To improve the power supply availability in an island microgrid, this paper proposes a new approach that integrates distributed energy sources economically, reliably and efficiently. In an island mode, a microgrid must ensure its self-sufficiency of energy production since it cannot make an energetic exchange with a main grid. However, in this mode, the random behavior of the resources affected by meteorological factors presents a major constraint. The challenge related to the power availability in microgrids is to find a solution that faces the operation of intermittent power sources. The microgrid should guarantee a useful power management in order to achieve a high availability of energy. In this paper, we present a mathematical model to describe the influence of the meteorological factors on the sources production. We propose a multi-agent control strategy based on the production forecasting and load shedding for a high availability of the microgrid power supply. The proposed multi-agent system uses the master-slave model in which the communication and negotiation between the defined agents are performed by a concept of tokens. The developed control system is implemented on Spartan 6 FPGA-Board. The paper's contribution is applied to a Tunisian petroleum platform where several blackouts are recorded between 2012 and 2014. Simulation and experimental results show clearly a high availability as a performance of the proposed control strategy.

1. Electric Power Systems Research 152 (2017) 411–423
Contents lists available at ScienceDirect
Electric Power Systems Research
journal homepage: www.elsevier.com/locate/epsr
Multi-agent oriented solution for forecasting-based control strategy
with load priority of microgrids in an island mode – Case study:
Tunisian petroleum platform
Mohamed Ghaieth Abidia,b
, Moncef Ben Smidac
, Mohamed Khalguia,b,∗
, Zhiwu Lid,e,∗∗
,
Naiqi Wud
a
School of Electrical and Information Engineering, Jinan University (Zhuhai Campus), Zhuhai 519070, China
b
LISI Lab, INSAT Institute, University of Carthage, Tunisia
c
LSA Lab, Tunisia Polytechnic School, University of Carthage, Tunisia
d
Institute of Systems Engineering, Macau University of Science and Technology, Macau
e
School of Electro-Mechanical Engineering, Xidian University, Xi’an 710071, China
a r t i c l e i n f o
Article history:
Received 10 August 2016
Received in revised form 9 July 2017
Accepted 10 July 2017
Keywords:
Microgrid
Forecasting of meteorological conditions
Load shedding
Multi-agent system
FPGA
Simulation
a b s t r a c t
To improve the power supply availability in an island microgrid, this paper proposes a new approach that
integrates distributed energy sources economically, reliably and efficiently. In an island mode, a microgrid
must ensure its self-sufficiency of energy production since it cannot make an energetic exchange with
a main grid. However, in this mode, the random behavior of the resources affected by meteorological
factors presents a major constraint. The challenge related to the power availability in microgrids is to
find a solution that faces the operation of intermittent power sources. The microgrid should guarantee a
useful power management in order to achieve a high availability of energy. In this paper, we present a
mathematical model to describe the influence of the meteorological factors on the sources production.
We propose a multi-agent control strategy based on the production forecasting and load shedding for a
high availability of the microgrid power supply. The proposed multi-agent system uses the master-slave
model in which the communication and negotiation between the defined agents are performed by a
concept of tokens. The developed control system is implemented on Spartan 6 FPGA-Board. The paper’s
contribution is applied to a Tunisian petroleum platform where several blackouts are recorded between
2012 and 2014. Simulation and experimental results show clearly a high availability as a performance of
the proposed control strategy.
© 2017 Elsevier B.V. All rights reserved.
1. Introduction
Nowadays, many human activities depend critically on secure
supplies of energy. For many energy consumers, such as hospitals,
research centers and military bases, any temporary absence of elec-
trical power can lead to material and human losses. The service
quality and mainly the power supply availability are regarded as
paramount factors [1]. Due to several technical and economic con-
straints of conventional electrical networks, using the distributed
∗ Corresponding author at: School of Electrical and Information Engineering, Jinan
University, China.
∗ ∗ Corresponding author at: Institute of Systems Engineering, Macau University of
Science and Technology, Macau.
E-mail addresses: khalgui.mohamed@gmail.com (M. Khalgui),
zhwli@xidian.edu.cn (Z. Li).
energy production becomes a necessity [2]. Microgrid is a new gen-
eration of electrical networks, which aims to integrate different
electrical power technologies efficiently and reliably [3–5] in order
to meet the power requirements of consumers [6]. A microgrid is
composed of networked generation sources, energy storage devices
and loads interconnected and controlled by an energy manage-
ment system [7,8]. The potential for improving the power supply
availability is one of the main motivations behind the develop-
ment and deployment of microgrids [9,10], especially in an isolated
mode [11,12]. In this operation mode, a microgrid becomes an
autonomous power system. It should have its self-sufficiency in the
power production and should be able to ensure an accepted qual-
ity of energy requested by consumers [13,14]. In an island mode
with the absence of renewable energy sources, the microgrid is
supplied by backup sources. The major constraint to ensure a high
power availability is the randomness and intermittent behaviour
http://dx.doi.org/10.1016/j.epsr.2017.07.013
0378-7796/© 2017 Elsevier B.V. All rights reserved.

2. 412 M.G. Abidi et al. / Electric Power Systems Research 152 (2017) 411–423
Nomenclature
PV photovoltaic cells
WT wind turbine
B battery
GE diesel generator
PSource(t) electrical power exchanged between the source
and the rest of the network at time t
PLoad1(t) electrical power consumed by a critical load at time
t
PLoad2(t) electrical power consumed by an uncritical load at
time t
En(t) insolation at time t
VV(t) wind speed at time t
Echarge(t) battery charge level at time t
Eclim(t) sea state at time t
NCharge(t) level of the fuel in the tank of the diesel generator
at time t
Eclim
0
nominal sea state
k(t) electrical power produced by source k at time t
SBWD number of the days during which the renewable
energy sources are unavailable
APS(%) availability rate of the power supply
of renewable energy sources [15], which can cause an imbalance
of the energy between the production and consumption. The avail-
ability of the electrical energy in an isolated site presents one of the
main problems to be solved.
BARAKA platform is an islanded petroleum platform located at
the Tunisian coast. It is supplied by: (i) photovoltaic generators, (ii)
wind turbines, (iii) batteries and (iv) diesel generators. This plat-
form presents a real case study for an island microgrid that has
suffered several power blackouts. In bad weather conditions, the
platform becomes inaccessible for refuelling diesel generators. In
this condition, a renewable sources unavailability can cause a total
absence of electrical power supply. Between 2012 and 2014, six
total stops of production are registered. These stops caused approx-
imately one million dollars of losses for the Tunisian government.1
These significant losses motivate us to find a solution that improves
the availability and minimizes the number of stops.
In order to reduce the influence of intermittent sources
behaviour and to ensure the balance between the produced energy
and the consumers’ demands, several studies are done to analyze
the adequate types and capacities of sources. This analysis presents
the first step to improve any availability [16,17]. These studies show
significant technical results, though their solutions are econom-
ically expensive. Some studies explore also how the availability
of microgrids is impacted by their topology design. These studies
focus on the effect of the system architecture and the converters
topology on the system availability [18]. They present acceptable
technical results though their influence on the power availability
is limited.
A power management strategy also has a significant impact on
the energy control optimization [19,20]. As a result, the system is
able to deal with dangerous situations. Several studies focus on the
impact of the power management strategies on the electrical sup-
ply quality in microgrids, especially on its stability and availability
[21,22]. The power management can be ensured by various tech-
niques, ranging from a centralized control approach as reported
in [23] to a fully decentralized one, depending on the responsi-
1
Official statistics provided by CIPEM.
bility rates assumed by the central and the distributed microgrid
controllers. The centralized control is widely used in connected
mode-oriented microgrids. With this type of control, the optimiza-
tion problems become extremely complex. In fact, any modification
of the installation (loads or sources) influences the global con-
trol strategy. The decentralized approach suggests that this kind
of constraints and sub-problems should be solved at the local level.
