Documente Academic
Documente Profesional
Documente Cultură
Jie Wu , Sing Kiong Nguang , Jiong Shen , Guangyu Justin Liu & Yi Guo Li
To cite this article: Jie Wu , Sing Kiong Nguang , Jiong Shen , Guangyu Justin Liu & Yi Guo
Li (2010) Guaranteed cost nonlinear tracking control of a boiler-turbine unit: an LMI approach,
International Journal of Systems Science, 41:7, 889-895, DOI: 10.1080/00207720903480683
This article addresses the problem of designing a guaranteed cost nonlinear state feedback tracking control for
a boiler-turbine unit. First, the nonlinear boiler-turbine is re-expressed as a linear system with norm bounded
uncertainties via a nonlinear transformation function. Then, based on this linear model a sufficient condition for
the existence of a guaranteed cost nonlinear state feedback tracking control is derived in terms of linear matrix
inequalities. The advantage of the proposed tracking control design is that only a simple nonlinear controller is
constructed and it does not involve feedback linearisation technique and complicated adaptive or fuzzy schemes.
An industrial boiler-turbine system is used to illustrate the effectiveness of the proposed design as compared with
a linearised approach.
Keywords: optimal control; energy and power systems
One approach to handle such a problem is the so-called is an unknown matrix which is bounded with Lebesgue
guaranteed cost approach, and the approach has the measurable elements and satisfies
advantage of providing an upper bound on a given FT ðtÞFðtÞ I, ð2:3Þ
performance index. In the past decade, for the
advantages of linear matrix inequalities (LMI) and where I denotes the identity matrix. The uncertainties
breakthrough of its solving approach, it has been are said to be admissible if they satisfy (2.2) and (2.3).
broadly used in the analysis and design of control Associating with the system (2.1), the quadratic
systems such as continuous-time uncertain system, performance cost is defined as follows:
Z1
discrete-time uncertain system and system with uncer-
tain time-delay. This article is mainly devoted to a J¼ ðxT ðtÞQxðtÞ þ uT ðtÞRuðtÞÞdt, ð2:4Þ
0
guaranteed cost nonlinear tracking state-feedback
where Q and R are given symmetric positive-definite
control for the boiler-turbine unit subject to input
weighting matrices.
constraints using an LMI approach. The main
contribution of this article is the nonlinear boiler- Definition 1: For the uncertain system (2.1) and the
turbine model re-expressed as a TS fuzzy system via a performance cost (2.4), if there exist a control law u(t)
nonlinear transformation function. This approach has and a positive scalar J such that for all admissible
never been done in the literature. The advantage of the uncertainties, the closed-loop system is asymptotically
proposed tracking control design is that only a simple stable with the performance cost satisfying J J. Then
nonlinear controller is constructed and it does not J is called a guaranteed cost, and u(t) is called a
involve feedback linearisation technique and compli- guaranteed cost control law for the uncertain
cated adaptive or fuzzy schemes. system (2.1).
The rest of this article is organised as follows.
The design of such control law u(t) guarantees
Section 2 gives preliminaries of a guaranteed cost state
feedback control. Section 3 focuses on the model of the not only the asymptotical stability of the closed-loop
boiler-turbine unit. The design of a guaranteed cost system, but also an adequate level of performance in
nonlinear tracking control for the constrained boiler- terms of a certain upper bound of the guaranteed
turbine unit is given in Section 4. Simulation studies performance cost J.
