126 IEEE SYSTEMS JOURNAL, VOL. 6, NO.

1, MARCH 2012

Optimal Control in a Cooperative Network of

Smart Power Grids
Riccardo Minciardi, Member IEEE, and Roberto Sacile, Member IEEE

Abstract—The possibility to store energy, to exchange power examples are: microgrids, and virtual utilities. According to
and information on demand and production among grids allows [4], microgrids are defined as low voltage networks with
us to achieve an active distribution which is of major interest for
distributed generation sources, together with local storage
cooperative smart power grids, that are grids which can forecast
demand and production and are able to exchange power in order devices and controllable loads. They have a total installed
to enhance the quality of the service. In this paper, a model to capacity in the range from a few hundred kilowatts to a
support optimal decisions in a network of cooperative grids is couple of megawatts. Microgrids mostly operate connected
formalized as an original discrete and centralized problem here to the distribution network. Within the main grid, a microgrid
defined as cooperative network of smart power grids (CNSPG)
can be regarded as a controlled entity which can be operated
problem. In the CNSPG problem, the control variables are the
instantaneous flows of power in the network of grids, which can as a single aggregated load or generator and as a small source
be obtained from the solution of a linear quadratic Gaussian of power supporting the main grid. In [4], virtual utilities
problem on a fixed time horizon. A simple case study showing have an Internet-like structure with information exchange and
the enhancement which may be obtained from the introduction trading capability. Power is purchased and delivered to nodes
of direct connections among microgrids according to a lattice
and it is determined by the supplier, e.g., using a conventional
network is shown and finally discussed.
generator, RES, or an energy storage device. EU microgrid
Index Terms—Energy storage, linear quadratic Gaussian and virtual utilities research go in the same direction of
(LQG) control, optimal control, power microgrids, renewable
United States Gridwise initiative (see [5], for more details).
energy, smart grids.
Among the possible approaches to allow a RES base active
I. Introduction distribution, the following options seem to be particularly
relevant to enhance the flexibility and the agility of a smart
T HE USE OF renewable energy sources (RESs) is one of
the fundamental strategies to fight against climate change
and to reduce the dependence on fossil fuels. Wind and solar
power microgrid: to integrate different RESs into a so-called
hybrid system, with the intent to provide a more stable energy
sources seem to be particularly promising and several countries supply; to provide a microgrid or a renewable energy system
are investing in technologies to exploit these RESs. However, with effective local storage devices [6], in order to store energy
due to the fact that the wind and solar energy sources cannot in periods when demand is lower than produced power, and,
be controlled and are dependent on meteorological conditions, vice versa, to provide power when demand is higher than the
an intermittent, stochastic behavior characterizes such sources. produced power; to cooperate either with other microgrids
On the other hand, the possibility to use modern smart or with the National grid itself to exchange information on
meters provides the knowledge of the current electric con- predicted demand and production, as well as power in order
sumption of each user in real-time. Besides, it is also possible to enhance the overall service provision.
to predict forthcoming user demand, although also in this case, In regards to the three options quoted above, several recent
due to unexpected user behaviors, the prediction is affected by contributions are available in the literature. Some of them are
an error which can be modeled as a stochastic process. These quoted hereinafter.
possibilities lead to the concept of active distribution [1]–[3]. In regards to hybrid systems, Anagnostopoulos and Papan-
Among RESs, wind and solar sources are not directly tonis [7] presented a numerical study for the optimum sizing
exploitable for active distribution. Technological and method- and design of a station unit in a hybrid wind-hydro plant (i.e.,
ological approaches are necessary to make smart power grids wind turbines and hydroelectric plant). Dagdougui et al. [8]
more flexible and agile to react promptly to the forecasted user introduced a dynamic decision model for the real time control
demand. of hybrid renewable energy production systems, which can be
In European Union (EU), new system concepts have been particularly suitable for autonomous systems, such as islands
adopted to implement such an active distribution [4]. Two or isolated villages. In their paper, the demand of energy is
coupled with the demand of water and hydrogen, where the
Manuscript received September 22, 2010; revised February 10, 2011;
accepted May 25, 2011. Date of publication August 30, 2011; date of current hydrogen and water reservoirs also work as storage energy
version February 23, 2012. systems. Korpas and Holen [9] presented a methodology for
The authors are with the Department of Communication, Computer and the definition of control strategies for a hybrid plant with wind
System Sciences, University of Genova, Genova 16145, Italy (e-mail:
[email protected]; [email protected]). power and hydrogen storage. Their objective is to maximize
Digital Object Identifier 10.1109/JSYST.2011.2163016 the expected profit from power exchange in a day-ahead
1932-8184/$26.00 
c 2011 IEEE
MINCIARDI AND SACILE: OPTIMAL CONTROL IN A COOPERATIVE NETWORK OF SMART POWER GRIDS 127