The main responsibility is given to the microsources to optimize
their production and to the local loads to control their consump-
tion. For this kind of control, the multi-agent theory presents an
interesting and useful solution that can ensure a self-monitoring
for each controllable element [24–26]. Whatever the centralized or
decentralized approach is applied, especially in an island mode, a
real-time control is insufficient. The microgrid should have a con-
trol strategy based on the proactive reaction that takes into account
production and consumption predictions to ensure the power bal-
ance of networks [27,28].
Although these research works are interesting, no one can solve
the BARAKA problem for significant reasons related to the loca-
tion or the size of the platform. In this paper, we present a novel
solution to the BARAKA problem. We propose a new multi-agent
power management strategy that can optimize real-time power
dispatches [29] in order to face unsafe situations [30,31]. A load
shedding strategy based on weather forecasting information is
developed. With this information, we predict the production insuf-
ficiency. Then, the load shedding method is used to ensure the
balance between the available and requested energy by promoting
high priority loads [32]. The use of forecasting information in the
load shedding decision gives rise to proactive control aspect. This
aspect allows the system to make the right decision about the refu-
elling of diesel generators and the choice between supplying total
loads or using the load shedding method in order to increase the
autonomy of backup sources (batteries and diesel generators) in the
unfavorable weather conditions [33–35]. Forecasting options may
have a direct impact on the economic viability and supply availabil-
ity of microgrids [36]. The proposed control strategy is expected
to minimize the negative influence of the intermittent behavior
of the renewable sources availability on the platform production
[37]. By the proposed model of power management, we develop a
new control strategy by which the energy management system is
subdivided into two main management parts: production and con-
sumption. For each part, a hierarchical multi-agent system with
the master-slave model is used to control load and source penetra-
tions. An agent is used to provide the meteorological forecasting
data. The production management is assured by a super master
agent, four master agents (master agent for each type of sources),
and several slave agents (an agent for a micro source). The super
master agent of production is used to choose the type of source to
be integrated into the network based on the information collected
by the master agents of production. These agents collect the use-
ful information about the availability and autonomy state of their
sources. They choose one, among them, that will be integrated into
the network while taking into account the decision made by the
super master agent. The consumption management is made in a
similar way. The communication is made by tokens of information
and control that allow to avoid the point-to-point high cost com-
munications. The implementation of this strategy requires several
input/output ports. The acquisition of weather forecasting data is
periodic [38] and the management of energy flow is real-time. For
technical and economic reasons, the FPGA Spartan 6 is chosen as
a perfect solution for the multiple input/output control strategy
implementations. The contributions of this paper are:
• Proposition of a new control strategy based on weather forecast-
ing and load shedding method. This strategy is supported by a
mathematical model that describes the relationship between dif-

3. M.G. Abidi et al. / Electric Power Systems Research 152 (2017) 411–423 413
ferent production and consumption components in a microgrid
and the influence of their yields by the meteorological factors,
• Presentation of a hierarchical multi-agent solution for the pro-
posed strategy [39]. To insure the information and control
exchanged between the various agents, we propose a commu-
nication protocol based on tokens. the multi-agent solution is
implemented in Field Programmable Gate Array (FPGA),
• Application of the paper’s contribution to the BARAKA platform
for evaluation of performance. The goal is to check the number of
stops in a period of time.
We test the developed control strategy in similar situations that
cause the stops of the platform. CIPEM company gave us the nec-
essary information to simulate these situations. The experimental
results demonstrate that the solution is effective and can avoid
short-term stops by increasing the autonomy of its backup sources,
i.e., a high improvement of power availability is achieved. The plat-
form can avoid losses estimated at least up to 200,000 US dollars
per year caused by the power unavailability.
This paper is organized as follows. Section 2 describes the prob-
lem and the contribution of this work. Section 3 proposes a new
multi-agent architecture for a microgrid. In Section 4, we present
the strategy of communication between agents and the imple-
mentation of the proposed architecture. Section 5 presents the
application of the proposed strategy on the BARAKA platform and
evaluates its performance. Finally, Section 6 summarizes this paper.
2. The case problem
This section describes the considered case problem from a
petroleum platform.
2.1. Tunisian petroleum platform
The microgrid investigated in this paper is an island petroleum
platform located at the Tunisian coast. The architecture of this
microgrid adopted for this case problem is composed of pho-
tovoltaic cells (PV), wind turbines (WT), batteries (B), diesel
generators (GE) and loads as shown in Fig. 1.
The distributed energy sources are designed as follows. (i) Each
renewable energy source (PV, WT) is sized to be able to generate
the electrical power supply required by both loads and batteries in
favourable weather conditions. (ii) Each diesel generator is dimen-
sioned to be able to produce the electrical energy required by the
loads. The autonomy of this source is proportional to the fuel level
in its tank and the power required by the loads. (iii) Batteries are
sized to be able to provide the electrical power supply required by
loads with an autonomy proportional to their charge levels and the
electrical power requested by loads. In its charging phase, a battery
is considered as a load.
The loads can be classified into two classes: (i) Critical loads
for which the high availability of electrical power supply must be
assured and (ii) Uncritical loads for which power can be switched
off in emergency cases.
The microgrid in this case study is composed of three PV
microsources {PV1, PV2, PV3}, two WT microsources {WT1, WT2},
four batteries microsources {B1, B2, B3, B4}, two diesel generators
microsources {GE1, GE2}, two sets of critical loads {CL1, CL2}, and
two sets of uncritical loads {UCL1, UCL2}. The different data about
sources and loads of the platform are listed in Table 1.
2.2. Problems
The petroleum platform can only operate in the island mode and
there is no recourse from a main electrical network. The microgrid
Fig. 2. Considered microgrid.
must produce the needed energy in order to ensure its energy self-
sufficiency. The microgrid has the intermittent nature for all the
renewable sources. (i) The renewable sources availability (photo-
voltaic cells and wind turbines) is related to the meteorological
terms (insolation and wind). The probability of these two meteo-
rological factors is in the acceptable margin and does not exceed
33% ((Table 2) for insolation). (ii) In the case of renewable sources
unavailability, the microgrid resorts to backup sources (batteries
and diesel generators). These sources can ensure the power supply
availability, but the availability of these sources is limited by their
capacity ratings. In the considered platform, the backup system can
ensure the energy demands of all the loads for a maximum duration
of three days.
If the downtime of the renewable sources exceeds the time that
could be covered by the backup sources (autonomy) in the platform,
then the electrical energy becomes totally unavailable, the control
and communication systems are shutdown and all the microgrid
loads would be off-services (Eq. (1)). Between 2012 and 2014, the
platform recorded six blackouts caused by the long-term climatic
fluctuations. These blackouts provoked approximately one million
dollars of losses for the Tunisian government. Therefore, it is nec-
essary to develop a control strategy to avoid or at least minimize
the downtime, especially for critical loads.
APV (t) + AWT (t) + AB(t) + AGE(t) = 0 (1)
The development and implementation of a multi-agent solu-
tion to control the case study (petroleum platform), based on field
programmable gate arrays (FPGAs), are presented in this paper. The
control strategy implemented on the FPGA has an objective to man-
age the connection of sources and loads to the microgrid network.
This strategy is based firstly on the real-time information about
the production and consumption state of various elements in the
platform, and secondly on the weather forecast information. The
real-time information concerns the production state of the renew-
able sources, the charge levels of the batteries, the fuel level in the
tanks of the diesel generators and load energy demands.
