Vous êtes sur la page 1sur 12

1688 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO.

4, APRIL 2013
Multiobjective Intelligent Energy
Management for a Microgrid
Aymen Chaouachi, Member, IEEE, Rashad M. Kamel, Ridha Andoulsi, and Ken Nagasaka, Member, IEEE
AbstractIn this paper, a generalized formulation for intelli-
gent energy management of a microgrid is proposed using articial
intelligence techniques jointly with linear-programming-based
multiobjective optimization. The proposed multiobjective intelli-
gent energy management aims to minimize the operation cost and
the environmental impact of a microgrid, taking into account its
preoperational variables as future availability of renewable ener-
gies and load demand (LD). An articial neural network ensemble
is developed to predict 24-h-ahead photovoltaic generation and
1-h-ahead wind power generation and LD. The proposed machine
learning is characterized by enhanced learning model and gen-
eralization capability. The efciency of the microgrid operation
strongly depends on the battery scheduling process, which cannot
be achieved through conventional optimization formulation. In
this paper, a fuzzy logic expert system is used for battery schedul-
ing. The proposed approach can handle uncertainties regarding to
the fuzzy environment of the overall microgrid operation and the
uncertainty related to the forecasted parameters. The results show
considerable minimization on operation cost and emission level
compared to literature microgrid energy management approaches
based on opportunity charging and Heuristic Flowchart (HF)
battery management.
Index TermsFuzzy logic (FL), microgrid, multiobjective in-
telligent energy management (MIEM), neural network ensemble
(NNE), short-term forecasting.
I. INTRODUCTION
T
HE INCREASING incidence of distributed generation
(DG) and active distributed networks, changing regular-
ities, and needs to improve the power system reliability and
clean power support is providing development of a new power
system perception commonly referred to as the smart grid. In
this regard, the microgrid can be considered as one of the
most promising concepts. A microgrid is essentially an active
distribution network dened as an integrated power delivery
Manuscript received April 30, 2011; revised November 27, 2011; accepted
December 30, 2011. Date of publication February 24, 2012; date of current
version November 22, 2012.
A. Chaouachi was with the Department of Electronic and Information
Engineering, Graduate School of Engineering, Tokyo University of Agriculture
and Technology, Koganei 184-8588, Japan. He is now with Hitachi, Ltd., Tokyo
100-8280, Japan (e-mail: a.chaouachi@gmail.com).
R. M. Kamel is with the Electrical Engineering Department, Faculty of
Engineering, Assiut University, Assiut 71518, Egypt (e-mail: r_m_kamel@
yahoo.com).
R. Andoulsi is with the Electrical Engineering Department, Commu-
nity College, University of Hail, Hail 81451, Saudi Arabia (e-mail: ridha_
andoulsi@yahoo.fr).
K. Nagasaka is with the Department of Electronic and Information Engi-
neering, Graduate School of Engineering, Tokyo University of Agriculture and
Technology, Koganei 184-8588, Japan (e-mail: bahman@cc.tuat.ac.jp).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TIE.2012.2188873
system. A microgrid consists of a low-voltage network com-
posed of loads, renewable energy (RE) sources, and DG units
operating as a single controllable load connected to the main
grid. Compared to the conventional power plants, a microgrid
is characterized by specic operation and constraints depending
on several critical stochastic parameters. So far, microgrids
have been mostly established as test-bed platforms in some
developed countries such as Japan, Canada, and the U.S. In
fact, such concept cannot be widely implemented in practice
without a prior successful establishment of an energy manager
to achieve optimal and reliable control of the microgrid for a
given site to minimize its operation cost while consolidating its
reliability and environment-friendly features [1].
As an essential element in the newera of smart power, several
approaches have been reported in the literature in relation to
microgrid intelligent energy management applicable within the
smart grid system [2][4]. A fuel consumption minimization
approach has been proposed in [5] based on power sharing ap-
proach, in which the optimized cost function includes a penalty
function for heat generation excess. The authors promoted
the importance of a microgrid communication infrastructure
allowing the coordination with RE source forecasting as a part
of the central control. However, the proposed approach does
not take into account the prediction of RE power generation
and the management of storage devices. In [6], mesh-adaptive-
direct-search-based optimization algorithm is applied for mi-
crogrid online energy management; however, this approach
does not provide information regarding the forecasting of the
RE sources or load demand (LD). Moreover, the microgrid
energy management problem has been reduced to a generation
scheduling without taking into consideration the storage facility
management as the battery is controlled based on conventional
opportunity charging approach. A mathematical framework for
environmental economic dispatch of microgrid energy man-
agement is proposed in [7], where advantages of intelligent
methods over traditional computational techniques for opti-
mization are discussed. However, the proposed formulations do
not take into consideration the energy storage management. An
energy management system for a microgrid using multilayered
perceptron articial neural networks (ANNs) is proposed in [8],
where hourly decisions of DG dispatch for cost optimization
are projected by the neural network. The training set is gen-
erated using a separate optimal power ow software package.
Unfortunately, complexity of the problem, eventual noisy sets
in the training data, and the generalization performances of the
multilayered perceptron network may result in serious reliabil-
ity issues in the microgrid operation. A one-day-ahead mul-
tiperiod optimization model under supplydemand balancing
0278-0046/$31.00 2012 IEEE
CHAOUACHI et al.: MULTIOBJECTIVE INTELLIGENT ENERGY MANAGEMENT FOR A MICROGRID 1689
constraints for a microgrid is proposed in [9] to determine
the amount of power that a microgrid should exchange with
a wholesale energy market so that its total benet is max-
imized. However, environmental aspects and RE generation
(REG) forecasting are not considered in their investigations. For
most of the microgrid energy management relevant literatures,
forecasting values of REG and LD are not treated in practice but
simply estimated or assumed as available data [2][10], which
implies that forecasting uncertainties are not considered during
the optimization process. Therefore, the energy management
process is regarded based on a deterministic approach which
naturally affects the result accuracy and, therefore, the overall
performances of the microgrid. In this context, an intelligent on-
line energy management approach was proposed in [11], where
the authors took into consideration the solar power generation
(SPG) forecasting and the battery management. However, the
proposed battery scheduling is considered in a deterministic
approach achieved based on a HF where uncertainty related to
