L
1
Adaptive Output Feedback Controller with Operating Constraints
for Solid Oxide Fuel Cells
Lei Pan
1
, Chengyu Cao
2
and Jiong Shen
1
1
School of Energy and Environment, Southeast University, 2 Si Pai Lou Road, Nanjing, China
2
Department of Mechanical Engineering, University of Connecticut, Storrs, U.S.A.
Keywords: L
1
Adaptive Controller, Solid Oxide Fuel Cell, Constraints, Disturbance Model Predictive Controller.
Abstract: Control on solid oxide fuel cells (SOFC) is challenging due to its nonlinearity, time-varying uncertainties,
tight operating constraints and modeling difficulties. The L
1
adaptive output feedback controller for systems
of unknown relative degree is introduced for the SOFC output voltage control in this paper. It allows for fast
and robust adaptation, and provides improved transient performance. Its advantages of not enforcing a
strictly positive real condition along with the low-pass filtered control signal bring it the potential to be
applied in wide industrial processes. In the study of the SOFC control, a dynamic SOFC model is first built;
then a L
1
adaptive output feedback controller is designed only using the nominal working conditions of the
SOFC model. Through setting the operating constraints at proper locations, the closed-loop stability is
maintained in the presence of hard constraints by the symmetric structure of the L
1
adaptive control loop. A
simulation comparison is made in the SOFC constant voltage control process between the L
1
adaptive
controller and a linear disturbance model predictive controller (DMPC) for their almost equal complexity in
designs. The result shows the advantage of the L
1
adaptive controller in disturbance rejections for its faster
transient response.
1 INTRODUCTION
Solid oxide fuel cell (SOFC) is a kind of high
efficiency, environment friendly power generation
assembly, which converts the chemical energy in
fuel and oxidant directly to electricity. Because of
the shortage of resources and increasing
environment pollutions, governments and
technicians all over the world pay more attentions on
the research and development of SOFC today. SOFC
in the application of massive distributed power
sources has been considered to be a potential
candidate to replace the traditional thermal cycle
power generation. The SOFC system has severe
output nonlinearity and tight operating constrains. It
also has time-varying uncertainties and is hard to
model. These features bring the major challenges on
the control methods of SOFC systems. Because
effective control on SOFC system can improve
operation efficiency, extend the stack lifespan, and
improve the quality of power, more and more
research has been taken on designing high-
performance controllers working with nonlinear and
uncertain dynamic characteristics of SOFC plants in
recent years.
Most of the research work is based on model
predictive control (MPC) methods (Pukrushpan et
al., 2002; Vahidi et al., 2004; Aguiar et al., 2005;
Stiller et al., 2006). The conventional MPC is a
receding-horizon linear quadratic control law, but it
can be extended for nonlinear control by
incorporating nonlinear prediction models. Fuzzy
prediction models and data-driven prediction models
are mainly used in nonlinear SOFC predictive
controls. A fuzzy Hammerstein model is used as the
predictive control model to achieve online control of
an SOFC system (Huo et al., 2008). In order to
control the stack temperature of a SOFC within a
safe range, an online nonlinear MPC scheme based
on an improved T-S fuzzy model is proposed (Yang
et al., 2009). Its control sequence could be obtained
by the branch-and-bound method. The nonlinear
predictive controller based on an improved radial
basis function neural network is applied (Wu et al.,
2008). It controls the voltage and guarantees fuel
utilization within a safe range and uses the genetic
algorithm for parameter optimizations. In order to
reduce the heavy computing load in nonlinear MPC
499
Pan L., Cao C. and Shen J..
L1 Adaptive Output Feedback Controller with Operating Constraints for Solid Oxide Fuel Cells.
DOI: 10.5220/0005043604990507
In Proceedings of the 11th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2014), pages 499-507
ISBN: 978-989-758-039-0
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
control, which is mainly caused by the nonlinear
optimization and on-line model identifications, the
disturbance model predictive control (Muske and
Badgwell, 2002) is introduced for SOFC (Pan and
Shen, 2012). It has less computing load and can deal
with some nonlinearity and uncertainty
characteristics. But it is non-adaptive and cannot
guarantee the closed-loop stability while achieving a
fast disturbance-rejection.
In this paper, we try to introduce another
advanced control approach, L
1
adaptive control, for
designing the SOFC control system. L
1
adaptive
control offers its own set of attractive features,
including fast and robust adaptation. In addition to
the conventional asymptotic performance
characterization, L
1
adaptive control permits
transient analysis for both control signal and system
