the
[P jDG , ref
, s, i, t |i = sec P jDG
, s, i, t |i = sec ]
commitment of converter based DGs remains constant during primary j DG(m )
= ; m MG, s NS, t NT
control level. Therefore: Dm, s, t +
uj, s, i, t |i = sec
j DG(m ) Rj
P jDG , ref
, s, i, t |i = pri = 0; j DG(m) , s NS, t NT (3) (14)
It should be noted that the generated powers of DGs during sec-
f mref, s, i, t |i = pri = 0; m MG, s NS, t NT (4) ondary control level can be obtained using power-supply balance
uj, s, i, t |i = pri = uj, , t ; j DG(m) , s NS , i CI , t NT (5) equation which is represented as follows:

Accordingly, Eq. (2) can be rewritten as follows: ( P jDG L

, s, i, t . uj, s, i, t )|i = sec = Pm, s, i, t |i = sec PlTie
, s, i, t |i = sec
j DG(m ) l Tie
(m : m')
P jDG
, s, i, t |i = pri
1 PwWT
, s, i, t |i = sec PpPV
, s, i, t |i = sec + Dm, s, t . fm, s, i, t
= ( f . uj, s, i, t )|i = pri ; m MG, j DG(m) , s NS, t NT
Rj m, s, i, t w WT(m) p PV(m)

(6) |i = sec LSm, s, i, t |i = sec ; m MG, j DG(m) , s NS, t NT (15)

Where,
2.3. Tertiary control level
P jDG DG
, s, i, t = P j, s, i, t |i = pri P jDG
,t ; j DG(m) , s NS , t NT (7)

f m0 ; The tertiary control level aims to manage exchanged powers

fm, s, i, t = fm, s, i, t |i = pri m MG, s NS, t NT (8)
Eq. (6) describes the static frequency control function of converter f
based DGs during the primary control level which is shown in Fig. 2. As
can be seen, the DG releases the scheduled primary reserve based on its
droop characteristic to compensate frequency excursion following f max
power imbalance in the MG.
The frequency excursion in the primary control level can be re- f 0
pri
trieved from Eqs. (1) and (6) as follows: f
fm, , i, t |i = pri = f min
PmL, s, i, t l Tie
PlTie
, s , i, t
(m : m')
|i = pri
w WT(m )
PwWT
, s, i, t p PV(m )
PpPV
, s, i, t LSm, s, i, t

Dm, s, t +
uj, s, i, t |i = pri
P DG , pri
j DG(m) Rj

; m MG, j DG(m) , s NS, t NT (9)

P
DG DG
Pmin P DG P DG , pri Pmax
2.2. Secondary control level
Fig. 2. Static frequency control function of converter based DGs in the primary
The MGCC should guarantee secure operation of the MG by control level.

5
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

3. Model formulation

The CEMS aims to schedule optimal energy and reserve in the IMMG
cluster to meet the operational objective in an economic and secure
way. This task is more challenging in the islanded systems in which the
P DG ,ref

frequency deviation
unforced uncertainties endanger the frequency security and therefore,

Secure range of
0
PsecDG ,ref the energy and reserve scheduling. To overcome this issue, a two-stage
f
f sec stochastic model for CEMS is utilized in this paper to minimize total
PpriDG ,ref operation cost of IMMG cluster while frequency security of the MGs is
f
f pri preserved in a predefined range. In this section, first the operational
uncertainties are modeled and then the detailed formulations of the
CEMS including objective function and constraints are presented.

3.1. Uncertainty modelling

DG
P
The forecasted uncertainties associated with power generation of
WTs and PVs as well as load demand of MGs are usually modeled by
P DG PsecDG PpriDG
continuous probability distribution functions (PDFs). However, using
P the PDFs in the continuous format makes the optimization problem very
Fig. 3. Static frequency control function of converter based DGs in the sec- complicated [31]. Therefore, the continuous PDF of each uncertain
ondary control level. variable is divided into discrete intervals to form multiple states and
then combined, to generate a scenario set. It should be mentioned that
each state represents an expected value for the uncertain variable with
between the MGs using BTB converters. As shown in Fig. 1, the BTB
its occurrence probability. The expected value is considered as the
converter includes two voltage source converters by which the fre-
middle of the interval while its probability can be easily calculated by
quency of each MG can be controlled separately. The first converter
integration.
regulates the voltage of dc-link, while the second converter controls the
To combine the different discrete PDFs, the scenario tree is used in
amount of power which is exchanged between the respective MGs. It
this paper [32]. As can be seen in Fig. 5, each scenario includes three
should be mentioned that the proportional–integral regulator can be
different states of wind power, solar power, and MG load at each hour
used for both inner control loops.
of the scheduling horizon. Likewise, a weight s is assigned to each
The static control function of BTB converter in the tertiary control
scenario that indicates the probability of occurrence. As the large
level is shown in Fig. 4. Initially, both MG1 and MG2 are operated in
number of scenarios makes the optimization problem insolvable, a
nominal frequency, i.e., f 0 . However, following a power shortage in
scenario reduction method should be applied. In this paper, the
MG1 which reduces its frequency to fMG1, the generated powers of
SCENRED tool which is conducted by the General Algebraic Modelling
converter based DGs in MG2 are increased and transferred power from
System (GAMS) is applies to reduce the number of scenarios with
MG2 to MG1 through BTB converter increases to compensate the power
keeping the properties of the original scenarios [32].
imbalance in cooperation with the converter based DGs of the MG1. As
a result, the frequency of MG1 increases to f MG1'
which is in the secure
range. On the other hand, the frequency of MG2 slightly decreases from 3.2. Objective function
f 0 to f MG2
'
which is also in the secure range. Therefore, the frequency of
both MGs are stabilized in the new steady state values which are in the The CEMS schedules energy and regulation reserves as well as the
predefined secure range. exchanged energy and reserves between MGs to minimize total opera-
tion cost of IMMG cluster while the frequency security is preserved

f
f
MG1 MG2

0
f f 0
f MG1 f MG 2

f MG1

P Tie P Tie

DG , ref
PMG1
DG
PMG1
DG
PMG1
P DG , ref
PMG 2
DG
PMG 2
P
Fig. 4. Static frequency control function of BTB converter in the tertiary control level.

