Você está na página 1de 6

Proceedings of the FrB09.

4
46th IEEE Conference on Decision and Control
New Orleans, LA, USA, Dec. 12-14, 2007

Hybrid Model Predictive Control of a Two–Generator Power Plant


Integrating Photovoltaic Panels and a Fuel Cell
Alejandro J. del Real, Alicia Arce, Carlos Bordons

Abstract— This paper describes the application of hybrid laboratory trial system. The overall control scheme is com-
modeling control techniques to a two–generator power system posed of three parts: a reference governor that estimates the
connected to the grid. The plant consists of a solar field and best power production based on measurements and economic
a secondary power source formed by an electrolyzer, hydrogen
tank and fuel cell stack. The system is inherently hybrid as objectives; a second level, in which this paper is focussed,
it combines both continuous and hybrid dynamics, since it that estimates the best way to achieve the desired power
can operate in four distinct modes, depending on the power production, calculating thus the references that each system
circuit configuration and the fuel cell stack state. Firstly, a component must follow; and the lower level part, which
Mixed–Logical–Dynamical (MLD) description of the system is composed by each device local controller. The presence
is obtained. A hybrid receding horizon finite–time optimal
controller based on on–line multiparametric programming of switching modes and constraints make the problem in-
techniques is then tuned. Finally, the effectiveness of such a herently hybrid, and the main contribution of this paper
control design is shown through the simulation results. resides in the application of a recently proposed framework
for modeling hybrid systems [5] and synthesizing optimal
I. INTRODUCTION controllers [6],[7]. Also, the Hybrid Systems Description
Language (HYSDEL) [8] was utilized to obtain an open–loop
Nowadays, renewable energy generation represents an im- system model suitable for control design from a simplified
portant part of the whole energy scenario in many countries. and linearized model of the plant.
One of the most important problems associated with this Section II describes the physical system and section III
kind of systems is the reliability and quality of power explains the overall control problem. Section IV presents
supply. Since the primary power source (wind, solar, etc) the non–linear model of the plant. The linearized system for
cannot be manipulated, unpredictable fluctuations may ap- hybrid modeling and the optimal control design are described
pear in power output. Also, electrical production is not in sections V and VI, respectively. Some simulated results
subject to the demand, which usually results in an unbal- are presented in section VII. Finally, the major conclusions
anced system. One way to overcome this problem is by are drawn in section VIII.
the inclusion of intermediate storage, which has been used
both in grid–connected and isolated cases [1],[2]. In recent II. SYSTEM DESCRIPTION
years, hydrogen–based technologies have been improved, so Fig. 1 shows the block diagram of the system. The primary
hydrogen production, storage and profiteering to produce energy source of the plant is the photovoltaic field (SP) which
electricity are nowadays economically beneficial [3]. Also, is connected to the grid through a switch (s) that opens and
the use of hydrogen as an intermediate storage has some closes the circuit. The electricity produced by these solar
advantages with respect to commonly utilized devices such as panels (Pav ) due to the solar radiation (rad) can then be
batteries, which are prohibitively costly in large scale power sent to the grid and totally or partially be derived to the
generation systems. For these reasons, many research groups electrolyzer (E). The energy consumed by the electrolyzer
are working on the integration of renewable energy sources (Pe ) is then used to produce hydrogen (h), which is stored
with hydrogen–based systems as the secondary power supply in the deposits placed in the hydrogen line. The fuel cell
[4]. stack (FC), when necessary, can then be switched on and
This work focusses on the integration of a solar field be fed by these deposits to produce electricity (Pf c ). Then,
connected to the grid and a hydrogen–based system as the the total power sent to the grid (Pg ) will be the sum of the
intermediate power source (as described in section II), which power produced by the fuel cell stack and the power from the
allows a much more flexible and efficient energy production. solar field not derived to the electrolyzer. Notice that all the
Also, the dimensions of the devices have been selected to dynamics of the power conditioning devices, such as DC/AC
make possible a future system implementation as a small converters, has been obviated. This simplification does not
affect to the solution of the problem, as the dynamics of these
This work was supported in part by the Spanish Ministry of Science and devices are much faster than the other system dynamics, such
Technology under Grant DPI 2004-07444-c04-01 and European Commis-
sion under grant 511368 (Hycon Network of Excellence). as the fuel cell stack or the hydrogen production rates.
All the authors are with Escuela Superior de Ingenieros, Departamento
de Sistemas y Automática, University of Seville, 41092 Camino de los III. CONTROL PROBLEM
Descubrimientos, Spain.
Corresponding Author: Alejandro J. del Real. Email address: The overall control scheme is divided into three parts
alejandro.delreal@gmail.com (see Fig. 2). The higher level corresponds to the reference

