Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
0~
Printed in Great Britain. Pergamon Journals Ltd.
~ 1987 International Federation of Automatic Control
Akstraet--Current self-tuning algorithms lack robustness to (1) a nonminimum-phase plant: most continuous-
prior choices of either dead-time or model order. A novel time transfer functions tend to exhibit discrete-time
method--generalized predictive control or GPC--is developed
which is shown by simulation studies to be superior to accepted zeros outside the unit circle when sampled at a fast
techniques such as generalized minimum-variance and pole- enough rate (see Clarke, 1984);
placement. This receding-horizon method depends on predicting (2) an open-loop unstable plant or plant with
the plant's output over several steps based on assumptions
about future control actions. One assumption--that there is a badly-damped poles such as a flexible spacecraft or
"control horizon" beyond which all control increments become robots;
zero--is shown to be beneficial both in terms of robustness and (3) a plant with variable or unknown dead-time:
for providing simplified calculations. Choosing particular values
of the output and control horizons produces as subsets of the some methods (e.g. minimum-variance self-tuners,
method various useful algorithms such as GMV, EPSAC, /~str6m and Wittenmark, 1973) are highly sensitive
Peterka's predictive controller (1984, Automatica, 20, 39-50) to the assumptions made about the dead-time and
and Ydstie's extended-horizon design (1984, IFAC 9th World
Congress, Budapest, Hungary). Hence GPC can be used either approaches (e.g. Kurz and Goedecke, 1981) which
to control a "simple" plant (e.g. open-loop stable) with little attempt to estimate the dead-time using operating
prior knowledge or a more complex plant such as nonminimum- data tend to be complex and lack robustness;
phase, open-loop unstable and having variable dead-time. In
particular GPC seems to be unaffected (unlike pole-placement (4) a plant with unknown order: pole-placement
strategies) if the plant model is overparameterized. Furthermore, and LQG self-tuners perform badly if the order of
as offsets are eliminated by the consequence of assuming a the plant is overestimated because of pole/zero
CARIMA plant model, GPC is a contender for general self-
tuning applications. This is verified by a comparative simulation cancellations in the identified model, unless special
study. precautions are taken.
The method described in this paper--General-
ized Predictive Control or GPC--appears to over-
1. INTRODUCTION come these problems in one algorithm. It is capable
ALTHOUGHSELF-TUNINGand adaptive control has of stable control of processes with variable parame-
made much progress over the previous decade, ters, with variable dead-time, and with a model
both in terms of theoretical understanding and order which changes instantaneously provided that
practical applications, no one method proposed so the input/output data are sufficiently rich to allow
far is suitable as a "general purpose" algorithm for reasonable plant identification. It is effective with
the stable control of the majority of real processes. a plant which is simultaneously nonminimum-
To be considered for this role a method must be phase and open-loop unstable and whose model
applicable to: is overparameterized by the estimation scheme
without special precautions being taken. Hence it
* Received 2 March 1985; revised 7 July 1985; revised 3 March is suited to high-performance applications such as
1986; revised 22 September 1986. The original version of this the control of flexible systems.
paper was not presented at any IFAC meeting. This paper was
recommended for publication in revised form by Associate
Hitherto the principal applied self-tuning
Editor M. Gevers under the direction of Editor P. C. Parks. methods have been based on the "Generalized
t Department of Engineering Science, Parks Road, Oxford Minimum-Variance" approach (Clarke and
OXI 3PJ, U.K. Gawthrop, 1975, 1979) and the pole-placement
Aluminum Company of America, Alcoa Technical Center,
Pittsburgh, PA 15069, U.S.A. algorithm (Wellstead et al., 1979; /~str6m and
AUT 23/2 - A
137
138 1). W. CLARKk et al.
Wittenmark, 1980). The implicit GMV self-tuner If the plant has a non-zero dead-time the leading
(so-called because the controller parameters are elements of the polynomial B(q- ~) are zero. In (1),
directly estimated without an intermediate calcul- u(t) is the control input, .~t) is the measured variable
ation) is robust against model order assumptions or output, and x(t) is a disturbance term.
but can perform badly if the plant dead-time varies. In the literature x(t) has been considered to be
The explicit pole-placement method, in which a of moving average form:
Diophantine equation is numerically solved as a
bridging step between plant identification and the x(t) = C(q- ~)~.(t) (2)
control calculation, can cope with variable dead-
time but not with a model with overspecified order.
where C ( q - 1 ) = I + clq ~ + ' + c,cq -"c.
