You are on page 1of 15

This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON SMART GRID 1

Coupling Neighboring Microgrids for Overload


Management Based on Dynamic
Multicriteria Decision-Making
Farhad Shahnia, Senior Member, IEEE, Soheil Bourbour, and Arindam Ghosh, Fellow, IEEE

AbstractA microgrid (MG) is expected to supply its local


loads independently; however, due to intermittency of wind and
solar-based energy resources as well as the load uncertainty, it is
probable that the MG experiences power deficiency (overloading).
This problem can be mitigated by coupling the overloaded MG
to another neighboring MG that has surplus power. Considering
a distribution network composed of several islanded MGs, defin-
ing the suitable MGs (alternative) to be coupled with the
overloaded MG is a challenge. An MG overload management
technique is developed in this paper, which first identifies the
overloaded MG(s) and then selects the most suitable alterna-
tive. The alternative selection is based on different criteria,
such as available surplus power, reliability, supply security,
power loss, electricity cost, and CO2 emissions in the alterna-
tive MGs. Moreover, the frequency and voltage deviation in
the system of coupled MGs are considered in the selection.
A dynamic multicriteria decision-making algorithm is devel-
oped for this purpose. To contemplate the uncertainties in the
considered distribution network, a cloud theory-based proba-
bilistic analysis is deployed as the research framework and Fig. 1. Remote area distribution network consisting of several isolated MGs
the performance of the developed technique is evaluated in with normally open ISSes among them.
MATLAB.
Index TermsCoupled microgrids (CMGs), decision-making,
load shedding, overload management. SLSA Selective load-shedding algorithm.
UPC Unused power capacity.
N OMENCLATURE
ALS Amount of load to be shed. I. I NTRODUCTION
APC Available power capacity.
LECTRIFICATION of remote and rural areas has been
CMGs
DER
Coupled microgrids.
Distributed energy resources. E always a major challenge due to a variety of constraints
such as the area accessibility and economic factors [1], [2].
DMA Decision-making algorithm.
The electricity demand in these areas can be supplied by the
ISS Interconnecting static switch.
help of DERs in an islanded scheme. Thus, the power system
MAIFI Momentary average interruption frequency index.
of remote areas can be considered as a microgrid (MG) that
OMT Overload management technique.
operates in islanded mode [3]. The MGs should be formed
PDF Probability density function.
such that there is enough generation capacity in their embed-
PDL Power deficiency level.
ded DERs to meet their local demands [4][6]. It is to be noted
PFA Power flow analysis.
that a remote area/town can be supplied by several indepen-
SAIDI System average interruption duration index.
dent MGs, where each may have a different owner (operator)
SAIFI System average interruption frequency index.
and each is responsible for supplying the loads of a specific
Manuscript received February 27, 2015; revised June 22, 2015 and region. Thus, the distribution network of a remote area can
August 10, 2015; accepted September 7, 2015. Paper no. TSG-00235-2015. resemble the system of Fig. 1. These types of islanded MGs are
(Corresponding author: Farhad Shahnia.) conventionally supplied by diesel generators. The fuel trans-
F. Shahnia is with the School of Engineering, Murdoch University, Perth,
WA 6150, Australia (e-mail: farhad.shahnia@curtin.edu.au). portation difficulties and the fuel cost adversely affect the
S. Bourbour and A. Ghosh are with the Department of Electrical and profits for the owners of each MG. If these areas are rich
Computer Engineering, Curtin University, Perth, WA 6102, Australia. in renewable energies, renewable-energy-based DERs can be
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. utilized for electricity generation to meet the local electrical
Digital Object Identifier 10.1109/TSG.2015.2477845 demand [7].
1949-3053 c 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON SMART GRID

The intermittency of nondispatchable (e.g., solar and 3) Define the different criteria required for DMA.
wind-based) DERs in addition to load uncertainties can 4) Qualify the selected MG(s) based on the deviations in
lead to imbalance between the instantaneous power gen- voltage and frequency after coupling the MGs.
eration and demand in an MG [8]. Any generation defi- 5) Define the portion of the nonessential loads to be shed
ciency (overloading) will lead to voltage/frequency drop. To from each MG based on the proposed DMA such that
address power imbalance problems in MGs, several solutions all essential loads of all MGs are always supplied.
can be considered as follows: 6) Develop a suitable PFA technique, applicable for MGs.
1) under frequency/voltage load-shedding [9]; The rest of this paper is organized as follows. Section II
2) utilization and control of battery energy storages [10]; briefly discusses the necessity of coupling the MGs and the
3) optimal capacity design of dispatchable DERs complexity of decision-making for this purpose. The proposed
(e.g., diesel generators) [11], [12]; OMT is presented in Section III while Section IV introduces
4) interconnection of the MG to utility [13]; the developed dynamic multicriteria DMA. The performance
5) coupling of one MG to one/more neighboring of the developed OMT is evaluated within a MATLAB-based
MG(s) [14]. stochastic analysis in Section V. The main conclusions of this
MGs coupling is introduced in [15] as a solution to prolif- paper are highlighted in Section VI while the developed PFA
erate the number of DERs in distribution networks. Each MG and the considered stochastic framework are presented in the
in Fig. 1 may be supported by one/more of its neighboring Appendixes.
MG(s) during power deficiency. This can be achieved by clos-
ing the normally open ISS which is located between every two II. N ECESSITY AND C OMPLEXITY OF
adjunct MGs. It is to be noted that the structure and control C OUPLING M ICROGRIDS
mechanism of the ISS is beyond the scope of this paper.
Wang and Wang [16] and Wang et al. [17] proposed a trans- This paper focuses on remote area networks, where a util-
formative architecture for coupling the neighboring MGs as ity connection is not available. Thus, the loads are supplied
a technique for improving the self-healing of the distribu- by small MG networks. Each of these MGs may have a dif-
tion system in case of short-circuit faults in the network. ferent owner, that has invested in the installation of the DERs
The trade of power among MGs in the system of CMGs is in that MG, and charges the loads based on their electricity
addressed in [18]. Optimal control of a distribution network consumption. Thereby, it is assumed that these MGs operate
composed of utility-connected MGs forming a CMG is also in islanded mode and independently from each other.
studied in [19]. Dynamic operation of DERs within CMGs is Due to load and generation uncertainty, it is highly probable
investigated in [20] and the dynamic security of the CMGs to expect power generation deficiency or overloading at some
is examined in [21]. The conditions under which two MGs portions of the operation stage. Under such conditions, except
are interconnected are addressed in [22]. The stability anal- load-shedding, the other possible option to control an MG is
ysis of a CMG prior to the interconnection of the MGs is external support in the form of importing power from one or
discussed in [23], as a preliminary step to prevent any inter- a group of neighboring MGs. Thereby, an OMT is proposed
connection that may lead to system instability. Selection of and developed in this paper to overcome the overloading issues
the suitable MG(s) among the available neighboring MGs of the MGs that operate in remote areas.
when interconnecting them during overloading has not been Consider the distribution network of Fig. 1 with N islanded
addressed in the previous literature and is the main focus of MGs, among which MG-1 is assumed to be overloaded. Each
this paper. of the other existing MGs may be able to support MG-1 indi-
This paper proposes an OMT, based on coupling the vidually or in combination with others. As an example, for
neighboring islanded MGs, and utilizes a dynamic multicrite- N = 3, the alternative MGs are [{2}, {3}, {2,3}] (see Fig. 2)
ria DMA. The proposed OMT assumes a data communication and for N = 4, the alternative MGs are [{2}, {3}, {4}, {2,3},
system is available to receive the power generation of all DERs {2,4}, {3,4}, {2,3,4}] while for N = 5, the alternative MGs are
and consumption of essential/nonessential loads in all MGs. [{2}, {3}, {4}, {5}, {2,3}, {2,4}, {2,5}, {3,4}, {3,5}, {4,5},
The communication system also transmits the command (out- {2,3,4}, {2,3,5}, {2,4,5}, {3,4,5}, {2,3,4,5}]. In general, assum-
put) of the OMT to the relevant ISS(es) to couple the MGs. ing N  MGs are overloaded, the alternatives are combinations
Under such a case, the power flow control in the considered of any single MGs, any two MGs, any three MGs, . . . , and
system is based on the proper operation of the ISS, i.e., if any N N  MGs out of the available N N  MGs. Hence, the
the OMT decides that some MGs should be interconnected, total number of alternatives, denoted by NA , is
the ISS of each of those MGs will be closed and thereby, the 
NN 

power flow will occur between the interconnected MGs auto- NA = CiNN
matically based on the dynamic operation of the DERs in each i=1
MG and no further power control is required.    
= C1NN + C2NN + C3NN + + CNN
NN
 (1)
The main contributions of this paper are as follows.
1) Develop an OMT to reduce load-shedding rate in MGs where Cab = b!/[(b a)! a!] and a! = a (a 1) . . . 1.
during overloading conditions. Equation (1) can be simplified as
2) Develop a dynamic multicriteria DMA to select the

suitable MGs. NA = 2NN 1. (2)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SHAHNIA et al.: COUPLING NEIGHBORING MGs FOR OVERLOAD MANAGEMENT 3

(a) (b)

Fig. 3. Schematic of the communication links between the developed OMT


(located within the network tertiary controller) and the central controller of
each MG as well as the ISSes.