3. Microgrid architecture
The addressed microgrid is composed of photovoltaic cells,
wind turbines, batteries, diesel generators, and loads. The control
strategy of the microgrid should solve many specific operational
problems and several decisions should be made locally [40]. For
each kind of sources or loads, the controller should have a degree of
autonomy and intelligence. Thus, a multi-agent solution is chosen
to provide the most suitable paradigm for this type of control strat-
egy due to its inherent advantages such as reactivity, proactivity,
and autonomy [41–43].
In this section, we describe the configuration of the microgrid
shown in Fig. 2, and explain the proposed multi-agent architec-
ture for a required high power availability by using a mathematical
model.

4. 414 M.G. Abidi et al. / Electric Power Systems Research 152 (2017) 411–423
Fig. 1. Microgrid network of the case problem.
Table 1
Loads and sources of the platform.
Loads Sources
Type Set Total power by set (kW) Loads of set Power rate by load (kW) Type Number of microsources Power by microsource
Critical loads CL1 4.1 CL11 1.4 Photovoltaic cells 3 33.2 kWp
CL12 2.7
CL2 3.9 CL21 1.1 Wind turbines 2 32 kW
CL22 0.5
CL23 2.3
Uncritical loads UCL13.8 UCL11 2.5 Batteries 4 5 kW
UCL12 1.3
UCL24.2 UCL21 1.7 Diesel generators 2 5 kW
UCL22 1.2
UCL23 1.3
Table 2
Insolation rate in Tunis.
Month Jan Feb Mar Apr May Jun
Insolation (h) 146 160 198 225 282 309
Month Jul Aug Sept Oct Nov Dec
Insolation (h) 357 329 258 214 174 149
Total 2804 h/year
3.1. Motivation
The goal of this paper is to develop a new automated and
intelligent control strategy based on real-time measurement and
power generation forecasting [44]. By minimizing the impact of the
fluctuating and intermittent behaviour of renewable sources, this
strategy is able to optimize the power supply availability. The pro-
posed idea is to use: (i) real-time information (measures) to ensure
the availability of electrical energy, (ii) forecasting data to estimate
the availability of sources in the future, and (iii) all the information
to generate proactive reaction control. In the case of renewable
energy source unavailability, this proactive reaction gives to the
system the possibility of minimizing the energy consumption by a
load shedding method. This method reduces the consumption and
increases the autonomy of backup sources. The choice of loads to
be shed is based on the production level and the load priority. A
detailed mathematical model of this strategy is described in the
next subsection.
3.2. Formalization of equipment
The platform P is composed of a set cons of several distributed
loads and a set prod of sources (photovoltaic cells, wind turbines,
batteries and diesel generators). The loads in cons can be classified
into two groups: Critical ˇp and Uncritical ˇnp loads.
On the platform, we consider NP critical loads {ˇ1
P
,. . .,ˇNP
P
} that
should be always connected to the grid and NNP uncritical loads
{ˇ1
NP
,. . .,ˇNNP
NP
} that can be disconnected in some cases. The micro-
grid is powered by four types of energy sources: (i) photovoltaic
cells (SPV), (ii) wind turbines (SWT), (iii) batteries (SB), and (iv) diesel
generators (SGE). The set of distributed sources prod is {SPV, SWT, SB,
SGE}.
The number of sources varies from one type to another. We con-
sider NPV photovoltaic cells defined by set SPV = {S1
PV
,. . .,SNPV
PV
}. We
denote by NWT the number of wind turbines defined by set SWT =
{S1
WT
,. . .,SNWT
WT
}. We denote by NB the number of batteries defined
by set SB = {S1
B
,. . .,SNB
WT
}, and we denote by NGE the number of diesel
generators defined by set SGE = {S1
GE
,. . .,SNGE
GE
}. The platform con-
tains also a meteorological database in which data are used for
production forecast.
3.3. Contribution: new multi-agent architecture for autonomous
microgrids
To construct a multi-agent system for the studied platform, the
energy management is provided mainly by various master and
slave agents as shown in Fig. 3. Agent MAprod is the super mas-

5. M.G. Abidi et al. / Electric Power Systems Research 152 (2017) 411–423 415
Fig. 3. Multi-agent system in the petroleum platform. (For interpretation of the references to color in the text, the reader is referred to the web version of this article.)
ter agent of production, its role is to: (i) maintain the balance
between the production and consumption, (ii) calculate and make
the prediction of the energy production based on Agent MAmeteo,
(iii) communicate with the consumption master agent and master
agents of each type of sources, (iv) collect the production power
information from Agents MAPV, MAWT, MAB and MAGE, and (v) con-
trol the state of penetration of sources. Agent MAprod is at the higher
level of {MAPV, MAWT, MAB, MAGE}.
MAPV, MAWT, MAB and MAGE are respectively the master agents
of photovoltaic cells, wind turbines, batteries (in the production
mode) and diesel generators. Each master agent can control and
communicate with its slave agents. AgentPV(MAPV) in charge of
{S1
PV
, . . ., S(NPV )
PV
} is responsible for the photovoltaic cells’ produc-
tion management. AgentWT(MAWT) in charge of {S1
WT
, . . ., S(NWT )
WT
}
is responsible for the wind turbines’ production management.
AgentB(MAB) in charge of {S1
B
, . . ., S(NB)
B
} is responsible for the
batteries’ production management. AgentGE(MAGE) in charge of
{S1
GE
, . . ., S(NGE)
GE
} is responsible for the diesel generators’ production
management.
These agents are responsible for collecting information from
their slaves and controlling their state of penetration. The super
master agent of consumption is Agent(MAcons) in charge of {MAP,
MANP}. Note that Agent(MAP) is responsible for critical load
consumption management and is in charge of {ˇ1
P
, . . ., ˇ(NP )
P
}.
Agent(MANP) is responsible for uncritical load consumption man-
agement and is at the higher level of {ˇ1
NP
, . . ., ˇ(NNP )
NP
}.
Agent MAcons is responsible for power demand management in
the system and it communicates with priority and non-priority
loads master agents to collect the power required by the loads. The
super master agent of consumption informs the super master agent
of production MAprod about the load request and receives thereafter
the information about the produced power. Finally, it communi-
cates with agents MAP and MANP to control the connection state of
their associated loads.
Agents MAP and MANP are the critical (priority), uncritical (non-
priority) load agents and batteries (in the consumption mode),
respectively. They are responsible for collecting information from
the associated slave loads and send this information to agent
MAcons. These slave agents are responsible for collecting informa-
tion about energy demand of loads and applying the load shedding
strategy.
Agent MAmeteo is responsible for storing the periodic meteoro-
logical forecasts for the next seven days. Nowadays, this type of
forecasts presents an acceptable precision [45,40]. MAmeteo pro-
vides this information to the super master agent of production to
estimate the production of sources. The meteorological forecasting
data are the inputs to the fixed problem and we suppose that they
are precise.
3.3.1. Slave source agents
These agents present the link between the control system and
the controlled sources. At this level, the platform sends the required
measure to the control system and gets an order as a feedback from
the same system about the microsources’ states. For each kind of
source, there are Nk microsources (Ms) (k ∈ {PV; WT; B; GE}). Each
microsource Msk can have both availability states AMs
k
: Available (1)
or Not Available (0). In the following equations, x represents the
function ceiling(x) i.e.,
⎧
⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨
⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩
AMs
PV
(t) =
En(t)
En0
− 1
AMs
WT
(t) =
VV (t)
VV0
− 1
AMs
B
(t) =
ECharge(t)
ECharge0
− 1
AMs
GE
(t) =
NCharge(t)
NCharge0
− 1
(2)
where En0
, VV0
, ECharge0
and NCharge0
are the nominal values from
which the sources are capable of producing energy.