Its good behaviour with variable dead-time is due
In this equation, ¢(t) is an uncorrelated random
to overparameterization of the numerator dynamics
sequence, and combining with (I) we obtain the
B(q - 1); this means that the order of the denominator
CARMA (Controlled Auto-Regressive and Mov-
dynamics has to be chosen with great care to avoid
ing-Average) model:
singularities in the resolution of the Diophantine
identity. The G P C approach being based on an
explicit plant formulation can deal with variable A(q- l)y(t) = B ( q ~ 1)u(t - I)
dead-time, but as it is a predictive method it can + C(q- l)~(t). (3)
also cope with overparameterization.
The type of plant that a self-tuner is expected to
Though much self-tuning theory is based on this
control varies widely. On the one hand, many
model it seems to be inappropriate for many
industrial processes have "simple" models: low
industrial applications in which disturbances are
order with real poles and probably with dead-time.
non-stationary. In practice, two principal disturb-
For some critical loops, however, the model might
ances are encountered: random steps occuring at
be more complex, such as open-loop unstable,
random times (for example, changes in material
underdamped poles, multiple integrators. It will be
quality) and Brownian motion (found in plants
shown that GPC has a readily-understandable
relying on energy balance). In both these cases an
default operation which can be used for a simple
appropriate model is:
plant without needing the detailed prior design of
many adaptive methods. Moreover, at a slight
increase of computational time more complex pro- x(t) = C ( q - 1)~(t)/A (4)
cesses can be accommodated by GPC within the
basic framework. where A is the differencing operator I - q - 1 .
All industrial plants are subjected to load-dis- Coupled with (1) this gives the CARIMA model
turbances which tend to be in the form of random- (integrated moving-average):
steps at random times in the deterministic case or
of Brownian motion in stochastic systems. To A ( q - l)y(t) = B(q l)u(t - 1)
achieve offset-free closed-loop behaviour given
these disturbances the controller must possess + C ( q - l)~(t)/A.
inherent integral action. It is seen that GPC adopts
an integrator as a natural consequence of its This model has been used by Tufts and Clarke
assumption about the basic plant model, unlike the (1985) to derive GMV and pole-placement self-
majority of designs where integrators are added in tuners with inherent integral action. For simplicity
an ad hoc way. in the development here C ( q - 1 ) is chosen to be 1
(alternatively C-1 is truncated and absorbed into
2. THE CARIMA PLANT MODEL AND OUTPUT the A and B polynomials; see Part II for the case
PREDICTION
of general C) to give the model:
When considering regulation about a particular
operating point, even a non-linear plant generally
admits a locally-linearized model: A ( q - 1 ) y ( t ) = B(q 1)u(t - l) + ~(tffA. (5)
+ Ej¢(t + j)
The polynomial R - E is of degree j and may be
split into two parts:
and substituting for EjAA from (6) gives:
R- E = ~ + rjq -j
y(t + j) = EjBAu(t + j - 1)
+ Fjy(t) + E~(t + j). (7)
so that:
rj -~ fo (1 la)
where Gj(q-1) = EjB.
Note that Gj(q- 1) = B(q- 1)[1 - q-~Fj(q- t)]/
A(q- t)A so that one way of computing G~ is simply Si = f i + 1 - - ~li+ l r i (1 lb)
to consider the Z-transform of the plant's step-
response and to take the first j terms (Clarke and for i = 0 to the degree of S(q-1);
Zhang, 1985).
In the development of the GMV self-tuning
and: R ( q - i ) = E(q-1) + q-Jrj (12)
controller only one prediction ~(t + kit) is used
where k is the assumed value of the plant's dead-
time. Here we consider a whole set of predictions Gj+ I = B(q- 1)R(q- 1). (13)
for which j runs from a minimum up to a large
value: these are termed the minimum and maximum Hence given the plant polynomials A(q -1) and
"prediction horizons". For j < k the prediction B(q-1) and one solution Ej(q-a) and Fj(q-1) then
process .P(t + J l t) depends entirely on available (11) can be used to obtain Fj+~(q -1) and (12) to
data, but for j >/k assumptions need to be made give E j+ l(q- 1) and so on, with little computational
about future control actions. These assumptions effort. To initialize the iterations note that forj = 1:
are the cornerstone of the GPC approach.