(c) (d)
The OMT continuously communicates with the central con-
Fig. 2. (a) Remote area distribution network composed of three islanded MGs troller of each MG to receive the information about the
assuming MG-1 is overloaded. (b) Alternative-1 for CMGs. (c) Alternative-2
for CMGs. (d) Alternative-3 for CMGs.
total active power generation of the dispatchable DERs as
well as the total active power consumed by all loads. For
MG-i, let us denote these quantities as Pdisp-DER(MG-i) and
Pload(MG-i) . The OMT will first identify the overloaded MG(s)
Thus, selecting the suitable MG(s) to couple with the
and then take action depending on the network conditions
overloaded MG among all alternatives is challenging due to
such as the number of overloaded MGs and the UPC that
the high number of alternatives. In addition, different crite-
is available in the other MGs. The OMT calculates the UPC
ria can be considered for the selection, each with a different
for MG-i as
weighting, which can further complicate the selection process. 
To overcome the problem of proper alternative selection UPCi = Pdisp-DER (MG-i)
among a large number of alternatives, while considering sev- 
eral criteria with different weighting, the developed OMT Pload (MG-i) i {1, . . . , NA }. (3)
utilizes a DMA to assess the alternatives and select the It is desired to maintain the UPC of every MG higher than
suitable one. a threshold as
It is worth mentioning that Fig. 1 depicts only one of the 
many possible topologies that the neighboring MGs can be UPCi 1 Pmax
disp-DER (MG-i)
(4)
connected. The proposed OMT in this paper does not depend  max
where Pdisp-DER (MG-i) is the total capacity of dispatchable
on neither the interconnection topologies of the MGs, nor the
topology of an individual MG. Thus, other interconnection DERs in MG-i and 0 < 1 < 1 (e.g., 1 = 0.1) imposes
topologies are also acceptable. a safety margin. In the rest of this paper, (4) is assumed as
the overloading condition of an MG.
If condition (4) is valid for all MGs of the distribution net-
III. P ROPOSED OVERLOAD M ANAGEMENT T ECHNIQUE work, no action needs to be taken by the OMT. However, if
it is invalid for one or more MGs, the OMT evaluates the
The developed OMT will be located as a module (agent) availability of surplus power in the network as
within the network tertiary controller and will continuously
communicate with the central controller of each MG as well as 
N 
N 

the ISSes (see Fig. 3). It is to be noted that the communication UPCi 1 Pmax
disp-DER (MG-i)
. (5)
system, data bandwidth, and topology are beyond the scope i=1 i=1
of this paper and are not discussed in this paper. In the rest of this paper, (5) is considered as the constraint for
The proposed OMT is composed of one main function and coupling the MGs.
three subfunctions, namely decision-making function, load- If condition (4) flags that one or more MGs are overloaded
shedding function, and PFA function. Two algorithms, namely and constraint (5) flags the availability of surplus power in the
DMA and SLSA, are used, respectively, within the decision- distribution network, the OMT proceeds to support the over-
making and load-shedding functions. The proposed OMT loaded MGs by coupling one or more of the nonoverloaded
and its subfunctions are described in detail in the following, MGs to the overloaded one(s).
while the algorithms are described in the next section. The If condition (4) is invalid for n1 = N 1 MGs but con-
PFA function is discussed in Appendix A. straint (5) is valid, the OMT flags that the only alternative
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON SMART GRID

is coupling all MGs. It then calls the PFA function to verify And if condition (10) is invalid for all MGs, it flags that load-
that coupling all MGs will not cause nonstandard voltage and shedding is not viable.
frequency deviation in the CMG system. From the PFA, that If condition (10) is invalid for 1 n2 < N MGs, the load-
is discussed in Appendix A, the maximum voltage deviation shedding function first sends the ALS level to the MG(s) which
in the buses of the system and the maximum frequency devi- satisfy (10) to shed the defined portion of their nonessential
ation in the system, respectively, denoted by V and F, are loads. Then, it evaluates whether the sum of ALSes in all MGs
calculated as of the distribution network is less than the sum of nonessential
loads in all MGs, that is
V = max(|Vi Vnom |)i=1,..., Nbus
F = | f fnom | (6) 
N 
N 
ALSi Pnon-essLoad (MG-i) . (11)
where Vi is the per unit (p.u.) voltage of bus-i, i i=1 i=1
{1, 2, . . . , Nbus } and f is frequency of the system and Nbus is If (11) is satisfied, the load-shedding function seeks the possi-
the number of buses in the system of CMG while it is assumed bility of interconnecting the MG(s) which do not satisfy (10)
that Vnom = 1 p.u. and fnom = 50 Hz. These two parameters with the one(s) that still have some nonessential loads. For
are then evaluated in this, it defines a second ALS level to compensate for the MGs
who did not satisfy (10). In the rest of this paper, (11) is
V 0.1 and F 0.5 (7)
considered as another constraint for coupling MGs, which is
to verify the possibility of MGs interconnection, assuming that used only when load-shedding is accompanying coupling. The
the maximum acceptable voltage deviation is 10% and the second ASL level, denoted by ALS2 , is defined as
maximum acceptable frequency deviation is 0.5 Hz. n2 
  
If condition (4) is invalid for 1 < n1 < N MGs, the ALS2 = ALSi Pnon-essLoad (MG-i) i {1, . . . , n2 }.
OMT flags that several alternatives are available and decision- i=1
making is required. It then calls the decision-making function (12)
to formulate all alternatives and to define the suitable alter-
native using the DMA. The proposed DMA is discussed in The load-shedding function then formulates the alternatives
details in the next section. If an alternative is selected by the which their total remaining nonessential loads are higher
DMA, the OMT initiates the proper command for the relevant than ALS2 , that is
 
ISS(es) to close. However, if no suitable alternative is cho- Pnon-essLoad (MG-i) ALSi
sen, the decision-making function flags that load-shedding is k
inevitable and requests the main function of the OMT to call  
ALS2 k 1, . . . , NA (13)
the load-shedding function.
If the OMT calls the load-shedding function, it first seeks where NA is the number of the formulated alternatives. The
the possibility of shedding a portion of the nonessential loads alternatives qualified in (13) are then assessed by the SLSA to
in one or more MGs such that the essential loads of all MGs select a suitable alternative. If a suitable alternative is selected
are not interrupted. For this, it first identifies the APC of each by the SLSA, the load-shedding function flags that a suitable
MG as alternative is selected. It then defines the portion of ALS2 for
  each MG of the selected alternative based on the ratios of their
APCi = Pdisp-DER (MG-i) + Pnondisp-DER (MG-i) (8)
nonessential loads and sends the amount of the second load-

where Pnondisp-DER(MG-i) denotes the total active power gen- shedding level to the central controller of each MG as well as
eration by nondispatchable DERs in MG-I at that moment. It the proper command to the relevant ISS(es) to close. However,
is to be noted that APC represents the maximum power that if the SLSA does not select an alternative, the load-shedding
can be supplied in MG-i to its loads plus the power loss in function flags that MGs coupling with further load-shedding
the lines. Based on (8), the load-shedding function defines the is not viable.
ALS from each MG as The developed OMT with its main load-shedding and
  decision-making functions is shown in Appendix B and has
ALSi = (1 + 2 ) PLoad (MG-i) APCi (9) a time complexity of O(N).
In this paper, the MGs are coupled to prevent overloading
where 0 < 2 < 1 (e.g., 2 = 0.1) imposes a safety margin of an MG. The proposed method can be further expanded to
to compensate for line losses. For successful load-shedding in facilitate coupling of MGs so that the surplus power of the
each MG, the ALS should be equal or smaller than the total DERs of an MG can be exported to other MGs with a lower
of
 nonessential loads in each MG at that moment, denoted by electricity price (as an incentive technique) [18] to be stored
Pnon-essLoad , that is in the energy storage [24], or to be consumed by the con-

