International Journal of Innovative

Computing, Information and Control c

ICIC International °2007 ISSN 1349-4198
Volume 3, Number 2, April 2007 pp. 371—384

DESIGN OF A PRESSURE CONTROL SYSTEM IN A FURNACE OF

A THERMAL POWER PLANT BOILER USING SELF-TUNING I-P
AND FEED-FORWARD COMPENSATOR

Takao Sato
Division of Mechanical System
Department of Mechanical Engineering
Graduate School of Engineering
University of Hyogo
2167, Shosha, Himeji, Hyogo 671-2201, Japan
tsato@eng.u-hyogo.ac.jp
Kazuhiro Takeda
Vibration & Control System Laboratory
Hiroshima Research & Development Center
Technical Headquarters
Mitsubishi Heavy Industries, Ltd.
6-22, 4-chome, Kan-on-shin-machi, Nishi-ku, Hiroshima 733-8553, Japan
kazuhiro takeda@mhi.co.jp
Akira Inoue
Division of Industrial Innovation Sciences
Graduate School of Natural Science and Technology
Okayama University
3-1-1, Tsushima-naka, Okayama 700-8530, Japan
inoue@suri.sys.okayama-u.ac.jp
Received April 2006; revised July 2006

Abstract. This paper proposes a new design method of a pressure control system in
a furnace of a thermal power plant boiler. To improve the control performance of the
pressure control system in a furnace controlled by using self-tuning I-P controllers, a
feed-forward compensator is introduced. Simulation results are illustrated to show the
effectiveness of our proposed method.
Keywords: Pressure control system, Furnace, Thermal power plant, Boiler, I-P control,
Feed-forward compensator, Self-tuning, Generalized predictive control

1. Introduction. This paper proposes a new design method of a pressure control system
in a furnace of a thermal power plant boiler. The control objective of the pressure control
system in the furnace is to keep the pressure in a thermal power plant boiler slightly
lower than atmosphere pressure and to make combustion in the furnace stabilize. Since
the inlet gas flow changes depending on the quantity of fuel supplied to combustion
control, the deviation in the pressure in the furnace caused by change of the supplied
fuel has to be controlled. In the thermal power plant boiler the pressure control in
the furnace, automatic combustion control, feed water control, and steam temperature

371
372 T. SATO, K. TAKEDA AND A. INOUE

control are complex systems and interfere with each other. Hence, it is not easy to design
control parameters optimally. Especially, the design of a feedforward compensator used
to improve control systems has been adjusted by engineers on work-sites and has not been
designed optimally.
A pressure control system in a furnace has been controlled by using a proportional
and integral compensator mainly. The control performance of I-P control is decided by
the selection of PI parameters, but there is no optimal method to select the PI param-
eters. Hence, using GPC-based design methods [1, 2, 3] and an autotuning method of
I-P controllers in the pressure control system has been proposed [4]. However, because
the pressure control system has multivariables, interactions between variables, nonlinear-
ity and uncertain time delays, an usual I-P control system is not enough to control the
pressure control system. To resolve the problem, the pressure control system is approx-
imated by serially connected two independent first-order plus dead-time models. Then,
two I-P controllers are designed for each of the first-order plus dead-time models, and
interactions between variables are treated as disturbances. Because the pressure control
system is time-variant and has nonlinearity, to approximate the pressure control system
by time-invariant linear models, the values of the coefficients of the models are up-dated
on-line.
Because the design method [4] with the interactions as disturbance needs to be de-
signed to reduce the influence of disturbance, the self-tuning I-P control system has to
be designed based on an extended GPC so that a disturbance response can be redesigned
independently to a reference response [5]. On the other hand, in a pressure control system
in a furnace, inlet gas flow can be used in pressure control in order to reduce the variation
in a pressure due to the variation in the inlet gas flow. Hence, to reduce the effect of the
interference, the self-tuning I-P control system is extended using a feed-forward compen-
sator in this paper. Two design methods of the feed-forward compensator are proposed;
one method is based on the statics of a pressure control system, and the other method is
based on the dynamics of valves. In section 6 the effectiveness of the proposed method
is confirmed by applying the proposed method to a simulator of an actual process. From
simulation results, two design methods of a feed-forward compensator are compared.