response. Furthermore, this methodology has been
extended to systems with unknown time-varying
parameters (Cao and Hovakimyan, 2007), to systems
with nonlinear uncertainties (Cao and Hovakimyan,
2008), to systems with un-modeled internal and
actuator dynamics (Cao and Hovakimyan, 2008), to
systems in the presence of non-zero trajectory
initialization error (Cao and Hovakimyan, 2008),
and to a certain output feedback framework (Cao
and Hovakimyan, 2009). L
1
adaptive control has
been very successfully applied in unmanned flight
controls which have nonlinearities, time-varying
disturbances, unknown parameters and un-modeled
dynamics.
An extension approach of the L
1
adaptive output
feedback control (Cao and Hovakimyan, 2009) to
systems of unknown relative degree may deal well
with the control problems of SOFC, e.g., time-
varying uncertainties with unknown rate of
variations caused by load disturbances. Compared to
other L
1
adaptive control methods, this approach
adopts a new piece-wise continuous adaptive law
along with the low-pass filtered control signal. It
allows for achieving arbitrarily close tracking of the
reference signals, and the transfer function of its
reference system is not required to be strictly
positive real (SPR). Stability of this system is
guaranteed by its design via small-gain type
argument. These features show that this L
1
adaptive
control approach may have great potential to be
applied in wide industrial processes.
In this paper, we reproduce a SOFC simulation
model as the plant; then take advantage of its
nominal working conditions to design a L
1
adaptive
output feedback controller. Through the analysis on
the controller framework, the operating constraints
are set to the proper position in the loop. It holds the
closed-loop stability in the presence of the hard
constraints. Simulations comparing to a linear
disturbance model predictive controller (DMPC) on
the SOFC model show that L
1
adaptive control has
better disturbance-rejection performance and much
faster temporary regulating-process on the SOFC
constant voltage.
The paper is organized as follows. Section 2
describes the dynamic SOFC model. In Section 3, L
1
adaptive output feedback controller with operating
constraints is designed. For making an evaluation on
the new controller, a linear DMPC controller is
designed in Section 4. The simulation results along
with some discussions on these two SOFC control
methods are presented in Section 5. Section 6
concludes the paper.
2 DYNAMIC MODEL OF SOFC
Many nonlinear dynamic models of SOFC with
detailed descriptions on cell internal processes are
too complicated to support a controller designing
process. The model reported in the paper (Padullés
and Ault, 2000) describes the key characteristics of
the SOFC dynamic process in the Laplace transform
domain. It shows challenging control problems
owing to SOFC’s nonlinear dynamics, tight
operating constraints and unexpected disturbances. It
has been taken as a benchmark commonly studied in
the SOFC control literature (Huo et al., 2008; Wu et
al., 2008; Yang et al., 2009). Therefore, this model
will be adopted as the SOFC plant for the L
1
adaptive
controller design. Some preconditions (Padullés and
Ault, 2000) are stated in the following: The gases
are ideal; The stack is fed with hydrogen and air;
The flow ratio of hydrogen to oxygen is kept at
1.145; Lump gas pressures are considered in the
channels along the electrodes; The temperature is
stable; The exhaust of each channel is via a single
orifice, and the ratio of pressures between the
interior and exterior of the channel is large enough
to consider that the orifice is choked; The Nernst
equation can be applied. This dynamic SOFC model
consists of the fuel processor and the fuel cell stack,
as shown in Fig. 1, where E denotes the stack output
voltage (V), q
f
the natural gas flow rate (mol/ s), and
I the external current load (A); p
H2
, p
O2
, and p
H2O
denote the partial pressures (Pa) of hydrogen,
oxygen, and water, respectively; q
H2,in
and q
O2,in
are
the input flow rates of hydrogen and oxygen (mol/s),
respectively. The model is described in the
following and all the parameters are annotated in
Table 1.
ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics
500
2.1 The Fuel Processor
The fuel processor converts fuels such as natural gas
to hydrogen and byproduct gases. From the
viewpoint of control analysis, a first-order transfer
function with time constant
f
can well describe the
dynamic reform process from the natural gas input
N
f
to the hydrogen-rich fuel q
H
2
,in
. The fuel processor
can be simply represented by
H2,in
1
1
ff
q
N
S
(1)
Hydrogen reacts with oxygen in SOFC and
generates water. The flow ratio between hydrogen
and oxygen is represented by
2,
2,
H
in
HO
Oin
q
q
(2)
For having a complete reaction by the excessive
oxygen, take the ratio
H-O
as 1.145 (Padullés and
Ault, 2000).
2.2 The Fuel Cell Stack
Applying Nernst’s equation and taking into account
ohmic, concentration, and activation losses (i.e.,
o,
c
and
a
), the stack output voltage is given by
22
00
2
( /101325)
ln
2
HO
oca
HO
pp
RT
ENE
Fp