6
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

p1 S1
p2 S2
w1 p3 S3
w2 p4 S4
d1 w3 p5 S5
w4 .
.
w5
.
w1 .
w2
.
.
d2 w3 .
w4 .
.
w5
.
w1 p1 S61
w2 p2 S62
d3 w3 p3 S63
w4 p4 S64

Scenario set {S1,S2,…,S175}

w5 p5 S65
.
w1
.
w2 .
d4 w3
.
.
w4 .
w5 .
.
w1 p1 S111
w2 p2 S112
d5 w3 p3 S113
w4 p4 S114
w5 p5 S115
.
w1
.
w2 .
d6 w3
.
.
w4 .
w5 .
.
w1 .
.
w2
.
d7 w3 p1 S171
w4 p2 S172
w5 p3 S173
p4 S174
p5 S175
Load demand states Wind power states Solar power states
Hour h
Fig. 5. Scenario tree for combination of wind power, solar power, and load forecasted values.

considering uncertainties of WTs, PVs, and load. Due to uncertainties in forecasted values of uncertain variables are used. Because the operation
load and renewable generations the performance of the proposed en- has not happened yet, the second-stage decisions (e.g., deployment
ergy management system should be evaluated under a wide range of reserves of controllable DGs in primary and secondary intervals, load
operation points which can be modeled using the stochastic optimiza- shedding, deployment reserves which are provide by MGs) are made
tion method. In the proposed EMS there are two different kinds of de- considering uncertainties throughout scenarios to compensate any bad
cision horizon (day-ahead scheduling and operation) which leads to a effects that might have been experienced as a result of the first-stage
two-stages stochastic optimization model. The first-stage decisions decision. The two-stage stochastic optimization method helps ensuring
(e.g., commitment and set points of controllable DGs, primary and the reserve requirements such a way the microgrid security and eco-
secondary reserves of controllable DGs, and exchanged power and re- nomic aspects preserved appropriately. Accordingly, the objective
serves between MGs) are made a day before operation which the function is defined through the following expression:

7
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

the MGs are calculated in fifth and sixth terms, respectively. The se-
Start venth term represent the expected cost of DGs. Meanwhile, the cost of
exchanged power between the MGs is considered in eighth term. The
last term indicates the expected cost of unavoidable load shedding.
Level 1: Scenario generation and reduction
The primary reserve is provided by the generating units through
Forecast average of wind speed, solar their local droop characteristic in response to system frequency devia-
irradiation, and load demand tions from nominal, a closed-loop process denoted as primary frequency
control reserves. This is the fastest of the reserve control strategies with
a response time of the order of seconds [5]. For more severe dis-
Estimate appropriate PDFs for uncertain turbances, secondary reserve provides generation to return the area-
variables control error to zero, as well as to ensure that all operational con-
straints, such as line flow, voltage magnitude limits and economic-en-
vironmental issues can be satisfied. Secondary control reserves are
Generate scenarios based on forecasted centrally implemented and have a response time of the order of minutes
Day-ahead scheduling

average of uncertain variables and estimated [33]. Thus, the MGCC in each hour of the proposed day-ahead security
PDFs using scenario tree method constrained energy management system should measure the last data
corresponding to the largest steady-state point of the frequency excur-
sions amongst the reduced scenarios. Then on the basis of the mon-
Reduce scenario to Ns using SCENRED tool
itored information during the real-time operation in the primary control
by GAMS
level, in an average 15 min time horizon, the MGCC is in charge to
optimize the optimal power settings of the droop controlled converter
Level 2: Execute the proposed energy management based DGs such not only the microgrid frequency is managed in a more
model secure range but also the economic requirements of the microgrid are
Time horizon

Minimize: (15) satisfied appropriately. Then the optimal power set-points are sent to
Subjected to: (16)-(49) the DGs. In fact, the MGCC monitors the microgrid frequency profile in
an online and real time manner. After measuring the requiring data, the
Power and commitment of DGs, MGCC runs a security constrained unit commitment problem in at most
exchanged power between MGs, reserve 15–30 min to ensure the microgrid techno-economic operational
of DGs, exchanged reserve between MGs planning. The MGCC through monitoring the microgrid energy flows
in primary and secondary control levels tries to schedule the droop based local controllers in such a way that
there will be enough reserve resources to cover the possible system
frequency excursions caused stochastically by the generation and load
Level 3: Deploy scheduled energy and reserve deviations. Hence, droop based local controllers automatically release
their primary reserve capacities in about 10–60 s, then, the MGCC by
Minimize: Imbalances between generation applying a security constraint unit commitment program adjusts the set
and supply points of the dispatchable DG units to minimize the microgrid steady-
Subjected to: (16)-(49) state frequency excursions caused by the local controllers’ primary
Real-time operation

control action.

Deploy scheduled reserve of DGs and 3.3. Constraints

exchange reserve between MGs in the
primary and secondary control levels to The CEMS should consider several technical constraints to ensure
manage frequency security of MGs in the the secure operation of the IMMG cluster which are presented in the
IMMG system. following of this subsection.