Feed-forward compensator
u1 Cf f u3

PC w2
F1 P 1
w1 FC
1 u2
u4
u1

Air
FDF FCV PCV IDF
Furnace Chimney
Fuel

Figure 1. Pressure control system in a furnace of a thermal power plant boiler

 DESIGN OF A PRESSURE CONTROL SYSTEM 373

2. Pressure Control System. A pressure control system in a furnace of a thermal

power plant boiler (herein after, abbreviated as “plant”) is shown in Figure 1.
The plant consists of a furnace, two valves FCV and PCV, two I-P controllers FC and
PC, a forced draft fan (FDF), and an induced draft fan (IDF). In the plant, fuel and
combustion air are sent to the furnace and are burned, then exhaust gas is discharged.
FDF sends required air to the furnace forcefully to obtain combustion in specified pressure,
and IDF induces exhaust gas from the furnace and discharges the exhaust gas to a chimney.
FC and PC operate FCV and PCV respectively in order to make inlet gas flow and the
pressure in the furnace follow each given reference value.
The dynamics of the plant is described by the following equations [4].
dH(t)
= F1 (t) − F2 (t) (1)
dt
RT
V P (t) = H(t) (2)
M
F1 (t) = Cv1 (P1 − P (t))(x1 (t) + 0.5) (3)
F2 (t) = Cv2 (P (t) − P2 )(x2 (t) + 0.5) (4)
where, the symbols in (1) ∼ (4) are described in Table 1 and Table 2, and flow[Nm3 /h] is
converted into mass[kg/s] using Alp shown in Table 2.

Table 1. Symbols in the pressure control system in the furnace (in equa-
tions (1) ∼ (4))

H(t)[kg] gas hold up

F1 (t)[kg/s] inlet gas flow
F2 (t)[kg/s] outlet gas flow
P (t)[mmH2 O] pressure in a furnace
V [m3 ] volume of a furnace
M [kg/kg-mol] gas molecular weight in a furnace
R[J/kg-mol · K] gas constant value
T [K] average temperature in a furnace
Cvi [ kg
s
/mmH2 O] flow coefficient
xi (t), Xi (s) valve opening rate

The dynamics of FCV and PCV are given as follows:

e−L1 s
(FCV) X1 (s) = U1 (s) (5)
1 + T1 s
e−L2 s
(PCV) X2 (s) = U2 (s) (6)
1 + T2 s
where, Li and Ti are a dead-time and time-constant respectively. Ui (s)(= L{ui (t)}) is the
output signal of a controller and Xi (s) = L{xi (t)}, where L denotes Laplace transform.

3. Design of I-P Controllers. The design method of I-P controllers are given in this
section. The PI parameters of the I-P controllers are designed based on GPC laws [4]. To
apply linear control theory to a plant having multivariables, interference between variables
374 T. SATO, K. TAKEDA AND A. INOUE

Table 2. Constant values of a pressure control system in a furnace

gas constant value (R) 8314.41

gas molecular weight (M ) 28.97
average temperature (T ) 600
volume of a furnace (V ) 1000
upstream pressure of FCV (P1 ) 180
downstream pressure of PCV (P2 ) −220
flow coefficient of FCV (Cv1 ) 110
flow coefficient of PCV (Cv2 ) 110
flow unit conversion coefficient (Alp) 0.000359

and nonlinearity, the plant is approximated by two independent first-order plus dead-
time models, first. And then, the plant is controlled by using self-tuning I-P controllers
including parameter identification.

3.1. Approximated plant model and I-P controllers. To obtain approximated lin-
ear plant models regarding the interference as disturbance, the following equations are
considered.
ξi [k]
Ai (z −1 )yi [k] = Bi (z −1 )ui [k − 1] + (7)
∆
Ai (z −1 ) = 1 + ai,1 z −1 (8)
−1 −1 −mi
Bi (z ) = bi,0 + bi,1 z + · · · + bi,mi z (9)
−1
∆ = 1−z (10)
(i = 1, 2)
where, y1 [k] is the sampled normalized value of F1 (t) at the range of 0 to 2.50 × 104 , y2 [k]
is the sampled normalized value of P (t) at the range of −100 to 100, u1 [k] and u2 [k] are
the control inputs to FCV and PCV respectively, ξi [k] is a disturbance which is Gaussian
white noise, and z −1 denotes the backward shift operator. When dead-times in discrete-
time domain are a positive integer, the leading elements of the polynomial Bi (z −1 ) are 0
[1].
The I-P controllers in a digital velocity form are expressed as follows.
∆u1 [k] = C1 (1)w1 [k] − C1 (z −1 )y1 [k] (11)
∆u2 [k] = C2 (z −1 )y2 [k] − C2 (1)w2 [k] (12)
Ts
Ci (z −1 ) = Kpi {(1 + ) − z −1 } (13)
TIi
where, wi [k] is a reference input to be followed by yi [k], and C1 (z −1 ) and C2 (z −1 ) are the
I-P compensators in FC and PC, respectively. Kpi is the proportional gain, TIi is the
integral time, and Ts is the sampling time. The PI parameters are designed so that GPC
laws are approximated by the I-P controllers.