1-4244-1498-9/07/$25.00 ©2007 IEEE. 5447


46th IEEE CDC, New Orleans, USA, Dec. 12-14, 2007 FrB09.4

rad
Pav s Pg corresponds to the hydrogen production and consumption
SP
rates, and the other dynamics are several orders of magnitude
Pe Pfc
lower. The nomenclature used is shown in Table I.
H2 H2
E FC
TABLE I
h
N OMENCLATURE
A surface area (m−2 )
Fig. 1. System description diagram
F Faraday constant (C mol−1 )
G radiation (W m−2 )
I current (A)
generation of the power delivered to the grid (Pg,re f ). This n number of cells
P Power (W)
calculation can be based on many variables: the system R electrical resistance (Ω)
state measurements (such as the stored hydrogen), the future T temperature (K)
electricity price, the grid capacity, the expected radiation, etc. V voltage (V)
W mass flow (kg s−1 )
Then, for example, an optimization of an economic criteria to
Subscripts
maximize the benefits can be done off–line, in order to obtain cons consumed
the most beneficial Pg . The next control level corresponds to e electrolyzer
the one this paper deals with. In this way, the controller ecell electrolyzer cell
H2 hydrogen
receives the calculated reference Pg,re f , the current radiation max maximum
measurement rad and the plant output variables h and Pg as n nominal
inputs (see Fig. 3). Then, the references used by the next oc open circuit
p parallel connection
level controllers are calculated: the discrete variables f c and prod produced
s, which correspond to the fuel cell stack (where 1 means s series connection
on, and 0 off) and switch states (where 1 corresponds to sc short circuit
scell solar cells
closed circuit and 0 to open), respectively; Pf c , the output sf solar field
power demanded to the fuel cell stack; and Pe , the power smod solar modules
derived to the electrolyzer. The lower level corresponds to T thermal effects
each system device controller, such as the power conditioning
and the on–board fuel cell stack controllers [9].
Expected radiation
The functional description and the fundamental equations
Energy price
Grid capacity governing the solar cell behaviour presented in this paper
Storaged hydrogen
...
are based on [10]. Also, in the exponential model discussed,
First level reference
generation
[ Continuous reference Pg ]
some suppositions were done as in [11], which allow the
calculation of the solar field characteristics from the data
Second level reference supplied by the manufacturer.
generation

Continuous references: Pfc, Pe The solar field polarization curve can be written as
Discrete references: s, fc

· µ ¶¸
Vs f − Voc,sf + Is f Rsf
Power Is f = Isc,s f 1 − exp (1)
Fuel cell
conditioning
Electrolizer Switch
VT,s f

Fig. 2. Overall control scheme


were Voc,sf = Voc,smod · ns,sf . The thermal voltage VT,s f is
expressed as

Pg,ref
Pfc
Pe
VT,sf = VT,scell · ns,sf · ns,smod (2)
h, Pg
Controller fc Plant
rad s
with
K · Tscell
Fig. 3. Second level controller VT,scell = (3)
e
Finally, the current Isc,s f dependency on the solar radiation
IV. NONLINEAR MODEL is calculated as
As the objective of this work is the design of a relatively
high level controller, the plant model does not have to be G
Isc,s f = Isc,smod · np,sf (4)
too detailed. In fact, simplicity should be sought, since 1000
this will improve the implementability of the controller. As Implementing the above equations drives to the solar cell
discussed above in section II, power conditioning devices polarization curves presented in Fig. 4. Also, notice there is
were omitted. Also, the models described in this section, a maximum peak power for each radiation value (see Fig. 5).
excepting hydrogen storage, will be considered statics. These Concerning the electrolyzer, the equations presented here
simplifications can be done since the dominant dynamics correspond to the ones described in [12].