3.3.2. Master source agents
In terms of availability A(t), all electrical energy sources (pho-
tovoltaic cells, wind turbines, batteries and diesel generators) can
have two states: (1) Available energy producer and (0) Unavailable
energy producer. In its charging phase, a battery acts as a load that

6. 416 M.G. Abidi et al. / Electric Power Systems Research 152 (2017) 411–423
may consume excess production. In this phase, the battery can have
a third load state (−1) of the battery. Different availability states of
the different sources are
⎧
⎪⎪⎪⎪⎨
⎪⎪⎪⎪⎩
APV (t) ∈ {1, 0}
AWT (t) ∈ {1, 0}
AB(t) ∈ {1, 0, −1}
AGE(t) ∈ {1, 0}.
(3)
Similar to sources, each microsource Ms has its availability state
AMs
i
. If Rk’s of Nk microsources are available, then this source is avail-
able, where Rk is the minimal number of microsources which can
assure the requested energy, i.e.,
Ak(t) =
⎧
⎪⎨
⎪⎩
1, if
Nk
i=1
AMs
i
≥ Rk
0, otherwise
(4)
where 1 ≤ Rk ≤ Nk, k ∈ {PV, WT, B, GE}.
The master production agent selects the source that supplies
the electrical energy to the microgrid. A selected source chooses
among its available microsources that should be connected while
respecting the rule Rk/Nk (Eq. (6)). By using these agents, the sys-
tem collects the real-time information on the energy production.
The information collected allows the system to choose the sources
to be penetrated to the grid. These agents are only responsible for
choosing the microsources which have to assure the energy produc-
tion requested by the corresponding master agent of production.
CMs
i
is the penetration state of the ith microsource. We have
Nk
i=1
CMs
i
= Rk (5)
The energy supplied to the microgrid by each source ( k) is the
sum of the electrical production (P) of its microsources (Ms) which
are connected to the microgrid, i.e.,
k =
Nk
i=1
CMsi
i
· PMsi
i
(6)
To represent the renewable source generation, some proba-
bilistic models were established [46]. According to [47], the wind
turbine power generation is given by
PWT ( ˜Vv) =
⎧
⎪⎪⎪⎪⎪⎨
⎪⎪⎪⎪⎪⎩
0, 0 ≤ ˜Vv ≤ Vv0
or Vco ≤ ˜Vv
Prated
˜Vv − Vv0
Vr − Vv0
, Vv0
≤ ˜Vv ≤ Vr
Prated Vr ≤ ˜Vv ≤ Vco
(7)
where ˜Vv, Vv0
, vr and Vco are the forecasted wind speed, cut in speed,
rated speed and cut-off speed of the wind turbine, respectively, and
PWT ( ˜Vv) is the forecasted output power of the wind turbine.
According to [48], the predicted photovoltaic power generation
is given by
PPV ( ˜En(t)) = Ápvg × Apvg × ˜En(t) (8)
where Apvg is the surface in (m2) of PV generator, Ápvg is the
efficiency of conversion and ˜En(t) is the forecasted insolation in
(W/m2), and Ápvg is given by
Ápvg = Ár × [1 − ˇ × (Tc − Tcref )] (9)
where Ár is the photovoltaic module reference efficiency, ˇ is the
temperature coefficient which is supposed to be a constant for sil-
icon solar cells, Tc is the solar cell temperature (C) and Tcref is the
reference solar cell temperature (C).
In the considered platform, the photovoltaic source (PV) is
available if at least two of the three photovoltaic fields are avail-
able. For other sources, they are available if one (at least) of
their microsources is available. We note that only the available
sources (and microsources) can be connected to the grid and
the most priority available source is penetrated to the grid. In
this study, the priority order of sources is (1) photovoltaic cells,
(2) wind turbines, (3) batteries, and (4) diesel generators. The
penetration management strategy of sources to the microgrid (con-
necting/disconnecting) is defined by
⎧
⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨
⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩
CPV (t) =
NPV
i=1
AMsPV
i
RPV
− 1
CWT (t) =
NWT
i=1
AMsWT
i
RWT
− 1 · ¯CPV (t)
CB(t) = C1
B
(t) − C2
B
(t)
C1
B
(t) =
NB
i=1
AMsB
i
RB
− 1 · ¯CWT (t) · ¯CPV (t)
C2
B
(t) = 2 −
NB
i=1
AMsB
i
RB
· (CPV (t) + CWT (t))
CGE(t) =
NGE
i=1
AMsGE
i
RGE
− 1 · ¯C1
B
(t) · ¯CWT (t) · ¯CPV (t)
(10)
where ¯C(t) is the logical complement of C(t) that produces 1 when
its operand is 0 and 0 when its operand is 1.
The supplied electrical power to the microgrid k(t) depends on
that produced by sources Pk(t) (Eq. (7)) and their penetration states
(Ck). These electrical powers are given by
⎧
⎪⎪⎪⎪⎨
⎪⎪⎪⎪⎩
PPV (t) = CPV (t) · PV (t)
PWT (t) = CWT (t) · WT (t)
PB(t) = CB(t) · B(t)
PGE(t) = CGE(t) · GE(t)
(11)
To ensure the availability of power supply, the electric produc-
tion delivered by the four sources should be equal (or superior) to
the consumption of the connected loads. The produced power can
be expressed by
PPV (t) + PWT (t) + PB(t) + PGE(t) ≥
NP
i=1
CP
i
(t) · PP
i
(t) + CNP
i
(t)
NNP
i=1
CNP
i
(t) · PNP
i
(t) (12)
where (a) NP is the number of critical loads, (b) NNP is the number
of uncritical loads, (c) CP
i
and PP
i
are the integration state and the
power rate of the ith critical load, respectively, (d) CN
i
P and PN
i
P
are the integration state and the power rate of the ith uncritical
load, respectively, and (e) CNP is the integration state of the uncrit-
ical loads. We consider two critical and two uncritical loads in the
platform. In the case of basic load shedding (without forecasting),
the load shedding method takes into account only the real-time
information about production and consumption, i.e.,
({CP
i
(t)}, {CNP
i
(t)}, {CNP
}) = f (PPV (t), PWT (t), Echarge(t), Ncharge(t), {PP
i
}, {PNP
i
}) (13)
During the use of the backup sources, we have to avoid the total
discharge of the batteries and the tanks of the diesel generators. To
avoid the phenomenon of sulfation [49], the batteries have to keep
a minimum level Echarge at which they should stop supplying the

7. M.G. Abidi et al. / Electric Power Systems Research 152 (2017) 411–423 417
microgrid. By analogy to batteries, the tanks of the diesel gener-
ators should have a minimum level Ncharge at which they should
disconnect from the microgrid in order to avoid any cavitation
problem. The load shedding can be done based on the classification
of loads. In this case, the system can act to connect or disconnect
uncritical loads (CNP) (Switch C in Fig. 1) by taking into account
their priorities. In this case, we may have a partial load shedding
and the control system can connect and disconnect loads belonging
to the same class ({CP
i
(t)}, {CNP
i
(t))}) (Switches S1, S2, S3 and S4 in
Fig. 1). The load shedding without a required forecasting can influ-
ence negatively the availability of the electrical energy as follows.
(i) If the system makes the decision of load shedding as soon as the
renewable sources become unavailable, then the uncritical loads
are disconnected at each short period of unavailability of renewable
sources. In this case, the availability of the uncritical loads would be
decreased in an unreasonable way. (ii) In the case of unavailability
of the renewable sources, any delay in the application of the load
shedding method decreases the autonomy of the backup sources
quickly. This decrease influences negatively the availability of the
critical loads in the case of a long downtime of renewable sources.