(3)
Where
1
2
22
2
(2)
1
H
HHr
H
K
p
qKI
S

(4)
1
2
22
2
()
1
O
OOr
O
K
p
qKI
S

(5)
1
2
2
2
2
1
HO
HO r
HO
K
pKI
S
(6)
log
a
I


(7)
ln 1
2
c
L
RT I
F
I




(8)
o
I
r
(9)
Equation (4)-(6) represent the dynamic
characteristics of the partial gas pressure of
hydrogen, oxygen and water inside the anode
channel associated with their molar flows through
the anode valve respectively. Take hydrogen as an
example to derive it. Consider the molar flow of
Hydrogen is proportional to its partial pressure
inside the channel and have
2
2
2
H
H
H
q
K
p
(10)
Take time derivative on the perfect gas equation of
hydrogen and obtain
2
2, 2, 2,
()
H
H
in H o H r
dp
RT
qqq
dt V

(11)
Where q
H2,in
is the input hydrogen flow, q
H2,o
is the
output hydrogen flow, and q
H2,r
is the hydrogen flow
that reacts. According to the basic electrochemical
relationships, q
H2,r
can be calculated by
2,
2
2
Hr r
NI
qKI
F

(12)
Apply (10) and (12) in (11) and then take the
Laplace transform, obtaining (4) and the equation
22
/( )
HH
VKRT
.
The transfer functions of oxygen and steam are
derived as well.
The fuel utilization which is one important
operating variable and may affect the performance
of SOFC is defined as
2, 2, 2,
2, 2, 2,
2
Hin HO Hr
f
H
in H in H in
qq q
NI
u
qqFq

(13)
The desired range of fuel utilization is from 0.7 to
0.9. The overused (u
f
> 0.9) and underused (u
f
< 0.7)
fuel conditions should be prevented. An overused
condition could lead to permanent damage to the
cells due to fuel starvation and an underused-fuel
situation results in unexpectedly high cell voltages
(Vahidi et al., 2004).
2
r
K
r
K
I
o
a
c
f
q
1
1
f
S
1/
HO
2,Hin
q
2,Ho
q
1
2
2
1
H
H
K
S
1
2
2
1
O
O
K
S
1
2
2
1
HO
HO
K
S
22
00
2
( /101325)
ln
2
HO
HO
pp
RT
NE
Fp