the forecasted parameters cannot be considered.
In this paper, a generalized formulation for microgrid energy
management approach is investigated and developed using
articial intelligence techniques. In Section II, the microgrid
model is briey introduced, where the goal is to minimize the
operating costs as well as the reduction of environmental impact
level of the microgrid that is formulated in a multiobjective
intelligent energy management (MIEM) control. The essential
module to implement efciently such energy management ap-
proach is the forecasting of renewable source generation as well
as the LD inside the microgrid. In Section III, the applicability
of ANNs is analyzed for short-term forecasting. A generalized
approach of a robust forecasting module based on a neural
network ensemble (NNE) is investigated and applied for 24-h-
ahead forecasting of SPG, 1-h-ahead forecasting of wind power
generation (WPG), and 1-h-ahead forecasting of microgrid
LD. Section IV is dedicated to the MIEM of the microgrid,
where the overall optimization process slightly depends on the
management of the storage facility. In fact, efcient battery
scheduling in a microgrid should take into consideration both
actual and future RE availability in relation to the actual LD
and the electricity market. However, such terminology cannot
be introduced in the form of a general linear optimization
programming or conventional optimization formulations. The
proposed MIEM approach consists in the joint formulation of
multiobjective optimization (MO) approach based on linear
programming and battery scheduling. The battery scheduling,
as a part of an optimal online energy management, can be
regarded as a decision-making process. In fact, one way to
solve this problem is to design an expert system that is able
to decide, referring to the actual and the forecasted entities,
whether the battery should be charged or discharged and at
which convenient rates. In this regard, fuzzy set theory offers a
good resolution as a mathematical approach designed to model
the vagueness and imprecision of a human cognitive process.
Such intelligent ability to categorize and manipulate entities
allows the proposed battery scheduling approach to underlie
the human aptitude to summarize and take decisions under
uncertainty or partial information. The experimental results and
discussions related to the 24-h-ahead SPG forecasting, 1-h-
ahead WPG and LD forecasting, and the MIEM simulations for
several cases are presented in Section V.
II. MODEL OF THE DEVELOPED MICROGRID
A microgrid normally operates in a grid-connected mode,
as power can be imported or exported from and to the main
grid. During disturbances, such as power quality events, IEEE
1547 events, or faults, the generation and corresponding loads
can be autonomously separated from the distribution system to
isolate the microgrids load from the disturbances (and thereby
maintaining high level of service) without harming the trans-
mission grids integrity via the sectionalizing circuit breaker
(SCB). In terms of power ow control, a DG unit can be either
a dispatchable or nondispatchable unit [12]. The power output
of a dispatchable unit is controlled externally over operation set
points provided by the MIEM as a part of the central controller
(CC) as the fuel cell and the microturbine that are controlled
over fuel-in-ow adjustments and governor speed control. On
the other hand, the RE sources are typical nondispatchable DG
units, and their generated power depends on the actual weather
conditions (wind speed, temperature, and irradiance) and is
controlled in a way to generate their maximum power [13].
The developed microgrid consists of six buses (Fig. 1) and is
composed of a 15-kW battery connected to bus 1, a 10-kW
rated wind electric generation system (WEGS) located at bus 2,
a 15-kW rated solid oxide fuel cell (SOFC) connected to
bus 3, a 20-kWp rated photovoltaic (PV) array connected to
bus 4, and a 15-kW rated single-shaft microturbine (SSMT)
connected to bus 6. In the microgrid simulation model, inverters
play a vital role by interfacing dc microsources with the ac
power side. Two kinds of control strategies are mainly used to
operate an inverter within a microgrid. The PQ inverter control
type is used to supply a given active and reactive power set
point. This type of inverter is used to interface the SSMT, the
SOFC, the PV array, and the battery during grid-connected
operation. The voltage source inverter (VSI) control type is
used to feed the load with predened values of voltage and
frequency. The VSI is used to interface the battery to the mi-
crogrid representing the reference bus during islanding mode.
When analyzing the dynamic behavior of a microgrid, inverters
are modeled based only on their control functions; therefore,
fast switching transients, harmonics, and inverter losses are
neglected. Further details of the microgrids simulation model
description are available in [13][15].
III. MICROGRID NONDISPATCHABLE GENERATION
AND LD FORECASTING
In this section, the applicability of ANNs for microgrid
environment short-term forecasting is investigated. Although
ANNs are providing a relevant methodology for solving several
types of nonlinear problems, they are still considered as an
unstable learning model [16]. In fact, the changes on training
data and the architecture of the network which incorporate the
number of hidden layers, the number of neurons, and the initial
connection weights have considerable effect on the network
training and predicting performances. Moreover, the tting to
the regularities of the data and the tting to the noise sets
during the training of a single neural network restrain it to
1690 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO. 4, APRIL 2013
Fig. 1. Architecture of the microgrid.
be trapped in a local optimum solution with low forecasting
accuracy. On the other hand, there is no systematic investigation
for those issues; mostly, researchers have adopted trial and error
methodology to deal with these inconsistencies [17]. In this
paper, a robust NNE based on regularized negative correlation
learning (NCL) (RNCL) is proposed for an enhanced learning
model performance by improving the generalization ability and
reducing the variance and the performance instability with a
relative simple implementation. Indeed, the major feature of the
proposed machine learning is redundancy; basing on combina-
tion of different learning machines, each neural network could
perform the task on its own, leading to better generalization
performance. The microgrids environment forecasting tasks
include the following: 24-h-ahead SPGforecasting of a 20-kWp
rated PV array, 1-h-ahead WPG forecasting of a 10-kW rated
WEGS, and 1-h-ahead LD forecasting of a building in Tokyo
University of Agriculture and Technology (TUAT).
A. Proposed NNE
Generalized machine learning for microgrid environment
forecasting is proposed based on RNCL for a heterogeneous
NNE composed by an aggregation of the following: multilay-
ered perceptron neural network (MLPNN), radial basis function
neural network (RBFNN), and recurrent neural network (RNN).
The aggregation of all predictors constitutes the network en-
semble {f
b
(x)}
B
b=1
, where the forecasting result of all the
networks is aggregated based on averaging to get the output
of the complete predictor ensemble, as shown in the following
equation:
f
B
(x
n
) =
1
B
B