3.2. Derivation of GPC laws. To obtain the PI parameters based on GPC, GPC laws
are derived. The GPC laws to be approximated by the I-P controllers are derived by
 DESIGN OF A PRESSURE CONTROL SYSTEM 375

minimizing the following cost functions:

"N Nu1
#
X 1 X
J1 [k] = E {y1 [k + j] − w1 [k]}2 + λ1 {∆u1 [k + j − 1]}2 (14)
j=1 j=1
"N Nu2
#
X2 X
2 2
J2 [k] = E {w2 [k] − y2 [k + j]} + λ2 {∆u2 [k + j − 1]} (15)
j=2 j=1

where, E[·] expresses an expected value over ξi [k], Ni is predictive horizon, Nui is control
horizon and λi is the weighting factor of the control input.
To derive the GPC laws, Diophantine equations (16) and (19) are solved, and the
polynomials Fi,j (z −1 ), Ri,j (z −1 ) and Si,j (z −1 ) are calculated.
1 = Ei,j (z −1 )∆Ai (z −1 ) + z −j Fi,j (z −1 ) (16)
Ei,j (z −1 ) = 1 + ei,j,1 z −1 + · · · + ei,j,j−1 z −(j−1) (17)
−1 −1
Fi,j (z ) = fi,j,0 + fi,j,1 z (18)
−1 −1 −1 −j −1
Ei,j (z )Bi (z ) = Ri,j (z ) + z Si,j (z ) (19)
Ri,j (z −1 ) = ri,0 + ri,1 z −1 + · · · + ri,j−1 z −(j−1) (20)
−1 −(mi −1)
Si,j (z ) = si,j,0 + · · · + si,j,mi −1 z (21)
Then, the predictive output at k + j step is obtained as
yi [k + j] = ŷi [k + j] + ²i [k + j] (22)
where
ŷi [k + j] = Ri,j (z −1 )∆ui [k + j − 1] + fi,j [k] (23)
fi,j [k] = Fi,j (z −1 )yi [k] + Si,j (z −1 )∆ui [k − 1] (24)
²i [k + j] = Ei,j (z −1 )ξi [k + j]. (25)
The vector form of the optimal predictive output ŷi [k + j] is given as follows.
ŷ i [k] = Ri ∆ui [k] + f i [k] (26)
where
⎡ ⎤ ⎡ ⎤
ŷi [k + 1] ∆ui [k]
⎢ ŷi [k + 2] ⎥ ⎢ ∆u [k + 1] ⎥
ŷ i [k] = ⎢ . ⎥ , ∆ui [k] = ⎢ . i ⎥ (27)
⎣ .. ⎦ ⎣ .. ⎦
ŷi [k + Ni ] ∆ui [k + Nui − 1]
⎡ ⎤
ri,0 0 ... 0
⎡ ⎤ ⎢ ... .. ⎥
fi,1 [k] ⎢ ri,1 ri,0 . ⎥
⎢ .. .. ⎥
⎢ fi,2 [k] ⎥ ⎢ . . 0 ⎥
f i [k] = ⎢⎣ ...
⎥ , Ri = ⎢
⎦ ⎢ r
⎥
⎥ (28)
r
⎢ i,Nui −1 i,Nui −2 r i,0 ⎥
fi,Ni [k] ⎢ .. .. .. ⎥
⎣ . . . ⎦
ri,Ni −1 ri,Ni −2 . . . ri,Ni −Nui
376 T. SATO, K. TAKEDA AND A. INOUE

