Vous êtes sur la page 1sur 6

Sliding Control Retrofit for a Thermal Power Plant

Rajiv Sreedhar Benito Fernhndez Glenn Y. Masada


Department of Mechanical Engineering
University of Texas at Austin
Austin, Texas - 78712

Abstract erally does not provide satisfactory performance at


other operating loads.
Preliminary results of a sliding mode controller
retrofit for a thermal power plant are presented. The Various control strategies have been successfully
proposed sliding mode controller replaces the existing implemented to provide optimal performance over
multi-loop PID controller on the water/steam-side of the entire range of operation of the power plant.
the thermal power plant. The choice of the sliding Tysspr [19, 201 and Cori [2] present results using lin-
mode control scheme is appropriate as it embodies ear quadratic techniques to optimally control a boiler
an inherent robustness to modeling errors and pa- system. Nakamura [lo, 111 uses an auto-regressive
rameter variations. The sliding mode controller is technique to obtain an accurate model and uses the
validated by simulations using a 2lSt order thermal model to design an optimal controller using linear
power plant model. quadratic techniques. Nomura [13] extends Naka-
mura's work and presents an adaptive steam tem-
Keywords : Robust control, power plants, sliding perature control system with on-line/off-line model
mode control, nonlinear control. identification. Amin [l] presents an adaptive control
scheme for a thermal power plant using a self-tuning
controller. A model reference adaptive control scheme
is presented by Mabius [8]. Improved control of the
1. Introduction power plant can also be achieved by selecting the
control scheme such that the coupling between the
The control of a thermal power plant presents a con- various control loops is reduced. Decoupling control
siderable challenge since the dynamic characteristics scheme for power plants using the inverse nyquist ar-
of the plant change significantly with operating con- ray (INA) technique are presented by Johansson [7],
ditions (Peiia [15]). A large majority of the existing Masada [9], and Nam [12].
thermal power plants are controlled by proportional,
integral, and derivative (PID) control systems. The The sliding mode control system presented in this pa-
popularity of the PID control system is mainly due per is selected because of its inherent robustness to
to its simple architecture and the ease with which it modeling errors and system faults. The robustness
can be implemented in hardware. It is possible to ob- properties of the sliding mode control scheme guar-
tain a satisfactory system response by appropriately antee stable and safe operation. The sliding mode
tuning the PID gains. The single loop PID control control scheme presented is easy to tune and has a
systems are usually tuned using the Zeigler-Nichols simple architecture. The controller is validated using
technique (Zeigler [22,23]). The Ziegler-Nichols tech- a 21St order thermal power plant model developed
nique provides satisfactory results even with an im- by Pefia [15]. The model represents the dynamics of
perfect model of the system. However the results of C.B. Jones Unit #2 thermal power plant operated by
the Ziegler-Nichols technique do not carry over to the Southwestern Public Services Company in Lubbock,
multi-input multi-output (MIMO) PID control sys- Texas. The Jones Unit #2 is a 235 MW, gas fired
tem. The MIMO PID control systems are consider- power plant.
ably more difficult to tune (Thompson [l8]). Pentti-
nen [16] and Peltomaa [14] present designs of multi-
variable PI controllers for unknown systems. In most 2. Multi-loop PID Control System
cases the PID gains are fixed after tuning the con-
troller; typically the tuning is performed at full load
Figure 1shows the water/steam-side schematic of the
conditions. In such tases the PID control system gen-
Jones #2 power plant. The existing control loops
'Corresponding author. (Hill [SI) in the water/steam-side are enumerated as

0-7803-2550-8/95$4.00O1995IEEE 409

Authorized licensed use limited to: Mohammad Safaei. Downloaded on July 22, 2009 at 04:08 from IEEE Xplore. Restrictions apply.
follows.
1. Drum level controller : regulates boiler feedwa-
ter flow to maintain drum level at setpoint. LP turbine HPturbine -.. ....