b=1
f
b
(x
n
). (1)
The developed NNEs are associated to a training set D
(composed of data recorded during scal years 2007 and 2008).
However, using the same training data set for all the component
networks (to estimate the same function) results in high cor-
related outputs that affect the efciency and the robustness of
the learning machine [18]. One way to mitigate the potential
collinearity problem is to add a negative collinearity penalty
to the error function of each predictor [19], [20], where the
training error cost function of each single predictor f
b
is
dened as
e
b
=
N

n=1
(f
b
(x
n
) y
n
)
2
+p
b
(2)
where p
b
is the correlation penalty term with a weighting
parameter expressed as
p
b
=
N

n=1
_
_
_
(f
b
(x
n
)f
B
(x
n
))

i=b
(f
i
(x
n
)f
B
(x
n
))
_
_
_
(3)
=
N

n=1
(f
b
(x
n
)f
B
(x
n
))
2
. (4)
The goal is minimizing p
b
to negatively correlate the predic-
tors error inside the overall ensemble. A null value of results
in an independently trained predictors, while for = 1, the
error function of the NNE is equal to averaging of all individual
networks expressed as
e
B
=
M

b=1
N

n=1
_
(f
b
(x
n
) y
n
)
2
(f
b
(x
n
) f
B
(x
n
))
2
_
=
N

n=1
(f
B
(x
n
) y
n
)
2
. (5)
To improve the NCL generalization performances in relation
to overtting issues, particularly for the case of nontrivial
data as microgrid LD, wind speed, and solar radiation [21],
RNCL has been proposed for homogeneous NNE composed of
RBFNN that is applicable to any nonlinear estimator [22]. In
this paper, the learning approach is extended to a heterogonous
NNE providing a more diversied output while guaranteeing
interaction between the ensemble networks. In fact, the ensem-
ble training objectives are decomposed into a set of subobjec-
tives while each subobjective is implemented by an individual
neural network. In comparison with the RNCL algorithm, a
CHAOUACHI et al.: MULTIOBJECTIVE INTELLIGENT ENERGY MANAGEMENT FOR A MICROGRID 1691
regularization term

B
b=1

b
w
T
b
w
b
is incorporated into the
ensemble error function
e
B
=
N

n=1
(f
B
(x
n
) y
n
)
2
+
B

b=1

b
w
T
b
w
b
(6)
where w
b
= [w
b,1
, . . . , w
b,n
b
] is the weight vector of a neural
network b in the NNE and n
b
is its total number of weights.
The regularization term is decomposed into B parts as for
the training of each network in the ensemble, where the error
function of a single network in the ensemble is dened as
e
b
=
1
B
N

n=1
(f
b
(x
n
) y
n
)
2

1
B
N

n=1
(f
b
(x
n
) f
B
(x
n
))
2
+
b
w
T
b
w
b
. (7)
The error function (7) imposes a regulation term to each
neural network to optimize the regularization parameter
b
in-
stead of the correlation parameter . Scaled conjugate gradient
algorithm [23] is implemented for the training process of the
neural networks, where the partial derivative of (7) with respect
to the neural network weights is expressed as
e
b
w
b,i
=
2
B
N

n=1
(f
B
(x
n
) y
n
)
f
b
(x
n
)
w
b,i

2
B
N

n=1
(f
b
(x
n
) f
B
(x
n
))
_
M 1
M
_
f
b
(x
n
)
w
b,i
+
b
w
T
b
w
b
. (8)
For a training data set D : {x
n
, y
n
}
N
n=1
, the optimal reg-
ularization parameter
opt
b
is calculated based on Bayesian
inference [24][26]. Assuming that the targets are sampled
with independent noise sample e
n
with zero-mean Gaussian
distribution and
1
variance where the weight vector related
to each network has a zero-mean Gaussian distribution and
1
b
variance, with
1
b
/
1
, then the target standard probabilistic
formulation is expressed as
y
n
= f
B
(x
n
) +e
n
. (9)
For the hyperparameters and , the weight vector w is ob-
tained by maximizing the posterior P(w|D) which, according
to the Bayesian theorem, is expressed as
P(w|D) =
P(D|w, )P(w|)
P(D|, )
(10)
where the prior of weight vector w = [w
T
1
, . . . , w
T
B
]
T
is ex-
pressed as
P(w|) =
B

b=1
_

i
2
_
n
i
2
e
(
1
2

b
w
T
b
w
b)
. (11)
As the noise e
n
follows a Gaussian distribution with zero mean
and variance
1
, the likelihood P(D|w, ) is expressed as
P(D|w, ) =
N

n=1
_

2
_1
2
e
(

2
e
n
2)
. (12)
Fig. 2. NNE bagging training owchart.
After omitting the constants and normalization factors, the
posterior P(w|D) can be approximated as
P(w|D) e
_

N
n=1
e
n
2
_
e
_

B
b=1

b
2
w
T
b
w
b
_
. (13)
Finally, the optimal regularization parameter
opt
b
is cal-
culated by maximizing the posterior of P(, |D) while the
marginal likelihood is expressed in (14); further calculation
details are available in [18] and [22]
P(D|, ) =
P(D|w, )P(w|)
P(w|D)
. (14)
Fig. 2 shows the owchart of the RNCL-based algorithm for
the developed NNE composed of B predictors. The original
training data set D = {(x
1
, d
1
), . . . , (x
N
, d
N
)} is normalized
(rescaled) using across-channel normalization [14]. The learn-
ing is conducted in an iterative mode, for the rst epoch
initial regulation parameters {
b
}
B
b=1
are set for the training
of the networks. Each single predictor training is executed for
a desired number of iterations j
Max
; once all the predictors
are trained (b = B), the NNE average error is calculated to
update the regularization parameter
opt
b
and start a new epoch.
1692 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO. 4, APRIL 2013
TABLE I
ANN INPUTS
The process is repeated until a suitable convergence criterion is
reached (goal).
B. Microgrid Environment Forecasting
Table I illustrates the input layers variables associated to the
proposed NNEs, where TUAT refers to the data recorded inside
the TUAT and JMA refers to the data collected from Japan
Meteorological Agency, measured in the local area of the TUAT
(Koganei city campus).
IV. MULTIOBJECTIVE INTELLIGENT
ENERGY MANAGEMENT
A. MO
The research eld of considering decision problems with
multiple conicting objectives (goals or criteria) is known
as multiple criteria decision making or MO [27][29]. Such
subject covers both discrete problems (with a nite set of
alternatives, also called actions or solutions) and continuous
problems. In other words, the MO is a methodology aimed
to nd the best solution among different objectives, usually
conicting objectives, where the problem can be handled as
follows:
minimize {f
1
(P), . . . , f
k
(P)} (k2) (15)
Subjected to :
equality constraints h
i
(P)=0, i =1, 2, . . . , q (16)
inequality constraints g
j
(P)0, j =1, 2, . . . , p (17)
lower and upper bounds p
min
i
P
i
P
max
i

i
=1, . . . , N. (18)
The conicting objective functions f
i
:
n
have to be
minimized simultaneously; the decision vector (variable) P =
(P
1
, P
2
, . . . , P
N
)
T
belongs to the nonempty feasible region
S
n
. A decision vector is regarded as optimal if none of its
components can be improved without deterioration to at least
one of the other components. More precisely, a decision vector
P