1 --- E1.4 + q-1Ft
2.1. Recursion of the Diophantine equation
One way to implement long-range prediction is
and as the leading element of ,4 is 1 then:
to have a bank of self-tuning predictors for each
horizon j; this is the approach of De Keyser
and Van Cauwenberghe (1982, 1983) and of the E1 = 1, F1 = q(1 - ,']).
MUSMAR method (Mosca et al., 1984). Altern-
atively, (6) can be resolved numerically for E i and The calculations involved, therefore, are straightfor-
F i for the whole range of js being considered. ward and simpler than those required when using
Both these methods are computationally expensive. a separate predictor for each output horizon.
Instead a simpler and more effective scheme is to
use recursion of the Diophantine equation so that 3. THE PREDICTIVE CONTROL LAW
the polynomials Ej+I and Fj+I are obtained given Suppose a future set-point or reference sequence
the values of Ej and Fj. [w(t + j); j = 1,2 .... ] is available. In most cases
Suppose for clarity of notation E = E j, R = E j . 1, w(t + j) will be a constant w equal to the current set-
F = Fj, S = Fj+ ~ and consider the two Diophantine point w(t), though sometimes (as in batch process
equations with $ defined as AA: control or robotics) future variations in w(t + j )
would be known. As in the IDCOM algorithm
1 = EA + q-iF (9) (Richalet et al., 1978) it might be considered that a
smoothed approach from the current output Xt) to
1 = R.4 + q-tJ+ 1~S. (10) w is required which is obtainable from the simple
140 D . W . CLARKE et al.
w set- point
jJ Predicted output
//-
J
Y
I t +INu I !.
t-2 t t~l t+N Time t
first-order lag model: polynomials, which means that the fast Diophantine
recursion is useful for adaptive control applications
w(t) = y(t) of the G P C method.
Consider a cost function of the form:
w(t + j ) = ~w(t + j - 1)
+(1 - a ) w j = l , 2 ....
t~j=N 1
where ~ ~ 1 for a slow transition from the current
measured-variable to the real set-point w. G P C is
capable of considering both constant and varying
future set-points.
+ ~ i(j)[Au(t + j - It] 2
j=l
} (14)
which shows that if y(t) is the constant y so that of dimension N U and the prediction equations
A3~t) = 0 then the component Fjlq-1)~t} reduces reduce to:
to f. This, together with the control given by (18),
ensures offset-free behaviour by integral action. ~=G~ii+f
with N 2 = 10 are similar to those of EPSAC Part TABLE l, TRANSIER-FUNCTIONS Ot THE SIMlr
LATED MODELS
lI of this paper gives a stability result for these
settings of the GPC horizons. Number Samples Model
60 o/~ Output (Y )
-la*/°
'~ 400
Systeml~ change System chonge
Controt signal
IOO*/~
0",
-I00"/@
400
60 °/° - Output ( Y
400
control is pole-placement (Wellstead et al., 1979) as 3 and 4, however, the algorithm is entirely ineffective
variations in plant dead-time can be catered for due to singularities in the Diophantine solution.
using an augmented B(q-1) polynomial. To cover This verifies the observations that pole-placement
the range of possibilities of dead-time here 2A(q- ~) is sensitive to model-order changes. In other cases,
and 6B(q -1) parameters were estimated and the even though the response is good, the control is
desired pole-position was specified by a polynomial rather "jittery".
p(q-l) = 1 - 0 . 5 q -1 (i.e. like the G M V case The G P C controller involved the same numbers
without detuning). Figure 4 shows how an adaptive of A and B (2 and 6) parameters as in the pole-
pole-placer which solves the corresponding placement simulation. Default settings of the output
Diophantine equation at each sample coped with and control horizons were chosen with N 1 = I,
the set of simulated models. For cases 1, 2 and 5 N 2 = 10 and NU = 1 throughout, as it has been
the output response is good with less overshoot found that this gives robust performance. Figure 5
than the G M V approach. For the first-order models shows the excellent behaviour achieved in all cases
146 D.W. CLARKEet al.
6 0 °,~ Output ( Y ]
, I
-IOO/o
i t ,
System change Systerr change System change System change 400
Control signal
lOG °/*
0 */*
-IOO°/o
0 400
6O*/o Output ( Y )
- i o °/o
o
Control signal
ooI I I o
. . . . . i I
aoo
by the GPC algorithm. For each new model only when used on a plant which has large dynamic
at most two steps in the set-point were required for variations.