market. The generation scheduling is based on forecasts of This paper focuses on the problem to evaluate quantitatively
electricity price, loads, and wind generation. During online the advantages of a cooperative network of power grids,
operation, a receding horizon strategy is applied to determine fully exchanging real-time information on energy demand
the set points for the electrolyzer and the fuel cell power. The and production. This problem is defined hereinafter as co-
market model is defined both for isolated and grid-connected operative network of smart power grids (CNSPG). CNSPG
systems. is specifically interesting for a network of microgrids where
In regards to storage devices, there are various possibilities direct connections might be used to exchange power and
for the storage of the produced electrical energy, such as to decrease the variability of the load of the power ex-
water pumping reversible hydro plants, batteries, compressed changed with the connection to the main grid. Section II
air storage, and hydrogen storage [6]. This possibility of describes the proposed model and shows how to achieve
energy storage can significantly reduce the drawbacks related an optimal control of the CNSPG, while Section III shows
to the fluctuating behavior of wind and solar energy resources an example with two microgrids and a related evaluation
and can provide cost-effective means to satisfy peak energy on the advantages of exchanging power by a direct connec-
demand. Maity and Rao [10] described a framework for an tion.
electrical power microgrid, modeled as a demand and supply
problem in a multiagent system. The resulting microgrid
system model is able to determine the optimum operation II. Optimal Control of the Energy Storage in a
of a solar-powered microgrid with respect to load demand, Cooperative Network of Smart Power Grids
environmental requirements, photovoltaic panel, and battery The simplified energy storage model proposed hereinafter
capacity. Among the energy storage devices, hydrogen based is based on the assumption that, in the power grid, there
systems have been the subject of several studies. Christopher is always the technology to store energy locally adopting a
et al. [11] presented a method for the evaluation of a wind– device with a given efficiency. The grid has the possibility
hydrogen energy system. The method includes simulations and either to give out or to acquire power from one or more
economic computations to define the size of the plants. Bernal- connections to other grids. The local power production of
Agustin and Dufo-Lopez [12] proposed a complete technical– the grid is RES based, specifically coming from wind and
economic analysis of the hourly energy management in wind– solar sources, whose power production is supposed to be
hydrogen systems. In particular, the authors proposed a method fully exploited. It is also supposed that the user demand
to adjust the generation curve to the demand curve. Their has to be fully satisfied. The main decisions are whether to
method consists of the generation of hydrogen and storing it in store exceeding instantaneous power production or to send
a hydrogen tank during off-peak (low demand) hours, while it to some of the grid connections. Alternatively, in case
during the rest of the hours (peak hours, high demand) the of energy shortage, to decide whether it is convenient to
stored hydrogen can be used to generate electricity. In Korpas acquire energy from some other grids or to utilize (if any) the
et al. [13], an operation strategy for a general energy storage energy stored in the local energy reservoir. These decisional
device connected to a wind farm is presented. A dynamic aspects, specifically under a collaborative framework which is
algorithm is applied for daily scheduling in a power market. of interest for example in a regional network of microgrids,
The objective of the online operation strategy is to follow lead to the formalization of an optimal control problem
a given generation schedule as closely as possible. A recent described in this section. In this framework, the following
comprehensive review to control local energy storage devices terminology will be adopted. Each grid will be referred to
in microgrids can be found in [14]. as a “smart power microgrid” (or simply as a “microgrid”),
In regards to cooperation with other grids, the current the connection to the local energy provider as a connection to
research trend is to provide agent based frameworks where the “main grid,” and the overall set of microgrids connected
decisions can be taken in a distributed way, in which micro- among them and to the main grid will be defined as the
grids can be inserted in a sort of “plug & play” mechanism. “network.”
The work by Hommelberg et al. [15] follows this research Specifically, each smart power microgrid is supposed to
approach. Mohod and Aware [16] described the problems due be connected to a regional network of similar grids, and, at
to the injection of several wind power systems into an electric least for one microgrid, to one main grid. This network is
grid and they presented an approach to mitigate these effects. modeled as a directed graph G = (V, E), where V is the set of
More recently, Ender et al. [17] developed a design tool vertex with cardinality S, representing either microgrids or the
that enables tradeoffs between various energy systems. Their main grid, and E is the set of directed links with cardinality
tool is based on a neural network surrogate model of a W, representing the power connections existing among the
publicly available power system modeling tool. It is important vertexes. As a convention, the Sth node is associated to the
to underline that, in this respect, power systems will be main grid.
required to interact with other quite different energy systems, The evolution over time of the energy stored in the network
such as transportation systems. In this respect, Aber and of microgrids is supposed to be described by the following
Venayagamoorthy [18] focused on RES integration and grid- discrete time state equation:
able vehicles to maximize emissions reduction: among the
three proposed models, the results obtained by the smart grid xt+1 = Axt + (int − outt + But )t, t = 0, ..., T − 1
(1)
model show the highest potential for sustainability. x0 , given
128 IEEE SYSTEMS JOURNAL, VOL. 6, NO. 1, MARCH 2012