S is called Pareto optimal if there is no other P S such


that f
i
(P) f
i
(P

)
i
= 1, . . . , k and f
j
(P) < f
j
(P

) for at
least one index j.
For the proposed microgrid energy management, the solution
of the optimal multiobjective process comes to nd an output
power generation vector P = (P
1
, P
2
, . . . , P
N
) providing the
generation set points of each controllable DG to guarantee the
minimum cost and gaseous emissions during the microgrids
operation while satisfying the load balance within the power
generation rating limit of each DG. Therefore, the general
model described in (15) is expressed in the following to dene
the proposed microgrid MO energy management involving two
objective functions:
f
1
(P) =
N

i=1
FC
i
(P
i
) +P
b
P
B
P
s
P
S
+M
i
(P
i
) (19)
f
2
(P) =
N

i=1
M

j=1
(EF
ij
P
i
) +GEF P
B
(GEF MGEF) P
S
(20)
where P
i
is the power output related to the DG unit i in
kilowatthours, FC
i
is the fuel consumption cost related to
the DG unit i in dollars per kilowatthour, P
b
is the actual
hour electricity buying price in dollars per kilowatthour, P
B
is the amount of purchased electricity, P
s
is the actual hour
electricity selling price in dollars per kilowatthour, P
S
is the
sold electricity in kilowatthours, M
i
is the maintenance cost
related to the DG unit i in dollars per kilowatthour, EF
ij
is
the emission factor related to the DG unit i in pounds per
kilowatthour where j is the emission type (NO
x
, CO
2
, and
SO
2
), MGEF is the microgrid average emission factor in
pounds per kilowatthour, and, nally, GEF is the main grid
average emission factor in pounds per kilowatthour.
The fuel consumption FC
i
for the controllable DG (fuel cell
and microturbine) is expressed in (21); more details can be
found in [30][32]
FC
i
(P
i
) = C
i
P
i

i
(21)
where C
i
is the fuel price and
i
is the efciency rate related to
the DG unit i.
The objective functions (19) and (20) are subjected to an
equality constraint for active power and load balance (22) and
restriction limits of the output related to each DG unit (23)
N

i=1
P
i
+P
PV
+P
WT
+P
batt
+P
B
L P
S
= 0 (22)
P
min
i
P
i
P
max
i

i
= 1, 2, . . . , N. (23)
CHAOUACHI et al.: MULTIOBJECTIVE INTELLIGENT ENERGY MANAGEMENT FOR A MICROGRID 1693
Fig. 3. Cycle expectancy for a leadacid battery.
The cost objective function expressed in (19) depends on
three parameters: The rst one is the power generation cost
related to each DG based on the fuel price and the consumption
rate of each unit described in (21), the second one is the energy
trade between the main grid and the microgrid, and the third
one is the maintenance cost assumed to be proportional to the
power generation of each DG unit [33]. On the other hand,
the emission objective function expressed in (20) depends on
the individual emission rate of each DG during autonomous
generation and on the main grid average emission rate (fossil,
coal, and nuclear) in case of a microgrids generation lack [30],
[34]. Furthermore, selling power to the grid is equivalent to
reducing the overall emission (microgrid + main grid) as the
microgrid emits less than conventional power plants; in this
case, the emission saving is approximated as the difference
between the microgrid average emission (MGEF) and the
main grid average emission (GEF).
B. Battery Scheduling
A 15-kWh leadacid battery is used as storage facility; such
batteries are relatively cheap but suffer from deep depth of
discharge (DOD) as it dramatically decreases its lifetime by
reducing the expected average cycles (Fig. 3). In this paper,
the lead-acid battery efciency is assumed to be 90% for both
charging and discharging efciencies. Based on the previous
assumptions, (22) is modied depending on the following two
cases:
1) charging
P
batt
= 0.9.
_

P
i
+P
PV
+P
WT
+P
B
L P
S
_
(24)
2) discharging
P
batt,k
=
1
0.9

_
L +P
S

P
i
P
PV
P
WT
P
B
_
.
(25)
C. FL-Based Expert System
The battery scheduling as a part of optimal online energy
management can be regarded as a decision-making process.
In fact, one way to solve this problem is to design an ex-
pert system that is able to decide, referring to the forecasted
entities, whether the battery should be charged or discharged
Fig. 4. Input membership functions.
and at which convenient rates. The whole process takes into
account the safe operation of the battery, the minimization of
the operation costs, and the gaseous emissions. Such approach
cannot be implemented using a deterministic analysis or a
linear-programming-based MO due to the sizeable scenario
combinations and their stochastic nature. In this paper, fuzzy
logic (FL)-based expert system is applied for microgrid battery
scheduling along with the multiobjective (cost and emission)
linear programming optimization. Indeed, fuzzy set theory can
encompass such subjective decision-making process due to its
ability to dene human reasoning that can handle uncertain-
ties regarding to the microgrid exogenous environment and
the uncertainty of the forecasted parameters. Moreover, such
approach can be easily extended to t the plug and play concept
of microgrids irrespectively of the generation rating and the
architecture of its components.
1) Fuzzication Process: The fuzzy-based expert system
applied to the battery scheduling is based on seven mea-
sured/forecasted parameters listed as follows: the actual time
(T), the battery state of charge (BSOC), electricity price (EP),
LD, REG, actual solar daily generation (ASDG) which is
equal to