5448
46th IEEE CDC, New Orleans, USA, Dec. 12-14, 2007 FrB09.4

−5
x 10
3.5 3
non−linear model
linearized model
−2
3 G = 1000 W m
2.5

hydrogen production rate (kg m )


−2
2.5
−2 2
G = 800 W m
cell current (A)
2

−2 1.5
G = 600 W m
1.5

1
1 −2
G = 400 W m

0.5
0.5
−2
G = 200 W m

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0 500 1000 1500 2000 2500 3000
cell voltage (V) electrolyzer power (W)

Fig. 4. Solar cell polarization curves for different radiation values Fig. 6. Hydrogen production vs electrolyzer power consumption

3.5
cell current (A)
cell power (W)
3
I = 3.08 A
sc,scell

2.5 Pe,min = 0 ≤ Pe ≤ 2500 = Pe,max (8)


2
Pe ≤ Pav (9)
1.5

1 [h ≥ hmax ] ⇒ [Pe = 0] (10)


0.5 V = 0.476 V
oc,scell
Concerning the fuel cell, the polarization curve presented
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 in Fig. 7 corresponds to a 1.2 kW Ballard (Nexa Power
cell voltage (V)
Module) PEM fuel cell stack. The stack is composed of 46
Fig. 5. Solar cell power characteristics
cells with a 110 cm2 membrane each one. The system is
autohumidified and air–cooled by a small fan. It is configured
in dead–end hydrogen feeding mode. Fig. 9 presents the stack
Thus, the power consumption is calculated as output power versus the hydrogen consumption rate.
Also, some constraints should be included in the model.
Thus, the output power limit is 1200 W. Furthermore, since
Vecell − Voc,ecell
Pe = ns,e · Voc,ecell · Ae (5) the fuel cell has both start up and shut down sequences,
Rs,e the control design should take into account not to change
In the same way, the electrolyzer voltage can be written the stack state continuously. So, a timer was included,
as establishing a minimum time of 3 hours since the fuel cell
µ ¶ is switched on until it can be switched off, and 120 seconds
Ie · Rs,ecell
Ve = ns,e + Voc,ecell (6) to do the reverse procedure. However, when the fuel stack is
Aecell switched on, a small hydrogen consumption is necessary to
Finally, the hydrogen production is given by produce electricity (around 35 W) in order to auto–maintain
its own electronic systems. Thus, when the hydrogen lower
pressure limit inside the deposits is reached, the fuel cell
ns,e · Ie
WH2 = 0.002 (7) stack must be switched off even without fulfilling the timer
2F constraint. All these constraints can be written as
The hydrogen production rate dependency on power con-
sumption can be seen in Fig. 6. Notice that it results in an
almost linear dependency. Pfc,min = 0 ≤ Pf c ≤ 1200 = Pfc,max (11)
Also, some functional constraints should be modeled. In
[tfc,off→on ≥ 120] ⇒ [ f c = 1] (12)
this way, the electrolyzer power consumption limit is 2500
W. Furthermore, this consumption cannot be higher than the ¯
[tfc,on→off ≥ 10800 ¯ h ≤ hmin ] ⇒ [ f c = 0] (13)
available power from the solar field (Pav ). Lastly, due to the
limited capacity of the hydrogen deposits, when the hydrogen Lastly, the hydrogen deposit was supposed to have a 125 L
upper pressure value is reached inside them, the electrolyzer capacity, and can store the gas at a maximum of 200 bar
should remain idle. Notice that there is no discrete variable (approximately 2 kg of hydrogen). A lower pressure of 20
associated to the electrolyzer state (this is, switched on or bar was imposed not to totally empty the deposit. In mass
off). This simplification can be done, since the electrolyzer terms, the storage constraints can be expressed as:
is always connected to a voltage source, in order to maintain
a minimum operational voltage needed by the electrolyzer
cells, so we can consider that its state is always on. hmin = 0.2 ≤ h ≤ 1.8 = hmax (14)