2H
p
2O
p
2HO
p
E
Figure 1: Dynamic model of SOFC.
Table 1: Parameters of the SOFC.
Para-
meter
Value Representation
T 1273 K Absolute temperature
F 96,485 C /mol Faraday’s constant
R
8.314
J/(mol·K)
Universal gas constant
E
0
1.18 V Ideal standard potential
N
0
384
Number of cells in
series in the stack
Kr
0.996×10
3
mol/(s·A)
Constant, Kr =N
0
/4F
L1AdaptiveOutputFeedbackControllerwithOperatingConstraintsforSolidOxideFuelCells
501
Table 1: Parameters of the SOFC. (Cont.)
Para-
meter
Value Representation
K
H2
8.32 × 10
6
mol/(s·Pa)
Valve molar constant
for hydrogen
K
H2O
2.77 × 10
6
mol/(s·Pa)
Valve molar constant
for water
K
O2
2.49 × 10
5
mol/(s·Pa)
Valve molar constant
for oxygen
H2
26.1 s
Response time of
hydrogen flow
H2O
78.3 s
Response time of water
flow
O2
2.91 s
Response time of
oxygen flow
HO
1.145
Ratio of hydrogen to
oxygen
r 0.126 Ohmic loss
f
5 s
Time constant of the
fuel processor
0.05 Tafel constant
0.11 Tafel slope
I
L
800 A
Limiting current
density
3 L
1
ADAPTIVE CONTROLLER
WITH INPUT CONSTRAINTS
ON SOFC
Several aspects should be considered in the design
of a controller for the SOFC system.
First, we know from the modeling work in
Section 2 that the nonlinear SOFC model composed
of (1) to (13) has the Wiener-type output
nonlinearity. Suppose the operating temperature and
pressure of the SOFC is kept constant, then the stack
terminal voltage E is mainly influenced by the inlet
hydrogen flow q
H2,in
and the current I. The operating
stack voltage usually shows significant changes at
low and high current loads, even shows a rapid
deterioration caused by overloaded current. Thus the
stack current is often taken as the main disturbance
variable to the process.
Second, the feasible operation area of SOFC
shows that it is impossible for SOFC to maintain a
simultaneous constant fuel utilization u
f
and constant
output voltage E operating regime for a range of
current I. The constant voltage control is much safer
than the constant fuel utilization control for the fuel
utilization performs (Wu et al., 2008).
In order to achieve a constant stack voltage
control under drastic current load disturbances, a L
1
adaptive output feedback controller is designed for
keeping both the SOFC output voltage at set point
and the fuel utilization within the safe range. Fig.2
shows the structure of the L
1
adaptive output
feedback control loop for a SOFC process.
ˆ
()yt
ˆ
()t
Figure 2: L
1
Adaptive output feedback control system for
SOFC.
3.1 Problem Formulation
Describe the controlled SOFC voltage dynamics as
the following single-input single-output system:
() ()(() ())ys As us ds
(14)
where yR is the SOFC terminal voltage, uR is the
hydrogen flow rate, A(s) is a strictly proper
unknown transfer function of unknown relative
degree nr, for which only a known lower bound 1<
dr <nr is available, d(s) is the Laplace transform of
the time-varying uncertainties and disturbance d(t) =
f(t, y(t)), where f is an unknown map, subject to the
following assumption:
Assumption 1 There exist constants L>0 and
L
0
>0 such that for all t0:
1212
(, ) (, )
f
ty fty Ly y
(15)
0
(, )
f
ty Ly L
(16)
Assumption 2 There exist constants L
1
>0, L
2
>0
and L
3
>0 such that for all t0:
123
() ()dLyt Lyt L

(17)
where the numbers L, L
0
, L
1
, L
2
, L
3
can be
arbitrarily large.
Assumption 3 The DC gain of the nominal
working point of SOFC is known.
Let r(t) be a given bounded continuous reference
input signal. The control objective is to design an
adaptive output feedback controller u(t) such that the
system output y(t) tracks the reference input r(t)
following a desired reference model
() ()()ys Msrs
(18)
where M(s) is a minimum-phase stable transfer
function of relative degree dr > 1. Thus we can
rewrite the system in (14) as:
ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics
502
() ()(() ())ys Ms us s
 (19)
() (( () ())() () ())/ ()
s
As Ms us Asds Ms
 (20)
Let (A
m
R
NN
, b
m
R
N
, c
m
R
N
) be the minimal
realization of M(s). Thus the system in (19) can be
rewritten as:
0
() () ( () ())
() (), (0)
mm
T
m
x
tAxtbut t
yt c xt x x