24
1
SPG
i
, and tomorrows solar daily availability
(TSDA) which is directly assigned in linguistic terms based on
Japan Meteorological Agency weather forecasting referred to as
sunny, mostly sunny, and cloudy. The input membership func-
tions forward a degree of membership related to its respective
input patterns to the fuzzy inference engine.
Fig. 4(a) shows the time membership function that takes as
input the actual time and assigns it to a degree of a membership
for each fuzzy set in the graph, where Mr stands for morning,
Md stands for midday, and N stands for night. Fig. 4(b) shows
the BSOC membership functions, where the input is the actual
state of charge of the battery assigned into four membership sets
listed as VL (very low), L (low), M (medium), and H (high).
Fig. 4(c) shows the EP membership functions, where the input
is the actual EP (in dollars per kilowatthour) assigned into three
membership sets listed as L (low), M (medium), and H (high).
Fig. 4(d) shows the LD membership functions, where the input
is the next-hour forecasted LD (in kilowatthours) assigned into
1694 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO. 4, APRIL 2013
TABLE II
PARAMETERS RELATIVE TO THE INPUT MEMBERSHIP FUNCTION SETS
Fig. 5. Typical membership function.
three membership sets listed as L (low), M (medium), and H
(high). Fig. 4(e) shows the REG membership function, where
the input is the next-hour forecasting of the SPG and WPG
(in kilowatthours) assigned into three membership sets listed
as L (low), M (medium), and H (high). Finally, Fig. 4(f) shows
the ASDG membership function, where the input is the average
of the 24-h-ahead SPG forecasting (in kilowatthours) assigned
into three membership sets listed as L (low), M (medium), and
H (high).
The parameters related to the input membership functions
are listed in Table II, where each variable coordinate is dened
based on the membership trapezoidal function model shown in
Fig. 5.
2) Inference Engine: Once the degrees of membership func-
tions of each fuzzy set have been determined for a particular
input, they are forwarded to the inference engine that denes
which rules should be evaluated; a typical fuzzy rule is ex-
pressed in this form
If (ASDG is H) and (TSDA is L) and (T is Md) and
(BSOC is H) and (LD is H) and (EP is M) and (REG
is L) Then (Discharge M).
The implemented fuzzy rules summarize an efcient battery
scheduling based on human expertise, and the goal is to decide
the most appropriate action whether to charge or discharge the
battery at a specic rate while mainly taking into consideration
the availability of the REG sources, the LD, and the EP. The
rule ensemble with charging consequent mainly tends to reach
the full BSOC with the slowest possible rate where the priority
for charging is associated to the low LD period and, eventually,
high generation potential of the renewable sources (i.e., high
REG). In case of potential lack of renewable energies (low REG
or ASDG), the battery scheduling tends to accelerate the charg-
ing process during low LD period (morning) so that the full
BSOC is reached antecedent to the midday load peak demand;
thereby, the dispatchable DG can be fully dedicated to meet
the load balance and therefore avoid expensive energy purchase
from the main grid, particularly in case of SPG unavailability.
On the other hand, the rule ensemble with discharging conse-
quent mainly tends to meet the LD balance with the highest
priority to avoid expensive energy purchases or trespass the
DOD limit. At the end of the day, BSOC is managed with
respect to the actual and next-day SPG availabilities (ASDA
and TSDA). Taking the case of an actual sunny day followed
by a cloudy day, the scheduling strategy will tend to store an
important quantity of energy during the actual day which obvi-
ously increases the actual operation cost; however, the overall
cost for the two days will noticeably decrease as important
energy purchases will be avoided on the next day during load
peak hours. The evaluated rules are selected based on nonzero
memberships of the input values. For instance, the previously
referred rule would only be selected if all of the inputs had
a membership value other than zero. Once this rule (and any
others meeting these criteria) is found, the output degrees of
membership would be asserted according to the memberships
values of the inputs.
3) Defuzzication: Once the degrees of membership have
been calculated via the inference engine, the next step is the
defuzzication into an output signal of charging or discharging
rate. This is done much like the fuzzication process but in
a reverse process. In the case where multiple rules have been
asserted (and, hence, multiple degrees of membership for the
outputs), the center of mass of the weighted outputs is used.
This output value is then multiplied by a normalizing value to
return it to the level of real-world outputs as a charging (Chg) or
a discharging (Dsc) rate. However, in case of possible linguistic
conicting output rules (i.e., charging and discharging in the
same time), then the rule with the higher membership output is
retained due to the fact that the two outputs are mutually ex-
clusive. Finally, the output corresponding to each input pattern
of the fuzzy expert system is forwarded to the multiobjective
algorithm and substituted in (24) or (25) where the energy
stored or restored to/from the battery is expressed as
P
batt
= (Chg +Dsc)BSOC. (26)
Compared to microgrids conventional formulations for bat-
tery scheduling, the proposed FL-based expert system approach
tends to be less conservative. Indeed, conventional binary-
logic-based methodologies handle decision making related to
the management of the battery basing on parameters that can
only have one of two states, i.e., true or false, symbolized
by 1 and 0. On the other hand, FL uses notions such as partially
true and partially false or completely true and completely
false. Such feature outcomes a more tolerant decision-making
process capable of adaptation as situations evolve as well as
taking consideration of potential uncertainties regarding to the
forecasted variables.
Fig. 6 summarizes the proposed microgrid MIEM control
strategy based on the FL battery scheduling. Local measure-
ment of the microgrid LD, WPG, and SPG as well as weather
conditions is prerequisite to forecast the RE power generation
CHAOUACHI et al.: MULTIOBJECTIVE INTELLIGENT ENERGY MANAGEMENT FOR A MICROGRID 1695
Fig. 6. MIEM control structure based on FL battery scheduling.
and the microgrids LD. The forecasted entities are forwarded
along with the BSOC, time, and cost factors (electricity and
fuel prices) to the FL-based decision-making system and the
MO linear programming to calculate the optimal operation set
points of the battery and the controllable DG sources.
V. SIMULATION RESULTS AND INTERPRETATIONS
A. RE Power Generation and LD Forecasting
For the validation of the proposed NNE-based forecasting
tasks, statistical error forecasting performance is investigated
basing on the mean average deviation (MAD) and the mean
absolute percentage error (MAPE). For each forecasting task,
the NNE forecasting performances are compared to conven-
tional single ANNs consisting of an MLPNN, an RBFNN, and
an RNN.
Table III summarizes the entire statistical performances of
the proposed forecasting tasks. The forecasting results are com-