5449
46th IEEE CDC, New Orleans, USA, Dec. 12-14, 2007 FrB09.4

50 2) The model is discretized with a sampling time Ts = 5 s,


45
and the piecewise affine (PWA) model is structured
dividing the dynamics into four distinct modes, which
stack current (A)
40
are functions of the switch state s and the fuel cell stack
35
state f c, following the automata shown in Table II.
30 TABLE II
AUTOMATA
25
s fc mode
20 0 0 i1
0 10 20 30 40 50
stack voltage (V) 1 0 i2
0 1 i3
1 1 i4
Fig. 7. Fuel cell stack polarization curve

−5
x 10
2.5
non−linear model
linearized model Denoting the continuos variables with the subscript £ ¤
2 r and the£ binary ones ¤with b, £let xr (t) ¤= h ,
hydrogen consumption rate (kg s )
−1

1.5
£xb (t) = i1 i2 ¤i3 i4 , yr (t) £= h P¤g , ur (t) =
Pe Pf c rad and ub (t) = s f c . Also, the
1 constant Pfc,on = 35 W is related to the parasitic power
consumed by the fuel cell when it is switched on (as
0.5
discussed in section IV). Then, the hybrid dynamics
0
can be written as:
0 200 400 600 800 1000 1200 1400
fuel cell stack power (W)

 £ ¤
 xr (t + 1)·= ¸m2 0 · 0 ur (t) ¸
Fig. 8. Fuel cell stack output power vs hydrogen consumption if i1 (t) then 1 0 0 0
 yr (t) = 0 xr (t) + 0 0 0 ur (t)
(18)
V. HYBRID MODEL FOR CONTROL  £ ¤
Due to the presence of both continuous dynamics and  xr (t + 1)·= ¸m2 0 · 0 ur (t) ¸
binary inputs, the hybrid systems framework is adopted to if i2 (t) then 1 0 0 0
 yr (t) = 0 xr (t) + −1 0 m ur (t)
1
describe the process [13]–[16]. Although many modeling (19)
formalisms can be found to describe hybrid systems [17], we

adopt here the mixed–logical (MLD) formulation [5]. The
£ ¤
 xr (t + 1)·= ¸m2 −m·3 0 ur (t)¸− m3 · Pfc,on
purpose of the hybrid systems description language [8] is if i3 (t) then 1 0 0 0
 yr (t) = 0 xr (t) + 0 1 0 ur (t)
the procurement of MLD models from a high–level textual
(20)
description of the hybrid dynamics. The model presented in
section IV is transformed into an equivalent discrete–time  £
 xr (t + 1)·= ¸m2 −m
¤
MLD model through the following steps. ·3 0 ur (t) − m
¸ 3 · Pfc,on
if i4 (t) then 1 0 0 0
1) The nonlinear equations are linearized. Concerning  yr (t) = 0 xr (t) + −1 1 m ur (t)
1
the solar field, the model can be considered linear (21)
at the maximum power point and below (see Fig. 5).
Then, the linear equation that follows can be obtained,
3) Lastly, all the constraints discussed in section IV are
relating the solar radiation with the available power
included, being written as:
from the solar field.

hmin ≤ xr (t) ≤ hmax (22)


Pav = m1 · rad = 2.9252 rad (15) £ ¤′ £ ¤′
0 0 0 ≤ ur (t) ≤ Pe,max Pfc,max radmax
The electrolyzer, as seen on Fig. 6, has an almost
(23)
linear behaviour, so the linear relation presented below
approximates the original curve very well. ur1 (t) ≤ m1 · ur3 (t) (24)
[tfc,off→on ≥ 120] ⇒ [ub2 (t) = 1] (25)
−8
¯
WH2 ,prod = m2 · Pe = 1.0018 · 10 Pe (16) [tfc,on→off ≥ 10800 ¯ xr (t) ≤ hmin ] ⇒ [ub2 (t) = 0]
In the same fashion, the fuel cell stack is linearized as (26)
the following equation (see also Fig. 9): [xr (t) ≥ hmax ] ⇒ [ub1 (t) = 0] (27)
The HYSDEL language [8] compiler translates the above
Pf c = m3 ·WH2 ,cons = 1.5267 · 10−8 WH2 ,cons (17) hybrid dynamics and constraints to MLD form:

5450
46th IEEE CDC, New Orleans, USA, Dec. 12-14, 2007 FrB09.4

0 0 0 0 0
 
x(t + 1) = A x(t) + B1 u(t) + B2 γ (t) + B3 z(t) (28a) · ¸ 0 0 0 0 0
0 0
 
y(t) = C x(t) + D1 u(t) + D2 γ (t) + D3 z(t) Q= , R= 0 0 0 0 0
 
(28b) 0 20 
 0 0 0 0 0 
E2 γ (t) + E3 z(t) ≤ E1 u(t) + E4 x(t) + E5 (28c) 0 0 0 0 5
£ ¤′ £ ¤′ £ ¤′
where x = xr xb , y = yr yb , u = ur ub , 
10 0 0 0 0
 
100 0 0 0 0

and δ ∈ {0, 1}rb and z ∈ ℜrr represents respectively  0 0 0 0 0   0 0 0 0 0 
auxiliary logical and continuous variables. S= 0 0 0 0 0  , ST =  0 0 0 0 0
   

 0 0 0 0 0   0 0 0 0 0 
VI. HYBRID MPC CONTROLLER WITH 0 0 0 0 0 0 0 0 0 0
CONSTRAINTS (33)

Hybrid MPC formulation has been successfully applied


to many industrial applications [18]-[21]. In this approach,
at each sampling instant, a finite horizon open–loop opti- VII. SIMULATION RESULTS
mization problem is solved, assuming the current state as Two different tests are presented in this section. The
the initial condition for the problem. Based on the receding simulations have been planned to show the behaviour of
horizon philosophy, only the first element of the control the controller in a wide range of situations, especially when
sequence obtained is applied to the hybrid system. Then, the constraints are reached. The first simulation is shown in
same process is repeated at each sampling time, providing in Figs. 9 and 10. The available power from the solar field was
this way a feedback and allowing disturbance rejection and obtained from real radiation data, since the experiment covers
reference tracking. Thus, the control problem is formulated an entire day. As can be seen, Pre f was set to a constant
as value of 1000 W. So, in the first stage, when no power is
available, the fuel cell is switched on to reach the reference,
N delivering 1000 W. As a result, the stored hydrogen falls as

¡ ¢
min J (ε , x(t)) = kR (uk − ure f )k∞ + kQ (yk − yre f )k∞ it is consumed. In the middle stage, as the available power
{u,δ ,z}0N−1 k=0
N−1 rises, the fuel cell stack delivered power declines, while at
+ ∑ kS(xk − xre f )k∞ + kST (xN − xre f k∞ the same time the switch is closed, thereby allowing us to
k=1 take advantage of the solar field instead of the stack (which
(29a) clearly is more efficient). When Pav rises above 1000 W,
subject to the stack is then switched off (as it has been working for
more than 3 hours) and the electrolyzer starts to use the

x = x(t) extra power to produce hydrogen. In a later stage, notice
 0

the fluctuations of the solar radiation. To keep the delivered
xk+1 = A xk + B1 uk + B2 γk + B3 zk
(29b)
 yk = C xk + D1 uk + D2 γk + D3 zk
 power constant, the stack is then switched on, injecting the
E2 γk + E3 zk ≤ E1 uk + E4 xk + E5 needed power, and remaining in this mode even when not
delivering power. This is due to the minimum working time
where N is the control horizon, x(t) is the state of the constraint, preventing in this way an iterative state change.
MLD system at time t, ε = [u0 , γ0 , z0 , . . . , uN−1 , γN−1 , zN−1 ] The final stage is analogous to the first, when no power is
is the optimization vector and Q, R, S and ST are weight available from the solar field.
matrices. Notice that there are two fixed references, xre f and
yre f , which in conjunction with the elected weight matrices, 4000

provide a weighted–multiobjective MPC controller. In fact, 2000


Pav (W)
P
ref
(W)

the main objective is following the output reference yre f 0


0 1 2 3 4 5 6 7 8 9
4

using the primary power source (this is, the solar field) 2000
x 10

Pg (W)
whenever possible. If the constraints or the current solar 1000 P
ref
(W)

radiation make this impossible, the second power source 0


0 1 2 3 4 5 6 7 8 9
4

would then be used (which is the fuel cell stack). If there 2


x 10

storage upper limit

is still available power unused, it would be then used to


h (Kg)