ALSi Pnon-essLoad (MG-i) . (10) trollable loads of the other MGs under the demand dispatch
concept [25].
If the load-shedding function defines that condition (10) is It is worth mentioning that the proposed OMT solely
valid for all MGs, it flags that load-shedding is successful and focuses on the situations that at least one of the MGs are over-
sends the ALS level of each MG to their central controllers. loaded and tries to select a suitable alternative to which the
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SHAHNIA et al.: COUPLING NEIGHBORING MGs FOR OVERLOAD MANAGEMENT 5

overloaded MG(s) will be interconnected, such that an accept-


able power balance is achieved. The proposed OMT does not
lead to an optimum generation in the DERs but tries to reduce
the load-shedding rate and prevent instability due to power
generation-demand imbalance. To achieve optimum generation
in the DERs of a CMG, an economic dispatch-type method
can be utilized.
It is also noteworthy that in this paper, overloading is
defined based on the balance of generation and consumption
of active power solely. Reactive power was not considered
in this approach assuming that reactive power support (in the
form of fixed or switched capacitors) is available in the net-
work. The proposed OMT can be further modified to consider
the interconnection of MGs for reactive power support, after
a detailed techno-economic analysis.

IV. DYNAMIC M ULTICRITERIA DMA


The multicriteria decision-making model prescribes
a method for prioritizing and selecting the most favor-
able alternative from a set of alternatives, denoted by
A {A1 , A2 , . . . , ANA }, and based on a set of criteria, denoted
by c {c1 , c2 , . . . , cNc }, where Nc being the number of
criteria. Each criterion may have a different weighting. The
corresponding normalized weightings forthe criteria are
denoted by w {w1 , w2 , . . . , wNc }, where j wj = 1. These
weightings are defined by the help of the experts, as discussed
later in this section.
The multicriteria decision-making problem can be modeled
in the form of a matrix as [26]

c1 (w1 ) c2 (w2 ) ... cNC (wNC )



A1 x11 x12 ... x1NC
A2 x21 x22 ... x2NC Fig. 4. Flowchart of the developed multicriteria dynamic DMA.

.. .. .. .. .. (14)
. . . . .
ANA xNA 1 xNA 2 ... xNA NC
product, maxmin, and Hurwitz in the form of [27][29]

where xuv represents the performance of alternative Au from 


the perspective of criterion cv . The matrix of (14) will then be 
NC
Xu = average (wv xuv ) = wv xuv NC
modified to include the weighting of each criterion as
u=1
NC
c1 c2 ... cNC Xu = product (wv xuv ) = wv xuv

A1 w 1 x11 w2 x12 ... wNC x1NC X1 u=1

A2 1 x21
w w2 x22 ... wNC x2NC Xu = min (wv xuv )v=1,..., NC
X2
.. .. .. .. .. = .. . Xu = Hurwitz(wv xuv )
. . . . . .
ANA w1 xNA 1 w2 xNA 2 ... wNC xNA NC XNA = 3 min(wv xuv ) + (1 3 ) max(wv xuv ) (17)
(15)
where min(.) and max(.) are, respectively, the minimum and
maximum functions and 3 [0, 1] (e.g., 3 = 0.75) is the
The evaluation results for the alternatives, denoted by X
optimist coefficient used in the Hurwitz method. Each of these
{X1 , X2 , . . . , XNA }, are calculated as
aggregators has advantages and disadvantages as discussed
in [27][29]. The alterative with the highest X has the highest
Xu = agg(wv xuv ) (16) priority and is selected by the DMA.
The flowchart of the developed multicriteria dynamic
where agg(.) is an aggregating function such as average, DMA is shown in Fig. 4.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON SMART GRID

A. Risk Index of either linguistic (extremely big/small, very big/small,


All of the aggregators given in (17) may not necessarily big/small, a little big/small, and neutral) or number (between
select the same alternative. To overcome this problem, the risk 0% and 100%). After this, the linguistic and numerical values
index, denoted by R {R1 , R2 , . . . , RNA }, is defined for each will be first mapped into a number in [0, 1] range and then
alternative in the form of a matrix as [26], [27] normalized. The weighting for each criterion will be defined
     max  as the average of all normalized values as
c1 x1max c2 x2max . . . cM xN
M

Ne 
A1 r11 r12 ... r1NM R1 wu = wiu Ne (21)
A2 r21 r22 ... r2NM
R2

i=1
.. . .. .. .. = . (18)
. .. . . . .. where Ne is the number of the experts. To achieve a high
ANA rNA 1 rNA 2 ... rNA NM RNA confidence in the weightings, the variation coefficient for
where ruv for criterion cv is the deviation of the performance data dispersion (v ) of criterion cv is calculated as
 
of each alternative versus the alternative with the highest  Ne 
 
v = 
performance and is calculated from 2
wiu wu Ne wu . (22)
ruv = wv xuv xvmax (19) i=1

If v is smaller than a predefined small threshold (e.g., ), it


where xvmax is the maximum of wv xuv for criterion cv among
can be concluded that wv represents the true weighting for cri-
all alternatives. From (18), the risk index for alternative Au ,
terion cv based on experiences and outlooks of all participated
denoted by Ru , is selected as the maximum of ruv .
experts. If v is not smaller than the predefined threshold,
In the case of difference among the selections by aggrega-
more experts should be invited to participate in the census so
tors of (17), the risk index of the alternatives selected by the
that the weightings fulfill the desired dispersion.
aggregators are analyzed only to speed up the process and
the selected alternative with a lower risk index is chosen by
the DMA as the suitable alternative. D. Qualifying Criteria
All possible alternatives may not be qualified to couple with
B. Dynamic DMA the overloaded MG(s) due to reasons such as lack of enough
surplus power in MGs of an alternative, nonstandard voltage,
DMA carries out dynamically. However, overtime the avail-
and frequency deviation in the system after an alternative is
able alternatives can be modified, e.g., newer alternatives get
coupled with the overloaded MG(s) or when one MG vetoes
added or previously existing alternatives vanish depending
supporting another MG.
on the overloaded MGs. Furthermore, the criteria and their
Hence, the DMA first qualifies the alternatives based on
weightings can dynamically change due to the preference of
these conditions. The first four criteria are defined as follows.
the network operator. Hence, a dynamic DMA is utilized in
1) Criterion 1: MG consent/veto for coupling.
which the historical decision-making results are considered in
2) Criterion 2: Availability of surplus power in MGs.
future decision-makings. In dynamic DMA, first the evalua-
3) Criterion 3: Voltage deviation.
tion result of each alternative is defined from (15). Then, a new
4) Criterion 4: Frequency deviation.
dynamic evaluation result, denoted by X d {X1d , X2,...,
d X d },
NA The performance of alternative Au versus these criteria is
is defined at iteration k based on the previous iteration as [28]
expressed as
Xud = Xuk + Xuk1 + Xuk Xuk1 . (20) 
0 if MG vetoes coupling
xu1 =
Note that in case an alternative appears for the first time, its 1 if MG consents coupling

current evaluation result is used instead of (20). In addition, if 1 if u 3
an alternative vanishes in a new decision-making iteration but xu2 = 0.5(u 1) if 1 u < 3

it was available in the previous iteration, its evaluation result 0 if u < 1 or xu1 = 0
is stored in memory and retrieved in the first decision-making
10V u + 1 if 0 V u 0.1
iteration that the alternative appears again.
xu3 = 10V u + 1 if 0.1 V u < 0

