Professional Documents
Culture Documents
00
Printed in Great Britain. PergamonJournalsLtd.
© 1987InternationalFederationof AutomaticControl
Abstract--The original GMV self-tuner was later extended example, in process control it is generally found
to provide a general framework which included feedforward that energetic control signals are undesirable, a
compensation and user-chosen polynomials with detuned mod-
el-reference, optimal Smith predictor and load-disturbance tail- slowly responding loop being prefered, and plant
oring objectives. This paper adds similar refinements to the models are poorly specified in terms of dead-
GPC algorithm which are illustrated by a set of simulations. time and order as well as their transfer function
The relationship between GPC and LQ designs is investigated
to show the computational advantage of the new approach. The parameters. Emphasis is therefore placed on robust
roles of the output and control horizons are explored for and consistent performance despite variations in
processes with nonminimum-phase, unstable and variable dead- quantities such as dead-time and despite sustained
time models. The robustness of the GPC approach to model
over- and under-parameterization and to fast sampling rates is load-disturbances. High-performance electromech-
demonstrated by further simulations. An appendix derives anical systems tend to have well-understood models
stability results showing that certain choices of control and though often with lightly-damped poles, and the
output horizons in GPC lead to cheap LQ, "mean-lever', state-
dead-beat and pole-placement controllers. control requirement is for fast response, accepting
the fact that the actuation might saturate. It is
1. I N T R O D U C T I O N doubtful whether a single criterion as in (1) can
THE BASIC GPC method developed in Part I is a deal with such a wide range of problems, so to
natural successor to the GMV algorithm of Clarke create an effective general-purpose self-tuner it is
and Gawthrop (1975) in which a cost-function of essential to be able to adapt the basic approach by
the form: the incorporation of "tuning-knobs".
The GMV design was developed into a useful
JoMv = E{(y(t + k) -- w(t))2 + 2uZ(t)[t} (1) self-tuning algorithm by the addition of user-chosen
transfer functions P(q-1), Q(q-1) and an observer
was first defined and minimized. The cost of (1) is polynomial T(q-1). These time-domain perfor-
single-stage and so it is found that effective control mance-oriented "knobs" allow the engineer to tac-
depends on knowledge of the dead-time k of the kle different control problems within the same
plant and for nonminimum-phase plant stability overall scheme. For example, Gawthrop (1977) and
requires a nonzero value of 2. The use of long- Clarke and Gawthrop (1979) showed that model-
range prediction and a multi-stage cost in GPC following, detuned model-following, and optimal
overcomes the problem of stabilizing a nonmini- Smith prediction were interpretations which could
mum-phase plant with unknown or variable dead- be invoked as well as the original control weighting
time. concept of (1). Clarke (1982) and Tufts (1984) give
The relative importance of controller perfor- further examples of the use of these polynomials in
mance criteria varies with the application area. For practice.
This paper introduces similar polynomials to
* Received 2 March 1985; revised 7 July 1985; revised 3 March GPC for specifying a desired closed-loop model
1986; revised 22 September 1986. The original version of this and for tailoring the controlled responses to load
paper was not presented at any IFAC meeting. This paper was
recommended for publication in revised form by Associate disturbances, and the derivation of GPC is
Editor M. Gevers under the direction of Editor P. C. Parks. expanded to include the more general CARIMA
"t Department of Engineering Science, Parks Road, Oxford model. The properties of these extensions are ver-
OX1 3PJ, U.K.
:~Aluminum Company of America, Alcoa Technical Center, ified by simulations, which also show the robustness
Pittsburgh, PA 15069, U.S.A. of the method to a range of practical problems such
149
150 D.W. CLARKEet al.
+ ~ 2(j)[Au(t + j -- 1)] 2 } 1
y(t) ~- -pW(t - k) = M ( q - 1)w(t - k)
2.2. Coloured noise C(q-1) and the design recursion equation outlined in Appendix A. Combi-
polynomial T(q - 1) ning (6) and (7) gives:
Most practical processes have more than one
disturbance or noise source acting on them to give j)(t +jlt) = GjAu(t + j - 1) + FjAuY(t - 1) (8)
an effective plant model: + Fyf(t).
0 °/* I ! !