are given in Section 5. Finally, conclusions are given in Lemma 1 [Petersen and McFarlane 1994, Yu, Xu, and
Section 6. Han 2004]: For the uncertain system (2.1) and the
performance cost (2.4), if there exist positive scalars "1
and "2, a matrix Y and a symmetric positive definite
2. Guaranteed cost state feedback control matrix X such that the following LMI holds for all
Consider a norm bounded continuous-time linear admissible uncertainties:
system which can be described by the following 28 9 3
T T
<XA þ AX þ BY >
> =
model with parameter uncertainties (Petersen 1987, 6 7
6 þYT BT þ "1 H1 HT1 YT ET2 XET1 Y X 7
Xu, Wang, and Wu 2006): 6> > 7
6 : þ"2 H2 HT ; 7
6 2 7
_ ¼ ðA þ 4AÞxðtÞ þ ðB þ 4BÞuðtÞ,
xðtÞ 6 7
6 "2 0 0 0 7
yðtÞ ¼ CxðtÞ, ð2:1Þ 6 7
6 "1 0 0 7
6 7
xðt0 Þ ¼ x0 , 6 7
4 R1 0 5
where x 2 Rn is the state vector, x0 is the initial state Q1
vector, u 2 Rm is the control input vector, y 2 Rq is the 0: ð2:5Þ
controlled output vector. A, B and C are known real
constant matrices that describe the nominal system. Then
iA and iB are matrix-valued functions representing
u ðtÞ ¼ YX1 xðtÞ ð2:6Þ
time-variant parameter uncertainties in the system
model, and are assumed to be of the following is a guaranteed cost controller for the uncertain
structure: system (2.1) and the performance cost of the closed-
4A ¼ H1 FðtÞE1 , loop system satisfies
ð2:2Þ
4B ¼ H2 FðtÞE2 , J traceðX1 Þ ¼ J , ð2:7Þ
where Hi and Ei (i ¼ 1, 2) are known real constant where trace is the sum of diagonal elements.
matrices with appropriate dimensions and F(t) 2 R The asterisk () in (2.5) stands for the corresponding
International Journal of Systems Science 891
elements below the main diagonal of a symmetric Table 1. Typical operating points of boiler-turbine.
block matrix.
#1 #2 #3 #4 #5 #6 #7
2.1. Guaranteed cost tracking control x01 75.60 86.40 97.20 108.00 118.80 129.60 135.00
x02 15.27 36.65 50.52 66.65 85.06 105.80 127.00
Consider the uncertain system (2.1) and assume that y(t)
is the controlled output signal, the desired tracked x03 492.20 469.30 450.90 427.90 398.40 355.40 289.10
signal is yref, then the error can be presented by u01 0.1191 0.2091 0.2705 0.3402 0.4182 0.5046 0.5878
(Czarkowski and Kazimierczuk 1994, Reza u02 0.3803 0.5519 0.6209 0.6900 0.7589 0.8279 0.9171
Moheimani and Petersen 1998, and Zhang and u03 0.1224 0.2556 0.3399 0.4358 0.5433 0.6625 0.7840
Hu 2008):
eðtÞ ¼ yðtÞ yref ðtÞ ¼ CxðtÞ yref ðtÞ: ð2:8Þ drum pressure (kg cm2), electric output (MW) and
The augmented system is given as follows: fluid density (kg cm3), respectively. The output y3 is
the drum water level deviation (m). acs and qe are steam
_
xðtÞ A 0 xðtÞ B 0
¼ Rt þ uðtÞ þ yref ðtÞ quantity and evaporation rate (kg/s), respectively and
eðtÞ C 0 0 eðtÞdt 0 I are given as follows:
ð2:9Þ
ð1 0:001538x3 Þð0:8x1 25:6Þ
or in a compact form acs ¼ ,
x3 ð1:0394 0:0012304x1 Þ
x_ ðtÞ ¼ Ae
e exðtÞ þ BuðtÞ
e e
þ dðtÞ, ð2:10Þ qe ¼ ð0:854u2 0:147Þx1 þ 45:59u1 2:514u3 2:096:
where ð3:2Þ
xðtÞ e ¼ 0 The control inputs are subject to
xðtÞ ¼ R t
e ,
dðtÞ ,
eðtÞdt y ðtÞ
0 ref 0 u1 , u2 , u3 1,
e¼ A 0 e¼ B :
A and B 0:007 u_ 1 0:007,
C 0 0 ð3:3Þ
Thus, the guaranteed cost control via an integral 2 u_ 2 0:02,
control can be formulated as a design of a guaranteed 0:05 u_ 3 0:05:
cost state feedback control law u ðtÞ ¼ Ke
xðtÞ.