0 if 0.1 < |V u | or xu1 xu2 = 0
C. Criteria Weightings
2F u + 1 if 0 F u 0.5
The outcome of the DMA highly depends on the assumed xu4 = 2F u + 1 if 0.5 F u < 0
weightings for each criterion. Therefore, these weightings
0 if 0.5 < |F u | or xu1 xu2 xu3 = 0
should be selected carefully. In complex systems such as (23)
power systems, there is not a systematic method to define
these weightings. An acceptable method is a census from the where u is the ratio of the UPC in MG(s) of alterna-
experts. For this purpose, a group of experts are required tive Au versus the PDL of overloaded MG-j (i.e., u =
to be asked to participate in defining the weightings for UPCu /PDLMG-j ).
these criteria. They will evaluate the weightings for each cri- If any of the above-mentioned four criteria define an alter-
terion based on their experience and outlook in the form native unqualified, the performance of that alternative versus
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SHAHNIA et al.: COUPLING NEIGHBORING MGs FOR OVERLOAD MANAGEMENT 7

the remaining criteria is neglected, that is Although the second method is not as accurate as considering
 PFA but it highly improves the speed of decision-making and
0 if xu1 xu2 xu3 xu4 = 0
xuv = v {5, . . . , NC }. is, therefore, preferred and used in this paper.
xuv if xu1 xu2 xu3 xu4 = 0
2) Criterion 6 (Electricity Price): Another important crite-
(24) rion in selecting an alternative is the price of electricity offered
by the owner of each MG. An MG owner may sell electric-
E. Other Criteria ity to neighboring MGs with a different price with respect
Six other criteria are considered for alternative selection to its own costumers. The price can also dynamically change
other than the four qualifying criteria. These criteria con- over time due to different reasons (e.g., the variations in the
sider the line loss, electricity price, reliability indices, supply price of fuel consumed in diesel generators or the availability
security, and CO2 emission when selecting a suitable alter- of power from intermittent nondispatchable DERs). Assuming
native. It is to be noted that defining and calculating these Ei as the electricity unit price offered by MG-i (in $/kWh),
indices for each of the MGs is beyond the scope of this paper. the total equivalent electricity cost to be paid by the owner of
Assuming these parameters are known and revealed to the the overloaded MG to the owners of MG(s) in alternative Au
OMT, the performance of each alternative for criteria 510 will is defined as
be calculated, as discussed in the following. 
p
1) Criterion 5 (Power Loss in Interconnecting Lines): Ecost,u = i,u Ei . (28)
One important criterion in the selection of an alternative is i=1
the power loss in the interconnecting lines (tie-lines) which
From (28), the performance of alternative Au is calculated for
depends on the distance between the overloaded MG and the
this criterion as
selected MG(s) and the impedance of these lines. The power
loss can be calculated from PFA for each alternative. Then, xu6 = 1 Ecost,u / max (Ecost,u )u=1,..., NA . (29)
it is normalized in the range of [0, 1] where 0 and 1 are,
respectively, for the alternative with maximum and minimum 3) Criteria 7 and 8 (Reliability): It is highly probable that
power loss. each MG may have a different reliability level. In addition,
To reduce the required computation time, a second method the failure rate of CMG is the sum of the failure rates of each
is utilized in this paper in which the distance and impedance participating individual MG. Therefore, as the number of MGs
of the line(s) among the overloaded MG and the MG(s) of increases in an alternative, the failure rate of the alternative
each alternative are considered. These values are predefined increases (i.e., the reliability decreases). It is highly desirable
for the DMA and the only unknown parameters are the power for the overloaded MG to couple with MG(s) in an alternative
flow from each MG to the overloaded MG. However, based which have higher reliability to reduce the possibility of inter-
on droop characteristic, each DER in the MG shares the load ruption to the essential load of the overloaded MG. Different
based on a predefined ratio (i.e., droop curve coefficients). reliability indices can be considered such as SAIFI, MAIFI,
Based on this concept, each MG shares a portion of the PDL and SAIDI. SAIFI and MAIFI represent the frequency of sup-
of the overloaded MG according to the maximum capacities of ply interruptions whereas SAIDI represents the duration of
f
its dispatchable DERs. Hence, assuming p MGs in alternative interruptions (in minutes). RBu and RBdu indices are defined
Au , coefficient i represents the ratio of power supplied from for alternative Au composed of p MGs as
MG-i to overloaded MG-j as 
p
 p RBuf = 1 SAIFIi,u + 2 MAIFIi,u
 
i,u = max
Pdisp-DER (MG-i) Pmax
disp-DER (MG-i)
i=1
i=1 p

i {1, . . . , p} (25) RBdu = SAIDIi,u (30)


 i=1
where Pmax
disp-DER (MG-i)
shows the maximum power capacity
where {1 , 2 } [0, 1] and 1 + 2 = 1 are the assumed
of all dispatchable DERs in MG-i. Coefficient i,u is calcu-
weightings for SAIFI and MAIFI, respectively. In this paper,
lated once only and is not repeated in each iteration. Thus,
it is assumed that 1 = 2 . From (30), the performance of
the computation time is significantly improved compared to
alternative Au is calculated for these criteria as
the PFA. After i is calculated, the total power loss parameter   
of alternative Au is calculated as xu7 = 1 RBuf max RBuf

p    u=1,..., NA
Ploss,u = i,u (Zi Di ). (26) xu8 = 1 RBdu max RBdu . (31)
u=1,..., NA
i=1

Note that (26) is not equal to the power loss of each alternative Alternatively, these criteria can be combined together with dif-
(in kVA) but represents a relationship among different alterna- ferent weightings to constitute one criterion only. However,
tives from line loss prospective. From (26), the performance since the performance of each alternative is easily compared
of each alternative is calculated for this criterion as when they are separate (due to dimension difference), reliabil-
 ity is considered as two criteria but with the same weightings
xu5 = 1 Ploss,u max(Ploss,u )u=1,..., NA . (27) in this paper.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON SMART GRID

4) Criterion 9 (Supply Security): If an MG has high power


generation by its nondispatchable DERs, it may have a higher
UPC in its dispatchable DERs. Once this MG is coupled
with an overloaded MG, the overloaded MG does not have
a high supply security as any unexpected drop in the power of
the nondispatchable DERs (due to environmental conditions)
may cause overloading of the CMG. Hence, it is important
to consider the security of supply when selecting an alterna-
tive. Supply security index SSu is defined for alternative Au
composed of p MGs as
 p
p  
SSu = Pdisp-DER(MG-i) Pload(MG-i) (32) Fig. 5. Assumed power system topology for the MGs in the distribution
i=1 i=1 network under consideration.

where Pload(MG-i) shows the average active power con- TABLE I
sumed by all loads in MG-i. From (32), the performance of A SSUMED C LOUD T HEORY-BASED P ROBABILISTIC P OWER DATA AND
THE C ALCULATED UPC AND PDL FOR THE D ISTRIBUTION
alternative Au is calculated for this criterion as N ETWORK [kW] (E XAMPLE 1)

xu9 = 1 SSu / max (SSu )u=1,..., NA . (33)

5) Criterion 10 (CO2 Emissions): Consider a distribution


network in which the distribution network operator penalizes
the MG owners based on their level of CO2 emission [30].
Hence, each MG owner may charge CO2 emission penal- TABLE II
ties from customers in the form of carbon tax, based on the A SSUMED C LOUD T HEORY-BASED P ROBABILISTIC DATA FOR E ACH MG
supplied electricity to each customer. When selecting an alter- OF THE D ISTRIBUTION N ETWORK U SED IN DMA (E XAMPLE 1)

native, it is desired to select an alternative with less CO2


emissions to minimize the penalties imposed to the overloaded
MG. Therefore, the level of CO2 emissions by each MG can
be considered as a criterion. Assuming the total CO2 emission
for MG-i as Emi , CO2 emission index is defined for alternative
It is worth mentioning that in reality, the MGs will have
Au composed of p MGs as
different topologies; however, for simplicity, the same struc-