2. Main steam pressure controller : regulates fuel


firing to maintain main steam pressure (sec-
ondary superheater outlet pressure) at setpoint.
3. Main steam temperature controller : regulates
I I
superheat attemperator spray to maintain main
steam temperature (secondary superheater out-
let temperature) at setpoint.
4. Turbzne demand controller : regulates governor
valve area in response to MW load demand.
5. Burner tilt controller : regulates burner tilt PO-
sition to maintain reheat steam temperature at
set point.
6. Reheat steam temperature controller : regulates
reheat attemperator spray flow to maintain re- I Burners
heat steam temperature at setpoint.
F"*l
The controlled variables (outputs) labeled yi, i = Fuelvahe

1,. . , 5 and the corresponding manipulated variables


+

ui, i = 1 , . . . , 6 are tabulated in Tables 1 and 2, re-


spectively. The temperature at the reheater outlet is Figure 1: Thermal power plant schematic
primarily controlled by manipulating the burner tilt
angle, however if the burner tilt is saturated at -30"
and the reheat temperature exceeds the setpoint by 5
O F , the reheat spray is activated to control the tem-

perature. The output setpoints are presented in Ta-


ble 3. The PID loops pair corresponding outputs and
inputs, and use feedforward signals to minimize the Table 2: Manipulated variables (inputs)
coupling between the loops. Complete details of the ~

PID control system for Jones Unit #2 thermal power Description Range
plant are presented by Peiia [15] and Hill [6]. u1 : Feedpump scoop position, ( X j p ) 0 1 I

u2 : Fuel valve area, (Cafv) 0-1


u3 : Superheat spray valve area, (CaSy) 0 - 1
Table 1: Controlled variables (outputs) u4 : Throttle valve area, (C,,,) 0-1
u5 : Burner tilt, ( X o 0 ) -0.5 - 0.5
1 Description I units 1 zl6 : Reheat spray valve area. (en,.?,) I0-1 I

Table 3: Output setpoints


3. Sliding Controller Design

Consider a nonlinear system described by

xi fi )(. 4- sij (.).j (1)


j
Yi = hi(.) (2)
where 2 E %", U E Sm, and y E %P are the states,
inputs, and outputs of the system, respectively. The
410

Authorized licensed use limited to: Mohammad Safaei. Downloaded on July 22, 2009 at 04:08 from IEEE Xplore. Restrictions apply.
functions f(.)? g ( . ) ? and h ( - )are smooth functions of The system input can be considered to be the sum
z. The derivative of the ith output yi is represented of the performance input up and the stabilizing input
as (sliding mode) us. The sliding mode gain matrix K
is defined as
yi c .Cfhi(z)+.Cg( :i)hi(z)ui +Ed,(.j)hi(z)uj (3) IC = -[$]-'q (12)
j#i
where V is an appropriate approximation of V ( x ) .
where C f h i ( z )represents the Lie derivative of hi(z) Assuming that $ is diagonally dominant, K is ap-
with respect to vector field f and 9( : j ) represents a proximated as a diagonal matrix. In the absence of
vector field corresponding to the j t h column of matrix accurate models the control system design involves se-
g(z). Equation 3 is written in matrix form as lecting the diagonal elements kid of matrix K ?the size
of the boundary layer p i , and the integration error
y = C(z) + V(z)u (4) weights Xi. The bandwidth of the controller within
the boundary layer is represented by k i i l p i . Since
where C(x) is a vector with ith element ,Cfhi(z)and
tanh(s) can take values of f 1 . 0 , the quantity l k i i / p i l ,
D ( z ) is a matrix with ijth element d,( :qhi(z).The
( p i = ith actuator range) determines the maximum
diagonal terms of matrix D ( z ) are not zero, since the
fraction of the stabilizing input.
boiler turbine model developed by Peiia [15] indicates
that the outputs yi are of relative order 1with respect
to the input ui.
Define a sliding surface

s(t) = {z E sJzn(s(z) = O } (5)


