Scilab Reference Manual |
---|
lqg2stan — LQG to standard problem
[P,r]=lqg2stan(P22,bigQ,bigR)
P22 | : syslin list (nominal plant) in state-space form |
bigQ | : [Q,S;S',N] (symmetric) weighting matrix |
bigR | : [R,T;T',V] (symmetric) covariance matrix |
r | : 1x2 row vector = (number of measurements, number of inputs) (dimension of the 2,2 part of P) |
P | : syslin list (augmented plant) |
lqg2stan returns the augmented plant for linear LQG (H2) controller design.
P22=syslin(dom,A,B2,C2) is the nominal plant; it can be in continuous time (dom='c') or discrete time (dom='d').
. x = Ax + w1 + B2u y = C2x + w2
for continuous time plant.
x[n+1]= Ax[n] + w1 + B2u y = C2x + w2
for discrete time plant.
The (instantaneous) cost function is [x' u'] bigQ [x;u].
The covariance of [w1;w2] is E[w1;w2] [w1',w2'] = bigR
If [B1;D21] is a factor of bigQ, [C1,D12] is a factor of bigR and [A,B2,C2,D22] is a realization of P22, then P is a realization of [A,[B1,B2],[C1,-C2],[0,D12;D21,D22]. The (negative) feedback computed by lqg stabilizes P22, i.e. the poles of cl=P22/.K are stable.
ny=2;nu=3;nx=4; P22=ssrand(ny,nu,nx); bigQ=rand(nx+nu,nx+nu);bigQ=bigQ*bigQ'; bigR=rand(nx+ny,nx+ny);bigR=bigR*bigR'; [P,r]=lqg2stan(P22,bigQ,bigR);K=lqg(P,r); //K=LQG-controller spec(h_cl(P,r,K)) //Closed loop should be stable //Same as Cl=P22/.K; spec(Cl('A')) s=poly(0,'s') lqg2stan(1/(s+2),eye(2,2),eye(2,2))
F.D.
<< lqe | lqg >> |