pared based on MAD and MAPE criteria for each of the devel-
oped ANNs, and testing data sets are composed of four seasons
with different weather conditions. Table III shows clearly that
the proposed NNE has the lowest forecasting error including
different testing data sets covering winter, spring, summer, and
fall. On the other hand, RBFNN, RNN, and MLPNN achieved
also an acceptable forecasting accuracy. Among the conven-
tional neural networks, the RBFNN presented the best overall
results, while the MLPNN achieved the lowest forecasting
accuracy. Indeed, the RBFNN can overcome several limitations
of MLPNN and RNN such as a highly nonlinear weight update
and slow convergence rate.
TABLE III
STATISTICAL ERROR FORECASTING PERFORMANCES
Fig. 7. SPG forecasting error of the proposed ANNs.
Fig. 7 shows a daily 2-D SPG prediction error graph in
kilowatthours of the developed ANNs, where it can be observed
that the proposed NNE achieved the lowest forecasting error.
Fig. 8 shows a comparison of the LD forecasting projected by
the developed ANNs during two days; the rst one is a holiday
with a low LD peak of 31 kWh, while the second one is a
working day with a high LD peak reaching 41 kWh. It can be
observed that the LD characteristics during the two days are
very different. In fact, for a holiday day type, the LD is almost
at, while the working day is characterized by low (6 A.M.) and
high peak (3 P.M.) demands. Because of the important dynamic
dissimilarities of the two daily LD characteristics, it can be
seen that the NNE accomplished a visible error attenuation
improvement resulting in stable forecasting performances.
1696 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO. 4, APRIL 2013
Fig. 8. LD forecasting for a working day and a holiday.
Fig. 9. Comparison of the forecasted and the real generated power of the PV
array.
Fig. 9 shows one-day measurement of the real power gen-
erated from the PV array that is used for the microgrid real-
time simulation model. The measured real SPG similitude with
the 24-h-ahead SPG forecasting used to predict the MIEM of
the microgrid corroborated the results obtained based on NNE
forecasting.
B. Multiobjective Intelligent Energy Management
To validate the MIEM, three approaches were compared: The
rst two approaches consist in a MIEM based on FL expert
system and HF battery scheduling, while the third one consists
in MO of the microgrid where the battery scheduling is man-
aged using conventional opportunity charging. The developed
HF is composed of six scenarios, and each scenario represents a
scheduling strategy of one-day typical operation. The choice of
a specic scenario depends on the actual and the next-day REG.
Dealing with each scenario is involving the time, the actual
state of charge of the battery, the actual load, and the actual
RE power generations where the time variable is correlated
to the EP (as the EP peak usually occurs during a specic
time period). Each scenario of the battery scheduling owchart
establishes operation set points for charging or discharging
the battery during operation hours, which comes to dene
the power stored or restored (P
batt
). After the execution of
Fig. 10. Battery scheduling process for one-day operation.
Fig. 11. Cost and emission comparison of the microgrid operation.
each iteration, (24) or (25) referring to charging or discharging
replaces the conventional equality constraint (23) while running
the MO process; further details can be found in [11] and [35].
Fig. 10 shows the battery scheduling for one-day operation
using the proposed FL-based expert system. The full BSOC is
reached before the peak of LD occurrence, then followed by a
gradual battery discharge (peak shaving). The delineated hourly
charging rates resulted in a smoother charging dynamic; there-
fore, there are less solicitations on the battery (overheating)
and the DG controllable sources. Similarly, during load peak
demand period, the proposed approach achieved a successful
load peak shaving with the slowest possible discharging rates.
Fig. 11 shows a comparison of the emission and operation
cost during one-day operation. The proposed MIEM (based on
FL battery scheduling) achieved important saving on opera-
tion cost as well as the gaseous emission level compared to
a conventional control operation (without MIEM). It can be
observed that the proposed methodology achieved considerable
cost saving during LD peak that coincides with nonavailability
of the PV generator and consequent rise of the EPs.
The simulation results cover two successive sunny and
cloudy days and one week using the HF, the FL, and op-
portunity battery management (no MIEM). Table IV summa-
rizes the simulation results that include the operation cost,
CO
2
, SO
2
, and NO
x
emissions, and the battery initial and nal
states of storage (beginning and end of the day). It can be seen
that the MIEM (using intelligent battery scheduling) achieved a
visible saving on the operation cost and the gaseous emissions.
The proposed MIEM based on FL battery scheduling accom-
plished 1.35% and 5.76% of cost saving and 2.96% and 6.1% of
CHAOUACHI et al.: MULTIOBJECTIVE INTELLIGENT ENERGY MANAGEMENT FOR A MICROGRID 1697
TABLE IV
OPERATION COST AND EMISSIONS OF THE MICROGRID
lower emissions compared to the HF and the conventional op-
portunity charging approaches, respectively, during the whole
referred simulation period. For the rst-day simulation, the cost
saving of the proposed MIEM approach compared to the op-
portunity charging was equal to 4.63% compared to 8.15% for
the next day as a result of the intelligent battery scheduling that
takes into account the next-day solar energy availability. In fact,
during the rst simulation day, the management strategy adapts
the charge, discharge, and the end-of-the-day storage so that the
saved energy can be restored during the next day characterized
by a poor SPG potential; therefore, more important operation
cost savings are achieved during the second day compared to
the rst one.
In addition to the saving on the operation cost and emissions,
indirect saving is also incited using the proposed intelligent
management strategy that consists in less maintenance and
replacement cost for the lead-acid battery. In fact, the lead-
acid battery lifetime and operation performance are affected
(therefore, maintenance cost and overall microgrid operation
cost) by fast charging rates that can lead to overcharge and/or
important temperature rise. Moreover, the life cycle of the
battery depends closely on the DOD in each cycle; the HF-
and FL-based scheduling approaches achieved DOD averages
equal to 64% and 61%, respectively, compared to 80% for the
opportunity battery management. According to Fig. 3, the FL
and HF battery scheduling approaches assure approximately
820 and 790 expected average cycles, respectively, compared to
515 cycles for the conventional opportunity charging approach.
Assuming that the battery accomplishes a charging and dis-
charging cycle once a day, then the battery needs to be approx-
imately replaced in 27 and 26 months for the case of FL- and
HF-based approaches, respectively, whereas it needs to be re-
placed in only 17 months for an opportunity charging approach.
For a replacement cost of the batteries equal to $180/kWh
($2700 for the whole system), extra daily maintenance costs