ei + X i J eidt
Si(.) = =O (6)
Pi
where e = y - yd is the difference between the out-
put and desired output, and Jedt is the integral of Figure 2: Output error schematic
the error (Fig. 2). The parameters X i and pi are
positive numbers which represent the bandwidth of
the system and the thickness of the boundary layer.
The plant inputs are selected so that e and Jedt are 4. Input Curve Fits
driven to the sliding surface defined by Eq. 5. This is
guaranteed by selecting the input U such that The performance input component up is considered
to be the steady state value of ith input when there
s(x)i(.) < 0, vt >0 (7) is perfect tracking (zero output error). The input
component up is obtained by a smooth curve fit of
This can be achieved by forcing the data obtained from the steady state operation
using the PID controller. The operating range 155-
si(,) = -17; tanh(si(t)) (8) 235 MW is divided in increments of 10 MW, and the
The derivative of the sliding surface parameter is plant is run at steady state at each load setting. The
curve fits for the inputs are presented in Eqs. 13 -
i(z) = C(z) + D(t)ui + i d + Xe (9) 17. Mathematica software (Wolfram [XI) is used to
calculate the curve fit equations.
The input which satisfies the sliding condition is cal-
culated from Eqs. 8 and 9. f 0.217 + 2.218e-03MW;
MW 5 180
U = 2)-l(z){-c(z) - y d - Xe - qtanh(s(z))} (10) xjP = {I0.353 +ifif 7.491e-04MW;
MW > 180
(13)

where r ) = diag(vi,...,qp) and tanh(s) =


diag(tanh(si), . ,tanh(sp)). If C(x) and D ( z ) are Casu = -4.258e-03 + 3.150e-03MW (14)
not known accurately, the sliding condition can still CaJy = -1.532 + 2.947e-02MW
be guaranteed (Slotine [17] and Fern6ndez [3, 41 ) by -9.548e-05MW2 (15)
selecting the system inputs to be of the form
C,,, +
= 0.373 1.972e-03MW (16)
U E up + us (11) X,, = -6.845 1.290e-1MW +
where up is an estimate of the inputs calculated using
-7.182e-4MW2 +
1.205e-6MW3 (17)
the best approximation of the system dynamics (Fil- The boiler feedpump scoop tube position curve fit
lipov's equivalent control [5]), and us = K tanh(s). has a hard nonlinearity at 180 MW, caused when
41 1

Authorized licensed use limited to: Mohammad Safaei. Downloaded on July 22, 2009 at 04:08 from IEEE Xplore. Restrictions apply.
the number of operational boiler feed pumps changes
from 1 to 2. The plots of the curve fits and the data
points are shown in Figs. 3 - 7.

150 160 170 180 190 200 210 220 230 240
Power output (Mw)

Figure 6: Throttle valve cross-section area curve fit


150 160 170 180 190 200 210 220 230 240
Powe ourput ( M w )

Figure 3: Boiler feedpump scoop position curve fit

$ . . . .
U

150 160 170 180 190 200 210 220 230 w


Power output (Mw)

150 160 170 180 190 200 210 220 230 240
power output
Figure 4: Fuel valve cross-section area curve fit

0.8
Figure 7: Burner tilt angle curve fit
0.7

0.6 ....... ....... ........ ....... .......


., .. ..
0.5 _ .. ......... .......... ..........
... i..... i.... i.... !:.>=:>
............... .,. \ < ........
..
8 0.4
.
a....:
. .. .. , .,........:.--
....j.........j......... i...........................
o',i
. .
j
.. .. .. .
: : :
........;.......... :.......... :.......... .j....... .:>! .....i.
... .,.. ... ... :. ,\ i \ .