0 400
05
Feed-forward signor
O *A I I I I . • , I I
400
The first plant simulated was the third-order hand, was unaffected by the change made in P.
oscillator: For the second example shown in Fig. 2, consider
the double-oscillator plant whose transfer function
(1 + sX1 + s2)Xt) = u(t) + d(t) + v(t), is given by:
is assumed where Av(t + j) = 0 for N2 > j > 0. 4.2. The effect of T(q -1)
For the second half of the simulation P(q-1) A second-order plant with time-delay was simu-
was set to ( 1 - 0.8q-1)/0.2 at t = 190. Both the lated where:
disturbance rejection and the set-point responses
were thereby detuned and the overshoot removed (1 + 15s + 50s2)3~t) = e-2Su(t) + lOd(t)
altogether. Feed-forward rejection, on the other + (1 + 15s + 50s2)dc~t) + (1 + 15s + 50s2)~(t)
AUT 2 3 / 2 - S
154 D . W . CLARKE et al.
~ A A ^ . , A A ^ . ~
-10%
, O v " v , VV'w . . . . l I
400
iooo/.,,.illControt signor
0* I
I
/
J
_100°/o ] I •
0 400
in which ~(t) was an uncorrelated random sequence whether G P C suffers from this problem:
with zero mean and RMS value of two units for
210 < t < 240. A step-disturbance dcy of three units (1 + 10s)3j~t) = u(t).
was added to the output at 150 < t < 180. Another
step-disturbance dcu, exciting all of the modes of Five A and five B parameters were estimated; N1
the plant, of + 10 units was added at 120 < t < 150 was chosen to be 1 and N 2 was initially set to 10
and 270 < t < 300. Two step changes in set-point but doubled at every upward-going step in w(t) to
at the end of a period of regulation were employed 50. A sampling time of 1 s was chosen; note that
to see the effect of disturbances on the servo the settling time of the plant is about 160s. Figure
performance of the controller, and three A and four 5 shows that the initial control was stable but had
B parameters were estimated with an assumed delay a small ringing mode and attained the imposed
of unity. saturation limits. When an output horizon of 20
The initial set-point response and subsequent samples was chosen this mode was removed. At
load-disturbance rejection were good, as seen in N 2 = 40 (the rise-time of the plant) the control was
Fig. 3. The rejection of dey, on the other hand, was much smoother. Increasing the horizon of output
very active initially and inconsistent in the second prediction to the settling time of the plant
change of load. This was due to dynamic parameter (N2 = 160) caused the speed of the closed-loop
changes caused by not estimating parameters asso- under this condition to be almost the same as that
ciated with the noise structure. Subsequent behavi- of the open-loop, verifying the "mean-level" theory
our based on the poor model was not very good of Appendix B. In all cases, then, the responses
as the control was far too active. were smooth despite the rapid sampling.
In the second simulation shown in Fig. 4, T(q - 1)
was chosen to be (1 - 0.8q-1). Note that although 4.4. Over-parameterization
T improved the disturbance rejection of the closed- One of the problems with many adaptive control
loop it had no effect on the set-point response. In schemes is that an exact knowledge of the model
addition, since the parameter estimator was better order is required; of particular interest is the ability
conditioned in the second case, the final set-point to over-parameterize the plant parameter estimator
responses were almost identical to the initial ones. in order to model the plant well in case of dynamic
changes.
4.3. The effect of N2 and the sampling period A first-order plant was simulated in discrete time:
One of the major criticisms of digital controllers
is that most designs only work well if the sampling (1 - 0.9q 1)34t) = u(t - 1).
period is chosen carefully (approximately 1/4 to
1/10 of the settling-time of the plant). A slow plant Estimation was disabled and the parameters were
with three real poles was chosen to investigate fixed a priori to the required values given below.
Generalized predictive c o n t r o l - - P a r t II 155
iO0 *A Output ( Y )
Set-point (WI
!50°/
0O/o I i I i I
0 40O
Control signal
_,oo
ii1
ioo'A
- t
i,-
I Pi
! I I I I I
0 400
FIG. 3. The control of a plant with additive disturbances (without the Tpolynomial).
I 0 0 */.
Output (Y)
Set-point ( W )
\ /k_
50 "/ W-'v =
0 °/o f, ! ! - I i I I I
400
Control signal
0 400
FIG, 4. The control of a plant with additive disturbances (with the Tpolynomial).