equal to $3.28, $3.4, and $5.4 for the FL, the HF, and the
opportunity charging battery management, respectively, need to
be considered as part of the overall operation cost. It is clear
that the developed MIEM approach based on FL expert system
for battery scheduling achieved the lowest operation cost and
emissions while paradoxically requiring a lower average DOD
of the battery compared to the two other approaches. This is
due to the efciency of the battery scheduling as for avoiding
important purchase of power from the main grid during the
Fig. 12. Comparison of the power ow between the main grid and the
microgrid.
electricity demand peaks. In fact, during load peak hours, the
EPs can dramatically increase in a deregulated electricity mar-
ket, resulting in an important increase of the overall operation
costs.
The proposed MIEM predictive control system is imple-
mented on the microgrid simulation model developed under
Matlab Simulink environment to validate the proposed ap-
proach over 24-h real-condition operation including experimen-
tal data measured in TUAT.
Fig. 12 shows the prole of power exchanged with the
main grid in real condition based on the microgrid simulation
model for a MIEM based on FL and HF battery scheduling
approaches. It is expected that the equality constraint on load
balance (22) may be partially violated in real-condition opera-
tion as the forecasted inputs of the MIEM are hourly averaged
values without considering the forecasting errors. However, the
proposed MIEM with FL expert system for battery scheduling
presented noticeably limited power import from the main utility
compared to the HF-based approach. As a result, considering
a microgrids long-term operation, important minimization on
operation cost and gaseous emissions can be achieved using the
proposed approach.
VI. CONCLUSION
In this paper, microgrid intelligent online energy manage-
ment under cost and emission minimization has been in-
vestigated; the problem formulation includes optimal battery
scheduling, taking into account the uncertainty of the microgrid
exogenous variables and forecasted entities.
In order to validate the NNE forecasting module, the fore-
casting reliability of the machine learning was evaluated in
terms of forecasting error. The experimental results showed
that the NNE achieved a higher forecasting accuracy than
conventional MLPNN, RBFNN, and RNN. In fact, the NNE can
improve the generalization and noise tolerance of the learning
machine effectively through aggregating numbers of neural
networks with different models by improving the generalization
ability and reducing the variance and the performance instabil-
ity. The forecasted entities are crucial data for the implementa-
tion of the online intelligent energy management.
The MIEM for microgrid was validated by comparing three
approaches for battery scheduling jointly with the MO for
dispatchable DG sources. The simulation results proved the
effectiveness of the proposed MIEM using FL-based expert
1698 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO. 4, APRIL 2013
system. The proposed intelligent energy management approach
achieved clearly important minimization on the operation cost
and emission level of the microgrid. Moreover, the fuzzy ex-
pert system for battery scheduling decreased considerably the
battery maintenance cost by extending the operation lifetime
through lower depth of battery discharge.
The proposed approach is nally implemented on a micro-
grid simulation model to validate the proposed methodology.
Simulation during 24-h real-condition operation including ex-
perimental data has been performed. The results of the sug-
gested MIEM compared with HF-based battery management
corroborated the results obtained based on the proposed control
system.
REFERENCES
[1] R. Firestone and C. Marnay, Energy manager design for microgrids,
CERTS, Calif. Energy Comm., Sacramento, CA, LBNL-54447, 2005.
[2] S. X. Chen and H. B. Gooi, Jump and shift method for multi-objective
optimization, IEEE Trans. Ind. Electron., vol. 58, no. 10, pp. 45384548,
Oct. 2011.
[3] S. Paudyal, C. A. Canizares, and K. Bhattacharya, Optimal operation of
distribution feeders in smart grids, IEEE Trans. Ind. Electron., vol. 58,
no. 10, pp. 44954503, Oct. 2011.
[4] H. Kanchev, D. Lu, F. Colas, V. Lazarov, and B. Francois, Energy man-
agement and operational planning of a microgrid with a PV-based active
generator for smart grid applications, IEEE Trans. Ind. Electron., vol. 58,
no. 10, pp. 45834592, Oct. 2011.
[5] C. A. Hernandez-Aramburo, T. C. Green, and N. Mugniot, Fuel con-
sumption minimization of a microgrid, IEEE Trans. Ind. Appl., vol. 41,
no. 3, pp. 673681, May/Jun. 2005.
[6] F. A. Mohamed and H. N. Koivo, System modelling and online optimal
management of microgrid using mesh adaptive direct search, Int. J. Elect.
Power Energy Syst., vol. 32, no. 5, pp. 398407, 2010.
[7] C. M. Colson, M. H. Nehrir, and S. A. Pourmousavi, Towards real-time
microgrid power management using computational intelligence methods,
in Proc. IEEE Power Energy Soc. Gen. Meeting, Minneapolis, MN, 2010,
pp. 18.
[8] G. Celli, F. Pilo, G. Pisano, and G. G. Soma, Optimal participation of a
microgrid to the energy market with an intelligent EMS, in Proc. IEEE
Int. Power Eng. Conf., Singapore, 2005, pp. 663668.
[9] E. Mashhour and S. M. Moghaddas-Tafreshi, Integration of distributed
energy resources into low voltage grid: A market-based multiperiod op-
timization model, Elect. Power Syst. Res., vol. 80, no. 4, pp. 473480,
Apr. 2010.
[10] F. A. Mohamed and H. N. Koivo, Multiobjective optimization using
modied game theory for online management of microgrid, Eur. Trans.
Elect. Power, vol. 21, no. 1, pp. 839854, Jan. 2011.
[11] S. Chakraborty, M. D. Weiss, and M. G. Simes, Distributed intelligent
energy management system for a single-phase high-frequency AC micro-
grid, IEEE Trans. Ind. Electron., vol. 54, no. 1, pp. 97109, Feb. 2007.
[12] F. Katiraei, R. Iravani, N. Hatziargyriou, and A. Dimeas, Microgrids
management, IEEE Power Energy Mag., vol. 6, no. 3, pp. 5465,
May/Jun. 2008.
[13] A. Chaouachi, R. M. Kamel, and K. Nagasaka, Microgrid efciency
enhancement based on neuro-fuzzy MPPT control for photovoltaic gener-
ator, in Proc. IEEE Photovol. Spec. Conf., Honolulu, HI, 2010, pp. 2889
2894.
[14] R. M. Kamel, A. Chaouachi, and K. Nagasaka, Wind power smoothing
using fuzzy logic pitch controller and energy capacitor system for im-
provement micro-grid performance in islanding mode, Energy, vol. 35,
no. 5, pp. 21192129, May 2010.
[15] R. Lasseter, K. Tomsovic, and P. Piagi, Scenarios for distributed tech-
nology applications with steady state and dynamic models of loads and
micro-sources, CERTS, U.S. Dept. Energy, Washington, DC, 2000.
[16] Z. Gao, F. Ming, and Z. Hongling, Bagging neural networks for predict-
ing water consumption, J. Commun. Comput., vol. 2, no. 3, pp. 1925,
2005.
[17] L. K. Hansen and P. Salamon, Neural network ensembles, IEEE Trans.
Pattern Anal., vol. 12, no. 10, pp. 9931001, Oct. 1990.
[18] B. Rosen, Ensemble learning using decor related neural networks, Con-
nection Sci., vol. 8, no. 3, pp. 373383, 1996.
[19] G. Brown and X. Yao, On the effectiveness of negative correlation learn-
ing, in Proc. 1st UK Workshop Comput. Intell., Edinburgh, U.K., 2001,
pp. 5762.