p ture is considered for all of the MGs in this section. It is to
CO2u = i,u Emi . (34) be highlighted that the developed OMT does not depend on
i=1 neither the MG topology nor the line parameters.
A cloud theory-based stochastic analysis is carried out in
From (34), the performance of alternative Au is calculated for
this paper which defines randomly the MG essential and
this criterion as
nonessential loads, capacity of the dispatchable and nondis-
xu10 = 1 Co2u /max (Co2u )u=1,..., NA . (35) patchable DERs, nominal and actual wind speed for wind
turbines, sun radiation for solar-based DERs, electricity price,
CO2 emission, and the reliability indices for each MG. The
V. P ERFORMANCE E VALUATION cloud theory-based stochastic analysis is discussed in details
To evaluate the performance of the developed OMT based in Appendix C and all uncertainty values are provided in
on the proposed DMA, first the network of Fig. 2, composed of Tables XX and XXI.
three isolated MGs, is considered. Later, a larger distribution First, let us consider the network of Fig. 2 with three MGs.
network composed of six MGs is focused. It is to be noted Examples 13 demonstrate three examples in which the OMT
that although in theory a network with infinite number of MGs flags that one MG is overloaded while there is surplus power in
can be assumed, in reality, the number of MGs in a remote the other MGs of the network [i.e., condition (4) is invalid for
area network will be limited. one of the MGs while constraint (5) is valid for the distribution
In the carried out analysis, all MGs are assumed to have network].
the power system topology of Fig. 5. In this topology, it is Table I provides the assumed cloud theory-based probabilis-
assumed that the nondispatchable solar and wind-based DERs tic power data for all three MGs as well as the calculated
are connected to buses 1 and 2, respectively, whereas the dis- UPC and PDL for example 1. This table illustrates that MG-1
patchable DER (i.e., diesel-generator) is connected to bus 4. is overloaded, while MG-2 and MG-3 have surplus power.
The essential and nonessential loads are assumed to be con- Thus, several alternatives (i.e., A1 = {MG-2}, A2 = {MG-3},
nected to bus 3, while bus 5 is the assumed interconnection and A3 = {MG-2, MG-3}) are available for MGs interconnec-
point of the MG to its neighboring MGs. The assumed line tion and thereby the OMT calls the decision-making function.
parameters are provided in Table XIX in Appendix A. Table II lists the assumed cloud theory-based probabilistic
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SHAHNIA et al.: COUPLING NEIGHBORING MGs FOR OVERLOAD MANAGEMENT 9

TABLE III TABLE X


C OMPARISON A MONG THE PARAMETERS OF AVAILABLE A LTERNATIVES N ORMALIZED W EIGHTED D ECISION -M AKING M ATRIX (E XAMPLE 3)
TO S UPPORT OVERLOADED MG-1 (E XAMPLE 1)

TABLE XI
TABLE IV S ELECTED A LTERNATIVE AND E VALUATION R ESULTS F ROM D IFFERENT
C ALCULATED D ECISION -M AKING M ATRIX (E XAMPLE 1) AGGREGATORS AND R ISK M ATRIX (E XAMPLE 3)

TABLE V
A SSUMED W EIGHTINGS FOR THE C RITERIA IN DMA

TABLE XII
R ISK M ATRIX (E XAMPLE 3)

TABLE VI
C ALCULATED W EIGHTED D ECISION -M AKING M ATRIX (E XAMPLE 1)

TABLE XIII
A SSUMED C LOUD T HEORY-BASED P ROBABILISTIC P OWER DATA AND
THE C ALCULATED UPC AND PDL FOR THE
N ETWORK [kW] (E XAMPLE 4)
TABLE VII
N ORMALIZED W EIGHTED D ECISION -M AKING M ATRIX (E XAMPLE 1)

TABLE VIII
S ELECTED A LTERNATIVE AND E VALUATION R ESULTS F ROM D IFFERENT
AGGREGATORS (E XAMPLE 1)
TABLE XIV
A SSUMED C LOUD T HEORY-BASED P ROBABILISTIC DATA FOR E ACH MG
OF THE D ISTRIBUTION N ETWORK U SED IN DMA (E XAMPLE 4)

TABLE IX
D ECISION -M AKING M ATRIX (E XAMPLE 2)

The normalized weighted decision-making matrix is then used


to define the evaluation result of each alternative, using the
aggregators of (17), as given in Table VIII. For example 1,
data for each MG of the alternatives for this example. These MG-1 is overloaded and alternative A3 is the preferred choice
data are utilized in DMA to define the performance of each of all aggregators and is selected to couple with MG-1. To
alternative for each criterion, as seen from Table III. Table IV investigate the reasons behind the selection of A3 and its supe-
demonstrates the calculated decision-making matrix of (14). riorities over the two other alternatives, the numerical values of
Let the normalized weightings for the considered crite- Table III should be investigated. From this table, it can be seen
ria based on the experts comments, defined from (21), be that A3 presents the best performance from the perspective of
as listed in Table V. From the data of Tables IV and V, the criteria 2 and 3, while it has an average performance from the
weighted decision-making matrix is calculated from (15) as perspective of criteria 46, 9, and 10. A3 also illustrates the
given in Table VI and then normalized as given in Table VII. worst performance from the perspective of criteria 7 and 8.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON SMART GRID

TABLE XV
C OMPARISON A MONG THE PARAMETERS OF AVAILABLE A LTERNATIVES TO S UPPORT OVERLOADED MG-1 (E XAMPLE 4)

TABLE XVI
D ECISION -M AKING M ATRIX (E XAMPLE 4)

Finally, the DMA has selected this alternative since crite- such scenarios. Table XIII provides the assumed cloud theory-
ria 24, and 6 had larger weightings versus criteria 7 and 8, based probabilistic power data for all six MGs as well as the
based on the experts. calculated UPC and PDL for Example 4. This table illus-
Example 2 demonstrates another scenario in which MG-2 is trates that MG-2 and MG-6 are overloaded simultaneously.
overloaded and alternative A1 (i.e., {MG-1}) fails to satisfy Thus, the OMT calls the decision-making function to for-
one of the qualifying criteria (i.e., criterion 4) as seen from mulate the possible alternatives to select the suitable one.
Table IX. Hence, the performances of this alternative for all Table XIV provides the assumed cloud theory-based proba-
other criteria are neglected based on (24) and are discarded bilistic data for each of the MGs of the distribution network
from the alternatives. for this example which are utilized to define the performance
Example 3 demonstrates another case in which MG-3 is of each alternative for each criterion, as seen from Table XV.
overloaded. The normalized weighted decision-making matrix For this example, a total of 15 alternatives are possible, as
is as given in Table X and the evaluation results of each listed in Table XV. The calculated decision-making matrix
alternative, based on the aggregators of (17), are given in for this example is provided in Table XVI. Table XVII illus-
Table XI. From this table, it is seen that alternatives A1 and trates the results of different aggregators for each of the
A3 are selected by the aggregators. Hence, a risk matrix is alternatives. Two alternatives (i.e., A6 and A7 ) are selected
defined for these alternatives from (18) and (19), as given in by the aggregators. Thus, their risk index is calculated by the
Table XII. From this table, alternative A3 (i.e., {MG-1, MG-2}) DMA. The risk index of both of these alternatives is the same
is selected as it demonstrates a lower risk index. (i.e., 0.47). So, the DMA selects one of them randomly. In
To illustrate the performance of the developed OMT for this example, alternative A6 is selected. Hence, the OMT sends
a distribution network with larger number of MGs, and a command to the ISSes of MG-1, 3, 5, and 6 to close and
thus a larger number of alternatives, another study is car- form a CMG.
ried out which assumes the network of Fig. 2 with six The stochastic analysis results of the network of Fig. 2 with
MGs. Thus, the OMT may evaluate up to 31 alternatives three MGs are summarized in Table XVIII which demon-
to define the suitable one. Example 4 illustrates one of strates the flags pointed out by the OMT and its embedded
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SHAHNIA et al.: COUPLING NEIGHBORING MGs FOR OVERLOAD MANAGEMENT 11

TABLE XVII
S ELECTED A LTERNATIVE AND E VALUATION R ESULTS F ROM of all three MGs is required among which 6.45% satisfy
D IFFERENT AGGREGATORS (E XAMPLE 4) PFA constraint (6) and interconnection is possible whereas
in 2.84% PFA constraint (6) is not satisfied. In 41.54% of
the cases, several alternatives are possible and the OMT calls
the decision-making function where in all of them, a suitable
alternative is selected. In 4.45% of the selected alternatives,
the DMA selects an alternative which is composed of one
MG while in 37.09% an alternative composed of two MGs
is selected. In 23.48% of the cases, the OMT calls the
load-shedding function where load-shedding is found to be
successful in all MGs.
From Table XVIII, it can also be seen that in 80.66% of
the cases, the OMT calls the load-shedding function where
load-shedding is found to be unsuccessful in 0.01% and suc-
cessful in 65.18% of them. Coupling of MGs accompanied
by load-shedding counts for 14.59% of the load-shedding sec-
tion, among which in 7.59% of them the selected alternative
is composed of one MG and in 7% of them the selected alter-
native is composed of two MGs. In 0.88% of the total cases,
TABLE XVIII it is found that coupling of MGs by further load-shedding is
S TOCHASTIC A NALYSIS R ESULTS D EMONSTRATING THE F LAGS not viable.
G ENERATED BY THE OMT AND D ECISION -M AKING AND
L OAD -S HEDDING F UNCTIONS FOR A N ETWORK
Note that the utilized stochastic analysis has a minimum of
C OMPOSED OF T HREE MG S U NDER T WO 10 000 trials to prevent any immature results. The stopping
C ONSIDERED S TUDY C ASES rule is defined to achieve a confidence level of 95% in the
mean and standard deviation of the results.