Using (26) ∼ (28), the vector forms of (14) and (15) are described by the following
equations.
J1 [k] = E[(R1 ∆u1 [k] + f 1 [k] − w1 [k])T (R1 ∆u1 [k] + f 1 [k] − w1 [k])
+λ1 ∆u1 [k]T ∆u1 [k]] + E[²T1 [k]²1 [k]] (29)
J2 [k] = E[(w 2 [k] − R2 ∆u2 [k] − f 2 [k])T (w2 [k] − R2 ∆u2 [k] − f 2 [k])
+λ2 ∆u2 [k]T ∆u2 [k]] + E[²T2 [k]²2 [k]] (30)
where
⎡ ⎤ ⎡ ⎤
wi [k] ²i [k + 1]
⎢ wi [k] ⎥ ⎢ ²i [k + 2] ⎥
wi [k] = ⎢
⎣ ...
⎥ , ²i [k] = ⎢
⎦ ⎣ .. ⎥
⎦ (31)
.
wi [k] ²i [k + Ni ]
Because the rearranged cost functions (29) and (30) are quadratic in ∆ui [k], the optimal
solutions with respect to ∆ui [k] are obtained by
∆u1 [k] = (RT1 R1 + λ1 I)−1 RT1 (w1 [k] − f 1 [k]) (32)
∆u2 [k] = (RT2 R2 + λ2 I) −1
RT2 (f 2 [k] − w 2 [k]). (33)
Using Receding Horizon, the following GPC laws are obtained.
Gp1 (z −1 )∆u1 [k] = P1 w1 [k] − Fp1 (z −1 )y1 [k] (34)
Gp2 (z −1 )∆u2 [k] = Fp2 (z −1 )y2 [k] − P2 w2 [k] (35)
where, the coefficient polynomials of the GPC laws are given by the following equations.
Pi = pi,1 + pi,2 + · · · + pi,Ni (36)
−1 −1 −1
Fpi (z ) = pi,1 Fi,1 (z ) + · · · + pi,Ni Fi,Ni (z ) (37)
Spi (z −1 ) = pi,1 Si,1 (z −1 ) + · · · + pi,Ni Si,Ni (z −1 ) (38)
Gpi (z −1 ) = 1 + z −1 Spi (z −1 ) (39)
£ ¤ £ ¤
pi,1 pi,2 . . . pi,Nui = 1 0 . . . 0 (RTi Ri + λi I)−1 RTi (40)

3.3. Design of PI parameters. To design the I-P controllers based on the GPC laws,
the PI parameters are calculated so as to satisfy the following relations.
P1
' C1 (1) (41)
Gp1 (z −1 )
Fp1 (z −1 )
' C1 (z −1 ) (42)
Gp1 (z −1 )
P2
' C2 (1) (43)
Gp2 (z −1 )
Fp2 (z −1 )
' C2 (z −1 ) (44)
Gp2 (z −1 )
However, these equations are not always satisfied because Gp1 (z −1 ) and Gp2 (z −1 ) are
high-order polynomials. Hence, Gp1 (z −1 ) and Gp2 (z −1 ) are replaced by steady state gains,
 DESIGN OF A PRESSURE CONTROL SYSTEM 377

and the PI parameters are calculated [4, 7]. νi is defined by

∆
νi = Gpi (1) (45)
(i = 1, 2)
where, Gpi (1) denotes the steady-state gain of Gpi (z −1 ). Using νi , the PI parameters are
calculated by solving the following equation.
Fpi (z −1 )
Ci (z −1 ) = (46)
νi
Consequently, the PI parameters are calculated as
Kpi = −f˜pi,1 (47)
f˜pi,1
TIi = Ts (48)
f˜pi,0 + f˜pi,1
where
Fpi = f˜pi,0 + f˜pi,1 z −1 (49)
Pi = Fpi (1). (50)

4. Design of a Feed-forward Compensator. To reduce the influence of interference,

a feed-forward compensator is designed. The pressure in a furnace often oscillates by
changing the reference value of the inlet gas flow. Although the variation in the pressure
is controlled by PC, the variation cannot be reduced quickly by PC only in the case that
a dead-time is long. Hence, the variation caused by changing the inlet gas flow is reduced
by using a feed-forward compensator.
It follows from Figure 1 that the control input u4 [k] to be sent to PCV consists of the
output signal u3 [k] of the feed-forward compensator Cf f (z −1 ) and the output signal u2 [k]
of PC.
u4 [k] = u2 [k] + u3 [k] (51)
u3 [k] = Cf f (z −1 )u1 [k] (52)
On the other hand, in the conventional method [4], because a feed-forward compensator
is not used, u4 [k] is equal to u2 [k] as follows.
u4 [k] = u2 [k]. (53)
The problem is how to design Cf f (z −1 ) to reduce the variation effectively. In this
section two design methods of Cf f (z −1 ) are proposed. One uses the static properties of
the plant, and the other uses the dynamics of the valves In section 6 the effectiveness of
the proposed design methods are compared.