3.3.1. Constraints of first stage optimization

Fig. 6. Flowchart of proposed CEMS. In the first stage, the energy and regulation reserves of each MG and
the exchanged energy and reserves between the MGs are determined.
The following constraints should be considered in this stage:

Min [( j. uj, t + j. P jDG

,t ) +
ON
j × uON
j, t +
OFF
j × uON
j, t ] + P jDG
,t + PlTie
,t + PwWT
,t + PpPV
,t
m MG t NT j DG(m) j DG(m ) l Tie w WT(m) p PV(m)
(m : m')

[ j, i × (RUjDG DG
, i, t + RD j, i, t )] +
WT
w . PwWT
,t +
PV
p . PpPV
,t + = PmL, t m MG, t NT (17)
j DG (m) i CI w WT(m) p PV(m )

µl . PlTie
,t + [ × (RUlTie Tie
, i, t + RDl, i, t )] + P jDG DG DG ¯ u ;
P jDG
l Tie l Tie i CI
l, i
, t + RUj, i, t |i = pri + RUj, i, t |i = sec j, t j DG(m) , t NT (18)
(m : m') (m : m')

P jDG
,t RDjDG
, i, t |i = pri RDjDG
, i, t |i = sec P jDGuj, t ; j DG(m) , t NT
s × j. P jDG
, s , i, t + l. PlTie
, s , i, t + VLLi,t . LSm,s, i, t _ (19)
s NS j DG(m ) i CI l Tie i CI i CI
(m : m')
P jDG P jDG rampjup (1 uON DG ON
j, t ) + P j uj, t ; j DG(m) , t NT
(16) ,t ,t 1
_ (20)
In (16), the first term is the fuel cost, start-up cost, and shut-down
cost of DGs. The second term represents the reserve provision cost. The P jDG
,t 1 P jDG
,t rampjdown (1 uOFF DG OFF
j, t ) + P j uj, t ; j DG(m) , t NT
third and fourth terms are maintenance cost of WTs and PVs, respec- _

tively. The costs of energy and reserve which are exchanged between (21)

8
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

f MG 3
p f
DG3 f p
p DG2
CEMS
DG1

MGCC3

MGCC1 MGCC2

MG1 MG 2
f f f

DG4
DG3

DG4

p p p
f

DG2
p

v abc 1 v abc 2
f f

DG1
DG1

p p

f
DG2

p f
f
DG5

DG3

p
p

Fig. 7. An IMMG test system with three MGs.

Table 2 u j, t + 1 u j, t uON
j, t + 1; j DG(m) , t NT (24)
Technical and economic data of converter based DGs.
DG1 DG2 DG3 DG4 DG5
u j, t u j, t + 1 uOFF
j, t + 1; j DG(m) , t NT (25)

¯
P DG
j (kW)
150 150 100 100 200 u j, t + 1 u j, t uON
j, t + 1 uOFF
j, t + 1; j DG(m) , t NT (26)
P jDG (kW) 50 50 20 30 35
PlTie Tie Tie ¯ ;
PlTie
_ , t + RUl, i, t |i = pri + RUl, i, t |i = sec l Tie(m :m') , t NT (27)
j ($/ h) 85.06 85.06 255.18 255.18 212
4.37 4.37 2.84 2.84 3.12 PlTie RDlTie RDlTie PlTie;
j ($/ kWh) ,t , i, t |i = pri , i, t |i = sec l Tie(m : m') , t NT
ON ($) 9 9 16 16 12 _ (28)
j
OFF ($) 8 8 9 9 8 ¯ ;
j
PwWT
,t PwWT w WT(m) , t NT (29)
j, i |i = pri ($/ kWh) 6 6 4 4 3.8
j, i |i = sec ($/ kWh) 2.2 2.2 1.5 1.5 1.7 ¯ ;
PpPV
,t PpPV p PV(m) , t NT (30)
rampup
j (kW / h)
250 250 250 250 280

ramp down
j (kW/h) 250 250 250 250 280 Eq. (17) indicates the supply-demand balance in each MG and each
UTj 1 1 1 1 1 hour of scheduling horizon. The scheduled power and reserves of DGs is
UTj 1 1 1 1 1 restricted to their upper and lower generation levels by (18) and (19).
Rj (mHz/kW) 0.01 0.01 0.015 0.015 0.0075 Likewise, ramp up/down limitations of DGs are presented in (20) and
(21) while, minimum up/down time limitations are described by (22)
and (23), respectively. Concurrent starting-up or shutting-down of DGs
h + UTj 1
are avoided by (24)-(26). The scheduled power and reserves which are
uj, t UTj uON
j, t ; j DG(m) , t NT exchanged between the MGs is restricted to the upper and lower limit of
t=1 (22)
tie-line by (27) and (28). In (29) and (30), the scheduled power of WTs
and PVs are restricted to their available generations.
h + DTj 1
(1 uj, t ) DTj uOFF
j, t ; j DG(m) , t NT
(23) 3.3.2. Constraint of second stage optimization
t=1
In the second stage, uncertainties related to generated power of

9
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

Fig. 8. Forecasted power generation of WTs, PVs, and load demand of MGs.