which can be rewritten introducing the following change of Under a simplifying hypothesis, let et be a vector whose
state variable: elements are represented by a Gaussian white noise, where
zt = xt − ẑ (2) E{et } = ηt = indt − outtd . (7)
as Thus, {ηt , t = 0, 1, ..., T }, with ηt ∈ RS−1 is a known
sequence of values over the interval [0, T ], while {ωt , t =
zt+1 = Azt + (A − I)ẑ + (int − outt + But )t, 0, 1, ..., T }, with ωt ∈ RS−1 being ωt defined as
t = 0, ..., T − 1 (3)
z0 = x0 − ẑ ωt = et − ηt (8)

where the following holds true. is a sequence of independent identically distributed zero-mean
Gaussian random vectors. It is also assumed that the sequences
1) xt ∈ RS−1 (kWh) is the state vector, whose ith entry is
{ωt } and {ηt } are independent.
the energy stored at microgrid i at instant t.
Under the assumption quoted above, and to simplify the
2) zt ∈ RS−1 (kWh) is the state vector, whose ith element is
notation assuming t = 1h, the following is the state equation
the energy stored at microgrid i at time instant t, with
of the energy storage in the microgrid network:
respect to an optimal working level ẑi . It is supposed
that each element of zt may assume positive as well as zt+1 = Azt + But + μt + ωt t = 0, ..., T − 1
negative values. z0 given (9)
3) ut ∈ RW (kW) is the vector of power flows (that are the μt = ηt + (A − I)ẑ
decision variables) in time interval (t, t + 1). Specifically, where:
the jth element of vector μt is the power flow in the 1) μt ∈ RS−1 is a known sequence of values over the
directed link j in time interval (t, t + 1). Such a flow interval [0, T ];
may be either positive (i.e., in accordance with the 2) ωt ∈ RS−1 is a vector whose ith element is the stochastic
conventional direction of that link) or negative. error affecting the prediction of the energy balance ηt of
4) int ∈ RS−1 (kW) is the vector of stochastic processes microgrid i in time interval (t, t + 1), modeled as above
corresponding to the power flow in input to each mi- specified.
crogrid, in time interval (t, t + 1), as given by the RES
It is supposed that there is a perfect knowledge of the
exploited in each microgrid.
state of each local storage. In addition, a cooperative strategy
5) outt ∈ RS−1 (kW) is the vector of the stochastic power
among the grids is followed, whose aim is to keep the level of
demand processes of each microgrid in time interval
energy in each local storage system as close as possible to the
(t, t + 1).
optimal working level, as well as to reduce, as far as possible,
6) A is a (S − 1) × (S − 1) diagonal matrix whose generic
the power flow among the grids. For these reasons, it seems
diagonal element αii is the efficiency of the energy
reasonable to choose the following objective function for the
storage technology in the ith microgrid (0 ≤ αii ≤ 1).
definition of the optimal control problem:
7) B is the (S − 1) × W incidence matrix, representing the  T −1 
network topology, where bi,j = −1 if link j exits the ith 
microgrid, bi,j = 1 if link j enters the ith microgrid, min J(z, u) = E c(zt , ut ) + zT MT zT (10)
t=0
and 0 otherwise. The direction of the link is purely
conventional, as the energy in each link is supposed to be c(zt , ut ) = zt Mzt + ut Nut (11)
allowed to flow in both directions. It is also worthwhile
where the following holds true.
underlining that real world power grids are sparsely
connected, and B will generally be associated to small- 1) M is a (S − 1) × (S − 1) diagonal matrix, related to the
world networks [19]. cost of an exceeding/lacking quantity of energy stored
8) t is the time discretization interval. in each energy storage device. This matrix is supposed
to be positive definite, i.e., M > 0 and constant for each
Under the hypothesis that int and outt can be forecasted on instant t = T .
a given interval [0, T ], with a certain degree of uncertainty, 2) MT , with MT > 0, has the same definition of M, but it
both of them can be split into their deterministic (respectively is only referred to instant t = T .
indt and outtd ) and stochastic (respectively win out
t and wt ) vector 3) N is a W × W matrix, N > 0, related to the cost of the
components giving rise to power sent on each edge of the network, whose elements
int = indt + win (4) are constant for each time interval t.
t
The CNSPG problem is so completely expressed in (10)
and (11), subject to the state equation (9) with known initial
outt = outtd + wout
t . (5)
state z0 . The problem is a linear quadratic Gaussian (LQG)
Let et be the vector of energy balance in each microgrid problem, that is “non-standard,” due to the presence of the
given by known input sequence {μt } in the state equation. However,
the optimal control strategy can be found in closed form as
et = int − outt . (6) demonstrated below.
MINCIARDI AND SACILE: OPTIMAL CONTROL IN A COOPERATIVE NETWORK OF SMART POWER GRIDS 129