4.1. Design based on valve opening ratio in steady state. Using the static re-
lationship between u1 [k] and u2 [k], the feed-forward compensator Cf f (z −1 ) is designed
as
u2s
Cf f (z −1 ) = (54)
u1s
378 T. SATO, K. TAKEDA AND A. INOUE

where, u1s and u2s represent the steady state values of u1 [k] and u2 [k] without a feed-
forward compensator, respectively. It follows from (54) that a feed-forward compensator
is designed as the ratio of the valve opening in steady state.
4.2. Design based on valve dynamics. It is assumed that the relationship among
u1 [k], u2 [k] and y2 [k] is illustrated in Figure 2, and the values of the time-constant of
the valves are known. It follows from Figure 2 that u1 [k] is utilized to make y1 [k] follow

feed-forward compensator
Cf f (s) u3
u1
w2 PC1
w1 FC1 u2 +
u4 +
u1
Gp (s)+ P
Gd (s) +
plant model
Figure 2. Block diagram including a feed-forward compensator

the reference value w1 [k], Gp (s) is regarded as the transfer function from u2 (t) to P (t),
and Gd (s) is regarded as the transfer function from u1 (t) to P (t). Hence, a feed-forward
compensator is designed using Gd (s) and Gp (s) [6]. To reduce the variation in the pressure
caused by changing the inlet gas flow, the feed-forward compensator is designed as follows.
Gd (s)
Cf f (s) = − (55)
Gp (s)
However, it is difficult to identify Gd (s) and Gp (s) directly because the plant is a
nonlinear system. Hence, Gp (s) and Gd (s) are approximated by first-order plus dead-
time. Then, the feed-forward compensator is designed using the approximated transfer
functions.
Using the step response of the pressure in a furnace, Gd (s) and Gp (s) are obtained. A
simulated step response is shown in Figure 3. u1 [k] and u2 [k] were changed after 20[s] and

Figure 3. Step response of the pressure by changing u1 and u2

 DESIGN OF A PRESSURE CONTROL SYSTEM 379

80[s], respectively. As can be seen, Gp (s) and Gd (s) can be approximated by first-order
plus dead-time. Hence, Gp (s) and Gd (s) are approximated by the dynamics of PCV and
FCV, respectively.
1
Gd (s) = (56)
1 + T1 s
1
Gp (s) = − (57)
1 + T2 s
where, the dead-times of valves are not included. Consequently using (56) and (57),
Cf f (s) can be designed by
1 + T2 s
Cf f (s) = . (58)
1 + T1 s
Since a controlled system is designed as a discrete-time control system, the designed
feed-forward compensator (58) is transformed into discrete-time domain.
Notice that the following assumption is required to design a feed-forward compensator
using the dynamics of valves.
Assumption A.
A.1: The values of the time-constant of valves are known.
The time-constant of valve can be obtained by pre-experiment.