1 storage lower limit

electrolyze water to produce hydrogen to store. In this way, 0


0 1 2 3 4 5 6 7 8 9
time (s) x 10
4

we let

£ ¤′ Fig. 9. Test (a) reference and model outputs


yre f = 0 Pg,re f (30)
£ ¤′
xre f = hmax 0 0 0 0 (31) The second simulation is shown in Figs. 11 and 12. Until
£ ¤′ the middle of the test, Pre f = 200 W and Pav = 3000 W. If
ure f = 0 0 0 0 1 (32)
the switch closed the circuit to connect the solar field to the
with the weight matrices grid, there would be an excess of 2800 W delivered to the

5451
46th IEEE CDC, New Orleans, USA, Dec. 12-14, 2007 FrB09.4

4000 show that the reference following is achieved quite well,

P (W)
2000 P upper limit
e

while respecting all the constraints. At the same time, the

e
0
0 1 2 3 4 5 6 7 8 9

2000
x 10
4
controller seeks the most efficient plan, which is mainly
P (W)
Pfc upper limit
1000 exploiting the solar field and producing hydrogen when there
fc
0
0 1 2 3 4 5 6 7 8
x 10
4
9
is an available power excess. As the plant is sized to make
2
1 real implementation possible, future works will also focus
fc

0
−1
0 1 2 3 4 5 6 7 8 9 on the experimental validation of the controller design.
4
x 10
2
1 R EFERENCES
s

0
−1
0 1 2 3 4 5 6 7 8 9
time (s) x 10
4 [1] H. Lund, “Large–scale integration of wind power into different energy
systems”, Energy, vol. 1, no. 11, 2004
[2] G.C. Bakos and N.F. Tsagas, “Technoeconomic assessment of a
hybrid solar/wind installation for electrical energy saving”, Energy
Fig. 10. Test (a) model inputs and Buildings, vol. 35, pp. 139–145, 2004
[3] E.I. Zoulias, R. Glokner, N. Lymberopoulosa, T. Tsoutsos, I. Vosseler,
O. Gavalda, H.J. Mydske, and P. Taylor, “Integration of hydrogen
grid. Then, the stack is switched on to supply the 200 W energy technologies in stand–alone power systems analysis of the
current potential for applications”, Renewable and Sustainable Energy
reference and the switch opens the circuit, while the power Review, vol. 1, no. 31, 2004
from the solar panels is used to produce hydrogen. But this [4] L. Ntziachristos, C. Kouridis, Z. Samaras, and K. Pattas, “A
power is limited to 2500 W, the upper limit of the electrolyzer wind–power fuel–cell hybrid system study on the non–interconnected
Aegean islands grid”, Renewable Energy, vol. 1, no. 17, 2004
capacity. As a result, 500 W are then unused. In the second [5] A. Bemporad and M. Morari, “Control of systems integrating logic,
half of the simulation, Pre f = 3600 W, while Pav remains dynamics, and constraints”, Automatica, vol. 35, no. 3, pp. 407–427,
equal to 3000 W. Then, to follow the reference, the switch Mar. 1999
[6] F. Borrelli, Constrained Optimal Control of Linear and Hybrid Sys-
connects the solar field to the grid, thus injecting 3000 W. tems. Berlin, Germany: Springer, 2003, vol 290, Lecture Notes in
The other 600 W are then supplied by the fuel cell, which COntrol and Information Sciences
it switched on. [7] F. Borrelli, M. Baotic, A. Bemporad, and M. Morari, “Dynamic
programming for constrained optimal control of discrete–time hybrid
systems”, Automatica, vol. 41, no. 1, pp. 1709–1721, Jan. 2005
4000 [8] F.D. Torrisi and A. Bemporad, “HYSDEL–A tool for generating
2000 P (W)
computational hybrid models”, IEEE Trans. Contr. Syst. Technol., vol.
av
P
ref
(W) 12, no. 2,pp. 235–249, Mar. 2004
0
0 2000 4000 6000 8000 10000 [9] C. Bordons, A. Arce, and A. del Real, “Constrained Predictive Control
4000
Strategies for PEM fuel cells”, IEEE proceedings of 2006 American
2000 P (W)
g
Control Conference, 2006
P
ref
(W) [10] M.C. Alonso Garcı́a, “Fundamentos, dimensionado y aplicaciones de
0
0 2000 4000 6000 8000 10000 la energı́a solar fotovoltaica”, Serie Ponencias, Editorial Ciemat, 2003
2
storage upper limit [11] M.A. Green, Solar cells. Operating principles, Technology and System
h (Kg)