VI. C ONCLUSION
An OMT is developed in this paper to reduce the load-
shedding rate of a remote area MG, during overloading con-
ditions, by interconnecting it with suitable neighboring MG(s).
A dynamic multicriteria DMA is presented to formulate the
possible alternatives, qualify them based on four proposed cri-
teria and then assess and select the most suitable one to achieve
the highest satisfaction and minimum risk considering another
six criteria. All criteria have different weightings. In case, sur-
plus power is not available in the other MGs, the developed
OMT proceeds to define the required ALS from each MG.
If this level is less than the nonessential part of the loads of
an MG, the developed technique defines which neighboring
MG(s) have extra nonessential loads. It then proceeds to iden-
tify those MGs, defines the portion of nonessential load to be
disconnected from them, and then interconnects those MGs.
The successful performance of the developed algorithms is val-
idated in a stochastic frame in MATLAB for a small network
composed of three MGs as well as a large network, composed
decision-making and load-shedding functions. The table shows of six MGs.
the results of two sets of stochastic analyses. It is worth mentioning that the proposed method of coupling
1) Case 1: The assumption that the dispatchable DERs the MGs can be expanded to be used when extra surplus power
have a capacity of 60% nominal load demand based is available in the DERs of an MG. Thus, the extra available
on [11] and [12]. power can be exported to other MGs with a lower electric-
2) Case 2: The assumption that the dispatchable DERs have ity price, to be stored in the energy storage units or to be
a capacity of 30% of the nominal load demand. consumed by the controllable loads of the other MGs, under
From this table, it is seen that for the considered uncer- demand dispatch concept.
tainties of case 1, no coupling of the MGs is required for It is also noteworthy that analyzing the effect of different
25.69% of the cases, as interconnection condition (4) is satis- weightings for each criterion and thereby the outcome of the
fied for all three MGs. In 9.29% of the cases, interconnection DMA can be another future research topic.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

12 IEEE TRANSACTIONS ON SMART GRID

It is to be noted that in this paper, it was assumed that the Then, the p.u. total power loss in the lines of MG, denoted
MGs are interconnected through one bus only. However, in by Sloss , is calculated as
general, it is possible to consider the interconnection of one  k  k 

Nbus 
Nbus
k 2
MG through different buses to different MGs. This can be loss p
S = Yi,k
bus
Vi p Vk p . (A4)
a future research topic and the power flow control under such i=1 k=i
conditions needs to be investigated.
Hence, the required power to be generated by the dispatchable
DERs is
 kp  
A PPENDIX A
Sdisp DER = Sload + (Sloss )kp Snon-disp DER . (A5)
D EVELOPED P OWER F LOW A NALYSIS FOR M ICROGRIDS
 disp DER
Standard PFAs have one slack bus with known voltage mag- The active and reactive power part of S is shared
nitude and angle and regulated buses with known voltage among the dispatchable DERs of the MG based on their
magnitudes and active powers. It also assumes that frequency predefined droop ratios, denoted by mD and nD , as

is constant. The MGs operating in islanded mode do not have   1 mD  
disp DER kp
slack bus. For an MG operating in islanded mode, the DERs Pi = N i
 Re S disp DER
disp DER
modify their output voltage magnitude and frequency, using 1 mD
i=1
 D i
the droop control, based on the network loading conditions,  kp 1 n  
as discussed in [31] and [32]. Hence, the PFA for an islanded Qdisp
i
DER
= 
i
Ndisp DER  Im S disp DER (A6)
MG needs to consider these characteristics, as highlighted i=1 1 nD
i
in [33]. It is also to be noted that as the voltage and frequency where i {1, . . . , Ndisp DER }, while Re(.) and Im(.) are,
are not constant in the MG under different loading conditions, respectively, the real and imaginary functions.
the dependency of the load and line parameters on voltage and Once the active powers generated by the dispatchable DER
frequency should also be considered in the PFA. buses are defined from (A6), the MG frequency is calculated
A GaussSeidel-based PFA is developed in this paper and from the droop curve of either of the dispatchable DERs as
used within the OMT to compute the maximum voltage  
DER kp
and frequency deviation for the considered CMG systems. f kp = fmax mi Pdisp
i . (A7)
Consider an MG with Nbus buses, Ndisp DER dispatchable In addition, the voltage magnitude of each dispatchable DER
DERs and Nnon-disp DER nondispatchable DERs (which may bus is calculated from the reactive power-voltage droop coef-
have a radial or mesh structure), and an admittance matrix of ficient of each DER as
Y bus (in p.u.). Assume that a bus is connected to either a load  
DER kp
or a DER. Bus-i when connected to a load (referred to as |Vi |kp = Vmax ni Qdisp
i . (A8)
load-bus in the rest of this section) has a power consumption
Due to lack of slack bus, the bus connected to one of the
of Siload = Pload
i + jQload
i (in p.u.) whereas when connected
dispatchable DERs (e.g., the first dispatchable DER) needs to
to a nondispatchable DER (referred to as nondispatchable
non-disp DER be assumed as the reference bus. The angle of the reference bus
DER bus) has a power consumption of Si =
non-disp DER is always zero and all other angles in the MG are considered
Pi (in p.u.). Assuming a set of initial values for with respect to this bus. Thus, the voltages of the dispatchable
the voltages of all buses (e.g., 1 0.01 p.u.), the PFA first DER buses are updated at each iteration based on the voltage
calculates the current drawn by each load in iteration kp as magnitudes calculated in (A8) and the angles of the voltages
k
  k 1  in the previous iteration as
Ii p = conj Si Vi p i {1, . . . , Nload } (A1) 
kp |Vi |kp 0  k 1  i = 1
Vi = (A9)
where conj(.) is the conjugate function. In a similar way, the |Vi | angle Vi p
kp
i {2, . . . , Ndisp DER }
current injected by each nondispatchable DER is calculated for where angle(.) is the function to derive the angle.
i {1, . . . , Nnon-disp DER }. Then, the voltages of all load and The considered loads in this paper are assumed to be
nondispatchable DER buses are calculated similar to classic a voltage- and frequency-dependent constant impedance type.
PFA as Thus, all loads are updated in each iteration of the PFA based
  on the voltage magnitude and frequency of the buses to which
Nbus
kp kp bus kp 1
Vi = Ii Yi,k Vk Yibus . (A2) they are connected at the previous iteration as [34]
 kp +1  k  
p kv 1 + k F kp
k=1
Pload
i = P0 |Vi | p
Once the voltages for all load buses and nondispatchable    k  
load kp +1
DER buses are updated from (A2) at each iteration, the Qi = Q0 |Vi | p kv 1 + kq F kp (A10)
voltages of the buses are slightly modified based on clas- where P0 and Q0 are, respectively, the assumed nominal active
sic GaussSeidel method with a correction/acceleration factor and reactive power of the load (in p.u.) while k v = 2, k p =
of 4 as 0.1, and k q = 0.1.
k k 1
 k k 1
 The assumed line parameters for the MG power system of
Vi p = Vi p + 4 Vi p Vi p . (A3) Fig. 5 are provided in Table XIX at fundamental frequency
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SHAHNIA et al.: COUPLING NEIGHBORING MGs FOR OVERLOAD MANAGEMENT 13