5. Algorithm of Proposed Design Method. To obtain the coefficient polynomials

Ai [z −1 ] and Bi [z −1 ] of the approximated linear model, a recursive least square identifica-
tion law [8] is utilized. And then, the PI parameters are calculated using obtained plant
parameters. The plant parameters âi,1 [k], b̂i,0 [k], · · · , b̂i,mi [k] are calculated by using the
following equations.
Γi [k − 1]ψ i [k − 1]
θ̂ i [k] = θ̂ i [k − 1] + T
εi [k] (59)
1 + ψ i [k − 1]Γi [k − 1]ψ i [k − 1]
λe,i Γi [k − 1]ψ i [k − 1]ψ Ti [k − 1]Γi [k − 1]
Γi [k] = Γi [k − 1] + (60)
1 + λe,i ψ Ti [k − 1]Γi [k − 1]ψ i [k − 1]
T
εi [k] = ∆yi [k] − θ̂ i [k − 1]ψ i [k − 1] (61)
θ̂ i [k] = [âi,1 [k], b̂i,0 [k], · · · , b̂i,mi [k]]T (62)
T
ψ i [k − 1] = [−∆yi [k − 1], ∆ui [k − 1], · · · , ∆ui [k − mi − 1]] (63)
Γi [0] = αe,i I (0 < αe,i < ∞) (64)
where, λe,i is the forgetting factor (0 < λe,i < 2), and Γi [k] is the estimated covariance
matrix.
The algorithm of the proposed design method is summarized as follows.
1. The design parameters of GPC and the parameter identification law are set, and a
feed-forward compensator is designed using (54) or (58).
2. Plant outputs are sampled.
3. Using (59) ∼ (63), the plant parameters of an approximated linear model are ob-
tained.
4. Using the obtained plant parameters, the PI parameters of I-P controllers are calcu-
lated based on GPC laws.
380 T. SATO, K. TAKEDA AND A. INOUE

5. Control inputs are calculated and are inputted into a plant.

6. Return to 2.

6. Simulation. In order to illustrate the effectiveness of the proposed method, simulation

results are shown.
The initial values of the inlet gas flow and the pressure in the furnace are set as 1.10 ×
10 [Nm3 /H] and −20[mmH2 O] respectively, and the initial value of the gas hold up is
4

404[kg]. The simulation length Tmax is set as 600[s], and the sampling time Ts is set as
1[s]. The reference value of the pressure (w2 [k]) is constant as −20[mmH2 O]. In order to
confirm the variation in the pressure caused by changing the inlet gas flow, the reference
value of the inlet gas flow (w1 [k]) is changed as shown in Table 3. The dynamics of the
valves are given in Table 4.

Table 3. Reference value of inlet gas flow

Time[s] 0 ∼ 50 50 ∼ 300 300 ∼ 600

w1 [k][Nm3 /H] 11,000 15,000 11,000

Table 4. Dynamics of valves

Time-constant Dead-time
FCV 2 (T1 ) 5 (L1 )
PCV 10 (T2 ) 1 (L2 )

To design self-tuning I-P controllers and a feed-forward compensator, the plant is ap-
proximated by two linear time-variant models, and the plant parameters of approximated
models are estimated by using a recursive least squares method. The PI parameters of the
I-P controllers based on GPC laws are calculated using the estimated plant parameters
at every sampling step. Because the main purpose of this simulation is to evaluate the
effectiveness of introducing a feed-forward compensator, the dead-times of the plant are
assumed to be known, and b1,j (j = 0, · · · , 4) and b2,0 are set as 0. The initial values
of the estimated plant parameters are set as â1,1 [0] = −0.5, b̂1,5 [0] = 0.1, â2,1 [0] = −0.5,
b̂2,1 [0] = 0.1. The recursive least square identification law having reset with the forget-
ting factor 0.99 is used, and the initial value of an estimated covariance matrix is set
as 10I. The design parameters of GPC are set as N1 = N2 = 10, Nu1 = Nu2 = 1,
λe,1 = λe,2 = 10. Simulation was conducted by using three control methods; one is the
conventional method [4], namely, self-tuning I-P controllers without a feed-forward com-
pensator (“Without FF”), the second is self-tuning I-P controllers with the feed-forward
compensator designed based on the statics of the plant (“With Statics-Based FF”), the
third is self-tuning I-P controllers with the feed-forward compensator designed based on
the dynamics of valves (“With Dynamics-Based FF”).
Figures 4 ∼ 13 show simulation results, and Table 5 explains the figures. To evaluate
 DESIGN OF A PRESSURE CONTROL SYSTEM 381

Table 5. Figure number of simulation result

Feed-Forward Inlet gas flow Control Proportional Integral

Compensator & Pressure Inputs Gains Times
Without FF Figure 4 Figure 5 Figure 6 Figure 7
With Statics-Based FF Figure 8 Figure 9 Figure 10 Figure 11
With Dynamics-Based FF Figure 12 Figure 13 Figure 14 Figure 15

control performance, the following performance indexes P Ii are utilized.