In order to guarantee the efficiency of this solution, the duration of
the load shedding must be justified, which is based on the current
state of sources and the duration of unavailability of the renew-
able sources (forecasting). In the case of a load shedding based on
the forecasting, the uncritical load can be disconnected. The load
shedding strategy is given by:
({CP
i
(t)}, {CNP
i
(t)}, {CNP
}) = f (PPV (t), PWT (t), ϕGE(t), ϕB(t), {PP
i
}, {PNP
i
}) (14)
where ϕGE(t) and ϕB(t) are the forecasted states of the backup
sources: diesel generators and batteries, respectively.
⎧
⎪⎪⎪⎨
⎪⎪⎪⎩
ϕGE(t) =
n
i=1
EClim(t + i)
n.EClim0
− 1 ·
n
i=1
NR(t + i)
n.NR0
− 1
ϕB(t) =
n
i=1
En(t + i)
n.En0
− 1 ·
n
i=1
VV (t + i)
n.VV0
− 1 ·
n
i=1
NCharge(t + i)
n.NCharge0
− 1
(15)
The estimated quantity of energy to be delivered by the renew-
able energy sources (PV and WT) in the forecasting horizon is given
by
⎧
⎪⎪⎨
⎪⎪⎩
EPV = PPV × PV where PV = Pr[APV (t) = 1] =
1
T
T
0
APV (t) =
TPV
T
EWT = PWT × WT where WT = Pr[AWT (t) = 1] =
1
T
T
0
AWT (t) =
TWT
T
(16)
where EPV and EWT are the quantity of energy to be delivered by the
photovoltaic cells and wind turbines, respectively. PV and WT are
the forecasted availability rates of these sources. TPV and TWT are
their total time durations, respectively, in which the sources are
expected to be available in the forecast horizon T.
3.3.3. Master consumption agent
The produced power in the microgrid may not be sufficient to
satisfy the totality of power demands at any time. For this reason,
the specified priority should be defined between loads. In the case
of an insufficient production, the loads with the highest priority
will be supplied. In the considered case, we have two classes of
priority: (i) priority loads which are critical and should be continu-
ously supplied in most of the time and (ii) non-priority loads which
are uncritical and can be disconnected in the load shedding phase
(Fig. 3).
3.3.4. Master load agent
We consider two master load agents (critical and uncritical
loads). To give more flexibility to the strategy of load shedding,
the loads should have a second priority level in each class of loads.
4. Control strategy for the microgrid high availability
In this section, we describe the developed the control strategy,
the communication protocol and present the implementation of
the control strategy in FPGA board.
4.1. Overview
As shown in Fig. 4, the control strategy can be divided into three
stages: (i) the collection of source production and load demand
information, (ii) the decision phase, in which the control strat-
egy adjusts the power generation level based on the information
collected in the previous phase by taking into account the meteo-
rological forecasting data, and (iii) the control phase, in this stage
the control system reacts according to the chosen production level
to connect or disconnect some sources and loads.
Renewable sources (photovoltaic cells and wind turbines) are
sized to meet the entire demands of loads. If one of these two
sources is available, then all the loads are powered. In the oppo-
site case, the control system can decrease the production level to
increase the autonomy of backup sources (batteries and diesel gen-
erators). In this case, the production level depends on the available
autonomy of these two sources and the time during which they
should operate. The minimization of production is surely followed
by a reduction in consumption. The control system has to eliminate
certain loads in order to guarantee the energy balance between the
consumption and production. The microgrid should allocate the
power to loads with high priority first. The control strategy should
allocate a specific priority for each load (load shedding). In the case
when we have several loads to facilitate the decision of the load
shedding, it is better to classify the loads responsibilities which
have a convergent priority degree. The load distribution by class
should be balanced and the number of loads by class should be
approximately equal to the number of classes.
In this study, we have two priority classes: CP for critical (prior-
ity) loads and CNP for uncritical (non-priority) loads. We consider
NP critical loads ˇi
P
(i ∈ {1, . . ., NP}), where each ˇi
P
requests PP
i
of
energy and NNP uncritical loads ˇi
NP
(i ∈ {1, . . ., NNP}), where each
ˇi
P
requests PNP
i
of energy. If at least one of renewable sources is
available, then the control system integrates the source which has
the highest priority and all loads (critical and uncritical loads) are
connected and powered. If these sources are not available, then
the backup sources (batteries and diesel generators) are used. The
system makes a time estimation in which these sources have to
ensure the production (SBWD). If these sources can supply the
requested power to all the loads during this period, then the pro-
duction level remains constant and the system continues to supply
all of the loads. On the contrary, the system minimizes the produc-
tion according to the autonomy of the available backup sources.
The produced energy is allocated to the loads which belong to the
classes with the highest priority. The rest of the produced power is
allocated to the higher priority loads of the next class (Fig. 5).
4.2. Communication protocol
The communication between agents is done by tokens as seen in
Fig. 3. A token is a data table where its size is dependent on the num-
ber of agents by which this token passes. Each of these agents, has

8. 418 M.G. Abidi et al. / Electric Power Systems Research 152 (2017) 411–423
Fig. 4. Control strategy.
Fig. 5. Control strategy flowchart.
its own cell in which it writes the information or receives orders.
This cell is accessible only by this agent and its master. The token
has two types. It is an information token if its first bit is ‘0’, other-
wise, it is a control token with the first bit being ‘1’. The ring token
communications protocol is chosen. This protocol is very flexible
(additional components do not affect the network performance)
and is organized as follows, all the traffic flows in only one direction
at very high speed to reduce chances of collision.
• At the beginning of each control cycle, the super master agents
(MAprod and MAcons) start to collect information about the state of
sources and loads. Initially, the batteries are considered as loads.
The master agent of batteries MAB is in a consumption mode
and it remains in this mode until the renewable sources become
unavailable. In this case, MAB becomes in a production mode
and the batteries are considered as sources until the renewable
sources become available again,
• The production super master agent (MAprod) sends a production
information token (blue arrow in Fig. 3) to its related master

9. M.G. Abidi et al. / Electric Power Systems Research 152 (2017) 411–423 419
Fig. 6. Production information token flow.
agents of production in order to determine the state of sources.
The token visits the production agents (MAPV, MAWT, MAB (if MAB
is in the production mode) and MAGE) at first and returns there-
after to super master agents of production (MAprod). Each master
production agent receives the token, sends an internal token to
its slaves to collect the information on the availability state of
their microsources. When the agent receives the internal token
again, it calculates the availability state of the source and fills
its own cell in the production token information. Fig. 6 shows the
UML sequence diagram of the production token information flow
among agents. In this diagram, the batteries are in the production
mode. The communication between a master agent and its slaves
is represented as a self-message,
• In the consumption management part, the super master agent of
consumption collects the information on the energy loads. The
super master agent sends a consumption token information (red
arrow in Fig. 3) to the load master agents (MAP, MANP and MAB (if
MAB is in the consumption mode)). The priority load agent (MAP)
sends an internal load information token to the related slaves in
order to determine their energy demands. It calculates the total
demand, fills and passes the token to the non-priority load agent
(MANP) that does the same thing. After that, the token returns to
the load master agents directly or passes by the master agent of
batteries MAB if it is in a consumption mode,
• The two super master agents (MAprod and MAcons) negotiate on the
production level, which will be supplied by available sources to
the connected loads, by taking into account the information col-
lectedbybothsupermasteragentandthemeteorologicalforecast
information provided by the meteo agent. These two super mas-
ter agents select the adequate operation mode of batteries for the
next control cycle,
• After choosing the production level, the super master agent of
production sends a control token to the master production agents
in order to integrate the highest priority available source and
disconnect the others,
• The master agents of sources to be disconnected send control
tokens to their slaves such that their microsources are discon-
nected. The master agent of the source to be connected has to
choose the microsources to be penetrated while meeting the
energy requirements. It then sends a control token to its slaves,
• In the same way, the super master agent of consumption coordi-
nates the load master agents (MAP and MANP) in order to connect
the highest priority loads by taking into account the production
level. If MAB is in a consumption mode (renewable sources are
available), then all the uncharged batteries are connected.
