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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492
|
function [Sk,rk,mu]=h_inf(P,r,mumin,mumax,nmax)
// H-infinity optimal control for the continuous-time plant P
// P is the plant (linear system)given in state-space form or in transfer form,
// e.g. P=syslin('c',A,B,C,D) with A,B,C,D = scalar matrices
// or P=syslin('c',H) with H a transfer matrix.
// r = size of the P22 plant i.e. 2-vector [#outputs,#inputs];
// mumin,mumax = bounds on mu with mu=1/gama^2; (mumin=0 usually)
// nmax = maximum number of iterations in the gama-iteration.
// Two possible calling sequences:
// [Sk,mu]=h_inf(P,r,mumin,mumax,nmax) returns mu and the central controller
// Sk in the same representation as P. (All calculations being done in state
// space).
// [Sk,rk,mu]=h_inf(P,r,mumin,mumax,nmax) returns mu
// and the parametrization of all stabilizing controllers:
// a stabilizing controller K is obtained by K=Fl(Sk,r,PHI) where
// PHI is a linear system with dimensions r' and satisfy h_norm(PHI) < gama.
// rk (=r) is the size of the Sk22 block and mu = 1/gama^2 after nmax
// iterations.
// Author: F.D. Inria (1990) adapted from Safonov-Limebeer.
//!
//
// Initialization : Uci and Yci normalize P
// mu_inf upper bound on mu = gama^-2
//P2 = normalized P.
//
// Copyright INRIA
P1=P(1);
[P2,mu_inf,Uci,Yci,D22]=h_init(P,r)
if mu_inf < mumax then
write(%io(2),mu_inf,'(3x,''romax too big: max romax= '',f10.5)');end
mumax=mini(mu_inf,mumax)
//
// Gama-iteration P6 = transformed P2 with D11 removed
[P6,Finf,mu,Uc#i,Yc#i]=h_iter(P2,r,mumin,mumax,nmax)
if mu==0 then
write(%io(2),[mumin,mumax],'(1x,''no feasible ro in bounds: '',2(e10.3,2x))');
rk=[];Sk=[];
return,end
//
// Optimal controller for P6
[Sk,polesH,polesJ]=h_contr(P6,r,1/mu,Uc#i,Yc#i);
[E,Ak,Bk1,Bk2,Ck1,Ck2,Dk11,Dk12,Dk21,Dk22]=Sk(:);
// Add the optimal controller at infinity
if norm(Finf,1) <> 0 then Dk11=Dk11+Finf;end
// Back-normalization of the controller
Bk1=Bk1*Yci;
Ck1=Uci*Ck1;
Dk11=Uci*Dk11*Yci;
Dk12=Uci*Dk12;
Dk21=Dk21*Yci;
//Convert to descriptor form:
Sk=des2ss(Ak,[Bk1,Bk2],[Ck1;Ck2],[Dk11,Dk12;Dk21,Dk22],E);
// Sk in transfer representation if P is.
[LHS,RHS]=argn(0);
if LHS<3 then Sk=Sk(1:r(2),1:r(1));rk=mu;
// Case D22 different from zero
if norm(coeff(D22),1) <> 0 then Sk=Sk/.D22;end
if P1(1)=='r' then Sk=ss2tf(Sk);end;
return;
end
if LHS==3 then rk=r;
if P1(1)=='r' then Sk=ss2tf(Sk);end;
end
function [P2,mu_inf,Uci,Yci,D22]=h_init(P,r)
//******************************
// Initialization of the standard plant
// P = standard plant; r=size of P22 (1X2 vector)
// P2 = normalized plant
// gama_inf :lower bound on gama
//
// [C1 D12]'*[C1 D12] = Q = [Q1 S;S' Q2]
// [B1;D21] *[B1;D21]'= R = [R1 L';L R2]
//!
P1=P(1);
if P1(1)=='r' then P=tf2ss(P);end
[A,B1,B2,C1,C2,D11,D12,D21,D22]=smga(P,r);
[na,na]=size(A);
[p1,m2]=size(D12),
[p2,m1]=size(D21),
//Useful indexes
l1=1:p1-m2;k1=1:m1-p2;
l2=1+p1-m2:p1;k2=1+m1-p2:m1;
//**************Check assumptions******************
//Stabilizability/detectability of P22 ?
P22=syslin('c',A,B2,C2);
[ns,Us,St]=st_ility(P22,1.d-10)
if ns<>na then write(%io(2),'Warning: P22 not stabilizable');end
if ns==na then write(%io(2),'P22 is stabilizable');end
[nd,Ud,Sd]=dt_ility(P22,1.d-10)
if nd <> 0 then write(%io(2),'Warning: P22 not detectable');end
if nd==0 then write(%io(2),'P22 is detectable');end
// rank P21=[A,B2,C1,D12] = m2 ?
