Professional Documents
Culture Documents
2007 3
March, 2007
.
.
.
TP273
XI Yu-Geng1
QIN Hui1
Abstract
Aggregation optimization strategies are effective approach to decreasing the on-line computation cost of
model predictive controllers. But most of the past aggregation strategies are heuristic, whose control quality cannot
keep up with that of the conventional predictive controllers. Starting with the fact that the predictive control algorithm
only applies the first moves to a system, this paper presents an equivalent aggregation strategy. We prove that for the
unconstraint and zero terminal constraint MPC, a proper choice of aggregation matrix can make the control quality of the
aggregation predictive controller comparable to that of the conventional predictive controller. The algorithm for obtaining
the equivalent aggregation matrix is presented, too.
Key words
1
(Model predictive control,
MPC) 70
[1] .
.
[24]
. Mayne [2]
4
.
MPC
2006-7-5
2006-10-16
Received July 5, 2006; in revised form October 16, 2006
(60474002, 60674041), 863
(2006AA04Z173)
Supported by National Natural Science Foundation of
P. R. China(60674002, 60674041) and National Hi-Tech Research
and Development Plan of P. R. China(2006AA04Z173)
1. 200240
1. Institute of Automation, Shanghai Jiao Tong Unversity,
Shanghai 200240
DOI: 10.1360/aas-007-0302
.
.
Zheng[5]
. Ricker [6]
Blocking
. [7]
(1)
A Rnn B Rnm . k
A
2
A
S =
.. ,
.
AN
B
0
B
AB
G=
..
..
.
.
N 1
A
B AB B
N Q > 0R > 0.
(2) N m.
.
.
.
[7]
0
0
..
.
H1s
..
.
HN s
T
Hi = Hi1
s.t.
H1
H11
. .
..
. .
H=
.
. = .
HN
HN 1
303
..
.
U (k) = HV (k)
(3)
T
V (k) = v1T (k) v2T (k) . . . vsT (k) (s <
N ) vi (k) Rm H N m sm
T
T
m sm .
His
(2)
(4)
. [7]
V (k) vi
.
.
vi
s .
[7] .
3
(1)
1. l N > l.
2. H
.
k
(2) .
(3) (2)
(4)
.
(m )
(m ) .
1.
.
304
.
1.
x = 0 Rkd
k < d = a1 a2 ai ad ai
i
xi 0.
. x = 0 a1 x1 + + ad xd = 0.
xi 6= 0
ai =
1
(a1 x1 + +ai1 xi1 +ai+1 xi+1 + +ad xd )
xi
ai xi
.
Rkd k < d
rank( ) = k
.
[ .. ] Rkk Rk(dk) rank( ) = k
1
.
.
0 = [1 ..0].
!
I 0
=
Rdg
0 1
I Rkk 1 R(dk)(gk) d > g > k
.
rank( ) = g. = [ ..0].
0
0 1
2. x = 0
X y = 0
Y X1 = Y .
.
0
x = !
11 2
X =
I
I
.
! y = 0 Y =
1
1 2 1
.
I
X1 = Y .
3. W RN mN m H RN ms
rank(W ) = N mrank(H) = s W
.
[W ..W ]W RN mm W RN m(N mm)
1
rank(H T W2 ) = s m H T W1
H T W .
. W H
rank(H T W ) = s.
.
W [W ..W ] W
RN mm W2 RN m(N mm) .
rank(H T W2 ) = s m H T W2
s m
. H T W2 =
33
[1 2 sm 1 N ms ] i s m
i i =
1i 1 + 2i 2 + + (sm)i sm .
H T W1 [1 2 m ]
1 H T W
1 = 1 2 + + m1 m + m 1 + + s1 sm
1 s 1 i
i rank(H T W ) = s 1
rank(H T W ) = s H T W1
H T W2 .
.
2
k (2) .
U (k) = HV (k)
V (k) (2) (4). H
H RN ms s
s < N m.
(2) (4)
.
(2)
J(k)
=0
U (k)
(GT QG + R)U (k) = GT QSx(k)
(5)
(4)
J(k)
=0
V (k)
(GH)T Q(Sx(k) + GHV (k)) = H T RHV (k)
((GH)T Q(GH) + H T RH)V (k)
= (GH)T QSx(k) (6)
(5) (6)
1. m
N
s > m .