Result: For the discrete time LQG problem with non-zero From (18), it follows that the LQG problem with non-
mean input noise, defined by (9), (10) and (11), the optimal zero input noise has a unique optimal solution given
control strategy is given by by
 
u∗t = Kt zt − zd2
t
g
+ Kt gt+1 t = 0, ..., T − 1 (12) z∗ = z∗,s + z∗,d (19)

where the following holds true. u∗ = u∗,s + u∗,d (20)

1) Kt is a W × (S − 1) matrix given by ∗,d ∗,d ∗,s ∗,s
being, of course, (z , u ) and (z , u ) the optimal so-
 −1 
Kt = −(N + B Pt+1 B) (B Pt+1 A) (13) lutions of the problem corresponding to the optimization of
the deterministic and stochastic costs, respectively. In fact, for
being Pt+1 a (S−1)×(S−1) matrix given by the discrete every admissible solution (z, u), the following relationship can
time Riccati equation be written:
     
Pt = M + A Pt+1 (I + BN −1 B Pt+1 )−1 A t = T − 1, ..., 0 J(z, u) = J zs , us + J zd , ud ≥ J z∗,s , u∗,s
   
P T = MT . +J z∗,d , u∗,d = J z∗ , u∗ . (21)
(14)
g The original problem can be so decomposed into a LQG
2) Kt is a W × (S − 1) matrix given by
regulation and a deterministic discrete linear quadratic (LQ)
Kt = (N + B Pt+1 B)−1 B .
g
(15) tracking problem. For the LQG regulation problem, the opti-
mal control is [21]
3) The vector zd2t is given by
u∗,s s
t = kt zt (22)
zt+1 = Azt + μt t = 0, ..., T − 1
d2 d2

zd2
0 = z0. (16) K = −(N + B P B)−1 (B P A) t = 0, ..., T − 1) (23)
t t+1 t+1

4) The vector gt is given by where Pt+1 is given by

gt = (A − Pt+1 (I + BN −1 B Pt+1 )−1 BN −1 B )gt+1 − Mzd2

t Pt = M+A Pt+1 (I+BN −1 B Pt+1 )−1 At = T −1, ..., 0 PT = MT .
gT = MT zd2
T . (24)
(17) The LQ tracking problem can be introduced as follows.
Proof: The proof follows the theorem proof relevant to Observe that zdt may be thought as composed by the two
an analogous problem in the continuous domain (see [20, components zd1 d2 d d1 d2
t and zt , such that zt = zt + zt , and
problem 1]). Due to its linearity, the system state equation
zd1 d1 d
t+1 = Azt + But t = 0, ..., T − 1
(9) can be written separating the stochastic and deterministic
components
zd1
0 =0 (9iv )

zst+1 +zdt+1 = A(zst +zdt )+B(ust +udt )+μt +ωt t = 0, ..., T −1. (9i ) zd2 d2
t = 0, ..., T − 1
t+1 = Azt + μt
System (9) can be so decomposed into a stochastic subsys-
tem zd2
0 = z0 . (9v )
Obviously, the vector zd2
t may be exactly determined for
zst+1 = Azst + Bust + ωt t = 0, ..., T − 1 zs0 = 0 (9ii ) each instant t. This means that the deterministic component
of the cost function can only be influenced by the values of
and into a deterministic subsystem
t . Then, putting zt = −rt , the optimization of J z , u
zd1 d2 d d
is
equivalent to the optimization of
zdt+1 = Azdt + Budt + μt t = 0, ..., T − 1 zd0 = z0 . (9iii )
  T−1     

Similarly, the cost function can be decomposed into its J zd1 , ud = −rt + zd1
t M −rt + zd1
t + udt Mudt
stochastic and deterministic components represented by the
t=0
   
+ −rt + zd1T MT −rT + zd1
T
functions J(zs , us ) and J(zd , ud ), defined in a straight- (25)
forward way. This means that for each admissible solu- which actually gives rise to a deterministic LQ tracking
tion (z, u), and for the corresponding stochastic (zs , us ) problem, whose solution is [22]
and deterministic (zd , ud ) components, the following identity
u∗,d
g
holds: t = Kt zd1
t + Kt gt+1 t = 0, ..., T − 1 (26)

J(z, u) = J(zs , us ) + J(zd , ud ). (18) where the feedback gain Kt is defined in (19), and the feed
g
forward gain Kt is given by
This immediately follows from the assumptions on the  −1 
Kt = N + B Pt+1 B
g
stochastic process ωt , that imply E{zst } = E{ust } = 0 ∀t. B t = 0, ..., T − 1 (27)
130 IEEE SYSTEMS JOURNAL, VOL. 6, NO. 1, MARCH 2012

TABLE I
Wi , φi , Di Parameters Defined for Each Microgrid
According to (28)