P12=syslin('c',A,B2,C1,D12);
[nt,dt]=trzeros(P12),rzt=real(nt./dt),
if size(nt,'*') > 0 then
if mini(abs(rzt)) < sqrt(%eps) then
write(%io(2),'Warning P12 has a zero on/close the imaginary axis'),
end,
end,
// rank P21=[A,B1,C2,D21] = p2 ?
P21=syslin('c',A,B1,C2,D21);
[nt,dt]=trzeros(P21),rzt=real(nt./dt),
if size(nt,'*')>0 then
if mini(abs(rzt)) < sqrt(%eps) then
write(%io(2),'Warning: P21 has a zero on/close the imaginary axis'),
end,
end,
//***********************end***************************
//Row compression of D12 (bottom)
[T1,r1]=rowcomp(D12),
if r1<>m2 then error('D12 not full column rank'),end,
T1=[T1(r1+1:p1,:);T1(1:r1,:)],
D12=T1*D12,
//Column compression of D21 (right)
[S1,r2]=colcomp(D21),
if r2<>p2 then error('D21 not full row rank'),end,
D21=D21*S1,
//Updating
B1=B1*S1,C1=T1*C1,
D11=T1*D11*S1,
// Scaling on U and Y
Uc=D12(l2,:);
Uci=inv(Uc);
B2=B2*Uci;
D12=D12*Uci;
Yc=D21(:,k2)
Yci=inv(Yc);
C2=Yci*C2;
D21=Yci*D21;
//P2=[A,B1,B2,C1,C2,D11,D12,D21,D22] with D12 and D21 scaled;
//Initialization
//Solve H-infinity problem at infinity
D1111=D11(l1,k1),
D1112=D11(l1,k2),
D1121=D11(l2,k1),
D1122=D11(l2,k2),
M11=[D1111,D1112];M22=[D1111',D1121'];
g1=-1;g2=-1;
if M11<>[] then g1=norm(M11);end
if M22<>[] then g2=norm(M22);end
gama_inf=maxi(g1,g2);
if gama_inf==0 then mu_inf=1/%eps/%eps, else mu_inf=1/(gama_inf*gama_inf);end
P2=syslin('c',A,[B1,B2],[C1;C2],[D11,D12;D21,0*D22]);
//P2 = standard plant with D22=0 and D12,D21 normalized;
function [P6ad,Finfad,muad,Uc#iad,Yc#iad]=h_iter(P2,r,mumin,mumax,nmax)
niter=0;muad=0;P6ad=[]; Finfad=[];Uc#iad=[];Yc#iad=[];
while niter < nmax
niter=niter+1;
mu=(mumin+mumax)/2;
[P6,Finf,tv,Uc#i,Yc#i]=h_test(P2,r,mu)
test=maxi(tv)
if test > 0 then
mumax=mu
else
mumin=mu,muad=mu;P6ad=P6;Finfad=Finf;Uc#iad=Uc#i;Yc#iad=Yc#i;
end
end //while
function [P6,Kinf,tv,Uc#i,Yc#i]=h_test(P2,r,mu)
//****************************
//To test if mu is feasable for the plant P2 :
//mu is feasible for P2 iff the three components of
//tv are negative
//!
//
// [C1 D12]*[C1 D12]'=Q=[Q1 S;S' Q2]
// [B1;D21]*[B1;D21]'=R=[R1 L';L R2]
tv=[1,1,1];
[A,B1,B2,C1,C2,D11,D12,D21,D22]=smga(P2,r);
[p1,m2]=size(D12),
[p2,m1]=size(D21),
//Useful indexes
l1=1:p1-m2;k1=1:m1-p2;
l2=1+p1-m2:p1;k2=1+m1-p2:m1;
//
D1111=D11(l1,k1),
D1112=D11(l1,k2),
D1121=D11(l2,k1),
D1122=D11(l2,k2),
if mu==0 then mu=%eps*%eps;end
mu1=1/mu;
gama=1/sqrt(mu);
gam2=1/(gama*gama);
err=(m1-p2)*(p1-m2);
if err==0 then
Kinf=-D1122
else
Kinf=-(D1122+D1121*inv(mu1*eye-D1111'*D1111)*D1111'*D1112);
end
//Kinf=admissible controller for mu
A=A+B2*Kinf*C2;
B1=B1+B2*Kinf*D21;
C1=C1+D12*Kinf*C2;
D11=D11+D12*Kinf*D21;
if norm(D11) >= gama then write(%io(2),'error : gamma too small');
P6=[]; Kinf=[];Uc#i=[];Yc#i=[];return;end
//P3=list(A,B1,B2,C1,C2,D11,D12,D21,D22) with norm(D11) < gama.