Some typical operating points of the boiler-turbine
model (3.1) while the drum water level deviation (y3)
3. The boiler-turbine model is kept at 0 are tabulated in Table 1.
A 160 MW oil-fired electrical power plant model is
used in this article constituted with a drum-type boiler
and a turbine. The model is based on the P16/G16 at 4. Design of guaranteed cost nonlinear
the Sydvenska Kraft AB plant in Malmö, Sweden. The tracking control
boiler dynamic model is provided by both physical and This section focuses on the design of a guaranteed
empirical methods based on the data acquired from a cost nonlinear tracking control for the boiler-turbine
series of experiments and identification which captures system. In order to apply the method outlined in
all the relevant characteristics of the process. The Section 2, the nonlinear boiler-turbine is first trans-
boiler-turbine dynamic nonlinear model is given in the formed into a linear system with norm bounded
following equations (Chen and Shamma 2004 and Tan uncertainties by defining
et al. 2005):
u2 ¼ u2 x1 , ð4:1Þ
x_1 ¼ 0:0018u2 x9=8
1 þ 0:9u1 0:15u3 ,
Applying (4.1), the boiler-turbine system (3.1) can be
x_2 ¼ ð0:073u2 0:016Þx9=8
1 0:1x2 , rewritten as
x_3 ¼ ½141u3 ð1:1u2 0:19Þx1 =85,
ð3:1Þ x_1 ¼ 0:0018u2 x1=8
1 þ 0:9u1 0:15u3
y1 ¼ x 1 ,
y2 ¼ x 2 , x_2 ¼ 0:073u2 x1=8 1=8
1 0:016x1 x1 0:1x2
2 3
0:2551 0:0001 0:0000 0:0003 0:0007 0:0002
6 7
Y ¼ 4 0:0088 0:3415 0:0373 0:0017 0:3910 0:0058 5
0:4475 0:0000 4:9490 0:0002 0:0003 0:0001
International Journal of Systems Science 893
and "1 ¼ 3 and "2 ¼ 4. Therefore, the integral action The cost function with weighting matrices Q and R
matrix KI and the state feedback control matrix K are are selected to be the same as in the above nonlinear
2 3 approach. The rest of the control strategy design
0:011 0:0003 0:0001
6 7 is identical. By solving the LMI (2.5), the following
KI ¼ 4 0:0019 0:0072 0 5 and
controller’s gain matrices are obtained:
0:0003 0:0002 0:0034
2 3 2 3
0:0731 0:0028 0:0063 0:011 0:0002 0
6 7 6 7
K ¼ 4 0:007 0:0721 0:0012 5: KI ¼ 64 0:0008 0:0028 0:0010 5 and
7
0:0069 0:0015 0:0851 0:0004 0:0007
0:0034
The guaranteed cost nonlinear tracking control law 2 3
0:0623 0:0018 0:0048
for the boiler-turbine unit is 6 7
Zt K¼6 7
4 0:0352 0:0163 0:0155 5:
~ ¼ KxðtÞ þ KI
uðtÞ eðtÞdt:
0 0:0205 0:0063 0:0837
Using (4.1), we have
In order to prevent the windup caused by the
u ðtÞ saturations of the actuators, the same tracking anti-
u2 ðtÞ ¼ 2 ,
x1 ðtÞ windup strategy has been used.
where e(t) ¼ x(t) xref(t) and the corresponding upper
bound of performance cost is
J J ¼ traceðX1 Þ ¼ 0:4974: 5. Simulation results
The control systems for the boiler-turbine model are
In order to prevent the windup caused by the developed in MATLAB. Simulations presented here
saturations of the actuators, the tracking anti-windup are to illustrate the performances of the designed
strategy (Tharayil and Alleyne 2002) is applied here nonlinear control for various kinds of electric load and
(Figure 1). drum pressure demands. In this section, the following
For comparisons, a linearised model at #4 opera- two cases will be considered.