4.3. Implementation of multi-agent architecture
For technical and economic reasons, we choose the Spartan 6
board (XC6LX16-CS324) for the implementation of the proposed
control strategy. This professional development board is ideal for
fast learning modern digital design techniques [50]. It presents a
perfect solution for multi-input/output control implementation.
The development of the control strategy is done by Xilinx Mat-
lab Simulink. This software gives us the ability to build and test
the control model (via a Xilinx library) and implement it in FPGA
[51]. The Simulink model of the proposed strategy is composed of:
(i) four subsystems that represent the master agents of the four
types of sources, (ii) a master agent for critical loads and another
one for uncritical loads, (iii) two super master agents which control
all other agents: the super master agent of production and that of
consumption, and (iv) an agent for meteorological forecasting data.
This model can be subdivided into two big communicating parts.
The first part groups the agents which manage the production of
various sources. The second part includes the agents responsible
for the energy consumption management of loads. These two parts
are connected to negotiate the production level that is provided by
the sources.
5. Application to Tunisian petrolium platform
In order to guarantee the performance of the better energy man-
agement that is theoretically proposed, the control strategy must be
tested in simulation scenarios similar to those that cause the stops
of the platform. CIPEM company (www.cipem.com.tn) gave us the
necessary information concerning dates and durations of the break-
downs. These simulations are based on climatic history (insolation,
wind speed) of the platform. The national institute of the meteo-
rology in Tunisia supplies us these data (www.meteo.tn). For our
experimental setup, a real scenario that causes a total power fail-
ure in Tunisia in April 2013 is used. Several simulation results that
highlight the influence of the control strategy on the power supply
availability are presented and discussed. In the results, we use two
power supply availability rates (APS(%)) for: (i) critical loads and (ii)
uncritical loads. The instantaneous availability may have only two
values, 1 in the case of availability and 0 in the opposite case. The
average availability AA(t) is the mean value of the instantaneous
availability between time=0 and time=t.
APS(t) =
1
t
t
0
A(x)dx (17)
In this section, we focus mainly on the production level choice
and its effect on the autonomy of the backup sources. Some exper-
imental results are presented to provide efficiency of the proposed
solution.
5.1. Numerical results
This subsection represents a comparison among the three
strategies of control:
• The first strategy consists in supplying all loads in the case of
availability of sources. In this case, the production level is fixed
(without a load shedding),
• The second strategy consists in the load shedding of uncritical
loads if the diesel generators are the only available sources in
order to increase their autonomy. The load shedding decision is
based only on the real-time information about the availability
state of sources,
• The third strategy presents the paper’s contribution that deals
with the load shedding method based on the forecasting informa-

10. 420 M.G. Abidi et al. / Electric Power Systems Research 152 (2017) 411–423
Fig. 7. Experimental results for the control strategy in case of a long unavailability of the renewable energy: (a) without load shedding, (b) with load shedding only, (c) with
load shedding and forecasting.
tion. If the system predicts a long unavailability of the renewable
sources, then the load shedding begins when the system uses the
backup sources.
The conditions under which we make the comparison are: (i)
the batteries can recover the energy demand of loads during two
units of time, and (ii) the diesel generators can recover the energy
demand of loads during only one unit of time. There are three levels
of production: (i) 100%, which assure a total power supply of loads,
(ii) 50%, which shows that only the critical loads (CL1 and CL2) are
connected, and (iii) 0%, which corresponds to a total absence of
the electrical energy in the platform. As we showed previously, the
penetration is equal to: (i) ‘1’ if the source is connected to the grid,
(ii) ‘0’ if the source is disconnected from the grid, and (iii) ‘−1’ for
the batteries in their charging phase.
5.1.1. Long absence duration of renewable sources
In this simulation, the renewable sources are unavailable for six
units of time (between t = 3 and t = 9).
The case (a) (without any load shedding): during the phase of
unavailability of renewable sources, the system continues to sup-
ply all of the loads. The backup sources assure the energy demand
during three units of time. The system becomes in a full stop (at

11. M.G. Abidi et al. / Electric Power Systems Research 152 (2017) 411–423 421
Fig. 8. Comparison between the control strategy in case of a short unavailability of the renewable energy: (b) with load shedding only, (c) with load shedding and forecasting.
t = 6) during three units of time (Fig. 7a). In this case, APS(%) is equal
to 70% for both types of loads (critical and uncritical loads).
The case (b) (with a load shedding): the production level is
maximal (100%) during the phase of availability of the renewable
sources or of the battery. When these sources become unavail-
able, the system uses the diesel generators to supply the loads and
reduces automatically the production level by using the load shed-
ding method. The reduction of the production (50%) doubles the
autonomy of this source. The system becomes in a full stop (at t = 7)
for only two units of time (Fig. 7b). In this case, APS(%) for critical
loads increases to 80% and APS(%) for uncritical loads decreases to
60%.
The case (c) (with a load shedding and a forecasting-based
control): the system predicts a long unavailability of renewable
sources. When these sources become unavailable, the control sys-
tem makes a decision for a load shedding. The production level is
reduced by a half in this case and the autonomy of batteries and
diesel generators is doubled. These sources can recover the energy
demand during the unavailability phase of the renewable sources
(Fig. 7c). In this case, APS(%) for critical loads increases and achieves
the total availability (100%) and APS(%) for uncritical loads decreases
to 40%.
These results are summarized in Table 3.
In the case of a long downtime of the renewable sources, the
system should promote the priority loads in order to avoid their
stops. The comparison shows that the system should make an early
decision for a load shedding. The load shedding strategy should be
based on a forecasting information.
5.1.2. Short absence duration of renewable sources
In this simulation, the renewable sources are unavailable for
three units of time (between t = 4 and t = 7).
In Fig. 8b, the control strategy uses only the real time infor-
mation. Without using the forecasting information, this strategy
makes the decision to disconnect the uncritical loads when diesel
groups become the only available sources (t = 6). The uncritical
loads are disconnected for one unit of time. APS(%) is equal to 100%
for critical loads, and 90% for uncritical loads.

12. 422 M.G. Abidi et al. / Electric Power Systems Research 152 (2017) 411–423
Table 3
Experimental results of three control strategies during a long absence duration of renewable sources.
Case Control strategy Availability rate of the power supply (%) Total stop (by unit of time)
Load shedding Forecasting For critical loads For uncritical loads
Case a No No 70 70 3
Case b Yes No 80 60 2
Case c Yes Yes 100 40 0
Table 4
Experimental results of load shedding strategies during a short absence duration of renewable sources.
Case Control strategy Availability rate of the power supply (%)
Load shedding Forecasting For critical loads For uncritical loads
Case b Yes No 100 90
Case c Yes Yes 100 100
By using the forecasting data (Fig. 8c), the control system
estimates the duration of renewable sources unavailability. The col-
lected information about state of backup sources shows that they
can produce the load demands throughout this period, the system
can avoid the un-needed load shedding. The uncritical loads remain
connected during this period. APS(%) is equal to 100% for both types
of loads (critical and uncritical loads). Compared to the control
strategy that uses only the real time information, the proposed
strategy increases the availability of electric power of uncritical
loads in this case (Table 4).