(21)
3.2 L
1
Adaptive Output Feedback
Controller
L
1
Adaptive controller consists of the state predictor,
the adaptation law and the control law.
The state predictor is given by:
0
ˆˆ ˆ
() () () ()
ˆˆ
ˆ
() (), (0)
mm
T
m
x
tAxtbut t
yt c xt x x


(22)
Where
ˆ
()
x
t R
N
and
ˆ
()yt
R are the state and
output of the predictor respectively;
ˆ
()t
R
N
compensates the system disturbances and model
mismatch. It is on-line estimated by the following
adaptation law:
1
ˆˆ
() ( ) [ ,( 1) ],
ˆ
() ()(), 0,1,2, ,
tiTtiTiT
iT T iT i




(23)
Where T>0 is the sampling time of the adaptation
law; and
1
1
()
0
1
()
( ) 1 ( ), 0,1, 2, .
m
m
T
AT
AT
Te d
iT e y iT i




(24)
Where 1
1
R
N
be the basis vector with first element
1 and all other elements zero;
ˆ
() () (); ,
T
m
NN
c
yt yt yt R
DP






,
where P=P
T
>0 satisfies the algebraic Lyapunov
equation A
m
T
P + PA
m
= -Q, Q>0; and let D R
(N-
1)N
satisfies
1
(( )) 0
TT
m
Dc P
.
The control law is defined via the output of the
low-pass filter C(s):
1
()
ˆ
() ()() ( ) ()
()
T
mm
Cs
us Csrs c SI A s
Ms

(25)
The selection of C(s) and M(s) must ensure that
() () ()/( () () (1 ()) ())H s AsM s Cs As Cs M s
(26)
is stable and that the L
1
-gain of the system is
bounded as follows:
1
()(1 ()) 1
L
Hs Cs L
(27)
(Cao and Hovakimyan, 2009).
The above piece-wise continuous adaptive law
with the low-pass filtered control signal allows for
achieving arbitrarily close tracking of the input and
the output signals of the reference system. The
performance bounds between the closed-loop
reference system and the closed-loop L
1
adaptive
system can be rendered arbitrarily small by reducing
the step size of integration. It can be represented by
the following equations.
00
lim ( ) ( ), lim ( ) ( )
ref ref
TT
yt y t ut u t


where T is the integration step of the L
1
adaptive
controller.
3.3 Operating Constraints for SOFC
Besides the design above, we need to put the input
constraints in the L
1
adaptive controller for the
SOFC voltage control, i.e., letting
min max
()uutu
hold for all t0, where u
min
=0, u
max
=1.7023mol/s
given in the paper (Padullés and Ault, 2000).
Considering the subtle symmetric structure of L
1
adaptive control, we cannot constrain u(t) directly.
ˆ
()t
is sent into both the plant and the state
predictor for cancellation. Its constraints can
influence the value of u(t) but cannot change the
stability of the closed loop. Thus, we have
min max
min min
max min
ˆ
ˆˆ ˆ
(), ()
ˆ
ˆˆ ˆ
() , ( )
ˆ
ˆˆ
,()
[,( 1)], 0,1,2,.
iT if iT
tifiT
if iT
tiTi T i






