Documentos de Académico
Documentos de Profesional
Documentos de Cultura
(1) a nonminimum-phase plant: most continuoustime transfer functions tend to exhibit discrete-time
zeros outside the unit circle when sampled at a fast
enough rate (see Clarke, 1984);
(2) an open-loop unstable plant or plant with
badly-damped poles such as a flexible spacecraft or
robots;
(3) a plant with variable or unknown dead-time:
some methods (e.g. minimum-variance self-tuners,
/~str6m and Wittenmark, 1973) are highly sensitive
to the assumptions made about the dead-time and
approaches (e.g. Kurz and Goedecke, 1981) which
attempt to estimate the dead-time using operating
data tend to be complex and lack robustness;
(4) a plant with unknown order: pole-placement
and LQG self-tuners perform badly if the order of
the plant is overestimated because of pole/zero
cancellations in the identified model, unless special
precautions are taken.
The method described in this paper--Generalized Predictive Control or GPC--appears to overcome these problems in one algorithm. It is capable
of stable control of processes with variable parameters, with variable dead-time, and with a model
order which changes instantaneously provided that
the input/output data are sufficiently rich to allow
reasonable plant identification. It is effective with
a plant which is simultaneously nonminimumphase and open-loop unstable and whose model
is overparameterized by the estimation scheme
without special precautions being taken. Hence it
is suited to high-performance applications such as
the control of flexible systems.
Hitherto the principal applied self-tuning
methods have been based on the "Generalized
Minimum-Variance"
approach (Clarke and
Gawthrop, 1975, 1979) and the pole-placement
algorithm (Wellstead et al., 1979; /~str6m and
1. INTRODUCTION
AUT
23/2
137
138
(1)
(2)
A(q- l)y(t)
I)
B ( q ~ 1)u(t
+ C(q-
l)~(t).
(3)
(4)
l)u(t -
1)
+ C ( q - l)~(t)/A.
1)u(t - l) + ~(tffA.
(5)
(6)
Generalized predictive c o n t r o l - - P a r t I
Subtracting (9) from (10) gives:
EjAAy(t + j) = EjBAu(t + j - 1)
+ Ej(t + j)
R-
y(t + j) = EjBAu(t + j - 1)
+ Fjy(t) +
139
E~(t + j).
(7)
(8)
E = ~ + rjq -j
so that:
~IR + q - J ( q - t S - F + ,4rj) = O.
Clearly then /~ = 0 and also S is given by
Sq( F - ,~rj).
As ,4 has a unit leading element we have:
rj -~ fo
(1 la)
Si = f i + 1 - - ~li+ l r i
(1 lb)
R ( q - i ) = E(q-1) + q-Jrj
(12)
(13)
(9)
(10)
F1 = q(1 - ,']).
The calculations involved, therefore, are straightforward and simpler than those required when using
a separate predictor for each output horizon.
3. THE PREDICTIVE CONTROL LAW
Suppose a future set-point or reference sequence
[w(t + j); j = 1,2 .... ] is available. In most cases
w(t + j) will be a constant w equal to the current setpoint w(t), though sometimes (as in batch process
control or robotics) future variations in w(t + j )
would be known. As in the IDCOM algorithm
(Richalet et al., 1978) it might be considered that a
smoothed approach from the current output Xt) to
w is required which is obtainable from the simple
D . W . CLARKE et al.
140
w set- point
//-
jJ
Predicted
output
t-2
t~l
t +INu
t+N
[--
!.
Time
Projected
controts
w(t) = y(t)
w(t + j ) = ~w(t + j - 1)
+(1 - a ) w
j = l , 2 ....
t~j=N 1
+ ~ i(j)[Au(t + j - It] 2
j=l
(14)
where:
N 1 is the minimum costing horizon;
is the maximum costing horizon, and
).(j) is a control-weighting sequence.
N2
Generalized predictive c o n t r o l - - P a r t I
N1 can be chosen as k or more to minimize
computations. It is found, however, that a large
class of plant models can be stabilized by G P C
with default values of 1 and 10 for N1 and N2. Part
II provides a theoretical justification for these
choices of horizon. For simplicity in the derivation,
below 2(j) is set to the constant 2, N1 to 1 and N 2
to N: the "output horizon".