5.2. Discussion
The experimental results show clearly that the proposed control
strategy increases APS(%) of critical loads. In cases of insufficient
production, the allocation of the available power becomes more
reasonable. According to the obtained results, it can be seen that:
• An adequate choice and size of sources increase the availabil-
ity of power supply. However, when we choose the sources, the
reconfiguration is costly and takes time. Economically, this kind
of solution is very expensive,
• The load shedding is a very important strategy to increase the
availability of electric power of critical loads in the case of insuf-
ficient production, but it can decrease the availability rate for the
non-priority loads,
• The load shedding strategy should be based on real-time infor-
mation and a forecasting-based control. By using this strategy,
the power supply availability can achieve a high level of critical
loads which can reach 100%. The proposed strategy allows the
platform to take the load shedding decision early in case of a long
absence duration of renewable sources. This decision increases
the availability of backup sources as well as the availability of the
platform. In the case of a short absence duration of renewable
sources, the forecasting information helps the control strategy to
avoid unjustified (un-needed) load shedding,
• The use of the multi-agent system in the power management of
a microgrid decreases the complexity of the control strategy. It is
an efficient way to solve several complex problems locally. This
way makes the control strategy more flexible and autonomous,
• The presence of individual agents for each category of units
reduces the complexity of the control strategy. It facilitates the
collection of information, the decision and the control of the var-
ious units of microgrid.
6. Conclusion
BARAKA is a Tunisian island petroleum platform that has critical
problems of power supply unavailability which reflects negatively
on its production capacitiesand consequently the corresponding
economic indicators. Since the related works cannot resolve these
problems, we propose a new solution where the resizing of the
platform sources is not feasible (such as islanded petroleum plat-
forms) under space and weight constraints. We propose a new
control strategy based on a forecasting oriented solution and load
priority. The predictive control can help the microgrid to improve
the power management by a proactive solution. The load shedding
increases the availability of electrical energy in the level of critical
loads which can ensure a continuity of production. The proposed
new solution presents several economical and technical benefits.
Indeed, it allows to increase the energy profitability of the plat-
form. This control strategy doubles also the autonomy of the backup
sources to guarantee the production continuity in the platform
which can avoid losses caused by the power unavailability. The pro-
posed strategy follows a decentralized control approach that uses a
token-based multi-agent model and it is implemented on an FPGA
board. By using the proposed solution, some experimental results
show that the platform can avoid losses estimated at least up to
200,000 US dollars per year caused by the power unavailability.
Nevertheless, the paper’s contribution does not guarantee a high
power supply availability without taking into account the commu-
nication faults. By using reconfigurable wireless sensor networks,
we can ensure a high availability level [52,53]. Moreover, the fault
detection and system reconfiguration can improve also the power
availability by minimizing the time of the breakdowns caused by
any failure in the microgrid components. These details will be the
objective of a future work.
References
[1] J. Song, V. Krishnamurthy, A. Kwasinski, R. Sharma, Development of a
markov-chain-based energy storage model for power supply availability
assessment of photovoltaic generation plants, IEEE Trans. Sustain. Energy 4
(2) (2013) 491–500.
[2] R.R. Bhoyar, S.S. Bharatkar, Renewable energy integration in to microgrid:
powering rural Maharashtra state of India, 2013 Annual IEEE India Conference
(INDICON) (2013) 1–6.
[3] A. Anastasiadis, A. Tsikalakis, N. Hatziargyriou, Operational and
environmental benefits due to significant penetration of microgrids and
topology sensitivity, IEEE PES General Meeting (2010) 1–8.
[4] K. Boroojeni, M. Amini, A. Nejadpak, S. Iyengar, B. Hoseinzadeh, C. Bak, A
theoretical bilevel control scheme for power networks with large-scale
penetration of distributed renewable resources, 2016 IEEE International
Conference on Electro Information Technology (EIT 2016) (2016) 0510–0515.
[5] W. Sheng, K. Liu, X. Meng, X. Ye, Y. Liu, Research and practice on typical
modes and optimal allocation method for PV-wind-ES in microgrid, Electr.
Power Syst. Res. 120 (2015) 242–255.
[6] X. Liu, B. Su, Microgrids – an integration of renewable energy technologies,
2008 China International Conference on Electricity Distribution (2008) 1–7.
[7] N. Hatziargyriou, Microgrids: Architectures and Control, 1st ed., Wiley/IEEE
Press, Chichester, West Sussex, UK, 2014.

13. M.G. Abidi et al. / Electric Power Systems Research 152 (2017) 411–423 423
[8] K. Boroojeni, M.H. Amini, A. Nejadpak, T. Dragicevic, S.S. Iyengar, F. Blaabjerg,
A novel cloud-based platform for implementation of oblivious power routing
for clusters of microgrids, IEEE Access 5 (2017) 607–619.
[9] J. Song, M.C. Bozchalui, A. Kwasinski, R. Sharma, Microgrids availability
evaluation using a markov chain energy storage model: a comparison study in
system architectures, Pes T&d 2012 (2012) 1–6.
[10] I.-S. Bae, J.-O. Kim, Reliability evaluation of customers in a microgrid, IEEE
Trans. Power Syst. 23 (3) (2008) 1416–1422.
[11] Q. Jiang, M. Xue, G. Geng, Energy management of microgrid in grid-connected
and stand-alone modes, IEEE Trans. Power Syst. 28 (3) (2013) 3380–3389.
[12] G. Liu, M. Starke, B. Xiao, X. Zhang, K. Tomsovic, Microgrid optimal scheduling
with chance-constrained islanding capability, Electr. Power Syst. Res. 145
(2017) 197–206.
[13] F. Kamyab, M. Amini, S. Sheykhha, M. Hasanpour, M. Jalali, Demand response
program in smart grid using supply function bidding mechanism, IEEE Trans.
Smart Grid 7 (3) (2016) 1277–1284.
[14] B. Zhao, X. Zhang, P. Li, K. Wang, M. Xue, C. Wang, Optimal sizing, operating
strategy and operational experience of a stand-alone microgrid on
Dongfushan island, Appl. Energy 113 (2014) 1656–1666.
[15] W. Hasselbring, D. Heinemann, J. Hurka, T. Scheidsteger, L. Bischofs, C. Mayer,
J. Ploski, G. Scherp, S. Lohmann, C. Hoyer-Klick, E. Thilo, S.-H. Marion, H. Gerd,
R. Stefan, Wisent: e-science for energy meteorology, 2006 Second IEEE
International Conference on e-Science and Grid Computing (e-Science’06)
(2006) 3134–3141.
[16] T. Logenthiran, D. Srinivasan, A.M. Khambadkone, T.S. Raj, Optimal sizing of an
islanded microgrid using evolutionary strategy, 2010 IEEE 11th International
Conference on Probabilistic Methods Applied to Power Systems (2010) 12–17.
[17] Y. Nian, S. Liu, D. Wu, J. Liu, A method for optimal sizing of stand-alone hybrid
PV/wind/battery system, 2nd IET Renewable Power Generation Conference
(RPG 2013) (2013).
[18] A. Kwasinski, Quantitative evaluation of DC microgrids availability: effects of
system architecture and converter topology design choices, IEEE Trans. Power
Electron. 26 (3) (2011) 835–851.
[19] H. Wang, X. Tong, F. Li, B. Ren, Research on energy management and its
control strategies of microgrid, 2011 Asia-Pacific Power and Energy
Engineering Conference (2011) 1–5.