0.2 ....L........ i........i........;......... i......... i.........i..al.i


ys-
.. . . . . . 1
. .. .. .. .. .. ... .. '.
0.1 1
150 160 170 180
1 I '
190
1 1
200 210 220 230 240
I
I
with the 21St order model of the Jones Unit #2 power
POWET output ( M w ) plant. Figures 9 - 10 show the deviations from set-
point using the PID controller. Figures 11 - 12 show
the performance of the sliding mode controller. The
Figure 5: Spray valve cross-section area curve fit
simulations show that the deviation from setpoint for
the drum level, megawatt, and seconda
pressure and temperature are greatly re
5. Simulations sliding mode controller. However the d
setpoint for the reheat temperature is
The performance of the sliding mode controller is sliding mode controller is used. The p
compared with the PID controller. Figure 8 shows tions of the reheat temperature are much less than the
the MW setpoint of the plant for the simulations, negative deviations because the reheat spray is only
with the transition between steady state loads at 10 effective in controlling positive deviations. The reheat
MW/min. The parameters selected for the sliding temperature deviations can be reduced by better tun-
mode control loops are tabulated in Table 4. These ing or by using more accurate model to calculate the
parameters were chosen after extensive simulations performance input up.

41 2

Authorized licensed use limited to: Mohammad Safaei. Downloaded on July 22, 2009 at 04:08 from IEEE Xplore. Restrictions apply.
Table 4: Sliding mode controller parameters
1- 0.1 ....i.......

..
...
.
i.......i....... !.......?........ i
. .. .. .. .......
....
. .. .. ..
, , . .. ,
i........i.?!V::.:.
.
I Control ~ O O D I h l f l I A l

5000 IMM l9x02owo 25000 Mwo 35000 40000 41000 1"


tlms (*)

Figure 11: Deviations using the sliding mode controller

B
B
170 t..,;. ..
.
~

.
... ...
$

..
I
.........j ...... +.j ......!.......................
I i
+ 1 1
...............i>.... {
160 ....... i.......4 .......i.......;
.__.;
:. ....... .... .... ......
I I
.. I I
.. .. .. ' '
110
9x0 low0 IMW 2MM 2Mw3MM 35000 400004WmWmO
tiam (0

Figure 8: Megawatt setpoint profile


Figure 12: Deviations using the sliding mode controller
(contd.)

6. Conclusion

A multivariable sliding mode controller was devel-


oped for the thermal power plant. The controller
was validated on a 213*order dynamic model of the
Jones #2 gas fired thermal power plant. Simulation
results indicate that the performance of the sliding
mode controller is better than the existing multi-loop
PID controller, with respect to minimizing deviations
from setpoints and greater stability robustness in the
Figure 9: Deviations using the PID controller event of faults. A more rigorous mathematical treat-
ment of the controller design, and optimization issues
will be addressed in future publications.

7. Acknowledgments

The authors appreciate the financial support of the


Texas Higher Education Coordinating Board Energy
Research Applications Progrum, Project #003658-
078, and National Science Foundation Neuro Engi-
neering Program, Project #ECS-911326.

References
Figure 10: Deviations using the PID controller (contd.) [l] Amin, M., Masada, G.Y., and Womak, B.F.,
"Application of a Multivariable Self-Tuning Con-
41 3