. (5) (6)
(7)
m
. Q R (GT QG + R)
. (GT QG + R)
.
[W ..W ]. 1 3
1
rank(H T W2 ) = s m.
W2 = [v1 v2 vN m ]T N m
(N m m) , . W2
N m m , v1T , v2T ,
T
vN
mm , vi ,
T
T
vi = i,1 v1T + + i,N mm vN
mm . ,
H j [i,1 i,2 i,N mm 0 0
1 0 0]T , j = 1, m, [0 0 1 0 0]T
m 1 , H T W2 j i,1 v1T +
T
T
+ i,N mm vN
= 0, i = N m
mm vi
T
m + 1, , N m. H W2 Rs(N mm)
N m m >> s
H rank(H T W2 ) = s m.
2. m
N W2
m m + 1.
. 1
. 1 W2
viT = i,1 v1T +
T
+ i,N mm vN
mm i = N m m + 1, , N m
s = m 1 H
rank(H T W2 ) = s m = 0
H s = m + 1 H
m + 1 1
rank(H T W2 ) = s m = 1.
m m + 1
.
mrank(H T (GT QG +
R)) < m H T (GT QG + R) = 0
N m m
m
m.
W2
()
m + 1. .
. 2
m m + 1.
m m + 1
.
.
H T (GT QG + R) m .
305
.
1.
1) W = (GT QG + R) W
m W2 = [am+1 ai aN m ].
L1
L2
2) W2 W2 =
.. W2
.
LN m
m {1 , , m } (N m m)
W3 .
3) {1 , , m }
W3
.
(2)
min J(k) = X T QX + U T RU
U
s.t.
X = Sx(k|k) + GU
x(k|k) = x(k)
x(k + N |k) = 0
(8)
AN x(k|k) = AN 1 B AN 2 B
B U
(9)
1 rank([Al1 B, Al2 B, , B]) = n
Al1 B, Al2 B, , B n
306
[Al1 B, Al2 B, , B]
= [h1 , h2 , , hn ].
(9)
[AN 1 B, AN 2 B, , B]
U = U0 K + U
(10)
U [AN 1 B, AN 2 B, , B]U =
AN x(k) U = [(1 AN x(k|k))T , 0, , 0]T
U U0
[AN 1 B, AN 2 B, , B]U0 = 0 2
K
.
[AN 1 B,
N 2
A
B, , B] U0
U k .
(10) (8)
min J(k) = X T QX + U T RU
K
(11)
G1 = GU0 . .
(11)
T
T
(GT
1 QG1 + U0 RU0 )K = G1 QSx(k|k) +
GT
(12)
1 QGU + U0 RU
4 2
H !
H =
I 0
rank([AN 1 B, AN 2 B, , B]H) =
0 H1
n [AN 1 B, AN 2 B, , B]H =
[h1 , h2 , , hn ].
s > n.
(8)
min J(k) = X T QX + U T RU
V (k)
X = Sx(k|k) + GU
x(k|k) = x(k)
x(K + N |k) = 0
U = HV (k)
33
A x(k|k) = [A
N 1
B, AN 2 B, , B]HV
s > n
0
0
V = V0 K +V V V0 K
.
H
V = [(1 AN x(k|k))T , 0, , 0]T V
s U = HV . 2
0
HV0 = U0 H1 HV = U0 H1 K + U .
(13)
min
J(k) = X T QX + U T RU
0
K
(14)
= H1T (GT
1 QSx(k|k) + G1 QGU + U0 RU ) (15)
= Sx(k|k) + GU + GU0 K
s.t.
T
H1T (GT
1 QG1 + U0 RU0 )H1 K
(13)
(8)
(15) (12).
3. m
n N
s n + m s n + m + 1
.
. (15) (12)
T
H1T (GT
1 QG1 + U0 RU0 )K
T
= H1T (GT
1 QG1 + U0 RU0 )H1 K
T
H1T (GT
1 QG1 + U0 RU0 )(H1 K K) = 0
(16)
T
T
(GT
1 QG1 + U0 RU0 ) U0 RU0
( U0 )
T
(GT
1 QG1 + U0 RU0 ) .
2 H1 ( H1
) m m + 1
n n + m
n + m + 1. s m m + 1
T
H1T (GT
1 QG1 + U0 RU0 ) m
.