Teta11=gam2*D11;
Teta22=gam2*D11';
W12=eye-gam2*D11*D11';
W21=eye-gam2*D11'*D11;
Teta12=(1/gama)*sqrtm(0.5*(W12+W12'));
Teta21=-(1/gama)*sqrtm(0.5*(W21+W21'));
//Teta*Teta'=(1/gama*gama)*eye
M=inv(eye-D11*Teta22);
N=inv(eye-Teta22*D11);
A=A+B1*Teta22*M*C1;
B2=B2+B1*Teta22*M*D12;
B1=B1*N*Teta21;
C2=C2+D21*Teta22*M*C1;
C1=Teta12*M*C1;
D11=0*D11; //By construction...
D22#=D22+D21*Teta22*M*D12;
D12=Teta12*M*D12;
D21=D21*N*Teta21;
//P4 =syslin('c',A,[B1,B2],[C1;C2],[D11,D12;D21,D22]
// with D11=0; P4=Fl(Teta,size(D11'),P3,r);
D22=0*D22#;
//P5 = [A,[B1,B2],[C1;C2],[D11,D12;D21,D22] with D11=0 and D22=0;
//Row compression of D12
[T1,r1]=rowcomp(D12);
if r1<>m2 then error('D12 not full rank! '),end
T1=[T1(r1+1:p1,:);T1(1:r1,:)],
D12=T1*D12,
//Column compression of D21
[S1,r2]=colcomp(D21),
if r2<>p2 then error('D21 not full rank! '),end,
D21=D21*S1,
//Updating
B1=B1*S1,C1=T1*C1,
D11=T1*D11*S1,
// Scaling on U and Y
Uc#=D12(l2,:);
Uc#i=inv(Uc#);
B2=B2*Uc#i;
D12=D12*Uc#i;
Yc#=D21(:,k2)
Yc#i=inv(Yc#);
C2=Yc#i*C2;
D21=Yc#i*D21;
//P6=[A,B1,B2,C1,C2,D11,D12,D21,D22] with D11=0,D22=0;D12 and D21 scaled.
//Standard assumptions now satisfied
//P6=[A,B1,B2,C1,C2,D11,D12,D21,D22];
// Test of mu for P6 <=> Test of mu^-1 for P2
//Optimal controller :
indic=0;
mu_test=1/mu;
R1=B1*B1';
S=D12'*C1;
C1#=(eye-D12*D12')*C1;
Ax=A-B2*S;
Qx=-C1#'*C1#;
Rx=mu_test*R1-B2*B2';
H=[Ax Rx;
Qx -Ax'];
dx=mini(abs(real(spec(H))));
//write(%io(2),dx);
if dx < 1.d-9 then
write(%io(2),'An eigenvalue of H (controller) is close to Imaginary axis !');
write(%io(2),dx);
indic=1;test=1;
end
if indic ==0 then
[X1,X2,errx]=ric_desc(H);
if errx > 1.d-4 then
write(%io(2),'Riccati solution inaccurate ');
write(%io(2),errx);
end
//Optimal observer :
Q1=C1'*C1;
L=B1*D21';
B1#=B1*(eye-D21'*D21);
Ay=A-L*C2;
Qy=-B1#*B1#';
Ry=mu_test*Q1-C2'*C2;
J=[Ay' Ry;
Qy -Ay];
dy=mini(abs(real(spec(J))));
//write(%io(2),dy);
if dy < 1.d-9 then
write(%io(2),'An eigenvalue of J (observer) is close to Imaginary axis !');
write(%io(2),dy);
indic=1 ;test=1;
end
if indic==0 then
[Y1,Y2,erry]=ric_desc(J);
if erry > 1.d-4 then
write(%io(2),'Riccati solution inaccurate ');
write(%io(2),erry);
end
//Tests
//
// E=(Y2'*X2-mu*Y1'*X1);
// write(%io(2),mini(svd(E)),'(5x,''mini(svd(E)) = '',f10.2)')
[al1,be1]=gspec(A*X2 -B2*(S*X2 +B2'*X1 ),X2);
[al2,be2]=gspec(Y2'*A-(Y2'*L+Y1'*C2')*C2,Y2');
[al3,be3]=gspec(mu_test*Y1'*X1,Y2'*X2);
//Here division by zero may appear...