WTs, PVs, and load are considered. Accordingly, the scheduled energy P jDGuj, s, i, t P jDG , ref
, s , i, t
¯ u
P jDG j , s , i, t ; j DG(m) , s NS, i CI , t NT
of the MGs and exchanged energy between them are readjusted con- _
sidering the scheduled reserves to preserve the frequency security of the (35)
MGs. The following constraints should be considered in this stage:
P jDG
, s, i, t |i = pri P jDG
,t RUjDG
, i, t |i = pri ; j DG(m) , s NS, t NT (36)
P jDG
, s, i, t . uj, , i, t
j DG(m)

L
= Pm , s, i, t PlTie
, s , i, t PwWT
, s , i, t PpPV
, s, i, t + Dm, s, t . fm, s, i, t L P jDG
,t P jDG
, s, i, t |i = pri RDjDG
, i, t |i = pri ; j DG(m) , s NS, t NT (37)
l Tie w WT(m) p PV(m)
(m : m')

Sm, s, i, t ; m MG, s NS, i CI , t NT (31) P jDG , ref

P jDG , ref
RUjDG
, s, i, t |i = sec , s, i, t |i = pri , i, t |i = pri ; j DG(m) , s NS, t NT

f¯m, i ; (38)
| fm, s, i, t| m MG, s NS, i CI , t NT (32)

P jDG , ref
, s, i, t |i = pri P jDG , ref
, s, i, t |i = sec RDjDG
, i, t |i = pri ; j DG(m) , s NS, t NT
LSm, s, i, t PmL, t ; m MG, s NS, i CI , t NT (33)
(39)
P jDGuj, s, i, t P jDG ¯ u
P jDG
, s , i, t j , s , i, t ; j DG(m) , s NS , i CI , t NT
_
P jDG
, s, i, t |i = pri P jDG
, s , i, t 1 |i = pri rampjup uj,s, i, t |i = pri ; j DG(m) , s NS , t NT

(34) (40)

10
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

(36)-(39) are used to adjust the reference power of committed DGs

considering the scheduled primary and secondary reserves. Ramp-up
and down limitations of DGs are modeled in (40)-(43) to cover primary
and secondary intervals and their interactions. Meanwhile, constraints
(44) and (45) are used to adjust the exchanged power between the MGs
considering the scheduled primary and secondary reserves.

3.4. Linearization

In order to implement a linear programming approach, the non-

linear terms namely (32), should be modified into their equivalent
linear formats. The linear expression of (32) can be illustrated by the
following substitutions [34]:
| fm, s, i, t| = f m+, s, i, t + f m, s, i, t ; m MG, s NS , i CI , t NT (46)

fm, s, i, t = f m+, s, i, t f m, s , i , t ; m MG, s NS, i CI , t NT (47)

f m+, s, i, t 0, f m, s, i, t 0; m MG, s NS, i CI , t NT (48)

The supply demand constraints establish a non-linear relationship

between power variations of converter based DGs and their commit-
ments. This non-linearity can be converted into linear expressions by
employing the lemma which is presented in [35]. To this end, each term
, s, i, t . uj, , i, t is replaced by an added variable Xj, , i, t and adjoined by
P jDG
the following constraints:

P jDG
, i × uj , s , i , t Xj, s, i, t P¯jDG
, i × u j, , i, t ;
_

j DG(m) , s NS, i CI , t NT (49)

P jDG
, s, i, t P¯jDG
, i × (1 u j , s , i, t ) Xj, s, i, t P jDG
, s , i, t P jDG
, i × (1 uj, s, i, t );
_

j DG(m) , s NS, i CI , t NT (50)

3.5. Solution algorithm

The 24 h based results are derived through a well-organized mixed

integer linear programming (MILP) based optimization model which
ensures achieving near-global solutions and will never change by same
inputs. Although the computing time is about some seconds and the
problem can be considered as optimization portfolios used in online
Fig. 9. The energy scheduling results of IMMG cluster. frameworks, however, the proposed energy management problem is
aimed to be solved in a day-ahead scheduling time horizon. Therefore,
P jDG
, s , i, t 1 |i = pri P jDG
, s, i, t |i = pri rampjdown uj, s, i,t |i = pri ; j DG(m) , s NS , t NT by modeling the uncertainties and considering provided energy and
(41) reserve during real-time operation, the real-time errors can be removed
by rerunning the proposed model for remaining hours of scheduling
P jDG
, s, i, t |i = sec P jDG
, s , i, t 1 |i = sec rampjup uj, s, i, t |i = sec ; j DG(m) , s NS , t NT horizon. The flowchart of solution algorithm for solving the proposed
(42) CEMS is shown in Fig. 6. As can be seen, the flowchart includes three
levels. In the first level, scenario generation and reduction process is
P jDG
, s, i, t 1 |i = sec P jDG
, s, i, t |i = sec rampjdown uj, s,i, t |i = sec ; j DG(m) , s NS , t NT accomplished to consider the uncertain variables. To this end, fore-
casted errors of uncertain variables are model by related PDFs which
(43)
are discretized and scenario tree method are applied to generate en-
PlTie
, s , i, t PlTie
,t RUlTie
, i, t ; l Tie(m :m') , s NS, i CI , t NT (44) ough scenarios over the scheduling horizon. Each scenario includes the
information of wind speed, solar irradiation, and load demand. To de-
PlTie PlTie RDlTie crease the difficulty of the optimization problem, the size of scenarios is
,t , s, i, t , i, t ; l Tie(m :m') , s NS, i CI , t NT (45)
reduced using SCENRED tool which is provided by GAMS software. In
The total generation and demand should be equal at primary and the second level, the proposed centralized energy management model is
secondary control levels which is met by (31). The frequency deviation solved by GAMS software and results of energy and reserve for each MG
in the primary and secondary intervals which are obtained by (9) and of IMMG cluster are retrieved in a day before the real-time operation
(14), are kept in the permissible range using constraint (32). Likewise, and send to the MGs. In the third level, while the MGs are fully aware of
constraint (33) limits the primary and secondary load shedding at each the day-ahead decisions, if the balance between generation and supply
MG to its hourly demand. Constraints (34) and (35) restrict the DGs’ are not met in real-time operation, the primary and secondary reserves
power in primary and secondary intervals as well as reference power of are employed to reduce disturbance and preserve the frequency of the
DGs within their lower and upper limits. The constraints described by system within the secure range.