Authorized licensed use limited to: Mohammad Safaei. Downloaded on July 22, 2009 at 04:08 from IEEE Xplore. Restrictions apply.
troller to a Power plant Boiler”, Proceedzngs of the 1141 Peltomaa, A. and Koivo H.N., “Tuning of a
1984 American Control Conference, pp. 339-347, San Multivariable Discrete Time PI Controller for Un-
Diego, California, June, 1984. known Systems”, International Journal of
[a] Cori, R. and Maffezzoni, C., “Practical-optimal Vol. 38, No. 4, pp. 735-745, October 1983.
Control of a Drum Boiler Power Plant”, Automatica, [15] Peiia, R., “Development of a Boiler-Turbine
Vol. 20, No. 2, pp. 163-173, March 1984. Mathematical Model for Power Plant Operation
[3] Fernindez, B. Control of Multivariable Nonlan- Studies”, Ph.D. Dissertation, Department of Electri-
ear Systems by the Sliding Mode Method, Ph.D. The- cal Engineering, University of Texas at Austin, De-
sis dissertation, Massachusetts Institute of Technol- cember 1993.
ogy, Cambridge, Massachusetts, 1988. [16] Penttinen, J . and Koivo, H.N., “Multivariable
Tuning Regulators for Unknown Systems”, Azltomat-
[4] Ferntindez, B. and Hedrick, J.K., “Generalized
Sliding Mode Control for MIMO Nonlinear Systems”, ica, Vol. 16, No. 4, pp. 393-398, July 1980.
International Journal of Control, Vol. 46, No. 3, pp [17] Slotine, J.J.E. and Li, W., Applied Nonlinear
1019-1040, 1988. Control, Prentice-Hall, Inc., 1991.
[5] Fillipov, A. F., “Differential Equations with [18] Thompson, S., “Multivariable PID Controller
Discontinuous Right-Hand Side” , American Mathe- for Unidentified Plant” , ASME Journal o f Dynamic
matical Society Translations, Series 2, Vol. 42, pp Systems, Measurement and Control, Vol. 104, pp.
199-231, 1964. 270-274, l982.
[6] Hill, AS., Thermal Optimization and Dzgi- [19] TyssG, A., Brembo, J.C., and Lind, K., “The
tal Control of a Fossil Fuel Power Plant through Design of a Multivariable Cont 1 System for a SHip
Boiler”, Automatica, Vol. 12, N
Computer Simulation, Master’s Thesis , Department
of Mechanical Engineering, University of Texas at 1976.
Austin , August 1992. [20] T y s s ~A.
, and Brembo, J .
Operation of a Multivariable Ship Boiler Control Sys-
[7] Johansson, L. and Koivo, H.N., ‘<Inverse
tem”, Automatica, Vol. 14, 1976.
Nyquist Array Technique in the Design of a Multivari-
able Controller for a Solid-Fuel Boiler”, International [all Wolfram, S., Mathematica A
Journal of Control, Vol. 40, No. 6, pp, 1077-1088, Math em at ics by Computer, Addison
December 1984. ing Company, Inc., 1991.
[22] Zeigler, J.G. and Nichols, N.B., “Optimum Set-
[8] Mabius, L., Kalnitsky, K., and Kaufman, H., tings for Automatic Controllers” , Transactions of the
“Model Reference Adaptive Control of Electric Gen-
ASME, Vol. 64, No. 11, 1942.
erating Plants”, Proceedings of the l g h IEEE Con-
ference on Decision & Control, pp. 286-291, Albu- [23] Zeigler, J.G. and Nichols, N.B., “Brocess Lags
querque, New Mexico, December 1980. in Automatic Control Circuits” , Transactions of the
ASME, Vol. 65, No. 5, 1943.
[9] Masada, G.Y. and Lee, M.S., “Inverse Nyquist
Array Control of a Coal Pulverizer”, 1989 Confer-
ence on Power Plant Control and Automation, Mi-
ami, Florida, 1989.
[lo] Nakamura, II. and Akaike, H., “Statistical Iden-
tification for Optimal Control of Supercritical Ther-
mal Power Plants”, Automatica, Vol. 17, No. 1, pp.
143-155, January 1981.
[ll] Nakamura, H., Toyota, Y., KushoHashi, M.,
and Uchida, M., “Optimal Control of Thermal Power
Plants”,ASME Journal of Dynamic Systems, Mea-
surement and Control, Vol. 111, No. 3, pp. 511-520,
September 1989.
[12] Nam, H., Masada, G.Y., and Hopkins, G.Y.,
“Modeling and Control of a Coal Fired Power Plant:
Part II-Control”, IFAC Symposium on Power Sys-
tems and Power Plant Control, Seoul, South Korea,
1989.
[13] Nomura, M. and Sato, Y., “Adaptive Optimal
Control of Steam Temperatures for Thermal Power
Plants” , IEEE Transactions on Energy Conwersion,
Vol. 4, NO. 1, pp. 25-33, March, 1989.

41 4

Authorized licensed use limited to: Mohammad Safaei. Downloaded on July 22, 2009 at 04:08 from IEEE Xplore. Restrictions apply.

Vous aimerez peut-être aussi