TABLE XIX
A SSUMED L INE PARAMETERS OF THE MG S YSTEM Algorithm 1: Overload Management Technique (OMT)
 max
OF F IG . 5 AT F UNDAMENTAL F REQUENCY Pre-defined Inputs: P
 disp-DER (MG
 -i) 
Dynamic Inputs:
 Pload(MG-i) , Pnon-essload(MG-i) , Pdisp-DER(MG-i)
Pnon-disp-DER(MG-i) ,
Dynamic Outputs:: Flag, command to ISSes, ALSi , ALSi2
Main Function:
1 Calculate UPCi from (3) for each microgrid;
2 if condition (4) is satisfied for all microgrids then
3 Flag: No interconnection is required;
4 else 
5 Calculate UPCii=1,...,N from (3);
6 if constraint (5) is not satisfied then
7 Flag: Coupling microgrids is not helpful and load-shedding is
inevitable;
8 Call Load-shedding Function to evaluate the possibility of
load-shedding;
9 else
10 if N 1 microgrids are overloaded based on condition (4) then
(i.e., 50 Hz). It is to be emphasized that the impedance of 11 Flag: The only alternative is coupling all N microgrids;
the lines and thus the Y bus of the system are updated in each 12 Call PFA Function;
13 if constraint (7) is satisfied then
iteration to consider the effect of the frequency changes in 14 Flag: All N microgrids can be coupled;
the MG. 15 Send the command to relevant ISS(es) to close;
16 else
At every iteration, the mismatch value is calculated as the 17 Flag: Coupling all N microgrids is not viable and
maximum of differences between all bus voltages of the MG load-shedding is inevitable;
18 Call Load-shedding Function to evaluate the possibility of
with their values in the previous iteration along with the dif- load-shedding;
19 end
ferences between the active/reactive power of all dispatchable 20 else
DERs with the values in the previous iteration, as 21 Flag: Several alternatives are possible; hence, decision-making is
  22
required;
Call Decision-making Function to assess the alternatives;
 = max |V| , P , Q (A11) 23 if an alternative is selected by Decision-making Function then
24 Flag: A suitable alternative is selected;
where 25 Send the command to relevant ISS(es) to close;
  26 else
|V| = max |Vi |kp |Vi |kp 1 i=1,..., N 27 Flag: A suitable alternative is not available and
load-shedding is inevitable;
   bus
  28 Call Load-shedding Function to evaluate the possibility of
disp DER kp disp DER kp 1
P = max Pi Pi load-shedding;
i=1,..., Ndisp 29 end
     DER
30 end
disp DER kp disp DER kp 1
Q = max Qi Qi . 31 end
i=1,..., Ndisp DER 32 end
Decision-Making Function:
Once  is smaller than a predefined value (e.g., 1010 ), the 1
2
Formulate the alternatives;
Call DMA;
PFA is deemed to be converged. 3 if an alternative is selected by DMA then
4 Flag: A suitable alternative is selected for coupling the microgrids;
Through the observations in this paper, it is revealed that 5 Send the command to relevant ISS(es) to close;
the developed PFA converges as far as the assumed demand 6 else
7 Flag: A suitable alternative is not available and load-shedding is inevitable;
is within the power generation capacity of the DERs. It is 8 Call Load-shedding Function to evaluate the possibility of load-shedding;
to be noted that since the size of the considered network is 9 end
Load-shedding Function:
small, the number of iterations is limited and no convergence 1 Calculate ALSi from (9);
problems were observed. However, analyzing the convergence 2 if condition (10) is satisfied for all microgrids then
3 Flag: Load-shedding is successful;
property, rate, and regions of the developed PFA is beyond the 4 Send ALS to each microgrid to shed its loads accordingly;
scope of this paer and can be a future research topic. 5 else
6 Send the ALS level to those microgrid(s) that satisfy condition (10);
7 if constraint (11) is satisfied then
8 Calculate ALS2 from (12) & formulate alternatives satisfying (13);
A PPENDIX B 9 Call SLSA;
D EVELOPED A LGORITHM FOR M ICROGRID OVERLOAD 10 if an alternative is selected by SLSA then
11 Flag: Coupling of microgrids accompanied by further
M ANAGEMENT T ECHNIQUE load-shedding is selected;
12 Send ALS2 to relevant microgrid(s) to shed their loads;
The developed algorithm of the proposed OMT is given in 13 Send the command to relevant ISS(es) to close;
Algorithm 1. 14 else
15 Flag: Coupling of microgrids accompanied by further
load-shedding is not viable;
16 end
A PPENDIX C 17 else
C LOUD T HEORY-BASED S TOCHASTIC A NALYSIS 18 Flag: Coupling of microgrids accompanied by further load-shedding is
not viable;
To consider the uncertainties in the wind and solar genera- 19 end
20 end
tion, load, capacity of dispatchable DERs, reliability indices,
CO2 emissions, and electricity price in each MG, a stochas-
tic framework is developed in this paper. Monte Carlo is
an approach for modeling uncertainties in power system above-mentioned uncertainties can be compromised by the his-
studies in which the uncertainties are modeled with an torical statistics, the standard deviation is another uncertain
appropriate PDF. Although the mean of the PDF for the parameter [35], [36]. This issue can be solved by considering
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

14 IEEE TRANSACTIONS ON SMART GRID

TABLE XX
A SSUMED U NCERTAINTIES IN THE G ENERATION C APACITY AND from [37], [38]
D EMAND FOR THE MG S   !
Top 20
PPV = Nc FF V oc
kv TA + s
0.8
"  sc #
s I + ki (Tc 25) (C2)
where Nc is the number of cells, FF = 0.7 is the fill factor,
Tc = 25 C and Top = 25 C are, respectively, the assumed
cell and normal operational temperatures, TA is ambient tem-
perature, V oc = 22 V and I sc = 3.7 A are the assumed
open-circuit voltage and short-circuit current of the PV cells,
and kv = 2.3 and ki = 6E 5 are, respectively, the assumed
voltage and current coefficients.
TABLE XXI The random value for the nominal (rated) wind speed of
A SSUMED U NCERTAINTIES IN THE PARAMETERS wind-based DERs is chosen from a normal PDF where the
OF THE C RITERIA IN DMA
value for the wind speed (vwind ) is chosen randomly from
a Rayleigh PDF as [37], [38]
2vwind
e(vwind /cv )
2
frayleigh (vwind ) = (C3)
cv
where cv 1.128 vm and vm is the average wind speed. Thus,
the probabilistic output power of a wind-based DER, denoted
by Pwind , is [37], [38]


0 vwind < vci or vwind > vco
v vci
max
Pwind = Pwind nominal vci vwind < vnominal
wind

v vci
max wind
Pwind vnominal
wind vwind vco
a stochastic framework based on cloud theory in which the (C4)
standard deviation for each uncertainty is defined as a hyper where vci = 3.5 m/s and vco = 25 m/s are, respectively, the
entropy index. Hence, the uncertainty of parameter x is defined assumed cut-in and cut-out speeds for the wind turbine.
by a cloud theory-based model, denoted by CL (Ex , En , He )
where Ex is the expected (mean) value, En is the entropy
R EFERENCES
(variation range), and He is the hyper entropy (divergence
of variation range) [35], [36]. He uses a normal distribution [1] R. Paleta, A. Pina, and C. A. S. Silva, Polygeneration energy container:
Designing and testing energy services for remote developing commu-
to model the entropy of En and CL uses a normal distribu- nities, IEEE Trans. Sustain. Energy, vol. 5, no. 4, pp. 13481355,
tion to model uncertainty x with the assumed mean of Ex Oct. 2014.
and standard deviation of En . Thus, the cloud theory-based [2] M. Arriaga, C. A. Canizares, and M. Kazerani, Northern lights: Access
to electricity in Canadas Northern and remote communities, IEEE
stochastic framework captures the uncertainty that exists in Power Energy Mag., vol. 12, no. 4, pp. 5059, Jul./Aug. 2014.
the determination of standard deviation. The values for cloud [3] K. Ubilla et al., Smart microgrids as a solution for rural electrifica-
theory-based models of load demand and its essential and tion: Ensuring long-term sustainability through cadastre and business
models, IEEE Trans. Sustain. Energy, vol. 5, no. 4, pp. 13101318,
nonessential portion, maximum (capacity) of power generated Oct. 2014.
by wind and solar-based and dispatchable DERs, electricity [4] Y. M. Atwa et al., Adequacy evaluation of distribution system including
price, CO2 emissions, SAIFI, SAIDI, and MAIFI in each MG wind/solar DG during different modes of operation, IEEE Trans. Power
Syst., vol. 26, no. 4, pp. 19451952, Nov. 2011.
are listed in Tables XX and XXI where floor(.) is the floor [5] J. M. Guerrero, J. C. Vasquez, J. Matas, L. G. de Vicua, and M. Castilla,
function. Hierarchical control of droop-controlled AC and DC microgrids
The probabilistic load of network is chosen from a uniform A general approach toward standardization, IEEE Trans. Ind. Electron.,
vol. 58, no. 1, pp. 158172, Jan. 2011.
distribution over [0, Pmax
load ] range. [6] B. Kroposki et al., Making microgrids work, IEEE Power Energy
The random output power of a solar-based DER in each MG Mag., vol. 6, no. 3, pp. 4053, May/Jun. 2008.
depends on the random solar irradiance parameter, denoted by [7] M. Arriaga, C. A. Canizares, and M. Kazerani, Renewable energy alter-
natives for remote communities in Northern Ontario, Canada, IEEE
s, and derived from a Beta PDF as [37], [38] Trans. Sustain. Energy, vol. 4, no. 3, pp. 661670, Jul. 2013.
[8] E. Pashajavid and M. A. Golkar, Non-Gaussian multivariate modeling
of plug-in electric vehicles load demand, Int. J. Elect. Power Energy

(5 + 5 )
fbeta (s) = s5 1 (1 s)5 1 (C1) Syst., vol. 61, pp. 197207, Oct. 2014.