full adaptation, but more importantly there is no Montague and Morris (1985) report a compar-
sign of instability, unlike all the other controllers. ative study of GPC, LQG, GMV and pole-place-
ment algorithms for control of a heat-exchanger
6. CONCLUSIONS with variable dead-time (12-85s, sampled at 6s
This paper has described a new robust algorithm intervals) and for controlling the biomass of a
which is suitable for challenging adaptive control penicillin fermentation process. The controllers
applications. The method is simple to derive and were programmed on an IBM PC in pro-
to implement in a computer; indeed for short Fortran/proPascal. Their paper concludes that the
control horizons GPC can be mounted in a micro- GPC approach behaved consistently in practice
computer. A simulation study shows that GPC is and that "the LQG and GPC algorithms gave the
superior to currently accepted adaptive controllers best all-round performance", the GPC method
Generalized predictive control--Part I 147
being preferred for long time-delay processes, and e = [El¢(t + 1),E2~(t + 2). . . . . EN¢(t + N)] r.
it confirms that GPC is simple to implement and
Clearly as e is a stochastic process the expected value of the
t o use. cost subject to appropriate conditions must be minimized. The
The GPC method is a further generalization of cost to be minimized therefore becomes
the well-known GMV approach, and so can be
Jl = E{(y - w)T(y - w) + )~firfi}.
equipped with design polynomials and transfer-
functions which have interpretations as in the GMV Two different assumptions will be made in the next section
case. The companion paper Part II, which follows, about the class of admissible controllers for the minimization
explores these ideas and presents simulations which of the cost above. Only models of autoregressive type are
considered:
show how GPC can be used for more demanding
control tasks. A(q- l)y(t) = B(q- ~)u(t - 1) + ¢(t)/A.
In calculating Au(t + N 2) the following e x p e c t a t i o n s are zero: For a c o n t r o l l e r of the form Rlq ~lAu{t) = w(t) - S(q ~l,~,t~
EIAult + N 2)~(t + N)] - 0,
EfLAu(t + N 2)~{t + N 1) ~, - 0. 'SO = 'ff'~g i . iO, L,~gt
B.I. E A N D F P A R A M E T E R S
and:
F r o m the recurrence relationships in the m a i n text it follows
that:
ARA + q-ISB~(I - 0.09q- 1).
e i = 1 -- a l e i_ G.lio - 1 -- JlG.til = eiaa
eo = l:[1.O];flo = (1 - aO:[1.9];.l;~ = a , : [ - 0 . 9 ]
The closed-loop pole is within the unit circle a n d therefore the
e 1 = 1 - a l : [ l . 9 ] ; J 2 o = 1 - al{1 - a 0 : [ 2 . 7 1 ] ; closed-loop is stable. In fact it can easily be d e m o n s t r a t e d that
J21 = a d l - a 0 : [ - 1.71] for all first o r d e r processes N2 > 6 B ( q - 1 ) + k - 1 stabilizes an
/ k
e 2 = 1 - a d l - al):[2.71]; o p e n - l o o p stable plant, for s i g n ( Z g i ] = s i g n ( D . C . g a i n ).
\ /
f3o = 1 a l ( l -- at(1 -- a0):[3.439]
.131 = a d 1 - al(1 - a 0 ) : [ - 2 . 4 3 9 ] -
N2=3
B.2. C O N T R O L L E R PARAMETERS
go = bo: [1.0]; gt 1 = b~: [2.0]
Au(t) = [11.41w - 34.857y{t) + 23.447y(t - 1)
gl = bo{l - al) + bl:[3.9];g22 = bt{1 - a 0 : [ 3 . 8 ]
- 52.104Au(t - 1)]/58.591
g2 = b0(l - al(1 - a~)) + bl(1 - at):[6.51];
g33 = bl( 1 - a d l - a0):[5.42]. and:
A s s u m i n g N U = 1, the effect of N 2 on the control c a l c u l a t i o n
and the closed-loop pole p o s i t i o n is n o w examined: ARA +q ISB=*.(1-0.416q-1).