Initially a c o m m o n factor of (1 + 2q-1) was set repeated. Figure 6 shows that in all cases the control
between the estimated A and B polynomials, giving: performance of G P C was unaffected by the c o m m o n
factor.
.~(q-1) = 1 + 1.1q -1 -- 1.8q -2
/~(q - 1) = 1 + 2q - 1.
4.5. Under-parameterization
Most industrial processes are nonlinear and
NU was set to one at the set-point change from 0 therefore may only be approximated by high-order
to 20, to two for the change 20-40, four for the linear models. A good choice of sample-rate, on the
change 4 0 - 2 0 and finally NU = 10 for the change other hand, enables the designer to use low-order
20-0. The c o m m o n root was then moved to - 0 . 5 , models for control: slow sampling masks the high-
0.5 and 2 in succession and the transient test order fast dynamics.
156 D . W . CLARKE et al,
I 0 0 °,~
Set-point (W) Output ( Y }
50 */
0 */* | I I I I I I ,I |
800
Control signal
I00
~-]r F L__/---I_
-IO0
I[I,II. ,Y. I I I
800
50 *A
0 */,
J Ltpul lI,J I
I
40C
Control signal
2 5 */*
J,. ;
ii 11 II-II
0 */
-25*/. I t I I I ! I
0 400
Consider the fourth-order plant: Fig. 7, offset-free control was achieved. The overall
performance was good despite the wrong parame-
(1 + s)2(1 + 3S)2)~t) = u(t) + d(t). terization; the overshoot could have been reduced
using P(q-1) as shown in the previous sections.
A second-order model was assumed and the plant
4.6. Unknown or variable time-delay
was sampled at 1 s intervals; note that the sampling
The G M V design is sensitive to choice of dead-
process was not masking the slightly faster poles.
time; G P C is however robust provided that 6B is
In this case two A and three B parameters were
chosen to absorb any change in the time-delay.
estimated and the assumed time-delay was unity.
Consider the plant:
N2 was set to 10 and N U was set to one. The set-
point sequence was a square wave with a period of (1 - 1.1q- 1))~t) = - ( 0 . 1 + 0 . 2 q - l)u(t - k)
40 samples. Load-disturbances of 10 and 20 units
were added at the marked times and as shown in where k = 1, 2, 3, 4, 5 at different stages in the trial.
Generalized predictive control--Part II 157
IO0%
Set-point (W) Output { Y )
5OO/
0 "/o
O 4o0
Control signal
0% " IL. _ .. __
I00 / I I I I I I
The value of k was changed at the downward- applied to a varying dead-time plant which is a
going steps in set-point increasing initially from requirement of GMV designs.
one to five and then decreasing from five back to The simulations show that GPC can cope with
one again. The adaptive controller estimated two the control of complex processes under realistic
A and six B parameters and a scalar forgetting- conditions. As it is relatively insensitive to basic
factor of 0.9 was employed to enable tracking of assumptions (model order, etc.) about the process,
variations in the dead-time. N 2 was set to 10 and GPC can be easily applied in practice without a
N U to one. The performance of GPC shown in prolonged design phase. These features ensure that
Fig. 8 is good; note that the plant was both the method provides an effective approach to the
nonminimum-phase and open-loop unstable with adaptive control of an industrial plant.
variable dead-time, yet stable control was achieved
with the default settings of this algorithm.
REFERENCES
5. CONCLUSIONS /~strrm, K. J. and B. Wittenmark, (1984). Computer Controlled
This paper has shown that GPC can be equipped Systems--Theory and Design. Prentice-Hall, Englewood
Cliffs, NJ.
with the design features of the well-known GMV Bierman, G. J. (1977). F actorization Methods for Discrete System
approach and given a wide range of possible Estimation. Academic Press, New York.
control objectives, which can be interpreted by its Clarke, D. W. (1982). The application of self-tuning control.
Trans. Inst. M.C., 5, 59-69.
relationship with LQ algorithms based on state- Clarke, D. W. and P. J. Gawthrop, (1975). Self-tuningcontroller.
space models. These results are summarized in Proc. IEE, 122, 929-934.
Table 1. It might seem that there are many possible Clarke, D. W. and P. J. Gawthrop, (1979). Self-tuning control.