Microgrid Wi φi Di
1 5 0 3
2 5 1 −2
3 4 2 −1
4 3 3 3
5 6 4 −5
6 2 5 1
7 5 6 1
Fig. 1. Logical structure of the network of independent microgrids: circles 8 6 7 −1
1–9 are representative of the microgrids, while circle 10 represents the main 9 5 8 1
grid. The arrows represent the independent power connections with the same
main grid.
A. Optimal Management with no Direct Connection Among
the Microgrids (Independent Microgrids)
and the vector gt is given by the following recursion: Here the nine microgrids (i.e., from 1 to 9) are supposed
 −1 to be just connected to the main grid (i.e., the tenth grid),
gt = A − APt+1 I + BN −1 B Pt+1 BN −1 B )gt+1
+Mrt t = T − 1, ..., 0 (17 ) while no connection is present among them (Fig. 1). In this
gT = MT rT . case, nine separate and independent simple control problems
are dealt with, each one with one state variable and one control
So, as the end of the proof, it has been demonstrated that variable, representing the power flow from the main grid. The
optimal strategy is also given by (12), taking into account that
u∗t = u∗,s ∗,d g
t  + ut = Kt zst + Kt zd1
t + Kt gt+1
g (12 ) in this case all variables and parameters are scalars.
= kt zt − zt + kt gt+1
d2
t = 0, ..., T − 1.
It is supposed that the storage devices in the nine microgrids
have the same efficiency, that is A = 85%, while B = 1.
The evaluation is performed on a period of 168 h (7 days).
The predictions of the local energy balance are available. For
III. Simple Example on a Lattice Network this fictional case study, the prediction ηit for each microgrid
i = 1, ..., 9, t = 0, ..., T − 1, is represented by a square wave
A simple fictional example follows illustrating one possible
function defined as follows:
application of the control scheme introduced in the previous π
section. In a region, nine microgrids can produce energy for ηit = Wi sgn(sin (t + φi ) + Di (28)
their user community but their current energy balance between 12
demand and production is often subject to intermittences where sgn is the sign function and Wi , φi , Di are parameters
mainly due to their wind/solar RESs as well as to a user defined as reported in Table I.
demand significantly varying according to the hours of the The true energy balance for each microgrid is supposed to
day. To mitigate this effect, all of them have a local energy result from the sum of the predicted energy balance and an
storage possibility, whose size is adequate to the microgrid additional Gaussian white noise ωti , which for the case study
needs so that it can be taken into account as an energy is ωti ∼ N(0, 1) for all the microgrids. The same realization of
storage of infinite capacity. The microgrids are also already such a stochastic stream has been used in all trials described
connected to the same main grid, with which they exchange hereinafter.
power. The decision maker (who has the responsibility of The scalar parameters in the cost function are set to the
managing all the microgrids) needs to evaluate how far the following values: M = MT = 1, and N = 1 in trial 1 and
introduction of the possibility of directly exchanging power N = 100 in trial 2. The reason for considering the two trials is
among the microgrids according to a lattice network can help to show results where the cost of the state is comparable to the
to improve the overall performance of the system (i.e., of the cost of the control (trial 1), and when the latter cost is higher
nine microgrids), for example by reducing the overall power (trial 2). For all the microgrids the initial state value is set
load on the main grid as well as limiting the variations of the to z0 = 0. The main statistical characteristics of the resulting
energy stocked in the local storages. optimal values u∗ and z∗ obtained in trials 1 and 2 on the
Two case studies are compared, the first one taking into whole set of microgrids are shown in Tables II and III.
account each microgrid as independent, and the second
one taking into account the microgrid connected among B. Optimal Management with Direct Connections Among the
them by a lattice network. In the latter case, it is supposed Microgrids According to a Lattice Network Model (Coopera-
that a unique decision maker (DM) can decide the optimal tive Microgrids)
strategy to control the storage level and the power flows in Here the network of nine microgrids is connected according
all microgrids, following a cooperative approach as described to a lattice graph, as shown in Fig. 2, where each vertex,
by the CNSPG problem. numbered (highlighted), represents a microgrid. With respect
MINCIARDI AND SACILE: OPTIMAL CONTROL IN A COOPERATIVE NETWORK OF SMART POWER GRIDS 131

TABLE II
Main Statistical Characteristics (Maximum, Minimum,
Average, and Standard Deviation) of the Resulting Optimal
Variables for Case A, Trial 1, M = MT = 1, N = 1