tion point is obtained as follows:
(1) Case 1:
_ ¼ AxðtÞ þ BuðtÞ,
xðtÞ ð4:6Þ 8 ref
>
> y ¼ 75:60, yref ref
2 ¼ 15:27, y3 ¼ 0, 0 t 500
where x ¼ x x0, u ¼ u u0, x0 and u0 are the < 1
corresponding steady-state values at #4 operating yref ref ref
1 ¼ 108:00, y2 ¼ 66:65, y3 ¼ 0, 500 t 1500
>
>
point, and the matrices A and B are : ref
y1 ¼ 75:60, yref ref
2 ¼ 15:27, y3 ¼ 0, 1500 t 2500
2 3
0:00251 0 0
6 7 (2) Case 2:
A ¼ 4 0:06941 0:1 0 5 and 8 ref
0:00669 0 0 >
> y ¼ 75:60, yref ref
2 ¼ 15:27, y3 ¼ 0, 0 t 500
2 3 < 1
0:9 0:349 0:15 yref ref ref
1 ¼ 135:00, y2 ¼ 127:00, y3 ¼ 0, 500 t 1500
6 7 >
>
B¼4 0 14:155 0 5: : ref
y1 ¼ 75:60, yref ref
2 ¼ 15:27, y3 ¼ 0, 1500 t 2500:
0 1:3976 1:6588
Remark 5.1: Case 1 simulates a change in the
reference signal from the operating point 1–4 and
Case 2 analyses how efficient the presented control
approach can be used to transfer the system to an
operating point that can be considered far away.
Simulation results for Cases 1 and 2 are shown in
Figures 2–5, respectively. Figure 2 shows the outputs
for Case 1. The figure shows that outputs of the linear
control design follow the reference signal faster than
the nonlinear control design, but with overshoots and
large control inputs. Also, the variations in the drum
Figure 1. Integral state-feedback control system with anti- water level are larger than in the nonlinear control
windup for boiler-turbine unit. design. Figure 4 shows the outputs for Case 2.
894 J. Wu et al.
(a) (a)
120 140
Drum pressure
Drum pressure
Nonlinear LMI Nonlinear LMI
120
(kg cm–2)
(kg cm–2)
Linear control Linear control
100
100
80 80
60 60
500 1000 1500 2000 2500 500 1000 1500 2000 2500
(b) Time (s) Time (s)
(b)
80 150
Electric output
Electric output
60
(MW)
100
(MW)
40
20 50
0 0
500 1000 1500 2000 2500 500 1000 1500 2000 2500
Time (s) Time (s)
(c) (c)
0.4 0.5
deviation (m)
deviation (m)
Water level
Water level
0.2
0 0
–0.2
–0.4 –0.5
500 1000 1500 2000 2500 500 1000 1500 2000 2500
Time (s) Time (s)
Figure 2. Boiler-turbine unit outputs for Case 1. Figure 4. Boiler-turbine unit outputs for Case 2.
(a) (a)
1 Nonlinear LMI 1 Nonlinear LMI
Fuel flow
valve (%)
valve (%)
0.5 0.5
0 0
500 1000 1500 2000 2500 500 1000 1500 2000 2500
Time (s) Time (s)
(b)
(b)
Steam control
1
Steam control
1
valve (%)
valve (%)
0.5
0.5
0
0 500 1000 1500 2000 2500
500 1000 1500 2000 2500
Time (s)
(c) Time (s) (c)
Feedwater flow
1
Feedwater flow
1
valve (%)
valve (%)
0.5
0.5
0
0
500 1000 1500 2000 2500
500 1000 1500 2000 2500 Time (s)
Time (s)
Figure 5. Boiler-turbine unit inputs for Case 2.
Figure 3. Boiler-turbine unit inputs for Case 1.
From the figure, one can conclude that, although the Conversely, the nonlinear designed system presents a
linear approach shows faster response in the drum good quality control in this case. It has been analysed
pressure tracking, it fails to track changes in electric in Tan et al. (2005) that linear control designs are not
output and the drum water level deviation. able to track large changes as given in Case 2.
International Journal of Systems Science 895