(28)
Another point is the possible different DC gains
between the plant and the state estimator. Because
the nominal parameters of SOFC are available, we
have Assumption 3. Dividing the output voltage
reference by the nominal DC gains of the SOFC
system, we get r(t) in control law (25).
4 DMPC CONTROLLER DESIGN
FOR A COMPARISON
In order to evaluate the performance of the L
1
Adaptive controller for SOFC, we try to introduce
L1AdaptiveOutputFeedbackControllerwithOperatingConstraintsforSolidOxideFuelCells
503
the linear disturbance model predictive controller
(Muske and Badgwell, 2002; Pannocchia and
Rawlings, 2003) to be an evaluating reference. It is a
kind of target-adaptive offset-free MPC with the
advantages in disturbance rejection and offset-free
tracking. The DMPC has been successfully applied
in CSTR (Pannocchia and Rawlings, 2003) and
become a fundamental approach where a variety of
MPC approaches have derived. The L
1
Adaptive
controller and the DMPC controller have almost
equal complexity in designs and computation load
online, therefore we will compare their performance
in the simulations. For clarity, a brief design of the
DMPC controller for SOFC is presented. First, the
augmented disturbance prediction model of SOFC is
built in term of the conditions for detectability, and
then the problem of estimating the augmented
disturbance states is solved. As a result, an
augmented observer is used to estimate the system
states and the lumped mismatch. Last, the
augmented disturbance model is adopted in the
predictive control algorithm to realize the control of
SOFC.
4.1 Disturbance Model and Estimator
We need to describe the SOFC plant approximately
by a linear model with augmented disturbance states
before the design of the DMPC controller. The
following linearized discrete state-space model
describes the controlled voltage system
1kkk
kk
x
Gx Hu
yCx

(29)
where y R is the SOFC terminal voltage, u R is
the hydrogen flow rate, x R
2
is the process state
and its rank represents the inertial of the process,
G R
22
, H R
21
, C R
12
, (G, H) is stabilizable
and (C,G) is detectable in the SOFC model. There
must be some model-plant mismatch in using the
linear model of Eq.29. We lump the mismatch along
with load disturbances into an augmented state to
make a disturbance model of SOFC

1
1
01 0
kk
d
k
kk
k
kd
k
xx
GG H
u
dd
x
yCC
d










(30)
where d R, G
d
R
21
, C
d
R
11
. Because the
lumped disturbance is unmeasured, an estimator is
needed for state-observing
1| | 1
1| | 1
1
|1 |1
2
ˆˆ
ˆˆ
01 0
ˆ
ˆ
()
kk kk
d
k
kk kk
kkk dkk
xx
GG H
u
dd
L
yCx Cd
L



 


 


 
 




(31)
where L
1
R
21
, L
2
R
11
are the predictor gain
matrices for the state and the disturbance. Since the
additional modes introduced by the disturbance are
unstable, detectability of the augmented system is a
necessary and sufficient condition for a stable
estimator to exist.
4.2 Detectability of the Augmented
State Space
The condition which ensures the observability of the
augmented disturbance system is given in the
following Lemma (Muske and Badgwell, 2002).
Lemma The augmented system presented in Eq.
30 is detectable if and only if (C,G) is detectable and
()
d
d
d
IG G
rank n n
CC





(32)
Where n is the number of the nonaugmented states,
n
d
is the number of the disturbances. In the SOFC
system, n is 2 and n
d
is 1. This Lemma implies that
the maximum dimension of the disturbance d in
Eq.30 such that the augmented system is detectable
is equal to the number of measurements y. That
gives us the guideline to design the augmented
system. Because (C, G) is detectable, the disturbance
model is defined by choosing the matrices G
d
and C
d
to hold Eq.32.
4.3 Target-adaptive MPC Algorithm
The goal of tracking the steady-state target is to
remove the effects of the estimated constant
disturbance states in the MPC control. It is a kind of
target-adaptive control. Given the current estimate of
the disturbance
|
ˆ
kk
d , the state and input target are
computed by solving the following quadratic
program
,
|
|
min max
min | max
min( ) ( )
..
ˆ
ˆ
0
ˆ
tt
T
ts ts
xu
dkk
t
t
dkk s
t
tdkk
uu Ruu
st
Gd
x
IG H
u
C
Cd y
uuu
yCxCdy











(33)
ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics
504
where x
t
R
2
, u
t
R, R
R and R
>0, y
s
R
and
u
s
R are the setpoints of the controlled and
manipulated variables respectively.
By tracking the steady-state target of the
manipulated variable, the MPC controller solves the
following optimization problem to obtain the input
sequence
01
, ,...
0
min max
min max
min ( ) ( )
()()
.. .30
T
ks ks
uu
k
T
kt kt
k
k
J
yyQyy
uuRuu
st Eq
uuu
yyy