//If such division appear try to uncomment the 3 following lines:
w1=find(be1==0);be1(w1)=%eps*ones(be1(w1));
w2=find(be2==0);be2(w2)=%eps*ones(be2(w2));
w3=find(be3==0);be3(w3)=%eps*ones(be3(w3));
test1=maxi(real(al1./be1));
test2=maxi(real(al2./be2));
test3=maxi(real(al3./be3))-1;
tv =[test1,test2,test3]
end
end
//write(%io(2),1/sqrt(mu),'(10x,'' Try gama = '',f18.10)');
[answer,no]=maxi(tv);
//if exists('tv')==1 then write(%io(2),[tv,maxi(tv)],'(4f15.10)');end
if exists('tv')==1 then
if answer>0 then
if no==1 then
write(%io(2),[1/sqrt(mu),answer],'('' gama = '',f18.10,'' Unfeasible (Hx hamiltonian) test = '',e15.5)');
end
if no==2 then
write(%io(2),[1/sqrt(mu),answer],'('' gama = '',f18.10,'' Unfeasible (Hy hamiltonian) test = '',e15.5)');
end
if no==3 then
write(%io(2),[1/sqrt(mu),answer],'('' gama = '',f18.10,'' Unfeasible (spectral radius) test = '',e15.5)');
end
else
write(%io(2),[1/sqrt(mu),answer],'('' gama = '',f18.10,'' OK test = '',e15.5)');
end;
end;
P6=syslin('c',A,[B1,B2],[C1;C2],[D11,D12;D21,D22])
function [Sk,polesH,polesJ]=h_contr(P,r,mu,U2i,Y2i)
// ****************************
// Computation of the optimal controller Sk for a standard
// plant which satisfies the assumption D11=0
//
// F.D.
//!
// [C1 D12]*[C1 D12]'=Q=[Q1 S;S' Q2]
// [B1;D21]*[B1;D21]'=R=[R1 L';L R2]
//
[A,B1,B2,C1,C2,D11,D12,D21,D22]=smga(P,r);
if norm(D11,1) > %eps then error('D11 <> 0'),return end
[p2,m1]=size(D21),
[p1,m2]=size(D12),
l1=1:p1-m2;k1=1:m1-p2;
l2=1+p1-m2:p1;k2=1+m1-p2:m1;
//Initialization : constants
R1=B1*B1';
S=D12'*C1;
C1#=(eye-D12*D12')*C1;
Ax=A-B2*S;
Qx=-C1#'*C1#;
Q1=C1'*C1;
L=B1*D21';
B1#=B1*(eye-D21'*D21);
Ay=A-L*C2;
Qy=-B1#*B1#';
// mu-dependent part
//Optimal controller
Rx=mu*R1-B2*B2';
H=[Ax Rx;
Qx -Ax'];
polesH=spec(H);
dx=mini(abs(real(polesH)));
//write(%io(2),dx);
if dx < 1.d-6 then
write(%io(2),'An eigenvalue of H (controller) is close to Imaginary axis !');
end
[X1,X2,errx]=ric_desc(H);
if errx > 1.d-4 then
write(%io(2),'Riccati solution inaccurate ');
write(%io(2),errx);
end
//Optimal observer :
Ry=mu*Q1-C2'*C2;
J=[Ay' Ry;
Qy -Ay];
polesJ=spec(J);
dy=mini(abs(real(polesJ)));
//write(%io(2),dy);
if dy < 1.d-6 then
write(%io(2),'An eigenvalue of J (observer) is close to Imaginary axis !');
end
[Y1,Y2,erry]=ric_desc(J);
if erry > 1.d-4 then
write(%io(2),'Riccati solution inaccurate ');
write(%io(2),erry);
end
//Controller in descriptor form
E=(Y2'*X2-mu*Y1'*X1);
A#=A-B2*S-L*C2;
Ak=Y2'*A#*X2+mu*Y1'*A#'*X1-Y2'*(mu*Qy+B2*B2')*X1-Y1'*(mu*Qx+C2'*C2)*X2;
Bk1=(Y2'*L+Y1'*C2');
Ck1=-(S*X2+B2'*X1);
Bk2=Y2'*B2+Y1'*S'
Ck2=-(C2*X2+L'*X1)
Dk11=0*Ck1*Bk1;
Dk22=0*Ck2*Bk2;
Dk12=eye(Ck1*Bk2);
Dk21=eye(Ck2*Bk1);
//Scaling back
Bk1=Bk1*Y2i;
Ck1=U2i*Ck1;
Dk21=Dk21*Y2i;
Dk12=U2i*Dk12;
// Dk11=U2i*Dk11*Y2i
Sk=list(E,Ak,Bk1,Bk2,Ck1,Ck2,Dk11,Dk12,Dk21,Dk22);
|