Microgrid Variable max min avg sd

u∗,1 3.1 −9.1 −2.9 4.5
u∗,2 7.8 −3.8 2.0 4.3
u∗,3 6.7 −4.2 1.0 3.6
u∗,4 1.1 −6.8 −2.9 2.7
u∗,5 11.7 −2.2 4.9 5.2
u∗,6 1.8 −4.3 −0.9 1.8
u∗,7 5.1 −6.7 −0.9 4.4
u∗,8 8.4 −6.1 1.1 5.2 Fig. 2. Lattice network model.
U ∗,9 5.1 −6.6 −1.0 4.3
z∗,1 5.7 −6.2 0.5 2.0
z∗,2 4.9 −6.3 −0.3 2.0 as a link numbered as 13. In this case, the 9 × 13 incidence
z∗,3 4.3 −5.8 −0.2 1.8 matrix B is as follows:
z∗,4 4.0 −4.7 0.4 1.4
B ⎡= ⎤
z∗,5 6.8 −7.2 −0.8 2.2
−1 0 −1 0 0 0 0 0 0 0 0 0 0
z∗,6 5.0 −3.2 0.0 1.2 −1 −1
z∗,7
⎢ 10 0 0 0 0 0 0 0 0 0 0
⎥
4.8 −5.7 −0.2 1.9 ⎢ 1 0 0 −1 0 0 0 0 0 0 0 0⎥
z∗,8 −6.2 −0.2
⎢ 0 −1 −1 0⎥
7.5 2.1 ⎢ 0 0 1 0 0 0 0 0 0 0
⎥
z∗,9 −5.4 ⎢ 0 0 0 0 1 −1 0 −1 0 0 0 0⎥
7.8 0.2 1.9 ⎢ 0 −1 0⎥
⎢ 0 0 0 0 0 0 1 0 0 0 0
⎥
⎣ 0 0 0 0 0 0 1 0 0 −1 0 0⎦
TABLE III 0 0 0 0 0 0 0 0 1 0 1 −1 0
Main Statistical Characteristics (Maximum, Minimum, 0 0 0 0 0 0 0 0 0 1 0 1 −1

Average, and Standard Deviation) of the Resulting Optimal (29)

Variables for Case A, Trial 2, M = MT = 1, N = 100
and the 9 × 9 matrix A is assumed to be diagonal with
ai,i = 0.85. It is assumed that the DM prefers to stress the
Microgrid Variable max min avg sd
minimization of the power load on the connection u13 , that
u∗,1 0.0 −1.7 −0.9 0.5
is he/she wishes to minimize the exchanges with the external
u∗,2 1.4 −0.2 0.6 0.5
u∗,3 0.9 −0.3 0.3 0.4
main grid. So, the 13 × 13 N matrix is a diagonal matrix
u∗,4 −0.1 −1.4 −0.9 0.3 whose elements ni,i = 1 ∀i = 13, and n13,13 = 100. Finally, it
u∗,5 2.4 0.3 1.4 0.6 is assumed M = MT , and to be 9 × 9 diagonal matrixes with
u∗,6 0.1 −0.7 −0.3 0.2 mi,i = 1.
u∗,7 0.4 −1.1 −0.3 0.5 The evaluation of the quality of the results obtained in
u∗,8 1.2 −0.6 0.3 0.6 connection with case study A and case study B cannot
u∗,9 0.48 −1.1 −0.3 0.5 be assessed according to the evaluation of the values of
z∗,1 37.6 −11.8 13.5 14.7 the respective objective functions since the configuration of
z∗,2 15.9 −33.2 −9.1 14.7 the regional network of microgrids is different in the two case
z∗,3 14.3 −25.9 −5.4 11.6 studies. On the other hand, some indications can be given
z∗,4 29.1 −2.9 13.1 9.2 to the DM according to the trends and the related statistical
z∗,5 7.5 −53.1 −22.1 17.1
features of the control and state variables in the different case
z∗,6 17.2 −6.7 4.6 6.0
studies. In addition, the trends of the overall power exchanges
z∗,7 29.7 −20.3 4.3 14.2
z∗,8 24.7 −33.9 −4.5 17.4
with the main grid may also represent an important factor for
z∗,9 29.9 −20.8 4.2 14.0 the DM, whose main goal was to reduce the variations of
power loads on the main grid due to the intermittency of the
RES based microgrids.
to case A, the state variables have the same physical meaning, The main statistical characteristics of the resulting optimal
while the control variables have different meanings due to the values u∗ and z∗ obtained are shown in Table IV. The lower
different network structure of case B. values of the maximum and minimum of u∗,13 with respect
The nine microgrids have the same storage efficiency of to the other control variables are due to the characteristics
case A, that is 85%. The 12 links, numbered in italics and of the N matrix which has been adopted. Fig. 3 shows the
conventionally directed (for positive power flow values) from pattern of the power exchanges with the external main grid,
left to right and from top to bottom, represent the power lines as resulting for case A (both trials 1 and 2) from the overall
connecting the microgrids among them. It is supposed that the algebraic sum of the power exchanges on the links of each
microgrid 9 is connected to the main grid, also modeled as a microgrid, and for case B from the power flow on link
vertex numbered as 10, by an additional power line modeled 13. It is not the aim of this example to assess whether it
132 IEEE SYSTEMS JOURNAL, VOL. 6, NO. 1, MARCH 2012

