Decoupling Control Of Multivariable System

Based On Adjoint MatrixP*

Kong YaGuang1 Zhang Fan2 Xue AnKe1
1）
Institute of Information and Control,
HangZhou DianZi University,
Hangzhou P.R.China 310037.
E-mail:[email protected]
2)
Hangzhou Electric Power Bureau,
Hangzhou P.R. China,310009.
E-mail:[email protected]

Abstract decoupling compensator proposed by this article has

satisfaction the dynamic performance.
A decoupling control method is given based on the
adjoint matrix of transfer function. This method can 2. Traditional decoupling control method
realize the entirely decoupling control of the
multivariable system, and needn’t solve the inverse of −1 −1
Let G(z ) , D(z ) respectively is the real object
transfer function, avert the infeasible problem of
~
traditional decoupling control method. −1
and the decoupling transfer function matrix，G(z ) is
−1
Keywords by the diagonal elements of matrix G(z ) composes
diagonal matrix. The traditional diagonal decoupling
Multivariable system,Decoupling control ,Adjoint control method namely must Satisfy such formula:
matrix ~
G(z −1 )D(z −1 ) = G(z −1 ) (1)
1. Introduction −1
If exists G(z ) ， then we can obtain the
The goal of multivariable control causes each decoupling compensator :
variable of the system to be accused only to its given ~

value change, not disturbed by other given values D(z −1 ) = G(z −1 )−1 G(z −1 ) (2)
influences, and has good dynamic performance and From(1),(2),It’s easy to get decoupling
static performance, namely realization decoupling mathematical model , but such decoupling compensator
control. Traditional decoupling control (for example has following several problems generally as follows:
traditional diagonal matrix decoupling law) has simple 1) the compensator which overthrows according to
principle, it’s also easy to get the decoupling the complete decoupling possibly is unable to realize;
mathematical model. But, they often have many 2) regarding certain objects (for example certain non
problems in the implementation process, the decoupling minimum phase systems), the obtained decoupling
possibly cannot be realized, either itself is unstable, −1
transfer function D(z ) possibly is essence unstable;
either model has slightly error namely system appears
3) the error of realizing the decoupling compensator
not stably. In order to avoid the problem in the complete
decoupling method, many articles have all adopted the D(z −1 ) to be very easy to cause the system unstable;
compromise algorithm, namely only realizes not 4) real system usually non linearly,time variant. For
incomplete decoupling or the static decoupling. Such achieve the optimization, the decoupling controller
controller although also can control, but all has the must certainly have the self adjustment ability.
certain applicable scope. This article then attempts to Based on the above question, many papers proposed
produce one complete decoupling method to have the each other decoupling compensator design method. This
universal significance. Through the adjoint matrix of the article introduces the adjoint matrix of opend loop
opened loop transfer function, thus may aim at the transfer function, thus may aim at the characteristic
characteristic polynomial to carry on the design of polynomial to carry on the controller design,transforms
controller . The simulation experiment indicates that the the coupling multivariable system to the same transfer
function N single variable system control, has avoided
*This work is supported by China National Science the above question completely.
Foundation Grant:60434020

0-7803-9514-X/06/$20.00 ©2006 IEEE ICIEA 2006

3. Design adjoint matrix decoupling Theorem 1: Real coefficient multinomial
^
compensator a = a 0 + a 1z −1 + ⋅⋅⋅ + a n z − n , ,all characteristic roots
^
Generally, only two inputs two output systems is of a in unit circle, then always has an enough big
considered to introduce the design method of the adjoint
matrix decoupling compensator. real number M0 > 0 , caused at that time,
Supposes the system as following transfer function multinomial
matrix: (In industry process, usually may through fall ^ ^ ^ ^
step processing to transform various single inputs list f = b z − d + Ma(1 − z −1 ), b = b 0 + b 1z −1 + ⋅⋅⋅ + b m z − m
output system as first order pure time delay system) d > 0 .all roots all in the unit circle sufficiently
⎡ b 11 − d 11 b 12 ⎤ ^ ^
⎢ 1 + a z −1 z −1
z −d 12 ⎥ condition are a(1) b(1) > 0 :
1 + a 12 z
G(z −1 ) = ⎢ ⎥
11
^
⎢ b 21 − d 21 b 22 ⎥ b(1)(1 + ε )d +1
⎢1 + a z −1 z z − d 22 ⎥
⎣ 21 1 + a 22 z −1
⎦ Also M0 = ^
,ε > 0
Then this system adjoint matrix is as follows: ε a(1)
⎡ b22 b ⎤ Thus we may know such front channel transfer
−d22
⎢ 1+a z−1 z − 12 −1 z−d12 ⎥ function of closed loop system as (3),the sufficient
1+a12z condition of all root of its characteristic polynomial
G*(z−1) = ⎢ ⎥
22