Recall that (7i models the future outputs:
y(t + 1) = GxAu(t) + Fly(t ) + E ~ ( t + 1)
y(t + 2) = G2Au(t + 1) + F2y(t) + E2~(t + 2)
y(t + N) = GNAu(t + N -
1)
+ FNy(t) + EN~(t + N).
go
gt
141
0
go
G=
gN- 1 gN- 2
go
(16)
i.e.
Jl = {(Gii + f - w)T(Gii + f - w) + 2iiTii}.
The minimization of J1 assuming no constraints
on future controls results in the projected controlincrement vector:
and
f ( t + 2) = q[G2(q- 1) _ q - tg21
-- g20]Au(t) + F2y(t),
(17)
etc.
Note that the first element of fi is Au(t) so that
the current control u(t) is given by:
(15)
(I 8)
1 = Ej(1)A(1)A(I) + Fj(1)
and as A ( 1 ) = 0
then F j ( 1 ) = I
so that
f ( t + j) = Fj)~t) is a signal whose mean value equals
that of y(t). Furthermore, defining F j ( q - i ) to be
Ej(q- 1).~(q- 1) gives:
Fj(q- 1)y(t) = (1 - F~(q- l)A)y(t)
= y(t) - F jAy(t)
142
D . W . CLARKE et al.
where:
i.e.
Au(t + j -
1) = 0
j > NU.
~=G~ii+f
-go
gl
0
go
GI=
0
0
go
gN- 1
gN- 2
isN xNU.
gN - Nu
(20)
(19)
143
144
I). W.
CLARKE
STUDY
et al.
TABLE l, TRANSIER-FUNCTIONS Ot THE SIMlr
LATED MODELS
Number
Samples
Model
1
1 79
I + 10s + 40s 2
2
80-159
160-239
e 2....
f + 10s + 40s 2
e .,7~
I +
240 -319
320-400
l
1 +
1
10s
10s
10s(l + 2.5s)
p(q- 1)
= (1 -
0.5q-1)/0.5
and
Generalized predictive c o n t r o l - - P a r t I
60
Output
o/~
145
(Y )
-la*/
Systeml~ change
'~
System chonge
400
Controt signal
IOO*/~
0",
-I00"/@
400
Output ( Y
60 / -
,1I 11
-ioo/ ~
System ~ change
System~ change
'
t'
System change
'
t
'
System change
'
~oo
Contro I. signaL
400
146
6 0 ,~
-IOO/o
System change
Systerr change
t ,
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
I
aoo
147
1)"r
+ 2 i l r ( G r ( f - w))
Au(t+N-
..... ~(t+N-1),
148
D . W . CLARKE et al.
In calculating Au(t + N
2)~{t + N
.~1 = Y.~,Ji,..'Zg,
ro = Z ( g a 2 ,.'~.~,
r, = ~,,glgli,"~gi
a n d the closed-loop
RAA + q-tSB.
This a p p e n d i x e x a m i n e s the relation of N 2 a n d the closedloop pole position for a simple example. C o n s i d e r a n o n m i n i m urn-phase first-order process (first-order + fractional dead-time):
(1 + a l q - 1 ) y ( t } = (b o + b l q - l l u ( t
0.gq-l)y{t)
(I
2 q - l)ult
--
RAA
+ q-ISB:::>(I
J21 = a d l - a 0 : [ -
1.71]
f3o = 1
a l ( l -- at(1 -- a0):[3.439]
of
+ 2q-1).
or~
16.82Au(t - 1)]/16.21
and:
- 0.09q- 1).
+ q-ISB~(I
e 2 = 1 - a d l - al):[2.71];
roots
ARA
eo = l:[1.O];flo = (1 - aO:[1.9];.l;~ = a , : [ - 0 . 9 ]
the
e 1 = 1 - a l : [ l . 9 ] ; J 2 o = 1 - al{1 - a 0 : [ 2 . 7 1 ] ;
at
I).
B.I. E A N D F P A R A M E T E R S
are
I)
pole-positions
N2=I
A P P E N D I X B. AN E X A M P L E F O R A F I R S T O R D E R
PROCESS
--
~l,~,t~
1) ~, - 0.
11
2)~(t + N)] - 0,
EIAult + N
EfLAu(t + N
PARAMETERS
N2=3
52.104Au(t - 1)]/58.591
and:
- lily(t - 2))
ARA
+q
ISB=*.(1-0.416q-1).
The pole is a g a i n within the unit circle. N o t e that the closedl o o p pole is t e n d i n g t o w a r d s the o p e n - l o o p pole as N 2 increases
(see Part II).