TABLE IV IV. Conclusion

Main Statistical Characteristics (Maximum, Minimum,
The proposed CNSPG problem is an original formulation
Average, and Standard Deviation) of the Resulting
which may be viewed as a preliminary attempt to model and
Optimal Variables for Case B
control the exchange of power in a network of microgrids and
grids considered as a system of systems [24], [25]. Its main
Microgrid Variable max min avg sd
originality is the use and the exchange of information and
u∗,1 6.9 −1.4 2.7 2.3
u∗,2 5.3 −4.6 0.4 2.8
forecast of energy production and consumption on the whole
u∗,3 3.4 −3.1 0.2 2.2 set of microgrids, to improve the overall quality of the power
u∗,4 3.9 −3.3 0.3 2.4 management and energy stocking.
u∗,5 4.9 −6.6 −0.6 3.3 The main limitation of the approach is the hypothesis to
u∗,6 5.7 0.2 2.9 1.5 know in real time the status of all storage devices in a central-
u∗,7 4.2 −5.2 −0.6 2.7 ized way. In the future, it is reasonable to assume that power
u∗,8 4.6 −3.9 0.4 2.7 networks will become more similar to an Internet network
u∗,9 5.4 −8.5 −1.1 4.8 where grids may be on-line or not, and will be controlled by
u∗,10 7.1 −7.1 −0.2 4.4 decentralized autonomous devices. In addition, even with the
u∗,11 6.5 −3.8 1.3 2.8 presence of a centralized controller, the communication system
u∗,12 5.5 −6.9 −0.9 3.9 can operate at a much slower speed: routine status information
u∗,13 0.2 −0.2 0.0 0.1
to the central controller will most likely be updated only once
z∗,1 6.0 −6.0 0.6 3.3
every several minutes [23]. Another limitation of the current
z∗,2 5.7 −6.9 0.1 2.9
z∗,3 4.9 −5.9 0.0 2.3
work is the fact that no lower and upper bounds are introduced
z∗,4 5.8 −4.4 0.6 2.6 for the control and state variables, as it may be required by
z∗,5 5.8 −5.1 0.1 2.5 the infrastructural elements. The adequacy of their values, as
z∗,6 4.6 −4.0 0.1 1.6 defined by the control law, could be obtained by an appropriate
z∗,7 6.8 −5.7 0.3 3.3 tuning of the parameters in the cost function.
z∗,8 6.9 −7.8 0.1 3.8 Thus, new research efforts should be devoted to model
z∗,9 10.2 −9.8 0.3 4.0 the information flow in the network, and the related effects
of delays in data acquisition, both in a centralized and in a
decentralized decision making approach.

Acknowledgment
The authors would like to thank Dr. H. Dagdougui and
Dr. A. Ouammi for their precious comments on the work.

References
[1] J. McDonald, “Adaptive intelligent power systems: Active distribution
networks,” Energy Policy, vol. 36, no. 12, pp. 4346–4351, Dec. 2008.
[2] P. M. Costa and M. A. Matos, “Assessing the contribution of microgrids
to the reliability of distribution networks,” Electr. Power Syst. Res., vol.
79, no. 2, pp. 382–389, Feb. 2009.
[3] A. Borghetti, M. Bosetti, S. Grillo, S. Massucco, C. A. Nucci, M.
Paolone, and F. Silvestro, “Short-term scheduling and control of active
distribution systems with high penetration of renewable resources,” IEEE
Syst. J., vol. 4, no. 3, pp. 313–322, Sep. 2010.
Fig. 3. Trend of the overall power exchanges with the main grid. In the
[4] European SmartGrids Technology Platform. Vision and Strategy for Eu-
case B, this is represented by the power flow on link 13, while in the cases A
rope’s Electricity Networks of the Future, Office for Official Publications
trial 1 and 2, this is represented by the overall sum of the power exchanges
of the European Communities, Luxembourg, 2006.
of all microgrids with the main grid. The x-axis is related to the temporal
[5] D. Coll-Mayor, M. Paget, and E. Lightner, “Future intelligent power
instants (in hours) and the y-axis to the power flows (expressed in kW).
grids: Analysis of the vision in the European union and the United
States,” Energy Policy, vol. 35, no. 4, pp. 2453–2465, Apr. 2007.
[6] K. C. Divya and J. Østergaard, “Battery energy storage technology for
is worthwhile or not connecting the microgrids in a lattice power systems: An overview,” Electr. Power Syst. Res., vol. 79, no. 4,
pp. 511–520, Apr. 2009.
network, since this fictional case has the aim to provide [7] J. S. Anagnostopoulos and D. E. Papantonis, “Pumping station design for
an example of the methodology described in the previous a pumped-storage wind-hydro power plant,” Energy Convers. Manage.,
session which may be subject of additional investigations, vol. 48, pp. 3009–3017, Aug. 2007.
[8] H. Dagdougui, R. Minciardi, A. Ouammi, M. Robba, and R. Sacile, “A
for example by a proper tuning of the N, M, and MT pa- dynamic decision model for the real time control of hybrid renewable
rameters. Besides, another issue to be taken into account is energy production systems,” IEEE Syst. J., vol. 4, no. 3, pp. 323–333,
the one related to the evolution of the cost for the setting Sep. 2010.
[9] M. Korpaas and A. T. Holen, “Operation planning of hydrogen storage
and the management of the physical network for the power connected to wind power operating in a power market,” IEEE Trans.
exchange. Energy Convers., vol. 21, no. 3, pp. 742–749, Sep. 2006.
MINCIARDI AND SACILE: OPTIMAL CONTROL IN A COOPERATIVE NETWORK OF SMART POWER GRIDS 133