11
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

Fig. 10. The reserve scheduling results of MGs.

4. Numerical results and discussion 10.63 cent/kW and 54.84 cent/kW , respectively [40]. The exchanged
power between the MGs of IMMG cluster is totally limited to 200 kW
4.1. Data which is limited to 100 kW in each of primary and secondary control
intervals. Each MG sells energy to connected MGs with a price equal to
To evaluate the effectiveness of the proposed model, it is im- the maximum marginal price of its DGs. Likewise, 30% of energy price
plemented for an IMMG cluster including three MGs which is shown in is considered for selling primary and secondary reserves to neighboring
Fig. 7. All MGs are interconnected by BTB converters and operated at MGs. The value of loss of load (VLL) which should be payed to con-
rated frequency of 50 Hz [36]. MG1 is equipped with five converter sumers of the MGs due to unavoidable load shedding is considered
based DGs, namely DG1 to DG5. Likewise, four converter based DGs 10,000 cents/kWh.
including DG1 to DG4 are installed in MG2 while MG3 has three con- The proposed EMS is a MILP problem that was solved using CPLEX
verter based DGs including DG1 to DG3. The technical and economic 12.5.1 under GAMS [41] on a 3.40-GHz Intel Core i5 CPU personal
data of controllable DGs are extracted from [37,38] and presented in computer with 4 GB of RAM memory. The gap tolerance for solving
Table 2. Based on the IEEE Std-1547, the maximum permitted devia- MILP problem was 0.05%. To model the uncertainties of forecasted
tions of frequency in the primary and secondary control levels are set power generation of WTs and PVs as well as load demand of MGs,
to ± 300 mHz [39]. The forecasted power generation of WTs and PVs Weibull, Bimodal, and normal PDF are used which their parameters can
as well as load demand of each MG is shown in Fig. 8. Meanwhile, the be found in [31]. Accordingly, 1000 scenarios are generated using
operation cost of WTs and PVs, i.e. wWT and pPV , are assumed to be scenario tree method and then, reduced to 100 by SCENRED tool which

12
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

Fig. 11. The exchanged primary and secondary reserves between MGs.

is provided by GAMS. The reduced scenarios have been applied to the frequency fluctuations. As can be seen, in the hours of 12–15 respec-
proposed EMS model. tively more primary and secondary reserves are scheduled. The reason
is that the production of renewable resources, i.e., WTs and PVs, are
increased and this poses more uncertainty. Therefore, more primary
4.2. Results
and secondary reserves are needed to compensate their uncertainties.
Since, the major part of MG frequency fluctuations is compensated in
The energy scheduling results of IMMG cluster are shown in Fig. 9.
the primary control interval and the rest of it is mitigated in the sec-
As can be seen, the cheaper DGs of each MG (e.g., DG3 in MG1, DG3 in
ondary control interval the scheduled primary reserve is always more
MG2 and MG3) are firstly dispatched to meet the load demand of MGs
than secondary reserve. Note that the amounts of primary reserves are
with the lowest operation cost. Moreover, exchanging power between
related to the frequency deviations of MG and the droop values of
the MGs can reduce the operation cost of IMGG and improve the se-
controllable DGs. Likewise, the reference power is determined by the
curity. As can be seen, in off-peak hours (i.e., 1–6), the power genera-
MGCC and the operational constraints which are considered by the EMS
tion of cheapest DGs are shared between MGs to reduce the operation
and determine the amount of secondary reserves.
cost of the system. The DGs’ installed capacity in MG1 is more than
The exchanged primary and secondary reserves are shown in Fig. 11
those of MG2 and MG3. Therefore, during on-peak hours of MG2 and
which the negative and positive values mean downward and upward
MG3, the power is sent from MG1 to MG2 and MG3 for reducing the
reserves, respectively. It should be mentioned that reserves are only
generation amount of these MGs. On the other hand, MG3 has the
provided by controllable DGs in the primary and secondary control
lowest installed capacity of DGs. Therefore, MG2 sends power to MG3
intervals. Accordingly, MG1 with the most controllable DGs sells re-
to supply a part of its demand during on-peak hours. These strategies
serves to MG2 and MG3, specifically during the hours of on-peak and
not only reduce the energy cost of IMMG, but also free the capacities of
high power generation of renewable generation. Moreover, MG2 which
DGs for providing primary and secondary reserves. Therefore, total
has respectively lower load demand sells primary and secondary re-
operation cost of IMMG is minimized while the security of the system is
serves to MG3. As can be seen in Fig. 10, downward reserves are pro-
improved during the scheduling horizon. It should be mentioned that
vided by MG1 and MG2 during off-peak hours, i.e., 1–6, while during
the runtime of proposed energy management model is 35.046 s which
other hours only upward reserves are exchanged between MGs. The
makes it applicable in practice.
reason is that MGs can reduce the power generations of their DGs with
The reserve scheduling results of MGs are shown in Fig. 10. Based
lower cost. However, during off-peak hours the generation power of
on the direction of imbalance power in each MG, the scheduled upward
DGs are respectively low and more reduction in their power generation
and downward reserves are deployed in the primary and secondary
may endanger the technical constraint of DGs.
control intervals to preserve frequency of the MGs in the pre-defined
The primary and secondary steady-state frequencies of MGs are
secure ranges. In other words, when the frequency of MG decreases
shown in Fig. 12. As can be seen, the frequency in the primary and
from the nominal value (i.e., 60 Hz), the scheduled upward reserves
secondary control intervals fluctuate within the pre-defined secure
should be used to mitigate the negative frequency fluctuations while
range (i.e., 49.7–50.3 Hz). It should be mentioned that the MGs can be
the scheduled downward reserves are used to alleviate the positive

13
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

Fig. 12. The primary and secondary steady-state frequencies of MGs.