⎢ b21 −d21 b11 −d11 ⎥ condition in the unit circle are: ∃k c ,

⎢−1+a z−1 z 1+a11z−1
z ⎥
kc *sgn(b12b21(1+a11)(1+a22 )−b11b22(1+a12 )(1+a21) >0 ,
⎣ 21 ⎦
Let Let
b 11b 22 kc = v0 *sgn(b12b21(1+a11)(1+a22 ) −b11b22(1+a12 )(1+a21))
E( z −1 ) = z −( d 11 + d 22 )
(1 + a 11 z −1 )(1 + a 22 z −1 ) Then when v 0 > 0 , 0 < v < v 0 ， All roots of
b 12 b 21 closed-loop system are in the unit circle.
− −1 −1
z −( d 12 + d 21 ) Then we select such v 0 according to above
(1 + a 12 z )(1 + a 21 z )
theorem:
Then:
⎡ E(z −1 ) 0 ⎤ ε(1−a3' )(1−a4' )
G(z −1 ) ⋅ G * (z −1 ) = ⎢ v0 = (4)
⎥ b12b21(1+a11)(1+a22)−b11b22(1+a12)(1+a21)(1+ε)d0+1
⎣ 0 E(z −1 ) ⎦
Thus we may transform the above multivariable And we know the maximum v 0 can be get at
system control question to two transfer functions
ε= 1 .
E(z −1 ) for single control of variable question.. d 0 +1
Supposes a 1' ,a 2 ' respectively the two polar of Here d 0 = max(d 11 + d 22 , d 12 + d 21 )
−1
transfer function E(z ) which most approaches unit
4. Simulation
circle,. a 3 ' ,a 4 ' are other two polar. Then we adopt the
following controller structure: In the papermaking process, paper weight and
' −1 ' −1 moisture and the paper pulp density and the steam
k c (1 − a 1 z )(1 − a 2 z )
C(z −1 ) = pressure are a close coupling variable, its transfer
1 − z −1 function expression as follows:
Then the front channel transfer function of ⎡ 45e −6.5s −11.6e −2.5s ⎤
closed loop system is:
⎡ y 1 ⎤ ⎢ 2.54s + 1 1.56s + 1 ⎥⎥ ⎡ u 1 ⎤
E(z−1)C(z−1) =kc *(b12b21(1+a12z−1)(1+a21z−1)z−(d12+d21) ⎢y ⎥ = ⎢ ⎢ ⎥
−6.5s
⎣ 2 ⎦ ⎢ 2e −4.08e −2.5s ⎥ ⎣ u 2 ⎦
−b11b22(1+a11z−1)(1+a22z−1)z−(d11+d22 ) )
⎣⎢ 2.54s + 1 1.56s + 1 ⎦⎥
(3)
−1 ' −1 ' −1
/(1−z )(1−a z )(1−a z )
3 4 Takes the sampling period is 0.5 second, after the Z
Thus we may obtain this single variable system the transformation obtains the following discrete transfer
characteristic polynomial. function expression:
Regarding such solid coefficient multinomial, has
the following theorem:
 ⎡ 8.04 −14 −3.18 ⎤
⎢1 − 0.8213z −1 z −1
z −6 ⎥
1 − 0.7258z
⎢ ⎥
⎢ 0.35 −14 −1.12 −6 ⎥
z z
⎣⎢1 − 0.8213z −1 1 − 0.7258z −1 ⎦⎥
select ε = 0.06 ,then obtain the controller
C(z −1 ) as follows:
−1 −0.385 *(1 − 0.8213z−1 )(1 − 0.8213z−1 )
C(z ) =
1 − z−1
Let set point y 1 = 2, y 2 =1 .Response curve of
closed loop system as Fig1.1,2.1,3.1.And consider
y 1 = 0, y 2 = 1 ,that’s when y 2 changes at one unit Fig2.2:Diagnal decoupling method
step, the response curve as Fig1.2,2.2,3.2.:

Fig3.1: Static gain decoupling method

Fig1.1: Adjoint Matrix Decoupling Method

Fig1.2: Adjiont Matrix Decoupling Method

Fig3.2: Static gain decoupling method
Here: A is set point of y 1 ,B is the output of y 1 ;C is
the set point of y 2 ,D is the output of y 2 .
From above curve ,we know the adjoint matrix
decoupling controller compared to the diagnal
decoupling as well as the static gain has better
decoupling effect and dynamic performance. At the
same time it does not need to request the transfer
function matrix to have the opposite angle superiority,
compared to the diagnal decoupling method has more
applicable scopes.

Fig2.1: Diagnal decoupling method

5. Conclusion
Based on the transfer function adjoint matrix
decoupling controller to the system complete
decoupling, the dynamic response and robustness all
have satisfaction the performance. At the same time this
design method can also be realized though the object is
pure time delay system or non minimum phase system,
the applicable scope is extremely broad.
After translate the multivariable system to N SISO
system, it will be easily to design the parameter of this
controller by using theorem 1,and can obtain
satisfactory performance. The design of controller is
extremely simple.

Refrences
[1] Liuxiang et.al.,Typical time delay uncertain
system robust PID control,Information &
control(Supplement),Vol 28,273-277,1999

[2] Huang daoping,zhuxuefeng，One kind of forecast

optimization decoupling compensator research
design,Acta Automatica Sinica,Vol.
26,No.2,250-253,2000