[10] I. Maity and S. Rao, “Simulation and pricing mechanism analysis of [24] L. R. Phillips, “The microgrid as a system of systems,” in Systems
a solar-powered electrical microgrid,” IEEE Syst. J., vol. 4, no. 3, pp. of Systems Engineering—Principles and Applications (SoS Engineering
275–284, Sep. 2010. Series, vol. 1), M. Jamshidi, Ed. London, U.K.: CRC/Taylor and Francis,
[11] J. G. Christopher, M. Korpas, and A. T. Holen, “A Norwegian case 2008, ch. 10, pp. 251–280.
study on the production of hydrogen from wind power,” Int. J. Hydrogen [25] P. Korba and I. A. Hiskens, “Operation and control of electrical power
Energy, vol. 32, pp. 1500–1507, Nov. 2007. systems,” in System of Systems Engineering—Innovations for the 21st
[12] J. L. Bernal-Agustın and R. Dufo-Lopez, “Hourly energy management Century, M. Jamshidi, Ed. New York: Wiley, 2009, ch. 16, pp. 385–408.
for grid-connected wind–hydrogen systems,” Int. J. Hydrogen Energy,
vol. 33, pp. 6401–6413, Sep. 2008.
[13] M. Korpaas, R. Hildrum, and A. T. Holen, “Operation and sizing
Riccardo Minciardi (M’82) received the Laurea
of energy storage for wind power plants in a market system,” Int.
degree in electronic engineering from the University
J. Electr. Power Energy Syst., vol. 25, no. 8, pp. 599–606, Oct.
of Genova, Genova, Italy, in 1975.
2003.
He is currently a Professor of System Modeling,
[14] R. Zamora and A. K. Srivastava, “Controls for microgrids with storage:
Simulation, and Control, and the Director of the De-
Review, challenges, and research needs,” Renewable Sustainable Energy
partment of Communication, Computer and System
Rev., vol. 14, pp. 2009–2018, Mar. 2010.
Sciences, University of Genoa. His current research
[15] M. P. F. Hommelberg, C. J. Warmer, I. G. Kamphuis, J. K. Kok, and G.
interests include modeling and control of automated
J. Schaeffer, “Distributed control concepts using multiagent technology
manufacturing systems, transport planning in urban
and automatic markets: An indispensable feature of smart power grids,”
systems, risk management, renewable energy, and
in Proc. IEEE Power Eng. Soc. General Meeting, Jun. 2007, pp. 1–7.
environmental decision support systems.
[16] S. W. Mohod and M. V. Aware, “A STATCOM-control scheme for grid
connected wind energy system for power quality improvement,” IEEE
Syst. J., vol. 4, no. 3, pp. 346–352, Sep. 2010.
[17] T. Ender, J. Murphy, and C. Haynes, “A framework for portfolio Roberto Sacile (M’11) received the Laurea degree
management of renewable hybrid energy sources,” IEEE Syst. J., vol. 4, in electronic engineering from the University of
no. 3, pp. 295–302, Sep. 2010. Genova, Genova, Italy, in 1990, and the Ph.D. degree
[18] A. Y. Saber and G. K. Venayagamoorthy, “Efficient utilization of from Milan Polytechnic, Milan, Italy, in 1994.
renewable energy sources by gridable vehicles in cyber-physical energy He joined the French National Institute for Re-
systems,” IEEE Syst. J., vol. 4, no. 3, pp. 285–294, Sep. 2010. search in Computer Science and Control, Sophia
[19] Z. Wang, A. Scaglione, and R. J. Thomas, “Generating statistically Antipolis, France, in 1995, as a Post-Doctorate. He
correct random topologies for testing smart grid communication and is currently with the Department of Communication,
control networks,” IEEE Trans. Smart Grid, vol. 1, no. 1, pp. 28–39, Computer and System Sciences, University of Gen-
Jun. 2010. ova, where, since 2000, he has been teaching the
[20] C. Bruni and D. Iacoviello, “Some results about the optimal LQG course of “models and methods for the management
tracking problem,” Int. J. Control, vol. 74, no. 10, pp. 977–987, 2001. of environmental systems.” His current research interests include decision
[21] P. Whittle, Risk-Sensitive Optimal Control. New York: Wiley Chichester, support methodologies based on optimal control techniques and on system of
1990. systems modeling, with specific applications to the environmental, transport,
[22] F. L. Lewis, Optimal Control. New York: Wiley, 1986. and energy systems.
[23] J. D. Kueck and B. J. Kirby, “The distribution system of the future,” Dr. Sacile has been an Associate Editor of the IEEE Systems Journal
Electricity J., vol. 16, no. 5, pp. 78–87, Jun. 2003. since 2011.