Table 3
Breakdown operation cost of IMMG cluster.
MG Energy Cost Scheduled Primary Scheduled Secondary Deployed Primary Deployed Secondary Energy and Reserve Total Operational Cost
(cent) Reserve Cost (cent) Reserve Cost (cent) Reserve Cost (cent) Reserve Cost ($) Exchanged Cost (cent) (cent)

MG1 35328.4 13435.7 6877.2 43496.9 44353.7 −41994 742792.4

MG2 26695.5 10698.4 5674.2 32083. 9 33715.4 −22834 635571.1
MG3 18585.9 8351.9 4458.6 22875.0 24475.1 40,199 971647.1
IMMG 80609.8 32486.0 17010.0 98455.8 102544.2 −3629 2350010.6

operated with the pre-defined secure range of frequency fluctuations. Table 3. As can be seen, MG3 has the highest total operation cost. The
Moreover, the frequency fluctuations in the primary control level are reason is that, the capacity of installed DGs in MG3 is low and therefore,
greater than the secondary one. The reason is that in the secondary it is forced to purchase energy and reserve from other MGs for sup-
control interval, the MGCC has more freedom to restore frequency to plying its load demand and preserve frequency security. On the other
nominal value by modifying the reference power of controllable DGs or hand, MG1 with the most installed capacity of DGs has the ability to sell
even their commitments. extra energy and reserve to other MGs which reduces its operation cost,
The breakdown operation cost of IMMG cluster is presented in considerably. Meanwhile, the peak load demand of MG2 occurs at final

14
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

Fig. 13. Total operation cost of MGs with respect to exchanged power limitation.

Fig. 14. Load shedding amount with considering exchanged power limitation equal to zero.

Table 4
Breakdown operation cost of IMMG cluster with f |i = sec = ± 100 mHz.

MG Energy Cost Scheduled Primary Scheduled Secondary Deployed Primary Deployed Secondary Energy and Reserve Total Operational
(cent) Reserve Cost (cent) Reserve Cost (cent) Reserve Cost (cent) Reserve Cost (cent) Exchanged Cost (cent) Cost (cent)

MG1 36029.3 13536.7 7888.0 44444.1 43864.6 −52237 839957.6

MG2 24679.7 9733.0 6176.0 29469.0 31787.0 −21981 568247.3
MG3 20091.5 9162.8 6478.1 24806.0 26975.2 42,782 924272.2
IMMG 80800.5 32432.6 20542.1 98719.2 102626.8 −3436 2332477.2

hours of scheduling which enables MG2 to sell energy and reserve to increasing the frequency security. To investigate this issue, the limita-
other MGs. tion of exchanged power between MGs is increased from zero to
As mentioned, the energy and reserve which are exchanged between 250 kW. Fig. 13 shows total operation cost of MGs with respect to
MGs in the IMMG cluster reduce total operation cost of MGs while limitation of exchanged power. As expected, total operation cost of MG

15
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

Fig. 15. The secondary steady-state frequency of MGs with considering f |i = sec equal to ± 300 mHz and ± 100 mHz.

has a decreasing trend. However, MG2 benefits more with increasing enough reserves to preserve frequency security.
the exchanged power limitation between MGs. The reason is that, the The values of pre-defined secure range effect on the total operation
installed capacity of DGs in MG3 is respectively low and therefore, it cost of IMMG cluster. The breakdown operation cost of IMMG cluster, whit
can purchase more energy and reserve from other MGs to reduce its considering f |i = sec equal to ± 100 mHz, is presented in Table 4. As ex-
operation cost and frequency security. As can be seen, with increasing pected, total operation cost of MGs increases by restricting the frequency
the exchanged power limitation from 150 kW to 250 kW, the operation security in the secondary control interval. Indeed, the MGCCs should
cost of MGs becomes almost fixed. The reason is that the capacity of schedule and deploy more reserves locally or by purchasing from other
DGs for providing energy and reserve are fully used and therefore, the MGs to ensure frequency security restricted range. This issue leads to more
MGs cannot buy/sell power energy from/to each other. power production of DGs and therefore increases energy cost of MGs.
If the limitation of exchanged power between MGs is considered zero, However, MG1 and MG2 can sell more power to MG3 and increases their
the connection between MGs are ignored. In this case, the total operation profit. On the other hand, MG3 is forced to purchase more energy and
cost of MGs increases significantly as shown in Fig. 13. The reason is that reserve from MG1 and MG2 which increases it operation cost significantly.
MG1 and MG2 cannot sell energy and reserve to MG3. Moreover, as shown Fig. 15 compares the secondary steady-state frequency of MGs with
in Fig. 14, MG2 and MG3 are forced to perform load shedding in the considering f |i = sec equal to ± 300 mHz and ± 100 mHz. As can be seen,
primary and secondary control intervals to preserve frequency security. It the MGs’ frequency fluctuations are properly controlled within the pre-
should be mentioned due to high installed capacity of DGs in MG1, it has defined ranges. To this end, optimal amounts of primary and secondary

16
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

Fig. 16. The effect of uncertainty on the operation cost of MGs.