(5 )
(5 ) [9] Y.-Y. Hong, M.-C. Hsiao, Y.-R. Chang, Y.-D. Lee, and H.-C. Huang,
Multiscenario underfrequency load shedding in a microgrid consisting
of intermittent renewables, IEEE Trans. Power Del., vol. 28, no. 3,
where 5 and 5 are the shape parameters of the beta distribu- pp. 16101617, Jul. 2013.
[10] S. Bahramirad, W. Reder, and A. Khodaei, Reliability-constrained opti-
tion (e.g., 5 = 2 and 5 = 2). The probabilistic output power mal sizing of energy storage system in a microgrid, IEEE Trans. Smart
of a solar-based DER, denoted by PPV , is then calculated Grid, vol. 3, no. 4, pp. 20562062, Dec. 2012.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SHAHNIA et al.: COUPLING NEIGHBORING MGs FOR OVERLOAD MANAGEMENT 15

[11] S. A. Arefifar and Y. A.-R. I. Mohamed, DG mix, reactive sources [29] G. Campanella and R. A. Ribeiro, A framework for dynamic multiple-
and energy storage units for optimizing microgrid reliability and sup- criteria decision making, Decis. Support Syst., vol. 52, no. 1, pp. 5260,
ply security, IEEE Trans. Smart Grid, vol. 5, no. 4, pp. 18351844, Dec. 2011.
Jul. 2014. [30] Y. Wu, S. Lou, and S. Lu, A model for power system interconnection
[12] S. A. Arefifar, Y. A.-R. I. Mohamed, and T. H. M. El-Fouly, Optimum planning under low-carbon economy with CO2 emission constraints,
microgrid design for enhancing reliability and supply-security, IEEE IEEE Trans. Sustain. Energy, vol. 2, no. 3, pp. 205214, Jul. 2011.
Trans. Smart Grid, vol. 4, no. 3, pp. 15671575, Sep. 2013. [31] J. M. Guerrero, L. G. D. Vicuna, J. Matas, M. Castilla, and J. Miret,
[13] R. Majumder, A. Ghosh, G. Ledwich, and F. Zare, Power management A wireless controller to enhance dynamic performance of paral-
and power flow control with back-to-back converters in a utility con- lel inverters in distributed generation systems, IEEE Trans. Power
nected microgrid, IEEE Trans. Power Syst., vol. 25, no. 2, pp. 821834, Electron., vol. 19, no. 5, pp. 12051213, Sep. 2004.
May 2010. [32] F. Katiraei and M. R. Iravani, Power management strategies for a micro-
[14] R. Minciardi and R. Sacile, Optimal control in a cooperative network of grid with multiple distributed generation units, IEEE Trans. Power Syst.,
smart power grids, IEEE Syst. J., vol. 6, no. 1, pp. 126133, Mar. 2012. vol. 21, no. 4, pp. 18211831, Nov. 2006.
[15] R. H. Lasseter, Smart distribution: Coupled microgrids, Proc. IEEE, [33] A. Elrayyah, Y. Sozer, and M. E. Elbuluk, A novel load-flow analysis
vol. 99, no. 6, pp. 10741082, Jun. 2011. for stable and optimized microgrid operation, IEEE Trans. Power Del.,
vol. 29, no. 4, pp. 17091717, Aug. 2014.
[16] Z. Wang and J. Wang, Self-healing resilient distribution systems based
[34] P. Kundur, Power System Stability and Control. New York, NY, USA:
on sectionalization into microgrids, IEEE Trans. Power Syst., vol. 30,
McGraw-Hill, 1993.
no. 6, pp. 31393149, Nov. 2015.
[35] D. Li, D. Cheung, X. Shi, and V. Ng, Uncertainty reasoning based
[17] Z. Wang, B. Chen, J. Wang, and C. Chen, Networked microgrids for on cloud models in controllers, Comput. Math. Appl., vol. 35, no. 3,
self-healing power systems, IEEE Trans. Smart Grid, to be published. pp. 99123, Feb. 1998.
[18] Z. Wang, B. Chen, J. Wang, M. M. Begovic, and C. Chen, Coordinated [36] A. Kavousi-Fard, T. Niknam, and M. Fotuhi-Firuzabad, A novel
energy management of networked microgrids in distribution systems, stochastic framework based on cloud theory and -modified Bat algo-
IEEE Trans. Smart Grid, vol. 6, no. 1, pp. 4553, Jan. 2015. rithm to solve the distribution feeder reconfiguration, IEEE Trans.
[19] H. Dagdougui, A. Ouammi, and R. Sacile, Optimal control of a network Smart Grid, to be published.
of power microgrids using the Pontryagins minimum principle, IEEE [37] S. A. Arefifar, Y. A. I. Mohamed, and T. H. M. El-Fouly, Supply-
Trans. Control Syst. Technol., vol. 22, no. 5, pp. 19421948, Sep. 2014. adequacy-based optimal construction of microgrids in smart distribution
[20] F. Shahnia, R. P. S. Chandrasena, S. Rajakaruna, and A. Ghosh, Primary systems, IEEE Trans. Smart Grid, vol. 3, no. 3, pp. 14911502,
control level of parallel distributed energy resources converters in system Sep. 2012.
of multiple interconnected autonomous microgrids within self-healing [38] Y. M. Atwa, E. F. El-Saadany, M. M. A. Salama, and R. Seethapathy,
networks, IET Gener. Transm. Distrib., vol. 8, no. 2, pp. 203222, Optimal renewable resources mix for distribution system energy loss
Feb. 2014. minimization, IEEE Trans. Power Syst., vol. 25, no. 1, pp. 360370,
[21] Y. Zhang and L. Xie, Online dynamic security assessment of microgrid Feb. 2010.
interconnections in smart distribution systems, IEEE Trans. Power Syst.,
vol. 30, no. 6, pp. 32463254, Nov. 2015.
[22] E. Pashajavid, F. Shahnia, and A. Ghosh, Overload management of
autonomous microgrids, in Proc. 11th IEEE Int. Conf. Power Electron.
Drive Syst. (PEDS), Sydney, NSW, Australia, 2015, pp. 7378.
[23] E. Pashajavid, F. Shahnia, and A. Ghosh, Interconnection of two neigh-
boring autonomous microgrids based on small signal analysis, in Proc. Farhad Shahnia (S10M11SM15) photograph and biography not avail-
9th IEEE Int. Conf. Power Electron. Asia (ICPE-ECCE Asia), Seoul, able at the time of publication.
Korea, Jun. 2015, pp. 213220.
[24] W. Su, J. Wang, and J. Roh, Stochastic energy scheduling in microgrids
with intermittent renewable energy resources, IEEE Trans. Smart Grid,
vol. 5, no. 4, pp. 18761883, Jul. 2014.
[25] A. Brooks, E. Lu, D. Reicher, C. Spirakis, and B. Weihl, Demand
dispatch, IEEE Power Energy Mag., vol. 8, no. 3, pp. 2029,
May/Jun. 2010. Soheil Bourbour, photograph and biography not available at the time of
[26] S.-M. Chen and C.-H. Wang, A generalized model for prioritized mul- publication.
ticriteria decision making systems, Expert Syst. Appl., vol. 36, no. 3,
pp. 47734783, Apr. 2009.
[27] P. Y. Ekel, J. S. C. Martini, and R. M. Palhares, Multicriteria analysis in
decision making under information uncertainty, Appl. Math. Comput.,
vol. 200, no. 2, pp. 501516, Jul. 2008.
[28] N. Fenton and W. Wang, Risk and confidence analysis for fuzzy
multicriteria decision making, Knowl.-Based Syst., vol. 19, no. 6, Arindam Ghosh (S80M83SM93F06) photograph and biography not
pp. 430437, Oct. 2006. available at the time of publication.

You might also like