(34)
where Q R
+
and R R
+
.
With an augmented disturbance prediction model
of SOFC and an online target-optimization
algorithm, the DMPC controller can deal with the
plant-model mismatch, unmodeled plant
disturbances and achieve the zero offset output
tracking. The design complexities of DMPC and L
1
adaptive controller are at the same level.
5 SIMULATIONS
AND DISCUSSION
Design the L
1
adaptive output feedback controller
based on (22)-(25) for the SOFC model shown in
Fig.1. The design information is shown as follows.
The reference model
2
1
()
1.4 1
Ms
s
s

; the filter
2
9
()
25 9
Cs
ss

; the sampling interval T=0.01s; the
offset range
min
ˆ
= -0.4,
max
ˆ
=0.4; the variables at
the nominal working point, I=300A, q
f
=0.746mol/s,
E=341.7V.Only the nominal DC gain of the SOFC
process is used for designing the L
1
adaptive
controller. Modeling is not needed for this control
algorithm.
Because of its successful and wide applications,
the model predictive control approach can act as a
reference to evaluate the L
1
adaptive output feedback
controller. Considering there are many kinds of
MPC approaches, we choose two ways to make
these comparisons.
First, the linear offset-free disturbance model
predictive controller (DMPC) presented in Section 4
is adopted for the comparison with the L
1
adaptive
output feedback controller. We apply L
1
adaptive
controller and DMPC controller respectively on the
SOFC model. We put two step disturbances into the
simulation experiments. Assuming at t = 400 s, a
load disturbance causes the stack current to have a
step change (from 300 to 280 A), and at t = 700 s, a
load disturbance causes the stack current to have
another step change (from 280 to 320 A). The step
disturbances on the stack current are shown in Fig.
3(a). Fig. 3(b) shows the fuel utilization by L
1
adaptive control. It is kept within the safe range. Fig.
3(c) and Fig. 3(d) compare the curves of the constant
voltage by L
1
and DMPC control. It shows that the
L
1
adaptive control has a shorter temporary
regulating-process in the constant voltage control.
For improving the robustness, the regulating of
DMPC is much slower than that of L
1
adaptive
control, which can be seen from the control signal
shown in Fig.3(e) and Fig.3(f). If we quicken the
DMPC regulating, the DMPC control system may
not be stable. Thus, L
1
adaptive control has obvious
advantages over DMPC control in the fast tracking
and disturbance rejection in this case. Something to
note, we cannot say the control performance of
DMPC shown in Fig.3 is its best one, but it is its
best in all our simulations.
400 500 600 700 800 900 1000
270
280
290
300
310
320
330
time (s)
Current
I(A)
(a) Disturbance of stack current.
400 500 600 700 800 900 1000
0.7
0.75
0.8
0.85
0.9
0.95
1
time (s)
Fuel Utilization
(b) From L
1
adaptive control.
L1AdaptiveOutputFeedbackControllerwithOperatingConstraintsforSolidOxideFuelCells
505
400 500 600 700 800 900 1000
330
335
340
345
350
Output Voltage
time (s)
E(V)
(c) From
L
1
adaptive control.
400 500 600 700 800 900 1000
330
332
334
336
338
340
342
344
346
348
350
Output Voltage
time (s)
E(V)
(d) From DMPC control.
400 500 600 700 800 900 1000
0.5
0.6
0.7
0.8
0.9
1
time (s)
Input Hydrogen flow
mol/s
(e) From L
1
adaptive control.
400 500 600 700 800 900 1000
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
Input Hydrogen flow
time (s)
mol/s
(f) From DMPC control.
Figure 3: The simulation results of the
L
1
adaptive control
and DMPC control on SOFC process.
Second, some comparisons are made with other
published results of a nonlinear MPC (Li et al.,
2011), we find that the L
1
adaptive controller has a
better rapidness under the guaranteed stability in the
nonlinear SOFC process control and has much less
online computation load.