reserves are scheduled and deployed which increases total operation costs To investigate the impact of ignoring frequency security, constraint
of MGs. (32) of the proposed model is removed. The primary and secondary
The effect of uncertainties on the operation cost of MGs are shown steady-state frequencies of MGs without considering frequency security
in Fig. 16. To this end, the uncertainty of renewable generation and constraint are shown in the following Fig. 17. As can be seen, the pri-
load of each MG is increased with respect to base case and total op- mary and secondary steady-state frequencies of MGs violate from the
eration cost of MGs are calculated. As can be seen, with increasing the security range (i.e., 49.7–50.3 Hz) and therefore, the load shedding is
uncertainty in MGs, their total operation costs are increased. The reason inevitable to preserve the stability of MGs. The reason is that enough
is that more reserve should be scheduled and deployed to compensate values of primary and secondary reserves are not scheduled to cope
uncertainty. With increasing the uncertainty in MG1, it forces to with uncertainties of renewable generation and load demand.
schedule local resources to preserve frequency security. Therefore, the
energy and reserve it can sell to MG2 and MG3 are decreased and their
5. Conclusion
operation cost are increased. On the other hand, with increasing the
uncertainty in MG2 and MG3, MG1 can sell more energy and reserve to
This paper developed a novel EMS for IMMG clusters, which aims to
them, and therefore, its operation cost decreases. Meanwhile, with in-
minimize total operation cost of MGs and preserve their frequency security
creasing the uncertainty in MG3, MG2 can also sell more reserve and
within a pre-defined range. Accordingly, the hierarchical frequency con-
energy to MG3 which reduce its operation cost.
trol levels of the IMMG cluster including primary, secondary, and tertiary

17
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

Fig. 17. The primary and secondary steady-state frequencies of MGs without considering frequency security constraint.

have been precisely formulated in a two-stage scenario-based stochastic Writing - review & editing, Methodology. Navid Rezaei: Writing -
framework. Furthermore, the proposed EMS provided optimal strategies original draft. Fatemeh Jahanbani Ardakani: Writing - original draft,
for the MGs to exchange energy and reserves during scheduling horizon. Writing - review & editing. Maryam Mohiti: . Josep M. Guerrero:
The simulation results indicated that with optimal scheduling of droop Writing - original draft, Writing - review & editing.
controlled DGs, the primary and secondary frequency securities of MGs
could be preserved within pre-defied security range in a cost effective
manner. Moreover, it was demonstrated that the exchanging energy and Declaration of Competing Interest
reserve between MGs helps MGCC to cope with inherent operational un-
certainties and improve frequency security. Likewise, the operation cost of The authors declare that they have no known competing financial
MGs is reduced with trading energy and reserve with each other. Indeed, interests or personal relationships that could have appeared to influ-
the implementation of a multi-microgrid system gave an insight capability ence the work reported in this paper.
to the MGCC for controlling the security of microgrid by properly trading-
off between economic and technical and issues.
Appendix A. Supplementary material
CRediT authorship contribution statement
Supplementary data to this article can be found online at https://
doi.org/10.1016/j.ijepes.2020.106134.
Mohammadreza Mazidi: Methodology, Writing - original draft,

18
M. Mazidi, et al. Electrical Power and Energy Systems 121 (2020) 106134

References [22] Li J, Liu Y, Wu L. Optimal operation for community-based multi-party microgrid in