[20] Z. S. H. Chan and N. Kasabov, Fast neural network ensemble learning via
negative-correlation data correction, IEEE Trans. Neural Netw., vol. 16,
no. 6, pp. 17071710, Nov. 2005.
[21] N. Amjady, F. Keynia, and H. Zareipour, Short-term load forecast of
microgrids by a new bilevel prediction strategy, IEEE Trans. SmartGrid,
vol. 1, no. 3, pp. 286294, Dec. 2010.
[22] H. Chen and X. Yao, Regularized negative correlation learning for neural
network ensembles, IEEE Trans. Neural Netw., vol. 20, no. 12, pp. 1962
1979, Dec. 2009.
[23] M. F. Mller, A scaled conjugate gradient algorithm for fast supervised
learning, Neural Netw., vol. 6, no. 4, pp. 525533, 1993.
[24] R. M. Neal, Bayesian Learning for Neural Networks. New York:
Springer-Verlag, 1996.
[25] D. J. C. MacKay, Bayesian interpolation, Neural Comput., vol. 4, no. 3,
pp. 415447, 1992.
[26] T. V. Gestel, J. A. K. Suykens, G. Lanckriet, A. Lambrechts, B. D. Moor,
and J. Vandewalle, Bayesian framework for least-squares support vector
machine classiers, Gaussian processes, and kernel Fisher discriminant
analysis, Neural Comput., vol. 14, no. 5, pp. 11151147, May 2002.
[27] M. Ehrgott and X. Gandibleux, Multiple Criteria Optimization: State of
the Art Annotated Bibliographic Surveys. Boston, MA: Kluwer, 2003.
[28] H. Li and Q. Zhang, Multiobjective optimization problems with compli-
cated Pareto sets, MOEA/D and NSGA-II, IEEE Trans. Evol. Comput.,
vol. 13, no. 2, pp. 284302, Apr. 2009.
[29] F. Kash, S. Hatami, and M. Pedram, Multiobjective optimization tech-
niques for VLSI circuits, in Proc. Int. Symp. Quality Electron. Des.,
2011, pp. 18.
[30] Emission Rates for New DG Technologies, the regulatory as-
sistance project. [Online]. Available: http://www.raponline.org/docs/
RAP_DGEmissions_2001_05.pdf
[31] S. Campanari and E. Macchi, Technical and tariff scenarios effect on
microturbine trigenerative applications, J. Eng. Gas Turbines Power,
vol. 126, no. 3, pp. 581589, Jul. 2004.
[32] S. Campanari, L. Boncompagni, and E. Macchi, Microturbines and
trigeneration: Optimization strategies and multiple engine congura-
tion effects, J. Eng. Gas Turbines Power, vol. 12, no. 1, pp. 92101,
Jan. 2004.
[33] F. Barbir and T. Gmez, Efciency and economics of proton exchange
membrane PEM fuel cell, Int. J. Hydrogen Energy, vol. 21, no. 10,
pp. 891901, Oct. 1996.
[34] S. Hggmark, V. Neimane, U. Axelsson, P. Holmberg, G. Karlsson,
K. Kauhaniemi, M. Olsson, and C. Liljegren, Aspects of different distrib-
uted generation technologies, CODGUNet WP 3, Vattenfall Utveckling
Ab, 2003.
[35] A. Chaouachi, R. M. Kamel, and K. Nagasaka, Multiobjective online
management of a microgrid, in Proc. Renew. Energy Conf., Yokohama,
Japan, 2010, pp. 8184.
Aymen Chaouachi (S11M12) was born in Tunis,
Tunisia, in 1981. He received the B.E. and M.S.
degrees in electrical engineering from the Ecole
Suprieure de Science et Technologie de Tunis,
Tunis, in 2004 and 2006, respectively, and the Ph.D.
degree in electrical engineering from Tokyo Univer-
sity of Agriculture and Technology, Koganei, Japan,
in 2011. His Ph.D. research was on smart microgrid
control and energy management with emphasis on
intelligent integration of renewable sources.
Before moving to Japan, he was a Lecturer with
the Tunisian Ministry of Education. He has also been involved with several
research projects on photovoltaic system control and integration with the
Tunisian Research and Technology Center of Energy (CRTEn), Bordj-cedria.
He is currently with Hitachi, Ltd., Tokyo, Japan, where his tasks involve
research and development of interoperability, smart control, and protection of
substation automation within the smart grid concept based on the IEC 61850
standards. His research interests include renewable energy, control systems,
articial neural networks, fuzzy logic, power electronics, and smart grid.
Dr. Chaouachi was the recipient of several awards, travel grants, and
scholarships from the Tunisian and Japanese governments for his scholastic
achievements.
CHAOUACHI et al.: MULTIOBJECTIVE INTELLIGENT ENERGY MANAGEMENT FOR A MICROGRID 1699
Rashad M. Kamel was born in Sohag, Egypt, in
1978. He received the B.Sc. and M.Sc. degrees
in electrical engineering from Assiut University,
Assiut, Egypt, in 2000 and 2005, respectively, and
the Ph.D. degree in renewable energy and microgrid
from Tokyo University of Agricultural and Technol-
ogy, Koganei, Japan, in 2011.
He is currently an Assistant Professor with the
Electrical Engineering Department, Faculty of En-
gineering, Assiut University. His research inter-
ests include power system dynamics, stability, and
protection.
Ridha Andoulsi was born in Tunisia in 1961. He
received the B.S. and M.S. degrees in electrical
engineering from the Ecole Suprieure de Science et
Technologie de Tunis, Tunis, Tunisia, in 1986 and
1991, respectively, and the Ph.D. degree in electrical
engineering (automatic and industrial computing)
from the University of Sciences and Technologies of
Lille, Villeneuve dAscq, France, in 2001.
In 1992, he occupied a permanent Research Scien-
tist position in the Research and Technology Center
of Energy (CRTEn), Bordj-cedria, where he led the
Photovoltaic System Group in the Photovoltaic Laboratory. He is currently an
Assistant Professor with the Electrical Engineering Department, Community
College, University of Hail (HCC), Hail, Saudi Arabia. Before joining HCC in
October 2009, he was a Visiting Professor for one year with Tokyo University of
Agriculture and Technology, Koganei, Japan. His research interests encompass
both theoretical foundations and system development of photovoltaic energy.
Ken Nagasaka (S86M12) was born in Kerman-
shah, Iran, in 1956. He received the B.S. degree in
electrical engineering from Nihon University, Tokyo,
Japan, in 1985, and the M.S. and Ph.D. degrees from
Tokyo Metropolitan University, Tokyo, in 1987 and
1990, respectively.
From 1990 to 1991, he was a Chief Researcher
for a computer software development company in
Tokyo, Japan. From 1991 to 1994, he was a Vis-
iting Professor with the University of Manitoba,
Winnipeg, MB, Canada. From 1994 to 1998, he was
a Scientist with the Central Research Institute of the Electric Power Industry,
Tokyo. From 1995 to 1998, he was an Associate Professor with the Department
of Electrical Engineering, Tokyo Metropolitan University. He is currently an
Associate Professor with the Department of Electrical and Electronics Engi-
neering, Graduate School of Engineering, Tokyo University of Agriculture and
Technology, Koganei, Japan. He has published more than 500 scientic papers,
one of which won a prize from the Institute of Electrical Installation Engineers
of Japan (IEIEJ) in 1991. His current research projects concern environmental
energy engineering, particularly power systems analysis, power deregulation,
wind power, microgrid, smart grid, energy saving, load forecasting, and appli-
cation of intelligent systems to power systems.
Dr. Nagasaka was a recipient of three awards from the Power System
Conference (PSC) in 2001, 2002, and 2006. Also, in 2006, he was a recipient
of a medal from the University of Tehran, Tehran, Iran, for his contribution
to the First Conference of Control and Management of Energy Systems. He is
a member of the Institute of Japan Society of Energy and Resources (JSER)
IEIEJ, and the International Neural Network Society of the U.S.

Vous aimerez peut-être aussi