. 2
n + m + 1
n + m.
3
.
2.
1) 1 H1 .
2) H1 h1 h2 , hn 1
H
..
H1 =
.. ..
.
..
..
.
0
.
.. ..
..
.
.
H = 0 0 1
.. ..
.
. .
.
[2, 2] 100
. 1
2
1.
1 2
. 1
307
Fig. 1
1
The state trajectory with aggregation
()
.
!
1.1
0
0.5 0.1
x(k + 1) =
x(k) +
u(k)
0 0.99
0.2 0.2
B 2
.
5 (13)
Q, R
N = 20. 5 2
(
2 )
!T
I22
0
H=
0
0
0 I22
, 2 36 .
40
4
.
Fig. 2
2
The state trajectory with no aggregation
308
Table 1
33
1
Comparing the control pulses
u1
u1
u2
u2
1
5
10
15
20
25
30
35
40
45
50
-1.883
0.613
-0.057
0.037
0.005
0.004
0.001
0.000
0.000
0.000
0.000
-1.886
0.615
-0.059
0.037
0.005
0.003
0.002
0.000
0.000
0.000
0.000
0.003
-0.002
0.002
0.000
0.000
0.001
-0.001
0.000
0.000
0.000
0.000
-1.293
-0.263
-0.188
-0.062
-0.027
-0.011
-0.004
-0.003
-0.001
0.000
0.000
-1.295
-0.261
-0.188
-0.062
-0.027
-0.011
-0.004
-0.002
-0.001
0.000
0.000
0.002
-0.002
0.000
0.000
0.000
0.000
0.000
-0.001
0.000
0.000
0.000
50 50 .
. 6
.
References
1 Xi Yu-Geng. Model Predictive Control. Beijing: National
Defense Publishing House, 1993
(. . , 1993)
2 Mayne D Q, Rawlings J B, Rao C V, Scokaert P O M. Constrained model predictive control: Stability and optimality.
Automatica, 2000, 36(6): 789814
3 Ohshima M, Hashimoto I, Takamatsu T, Ohno H. Robust
stability of model predictive control. International Chemical
Engineering, 1991, 31(1): 119127
4 Camacho E F. Constrained generalized predictive control.
IEEE Transactions on Automatic Control, 1993, 38(2):
327332
5 Zheng Alex. Reducing on-line computational demands in
model predictive control by approximating QP constraints.
Journal of Process Control, 1999, 9: 279290
6 Ricker N L. Use of quadratic programming for constrained
internal model control. Industrial and Engineering Chemisty
Process Design and Development, 1985, 24(4): 925938
7 Du Xiao-Ning. The Study and Analysis of New Optimization Strategies in Model Predictive Control [Ph.D. dissertation], Shanghai Jiao Tong University, 2001
(. [],
, 2001
8 Liu Bin, Xi Yu-Geng. An aggregation based robust model
predictive controller. Acta Automatica Sinica, 2003, 29(6):
801808
(. MPC .
200329(6): 801808)
9 Zhang Qun-Liang. Study on Design Method for Constraint
Model Predictive Control [Ph.D. dissertation], Shanghai
Jiao Tong University, 2006
(. [],
, 2006)
, 1993
,
. . E-mail:
dwli@sjtu.edu.cn
(LI De-Wei Ph. D. candidate in Automation Institute, Shanghai Jiao Tong
University. He received his B. S. degree
from Shanghai Jiao Tong University in 1993. His research
interest covers the theory and algorithm of model predictive
control, etc. Corresponding author of this paper.)
,
. E-mail:
ygxi@sjtu.edu.cn
(XI Yu-Geng Professor in Automation Institute, Shanghai Jiao Tong University. His research interest covers
the optimization control in the complex industry process, intelligent robot control, and rolling
scheduling.)
,
. E-mail: huiqin@sjtu.edu.cn
(QIN Hui Master student in Automation Institute, Shanghai Jiao Tong
University. His research interest covers
model predictive control and large scale
system theory.)