File: abinv.man

package info (click to toggle)
scilab 2.6-4
  • links: PTS
  • area: non-free
  • in suites: woody
  • size: 54,632 kB
  • ctags: 40,267
  • sloc: ansic: 267,851; fortran: 166,549; sh: 10,005; makefile: 4,119; tcl: 1,070; cpp: 233; csh: 143; asm: 135; perl: 130; java: 39
file content (193 lines) | stat: -rw-r--r-- 7,048 bytes parent folder | download
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
.TH abinv 1 "April 1993" "Scilab Group" "Scilab Function"
.so ../sci.an 
.SH NAME
abinv -  AB invariant subspace
.SH CALLING SEQUENCE
.nf
[X,dims,F,U,k,Z]=abinv(Sys,alfa,beta,flag)
.fi
.SH PARAMETERS
.TP
Sys
: \fVsyslin\fR list containing the matrices \fV[A,B,C,D]\fR.
.TP
alfa 
: (optional) real number or vector (possibly complex, location of closed loop poles)
.TP
beta
: (optional) real number or vector (possibly complex, location of closed loop poles)
.TP
flag
: (optional) character string \fV'ge'\fR (default) or \fV'st'\fR or \fV'pp'\fR
.TP
X
: orthogonal matrix of size nx (dim of state space).
.TP
dims
:
integer row vector \fVdims=[dimR,dimVg,dimV,noc,nos]\fR with \fVdimR<=dimVg<=dimV<=noc<=nos\fR. If \fVflag='st'\fR, (resp. \fV'pp'\fR), \fVdims\fR has 4
(resp. 3) components.
.TP
F
: real matrix (state feedback)
.TP
k
: integer (normal rank of \fVSys\fR)
.TP
Z
: non-singular linear system (\fVsyslin\fR list)
.SH DESCRIPTION
Output nulling subspace (maximal unobservable subspace) for
\fVSys\fR = linear system defined by a syslin list containing
the matrices [A,B,C,D] of \fVSys\fR.
The vector \fVdims=[dimR,dimVg,dimV,noc,nos]\fR gives the dimensions
of subspaces defined as columns of \fVX\fR according to partition given
below.
The \fVdimV\fR first columns of \fVX\fR i.e \fVV=X(:,1:dimV)\fR, 
span the AB-invariant subspace of \fVSys\fR i.e the unobservable subspace of 
\fV(A+B*F,C+D*F)\fR. (\fVdimV=nx\fR  iff C^(-1)(D)=X).
.LP
The \fVdimR\fR first columns of \fVX\fR i.e. \fVR=X(:,1:dimR)\fR spans  
the controllable part of \fVSys\fR in \fVV\fR, \fV(dimR<=dimV)\fR. (\fVdimR=0\fR
for a left invertible system). \fVR\fR is the maximal controllability
subspace of \fVSys\fR in \fVkernel(C)\fR.
.LP
The \fVdimVg\fR first columns of \fVX\fR spans 
\fVVg\fR=maximal AB-stabilizable subspace of \fVSys\fR. \fV(dimR<=dimVg<=dimV)\fR.
.LP
\fVF\fR is a decoupling feedback: for \fVX=[V,X2] (X2=X(:,dimV+1:nx))\fR one has 
\fVX2'*(A+B*F)*V=0 and (C+D*F)*V=0\fR.
.LP
The zeros od \fVSys\fR are given by : \fVX0=X(:,dimR+1:dimV); spec(X0'*(A+B*F)*X0)\fR
i.e. there are \fVdimV-dimR\fR closed-loop fixed modes.
.LP
If the optional parameter \fValfa\fR is given as input, 
the \fVdimR\fR controllable modes of \fV(A+BF)\fR in \fVV\fR 
are set to \fValfa\fR (or to \fV[alfa(1), alfa(2), ...]\fR.
(\fValfa\fR can be a vector (real or complex pairs) or a (real) number).
Default value \fValfa=-1\fR.
.LP
If the optional real parameter \fVbeta\fR is given as input, 
the \fVnoc-dimV\fR controllable modes of \fV(A+BF)\fR "outside" \fVV\fR 
are set to \fVbeta\fR (or \fV[beta(1),beta(2),...]\fR). Default value \fVbeta=-1\fR.
.LP
In the \fVX,U\fR bases, the matrices 
\fV[X'*(A+B*F)*X,X'*B*U;(C+D*F)*X,D*U]\fR 
are displayed as follows:
.nf