[20] S.J. Yuan, Z. Hou, D.G. Li, W. Gao, X.S. Hu, Optimal energy control strategy
design for a hybrid electric vehicle, Discrete Dyn. Nat. Soc. J. 2013 (2013) 1–8.
[21] C. Colson, M. Nehrir, C. Wang, Ant colony optimization for microgrid
multi-objective power management, 2009 IEEE/PES Power Systems
Conference and Exposition (2009) 1–7.
[22] Y. Pan, P. Li, X. Li, B. Lei, Z. Xu, Strategy of research and application for the
microgrid coordinated control, 2011 International Conference on Advanced
Power System Automation and Protection (2011) 873–878.
[23] H.C. Liu, J.X. Yuan, Z. Li, G.D. Tian, Fuzzy petri nets for knowledge
representation and reasoning: A literature review, Eng. Appl. Artif. 60 (1)
(2017) 45–56.
[24] T. Logenthiran, D. Srinivasan, Multi-agent system for the operation of an
integrated microgrid, J. Renew. Sustain. Energy 4 (1) (2012) 013116.
[25] C. Huang, S. Weng, D. Yue, S. Deng, J. Xie, H. Ge, Distributed cooperative
control of energy storage units in microgrid based on multi-agent consensus
method, Electr. Power Syst. Res. 147 (2017) 213–223.
[26] M. Amini, B. Nabi, M. Haghifam, Load management using multi-agent systems
in smart distribution network, 2013 IEEE Power & Energy Society General
Meeting (2013) (2013) 1–5.
[27] D.E. Olivares, C.A. Canizares, M. Kazerani, A centralized energy management
system for isolated microgrids, IEEE Trans. Smart Grid 5 (4) (2014)
1864–1875.
[28] J. Ma, F. Yang, Z. Li, S.J. Qin, A renewable energy integration application in a
microgrid based on model predictive control, 2012 IEEE Power and Energy
Society General Meeting (2012) 1–6.
[29] B. Zhao, M. Xue, X. Zhang, C. Wang, J. Zhao, An MAS based energy
management system for a stand-alone microgrid at high altitude, Appl.
Energy 143 (2015) 251–261.
[30] E. Kuznetsova, C. Ruiz, Y.-F. Li, E. Zio, Analysis of robust optimization for
decentralized microgrid energy management under uncertainty, Int. J. Electr.
Power Energy Syst. 64 (2015) 815–832.
[31] S. Grosswindhager, M. Kozek, A. Voigt, L. Haffner, Fuzzy predictive control of
district heating network, Int. J. Model. Identif. Control 19 (2) (2013) 161.
[32] K. Balasubramaniam, P. Saraf, R. Hadidi, E.B. Makram, Energy management
system for enhanced resiliency of microgrids during islanded operation,
Electr. Power Syst. Res. 137 (2016) 133–141.
[33] S. Xing, Microgrid emergency control based on the stratified controllable load
shedding optimization, International Conference on Sustainable Power
Generation and Supply (SUPERGEN 2012) (2012) 59.
[34] H. Zhang, C.S. Lai, L.L. Lai, A novel load shedding strategy for distribution
systems with distributed generations, in: IEEE PES Innovative Smart Grid
Technologies, Europe, 2014, pp. 1–6.
[35] M.G. Abidi, M. Ben Smida, M. Khalgui, New forecasting-based solutions for
optimal energy consumption in microgrids with load shedding – case study:
petroleum platform, vol. 1, 2015 International Conference on Pervasive and
Embedded Computing and Communication Systems (PECCS) (2015) 289–296.
[36] Y. Tang, W. Qi, Q. Sha, N. Chen, L. Zhu, A combination forecast method based
on cross entropy theory for wind power and application in power control,
Trans. Inst. Meas. Control 36 (7) (2014) 891–897.
[37] R.B. Hytowitz, K.W. Hedman, Managing solar uncertainty in microgrid
systems with stochastic unit commitment, Electr. Power Syst. Res. 119 (2015)
111–118.
[38] X. Wang, I. Khemaissia, M. Khalgui, Z. Li, O. Mosbahi, M. Zhou, Dynamic
low-power reconfiguration of real-time systems with periodic and
probabilistic tasks, IEEE Trans. Autom. Sci. Eng. 12 (1) (2015) 258–271.
[39] T. Logenthiran, D. Srinivasan, A. Khambadkone, Multi-agent system for energy
resource scheduling of integrated microgrids in a distributed system, Electr.
Power Syst. Res. 81 (2011) 138–148.
[40] M. Paulescu, E. Paulescu, P. Gravila, V. Badescu, Weather Modeling and
Forecasting of PV Systems Operation, 1st ed., Springer London, London, 2013.
[41] S. BenMeskina, N. Doggaz, M. Khalgui, Z. Li, Multiagent framework for smart
grids recovery, IEEE Trans. Syst. Man Cybern. 47 (7) (2017) 1284–1300.
[42] E. Karfopoulos, L. Tena, A. Torres, P. Salas, J.G. Jorda, A. Dimeas, N.
Hatziargyriou, A multi-agent system providing demand response services
from residential consumers, Electr. Power Syst. Res. 120 (2015) 163–176.
[43] J. Hu, H. Morais, M. Lind, H. Bindner, Multi-agent based modeling for electric
vehicle integration in a distribution network operation, Electr. Power Syst.
Res. 136 (2016) 341–351.
[44] X. Wang, Z. Li, W. Wonham, Dynamic multiple-period reconfiguration of
real-time scheduling based on timed des supervisory control, IEEE Trans.
Autom. Sci. Eng. 12 (1) (2016) 101–111.
[45] J. Kleissl, Solar Energy Forecasting and Resource Assessment, 1st ed., Elsevier,
AP, Academic Press in an imprint of Elsevier, Kidlington, Oxford, 2013.
[46] W. Alharbi, K. Raahemifar, Probabilistic coordination of microgrid energy
resources operation considering uncertainty, Electr. Power Syst. Res. 128
(2015) 1–10.
[47] Y. Liu, C. Yuen, N.U. Hassan, S. Huang, R. Yu, S. Xie, Electricity cost
minimization for a microgrid with distributed energy resource under different
information availability, IEEE Trans. Ind. Electron. 62 (4) (2015) 2571–2583.
[48] H. Belmili, M. Haddadi, S. Bacha, M.F. Almi, B. Bendib, Sizing stand-alone
photovoltaic-wind hybrid system: techno-economic analysis and
optimization, Renew. Sustain. Energy Rev. 30 (2014) 821–832.
[49] Y. Shi, C. Ferone, C. Rahn, Identification and remediation of sulfation in
lead-acid batteries using cell voltage and pressure sensing, J. Power Sources
221 (2013) 177–185.
[50] C. Ekaputri, A. Syaichu-Rohman, Model predictive control (MPC) design and
implementation using algorithm-3 on board SPARTAN 6 FPGA sp605
evaluation kit, 3rd International Conference on Instrumentation Control and
Automation (ICA 2013) (2013) 115–120.
[51] M. Petko, T. Uhl, Smart sensor for operational load measurement, Trans. Inst.
Meas. Control 26 (2) (2004) 99–117.
[52] M. Gasmi, O. Mosbahi, M. Khalgui, L. Gomes, Z. Li, R-node: new pipelined
approach for an effective reconfigurable wireless sensor node, IEEE Trans.
Syst. Man Cybern. PP (99) (2016) 1–14.
[53] H. Grichi, O. Mosbahi, M. Khalgui, Z. Li, New power-oriented methodology for
dynamic resizing and mobility of reconfigurable wireless sensor network,
IEEE Trans. Syst. Man Cybern. (2017), http://dx.doi.org/10.1109/TSMC.2016.
2645401.