6 CONCLUSIONS
This paper illustrates the fast-adaptation L
1
adaptive
controller design for the nonlinear SOFC process
control. An output feedback controller is designed
for SOFC system with unknown dynamics. Unlike
model-based control, it only needs a few system
parameters to design. The simulation results show
that it has good capability of disturbance rejection
and fast reference-tracking.
ACKNOWLEDGEMENT
The authors are grateful to the support of the
National Natural Science Foundation of China under
Grants 51106024 and 51036002. The authors would
like to express their appreciations to all the
reviewers for their invaluable comments.
REFERENCES
Aguiar, P., Adjiman, C., Brandon, N., 2005. Anode-
supported intermediate temperature direct internal
reforming solid oxide fuel cell: II. Model-based
dynamic performance and control. J. Power Sources,
147(1-2), 136-147.
ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics
506
Cao, C., Hovakimyan, N., 2007. Guaranteed transient
performance with L
1
adaptive controller for systems
with unknown time-varying parameters: part I.
Proceedings of American Control Conference, New
York.
Cao, C., Hovakimyan, N., 2008. L
1
adaptive controller for
a class of systems with unknown nonlinearities: part I.
American Control Conference, Seattle, WA.
Cao, C., Hovakimyan, N., 2008. L
1
adaptive controller for
nonlinear systems in the presence of unmodelled
dynamics: Part II. American Control Conference,
Seattle, WA.
Cao, C., Hovakimyan, N., 2008. L
1
adaptive controller for
systems with unknown time-varying parameters and
disturbances in the presence of non-zero trajectory
initialization error. International Journal of Control,
81(7), 1147–1161.
Cao, C., Hovakimyan, N., 2009. L
1
adaptive output
feedback controllers for non-strictly positive real
reference systems: missile longitudinal autopilot
design. Journal of Guidance, Control, and Dynamics,
32(3), 717-726.
Huo, H. B., Zhu X.J., Hu, W. Q., Tu, H. Y., Li, J., Yang,
J., 2008. Nonlinear model predictive control of SOFC
based on a Hammerstein model. J. Power Sources,
185(1), 338-344.
Li Y. G., Shen J., Lu J. H., 2011. Constrained model
predictive control of a solid oxide fuel cell based on
genetic optimization. J. Power Sources, 196, 5873-
5880.
Muske, K. R., Badgwell, T. A., 2002. Disturbance
modeling for offset-free linear model predictive
control. J. process control, 12(5), 617-632.
Padullés, G. W., Ault, J. R., 2000. An integrated SOFC
plant dynamic model for power systems simulation. J.
Power Sources, 86(1-2), 495-500.
Pan, L., Shen, J., 2012. Disturbance modeling and offset-
free predictive control for solid oxide fuel cell. 55th
ISA POWID Symposium, 492, 293-307.
Pannocchia, G., Rawlings, J. B., 2003. Disturbance
models for offset-Free model-predictive control.
AIChE Journal, 49(2),426-437.
Pukrushpan, J., A. Stefanopoulou, Peng, H., 2002.
Modeling and control for pem fuel cell stack system.
In Proc. of the 2002 American Control Conf.,
Anchorage, AK, pp. 3117-3122.
Stiller, C., Thorud, B., Bolland, O., Kandepu, R., Imsland,
L., 2006. Control strategy for a solid oxide fuel cell
and gas turbine hybrid system. J. Power Sources,
158(1), 303-315.
Vahidi, A., Stefanopoulou, A., Peng, H., 2004. Model
predictive control for starvation prevention of a hybrid
fuel cell system. In Proc. of the 2004 American
Control Conf., Boston, MA, 834-839.
Wu, X. J., Zhu, X.J., Cao, G.Y, Tu, H.Y., 2008. Predictive
control of SOFC based on a GA-RBF neural network
model. J. Power Sources, 179(1) , 232-239.
Yang, J., Mou, H. G., Li, J., 2009. Predictive control of
solid oxide fuel cell based on an improved Takagi-
Sugeno fuzzy model. J. Power Sources. 193(2), 699-705.
L1AdaptiveOutputFeedbackControllerwithOperatingConstraintsforSolidOxideFuelCells
507