Tmax /Ts
X
P Ii = |wi [k] − yi [k]|Ts (65)
k=1
(i = 1, 2)
Calculated index values are shown in Table 6. P I1 using a feed-forward compensator de-
teriorate slightly, but because P I2 using a feed-forward are reduced considerably. Table 6
shows the fact that the use of a feed-forward compensator reduces effectively the varia-
tion in the pressure caused by changing the inlet gas flow. Comparison of two simulation
results using a feed-forward compensators shows that P I1 of With Statics-Based FF is
slightly smaller than P I1 of With Dynamics-Based FF, on the other hand P I2 of With
Dynamics-Based FF is slightly smaller than P I2 of With Statics-Based.

Table 6. Performance Indexes

P I1 P I2
(Inlet gas flow) (Pressure)
Without FF 10.98 9.60
With Statics-Based FF 11.31 3.78
With Dynamics-Based FF 11.81 3.42

Figure 4. Inlet gas flow Figure 5. Control in-

and Pressure without a puts u1 and u2 without a
feed-forward compensator feed-forward compensator
382 T. SATO, K. TAKEDA AND A. INOUE

Figure 6. Kp1 and Kp2 Figure 7. TI1 and TI2

without a feed-forward without a feed-forward
compensator compensator

Figure 8. Inlet gas Figure 9. Control input

flow and Pressure with a u1 and u2 with a statics-
statics-based feed-forward based feed-forward com-
compensator pensator

Figure 10. Kp1 and Kp2 Figure 11. TI1 and TI2
with a statics-based feed- with a statics-based feed-
forward compensator forward compensator
 DESIGN OF A PRESSURE CONTROL SYSTEM 383

Figure 12. Inlet gas Figure 13. Control

flow and Pressure with input u1 and u2 with
a dynamics-based feed- a dynamics-based feed-
forward compensator forward compensator

Figure 14. Kp1 and Kp2 Figure 15. TI1 and TI2
with a dynamics-based with a dynamics-based
feed-forward compensator feed-forward compensator

7. Conclusion. This paper has proposed a new design method of a pressure control sys-
tem in a furnace of a thermal power plant boiler. To reduce the variation in inlet gas
flow caused by changing the pressure in a furnace, a feed-forward compensator has been
introduced. In this paper we have proposed two design methods of the feed-forward com-
pensator; statics-based method and dynamics-based method. To confirm the effectiveness
of the proposed method, simulation results have been shown, and the proposed method
has been compared with a conventional design method.
In this paper, in order to confirm the advantage of the introduction of a feed-forward
compensator and in order to be compared with the conventional design method, I-P
controllers have been designed without future reference trajectories of GPC. Hence, to
improve transient response, the future work is to design a control system including the
future reference trajectories [9].

Acknowledgment. The first author would like to thank Mr. Kazuhiro Kondo for his
help.
384 T. SATO, K. TAKEDA AND A. INOUE

REFERENCES
[1] Clarke, D. W., C. Mohtadi and P. S. Tuffs, Generalized predictive control — Part I and II, Automatica,
vol.23, no.2, pp.137-160, 1987.
[2] Camacho, E. F. and C. Bordons, Model Predictive Control, Springer-Verlag, 1999.
[3] Asano, M. and T. Yamamoto, A design of self-tuning predictive PID controllers, IEICE TRANS.
FUNDAMENTALS, vol.E84-A, no.7. pp.1779-1783, 2001.
[4] Sato, T., K. Takeda and A. Inoue, Self-tuning I-P control based on generalized predictive control
using coprime factorization for a thermal power plant boiler, IEEJ Transactions on Industry Appli-
cations, vol.123, no.10, pp.1132-1140, 2003 (in Japanese).
[5] Inoue, A., A. Yanou and Y. Hirashima, A design of a strongly stable generalized predictive control
using coprime factorization approach, Proc. of the 1999 American Control Conference, pp.652-656,
1999.
[6] Yamamoto, S. and N. Kato, Basis and Application of PID Control, Asakura Syoten, 1997 (in Japan-
ese).
[7] Asano, M., T. Yamamoto, T. Oki and M. Kaneda, A design of neural-net based predictive PID
controllers, Proc. of the IEEE International Conference on Systems, Man and Cybernetics, pp.1113-
1118, 1999.
[8] Goodwin, G. C. and K. S. Sin, Adaptive Filtering Prediction and Control, Prentice-Hall, 1984.
[9] Sato, T. and A. Inoue, Improvement of tracking performance in self-tuning PID controller based
on generalized predictive control, International Journal of Innovative Computing, Information &
Control, vol.2, no.3, pp.491-503, 2006.