Proc. IEE, 126, 633-640.
choices of design parameters in GPC, but the Clarke, D. W., P. P. Kanjilal and C. Mohtadi, (1985). A
table shows that many combinations lead to well- generalised LQG approach to self-tuning control. Int. J.
understood control laws. In practice not all this Control, 41, 1509-1544.
Gawthrop, P. J. (1977). Some interpretations of the self-tuning
flexibility would be required and many processes controller. Proc. lEE, 124, 889-894.
can be effectively controlled using default settings. Kwakernaak, H. and R. Sivan, (1972). Linear Optimal Control
Closer inspection of Table 1 shows that a "large" Systems. Wiley, New York.
Lam, K. P. (1980). Implicit and explicit self-tuning controllers.
value of N 2 is generally recommended and that D. Phil Thesis, Oxford University.
N U and P can then be chosen according to the Peterka, V. (1984). Predictor-based self-tuning control. Automa-
control philosophy appropriate for the plant and tica, 20, 39-50.
Tufts, P. S. 0984). Self-tuning control: algorithms and applica-
the computing power available. Hence the "knobs" tions. D. Phil. Thesis, Oxford University.
can be used to tailor an adaptive controller to Tufts, P. S. and D. W. Clarke, (1985). FAUST: a software
precise specifications, which is of great value in the package for self-tuning control. IEE Conf. "'Control 85",
Cambridge.
high-performance role. In particular, the method Wellstead, P. E., D. Prager, and P. Zanker, (1979). Pole
no longer needs to employ control weighting when assignment self-tuning regulator. Proc. IEE, 126, 781-787.
158 D.W. CLARKE et al.
ioo*/o 7 ~ - •
50 *A
O */*
0 800
Control signoL
too
• ' I t I I I , •
-io¢
800
NU N~ N2 P 2 Plant Controller
?o+ 1~i = Yoi + e~bi + gj+ ltl (A.5)
1 1 10 1 0 s,d "Default"
where ?u~ denotes the ith coefficient of the polynomial F~
1 1 ---,oc 1 0 s,d "Mean-level"
associated with q - 1 Note that the coefficients of B or T with
N2 1 />k P 0 mp Exact model-
indices greater than their respective degrees are zero.
following P = I / M
<N 2 1 ~>k P 0,2 "Detuned" model-
following A P P E N D I X B. S O M E STABILITY RESULTS F O R
N2 1 - , oe 1 >0 s,d LQ infinite-stage L I M I T I N G CASES O F G P C
N 2 -- n + 1 1 -,o~ 1 0 s,d Cheap LQ Consider the plant given in shift-operator form by:
n n >~2n - 1 1 0 o,c State-dead-beat
n n />2n - 1 P 0 o,c Pole-assignment
A(q- l)Ay(t) = B(q - l)Au(t -- 1). (B.I)
2 s,d "Detuned" pole-
assignment
A state-space model of this plant can be written as:
s: stabilizable; d: detectable; o: observable; c: controllable; mp:
minimum-phase. x(t + 1) = Ax(t) + hAu(t) (B.2)
Gj+ l = G'j+IT+ q - J - l F j ÷ r (A.2) then the matrices and vectors in the state-space form are:
-a2 0 1 ... 0
Ej(q l ) = e o + e l q l +...+ei_lq-J+l. 0
The cost-function in the state-space formulation can be where 2 i are the eigenvalues and IAil < 1 for all i # ! and 2~ = 1
written as: and ql and ri are right and left eigenvectors associated with the
particular eigenvalue of A.
N2 The right and left eigenvectors associated with the eigenvalue
J = 3-'~ [x(t + i - 1)TQx(t + i - 1) at 1 are given by:
Measurement update: Note, however, that the choice N U = 1 implies that there will
P*(i) = P(i + 1) - P(i + l)b(2(i) be N2 - 1 time-updates followed by a single full update at the
+ hrp(i + 1)h)-1hTp(i + 1). (B.5) last iteration, giving:
Remark 1. A mean-level controller provides a step in control Clearly if the matrix G is of full rank then Au(t) is unique.
following a step in the set-point which will drive the plant
output exactly to the set-point and hence provide the same
closed-loop dynamics as of the open-loop. Note that steps in
load-disturbance are, however, rejected since the controller
includes an integrator.