[A11,*,*,*,*,*]  [B11 * ]
[0,A22,*,*,*,*]  [0   * ]
[0,0,A33,*,*,*]  [0   * ]
[0,0,0,A44,*,*]  [0  B42]
[0,0,0,0,A55,*]  [0   0 ]
[0,0,0,0,0,A66]  [0   0 ]

[0,0,0,*,*,*]    [0  D2]
.fi
where the X-partitioning is defined by \fVdims\fR and 
the U-partitioning is defined by \fVk\fR.
.LP
\fVA11\fR is \fV(dimR x dimR)\fR and has its eigenvalues set to \fValfa(i)'s\fR.
The pair \fV(A11,B11)\fR is controllable and \fVB11\fR has \fVnu-k\fR columns. 
\fVA22\fR is a stable \fV(dimVg-dimR x dimVg-dimR)\fR matrix.
\fVA33\fR is an unstable \fV(dimV-dimVg x dimV-dimVg)\fR matrix (see \fVst_ility\fR).
.LP
\fVA44\fR is \fV(noc-dimV x noc-dimV)\fR and has its eigenvalues set to \fVbeta(i)'s\fR.
The pair \fV(A44,B42)\fR is controllable. 
\fVA55\fR is a stable \fV(nos-noc x nos-noc)\fR matrix.
\fVA66\fR is an unstable \fV(nx-nos x nx-nos)\fR matrix (see \fVst_ility\fR).
.LP
\fVZ\fR is a column compression of \fVSys\fR and \fVk\fR is
the normal rank of \fVSys\fR i.e \fVSys*Z\fR is a column-compressed linear
system. \fVk\fR is the column dimensions of \fVB42,B52,B62\fR and \fVD2\fR.
\fV[B42;B52;B62;D2]\fR is full column rank and has rank \fVk\fR.
.LP
If \fVflag='st'\fR is given, a five blocks partition of the matrices is 
returned and \fRdims\fR has four components. If \fVflag='pp'\fR is 
given a four blocks partition is returned. In case \fVflag='ge'\fR one has
\fVdims=[dimR,dimVg,dimV,dimV+nc2,dimV+ns2]\fR where \fVnc2\fR 
(resp. \fVns2\fR) is the dimension of the controllable (resp. 
stabilizable) pair \fV(A44,B42)\fR (resp. (\fV[A44,*;0,A55],[B42;0])\fR).
In case \fVflag='st'\fR one has \fVdims=[dimR,dimVg,dimVg+nc,dimVg+ns]\fR
and in case \fVflag='pp'\fR one has \fVdims=[dimR,dimR+nc,dimR+ns]\fR.
\fVnc\fR (resp. \fVns\fR) is here the dimension of the controllable
(resp. stabilizable) subspace of the blocks 3 to 6 (resp. 2 to 6).
.LP
This function can be used for the (exact) disturbance decoupling problem.
.nf
DDPS:
   Find u=Fx+Rd=[F,R]*[x;d] which rejects Q*d and stabilizes the plant:

    xdot = Ax+Bu+Qd
       y = Cx+Du+Td

DDPS has a solution if Im(Q) is included in Vg + Im(B) and stabilizability
assumption is satisfied. 
Let G=(X(:,dimVg+1:$))'= left annihilator of Vg i.e. G*Vg=0;
B2=G*B; Q2=G*Q; DDPS solvable iff [B2;D]*R + [Q2;T] =0 has a solution.
The pair F,R  is the solution  (with F=output of abinv).
Im(Q2) is in Im(B2) means row-compression of B2=>row-compression of Q2
Then C*[(sI-A-B*F)^(-1)+D]*(Q+B*R) =0   (<=>G*(Q+B*R)=0)
.fi
.SH EXAMPLE
.nf
nu=3;ny=4;nx=7;
nrt=2;ngt=3;ng0=3;nvt=5;rk=2;
flag=list('on',nrt,ngt,ng0,nvt,rk);
Sys=ssrand(ny,nu,nx,flag);alfa=-1;beta=-2;
[X,dims,F,U,k,Z]=abinv(Sys,alfa,beta);
[A,B,C,D]=abcd(Sys);dimV=dims(3);dimR=dims(1);
V=X(:,1:dimV);X2=X(:,dimV+1:nx);
X2'*(A+B*F)*V
(C+D*F)*V
X0=X(:,dimR+1:dimV); spec(X0'*(A+B*F)*X0)
trzeros(Sys)
spec(A+B*F)   //nr=2 evals at -1 and noc-dimV=2 evals at -2.
clean(ss2tf(Sys*Z))
// 2nd Example
nx=6;ny=3;nu=2;
A=diag(1:6);A(2,2)=-7;A(5,5)=-9;B=[1,2;0,3;0,4;0,5;0,0;0,0];
C=[zeros(ny,ny),eye(ny,ny)];D=[0,1;0,2;0,3];
sl=syslin('c',A,B,C,D);//sl=ss2ss(sl,rand(6,6))*rand(2,2);
[A,B,C,D]=abcd(sl);  //The matrices of sl.
alfa=-1;beta=-2;
[X,dims,F,U,k,Z]=abinv(sl,alfa,beta);dimVg=dims(2);
clean(X'*(A+B*F)*X)
clean(X'*B*U)
clean((C+D*F)*X)
clean(D*U)
G=(X(:,dimVg+1:$))';
B2=G*B;nd=3;
R=rand(nu,nd);Q2T=-[B2;D]*R;
p=size(G,1);Q2=Q2T(1:p,:);T=Q2T(p+1:$,:);
Q=G\Q2;   //a valid [Q;T] since 
[G*B;D]*R + [G*Q;T]  // is zero
closed=syslin('c',A+B*F,Q+B*R,C+D*F,T+D*R); // closed loop: d-->y
ss2tf(closed)       // Closed loop is zero
spec(closed('A'))   //The plant is not stabilizable!
[ns,nc,W,sl1]=st_ility(sl);
[A,B,C,D]=abcd(sl1);A=A(1:ns,1:ns);B=B(1:ns,:);C=C(:,1:ns);
slnew=syslin('c',A,B,C,D);  //Now stabilizable
//Fnew=stabil(slnew('A'),slnew('B'),-11);
//slnew('A')=slnew('A')+slnew('B')*Fnew;
//slnew('C')=slnew('C')+slnew('D')*Fnew;
[X,dims,F,U,k,Z]=abinv(slnew,alfa,beta);dimVg=dims(2);
[A,B,C,D]=abcd(slnew);
G=(X(:,dimVg+1:$))';
B2=G*B;nd=3;
R=rand(nu,nd);Q2T=-[B2;D]*R;
p=size(G,1);Q2=Q2T(1:p,:);T=Q2T(p+1:$,:);
Q=G\Q2;   //a valid [Q;T] since 
[G*B;D]*R + [G*Q;T]  // is zero
closed=syslin('c',A+B*F,Q+B*R,C+D*F,T+D*R); // closed loop: d-->y
ss2tf(closed)       // Closed loop is zero
spec(closed('A'))
.fi
.SH AUTHOR
F.D.
.SH SEE ALSO
cainv, st_ility, ssrand, ss2ss, ddp