grid-connected and islanded modes. IEEE Trans Smart Grid 2018;9:756–65.
[23] Liu Y, Li Y, Xin H, Gooi HB, Pan J. Distributed optimal tie-line power flow control
[1] Lopes JP, Hatziargyriou N, Mutale J, Djapic P, Jenkins N. Integrating distributed for multiple interconnected AC microgrids. IEEE Trans Power Syst
generation into electric power systems: a review of drivers, challenges and oppor- 2019;34:1869–80.
tunities. Electr Power Syst Res 2007;77:1189–203. [24] Hu M, Wang Y-W, Xiao J-W, Lin X. Multi-energy management with hierarchical
[2] Lasseter RH, Piagi P. Microgrid: a conceptual solution. In: IEEE power electronics distributed multi-scale strategy for pelagic islanded microgrid clusters. Energy
specialists conference; 2004. p. 4285–91. 2019;185:910–21.
[3] Ipakchi A, Albuyeh F. Grid of the future. IEEE Power Energ Mag 2009;7:52–62. [25] Ahmadi SE, Rezaei N. A new isolated renewable based multi microgrid optimal
[4] Jiayi H, Chuanwen J, Rong X. A review on distributed energy resources and energy management system considering uncertainty and demand response. Int J
MicroGrid. Renew Sustain Energy Rev 2008;12:2472–83. Electr Power Energy Syst 2020;118:105760.
[5] Hossain MA, Pota HR, Issa W, Hossain MJ. Overview of AC microgrid controls with [26] Purage MISL, Krishnan A, Foo YSE, Gooi HB. Cooperative bidding-based robust
inverter-interfaced generations. Energies 2017;10:1300. optimal energy management of multi-microgrids. IEEE Trans Industr Inform; 2019.
[6] Katiraei F, Iravani MR. Power management strategies for a microgrid with multiple p. 1–1.
distributed generation units. IEEE Trans Power Syst 2006;21:1821–31. [27] Liu Y, Gooi HB, Li Y, Xin H, Ye J. A Secure distributed transactive energy man-
[7] Guo L, Wang N, Lu H, Li X, Wang C. Multi-objective optimal planning of the stand- agement scheme for multiple interconnected microgrids considering misbehaviors.
alone microgrid system based on different benefit subjects. Energy IEEE Trans Smart Grid 2019;10:5975–86.
2016;116:353–63. [28] Lopes JP, Gil NJ. Exploiting automated demand response, generation and storage
[8] Ng EJ, El-Shatshat RA. Multi-microgrid control systems (MMCS). In: IEEE PES capabilities for hierarchical frequency control in islanded multi-microgrids; 2008.
general meeting; 2010. p. 1–6. [29] Nguyen T-T, Yoo H-J, Kim H-M. A Droop frequency control for maintaining dif-
[9] Arefifar SA, Ordonez M, Mohamed YAI. Energy management in multi-microgrid ferent frequency qualities in a stand-alone multimicrogrid system. IEEE Trans
systems—development and assessment. IEEE Trans Power Syst 2017;32:910–22. Sustain Energy 2017;9:599–609.
[10] Nikmehr N, Ravadanegh SN. Reliability evaluation of multi-microgrids considering [30] Yoo H-J, Nguyen T-T, Kim H-M. Multi-frequency control in a stand-alone multi-
optimal operation of small scale energy zones under load-generation uncertainties. microgrid system using a back-to-back converter. Energies 2017;10:822.
Int J Electr Power Energy Syst 2016;78:80–7. [31] Mazidi M, Rezaei N, Ghaderi A. Simultaneous power and heat scheduling of mi-
[11] Sattarpour T, Golshannavaz S, Nazarpour D, Siano P. A multi-stage linearized in- crogrids considering operational uncertainties: a new stochastic p-robust optimi-
teractive operation model of smart distribution grid with residential microgrids. Int zation approach. Energy 2019;185:239–53.
J Electr Power Energy Syst 2019;108:456–71. [32] Zakariazadeh A, Jadid S, Siano P. Economic-environmental energy and reserve
[12] Marzband M, Parhizi N, Savaghebi M, Guerrero JM. Distributed smart decision- scheduling of smart distribution systems: A multiobjective mathematical pro-
making for a multimicrogrid system based on a hierarchical interactive archi- gramming approach. Energy Convers Manage 2014;78:151–64.
tecture. IEEE Trans Energy Convers 2015;31:637–48. [33] Hossain MA, Pota HR, Hossain MJ, Blaabjerg F. Evolution of microgrids with
[13] Li Y, Zhao T, Wang P, Gooi HB, Wu L, Liu Y, et al. Optimal operation of multi- converter-interfaced generations: challenges and opportunities. Int J Electr Power
microgrids via cooperative energy and reserve scheduling. IEEE Trans Ind Inf Energy Syst 2019;109:160–86.
2018;14:3459–68. [34] Rezaei N, Kalantar M. Economic–environmental hierarchical frequency manage-
[14] Li Z, Shahidehpour M. Privacy-preserving collaborative operation of networked ment of a droop-controlled islanded microgrid. Energy Convers Manage
microgrids with the local utility grid based on enhanced benders decomposition. 2014;88:498–515.
IEEE Trans Smart Grid; 2019. p. 1–1. [35] Wang MQ, Gooi H. Spinning reserve estimation in microgrids. IEEE Trans Power
[15] Mohiti M, Monsef H, Anvari-moghaddam A, Guerrero J, Lesani H. A decentralized Syst 2011;26:1164–74.
robust model for optimal operation of distribution companies with private micro- [36] Niiranen J, Komsi R, Routimo M, Lähdeaho T, Antila S. Experiences from a back-to-
grids. Int J Electr Power Energy Syst 2019;106:105–23. back converter fed village microgrid. In: 2010 IEEE PES Innovative Smart Grid
[16] Aboli R, Ramezani M, Falaghi H. A hybrid robust distributed model for short-term Technologies Conference Europe (ISGT Europe); 2010. p. 1–5.
operation of multi-microgrid distribution networks. Electr Power Syst Res [37] Mohiti M, Monsef H, Anvari-Moghaddam A, Lesani H. Two-Stage robust optimi-
2019;177:106011. zation for resilient operation of microgrids considering hierarchical frequency
[17] Marzebali MH, Mazidi M, Mohiti M. An adaptive droop-based control strategy for control structure. IEEE Trans Industr Electron; 2019. p. 1–1.
fuel cell-battery hybrid energy storage system to support primary frequency in [38] Rezaei N, Ahmadi A, Khazali AH, Guerrero JM. Energy and frequency hierarchical
stand-alone microgrids. J Energy Storage 2020;27:101127. management system using information gap decision theory for islanded microgrids.
[18] Yazdaninejadi A, Hamidi A, Golshannavaz S, Aminifar F, Teimourzadeh S. Impact IEEE Trans Ind Electron 2018;65:7921–32.
of inverter-based DERs integration on protection, control, operation, and planning [39] IEEE Standard for Interconnection and Interoperability of Distributed Energy
of electrical distribution grids. The Electricity J 2019;32:43–56. Resources with Associated Electric Power Systems Interfaces. IEEE Std 1547-2018
[19] John T, Lam SP. Voltage and frequency control during microgrid islanding in a (Revision of IEEE Std 1547-2003); 2018. p. 1–138.
multi-area multi-microgrid system. IET Gener Transm Distrib 2017;11:1502–12. [40] Zhang C, Xu Y, Dong ZY, Wong KP. Robust coordination of distributed generation
[20] Bui V, Hussain A, Kim H. A strategy for flexible frequency operation of stand-alone and price-based demand response in microgrids. IEEE Trans Smart Grid
multimicrogrids. IEEE Trans Sustain Energy 2018;9:1636–47. 2017;9:4236–47.
[21] Amoateng DO, Hosani MA, Elmoursi MS, Turitsyn K, Kirtley JL. Adaptive voltage [41] “The General Algebraic Modeling System (GAMS) Software. Available at: http://
and frequency control of islanded multi-microgrids. IEEE Trans Power Syst www.gams.com.
2018;33:4454–65.

19