1 Applications. Prentice Hall, New Jersey, 1982


storage lower limit
[12] A. Molina, Development of simulation tools for renewable energy
0
0 2000 4000
time (s)
6000 8000 10000 systems, Msc Thesis, University of Seville, 2003
[13] P.J.Antsaklis, “A brief introduction to the theory and applications of
hybrid systems,” in Proc. IEEE, Spec. Issue Hybrid Syst.: Theory
Fig. 11. Test (c) reference and model outputs Appl., Jul. 2000, vol. 88, no. 7, pp. 879–886
[14] J. Lygeros, C. Tomlin, and S. Sastry, “Controllers for reachability
specifications for hybrid systems,” Automatica, vol. 35, no. 3, pp.
349–370, 1999
4000
[15] K. Gokbayrak and C.G. Cassandras, “A hierarchical decomposition
P (W)

2000
method for optimal control of hybrid systems,” in Proc. 38th IEEE
e

P upper limit
e

0
0 2000 4000 6000 8000 10000 Conf. Decis. Control, Phoenix, AZ, Dec. 1999, pp. 1816–1821
2000
P upper limit [16] M.S. Branicky, “Studies in hybrid systems: Modeling, analysis, and
P (W)

fc
1000
control” Ph.D. dissertation, LIDS–TH 2304, Massachusetts Inst. Tech-
fc

0
0 2000 4000 6000 8000 10000 nol., Cambridge, MA, 1995.
2
1 [17] W.P.M.H. Heemels, B. De Schutter, and A. Bemporad, “Equivalence of
fc

0 hybrid dynamical models,” Automatica, vol. 37, no. 7, pp. 1085–1091,


−1
2
0 2000 4000 6000 8000 10000 Jul. 2001
1 [18] F. Borrelli, A. Bemporad, M. Fodor, and D. Hrovat, “An MPC/hybrid
s

0
−1
system approach to traction control,” IEEE Trans. Control Syst.
0 2000 4000 6000 8000 10000
time (s) Technol., vol. 14, no. 3, pp.541–552,2006
[19] N. Giorgetti, A. Bemporad, H.E. Tseng, and D. Hovrat, “Hybrid
model predictive control application towards optimal semi–active
Fig. 12. Test (c) model inputs suspension,” Int. J. Control, vol. 79, no. 5, pp. 521–533, 2006
[20] S. Di Cairano, A. Bemporad, I. Kolmanovsky, and D. Hovrat, “Model
predictive control of nonlinear mechatronic systems: An application
VIII. CONCLUDING REMARKS to a magnetically actuated mass spring damper,” in Proc. 2nd IFAC
Conf. Anal. Des. Hybrid Syst., Alghero, Italy, 2006, pp.241–246
In this paper, we developed a hybrid model of a [21] N. Giorgetti, G. Ripaccioli, A. Bemporad, I. Kolmanovsky, and D.
two–generator power system plant, and also designed a hy- Hovrat, “Hybrid model predictive control of direct injection stratified
charge engines”, IEEE/ASME Trans. on mechatronics, vol. 11, no. 5,
brid controller based on recent state–of–the–art developments Oct. 2006
in this field. As can be seen in the last section, the simulations

5452

Você também pode gostar