問題設定

問題設定…Homework

[1] H_\infty制御系を設計する際の2ポートシステムの構成法について述べます。


図1 2ポートシステム

次の制御対象を考えます。

\displaystyle{(1)\quad G: \left\{\begin{array}{l} \dot{x}=Ax+Bu\\ y=Cx \end{array}\right.}

これに対する評価出力として、次を考えます。

\displaystyle{(2)\quad W_S: \left\{\begin{array}{l} \dot{x}_I=r-y\\ y_{11}=\omega_Ix_I \end{array}\right. }

\displaystyle{(3)\quad W_T: y_{12}=\omega_D\dot{y}=\omega_DCAx+\omega_DCBu}


図2 積分器を含める場合の2ポートシステム

このとき、次の2ポートシステムを構成します。

\displaystyle{(4)\quad \boxed{P: \left\{\begin{array}{l} \left[\begin{array}{c} \dot{x} \\ \dot{x}_I \end{array}\right]= \underbrace{ \left[\begin{array}{cc} A& 0 \\ -C & 0 \end{array}\right] }_{A} \left[\begin{array}{c} x \\ x_I \end{array}\right] + \underbrace{ \left[\begin{array}{c} 0 \\ 1 \end{array}\right] }_{B_1} r + \underbrace{\left[\begin{array}{c} B \\ 0 \end{array}\right] }_{B_2} u\\ \underbrace{ \left[\begin{array}{c} y_{11} \\ y_{12} \end{array}\right] }_{y_1} = \underbrace{ \left[\begin{array}{cc} 0 &\omega_I\\ \omega_DCA & 0 \end{array}\right] }_{C_1} \left[\begin{array}{c} x \\ x_I \end{array}\right] + \underbrace{ \left[\begin{array}{c} 0 \\ 0 \end{array}\right] }_{D_{11}} r + \underbrace{ \left[\begin{array}{c} 0 \\ \omega_DCB \end{array}\right] }_{D_{12}} u\\ \underbrace{ \left[\begin{array}{c} y \\ x_I \end{array}\right] }_{y_2} = \underbrace{ \left[\begin{array}{cc} C & 0\\ 0 & 1 \end{array}\right] }_{C_2} \left[\begin{array}{c} x \\ x_I \end{array}\right] + \underbrace{ \left[\begin{array}{c} 0 \\ 0 \end{array}\right] }_{D_{21}} r \end{array}\right.} }

これに対して、次のコントローラを設計します。

\displaystyle{(5)\quad K_0: \left\{\begin{array}{l} \dot{x}_K=A_Kx_K+ \underbrace{ \left[\begin{array}{cc} B_{K1} & B_{K2} \end{array}\right] }_{B_K} \left[\begin{array}{c} y \\ x_I \end{array}\right] \\ u=C_Kx_K + \underbrace{ \left[\begin{array}{cc} D_{K1} & D_{K2} \end{array}\right] }_{D_K} \left[\begin{array}{c} y \\ x_I \end{array}\right] \end{array}\right.}

これによる閉ループ系

\displaystyle{(6)\quad P_{CL}: \left\{\begin{array}{l} \left[\begin{array}{c} \dot{x} \\ \dot{x}_K \end{array}\right] = \underbrace{ \left[\begin{array}{ccc} A+B_2D_KC_2 & B_2C_K \\ B_KC_2 & A_K \end{array}\right] }_{A_{CL}} \left[\begin{array}{c} x \\ x_K \end{array}\right] + \underbrace{ \left[\begin{array}{ccc} B_1+B_2D_KD_{21} \\ B_KD_{21} \end{array}\right] }_{B_{CL}} r\\ \underbrace{ \left[\begin{array}{c} y_{11} \\ y_{12} \end{array}\right] }_{y_1} = \underbrace{ \left[\begin{array}{ccc} C_1+D_{12}D_KC_2 & D_{12}C_K \end{array}\right] }_{C_{CL}} \left[\begin{array}{c} x \\ x_K \end{array}\right] + \underbrace{ (D_{11}+D_{12}D_KD_{21}) }_{D_{CL}} r \end{array}\right. }

において、そのH_\inftyノルムが\gammaより小、すなわち

\displaystyle{(7)\quad \sup_{\omega\in{\rm\bf R}} \underbrace{||\hat{P}_{CL}(j\omega)||_2}_{\sigma_1(\hat{P}_{CL}(j\omega))} <\gamma }

ただし

\displaystyle{(8)\quad \hat{P}_{CL}(s)= \left[\begin{array}{c} \hat{G}_{y_{11}}(s)\\ \hat{G}_{y_{12}}(s) \end{array}\right]= C_{CL}(sI-A_{CL})^{-1}B_{CL}+D_{CL} }

となるように出力フィードバックのゲインA_K,B_K,C_K,D_Kを求める問題を考えます。

これをLMIベース設計問題として定式化すると次のようになります。

Minimize \gamma on R=R^T,S=S^T,{\cal A}_K,{\cal B}_K,{\cal C}_K,D_K subject to

\displaystyle{(9)\quad \left[\begin{array}{cc} R & I \\ I & S \end{array}\right] >0 }

\displaystyle{(10)\quad \left[\begin{array}{ccc} \left[\begin{array}{cc} AR+B_2{\cal C}_K & A+B_2D_KC_2\\ {\cal A}_K & SA+{\cal B}_KC_2 \end{array}\right]+(*)^T & \left[\begin{array}{c} B_1+B_2D_KD_{21} \\ SB_1+{\cal B}_KD_{21} \end{array}\right] & (*)^T \\ (*)^T & -\gamma^2 I & (*)^T \\ \left[\begin{array}{cc} C_1R+D_{12}{\cal C}_K & C_1+D_{12}D_KC_2 \end{array}\right] & D_{11} & -I \end{array}\right]<0 }

Determine the output feedback controller A_K,B_K,C_K by

\displaystyle{(11)\quad \begin{array}{lll} A_K&=&N^{-1}({\cal A}_K-S(A-B_2D_KC_2)R-{\cal B}_KC_2R-SB_2{\cal C}_K)M^{-T} \\ B_K&=&N^{-1}({\cal B}_K-SB_2D_K) \\ C_K&=&({\cal C}_K-D_KC_2R)M^{-T} \\ &&(I-SR=NM^T) \end{array} }

実際には、次のLMIも追加することが行われます。

\displaystyle{(12)\quad \begin{array}{l} \left[\begin{array}{cc} AR+B_2{\cal C}_K & A+B_2D_KC_2\\ {\cal A}_K & SA+{\cal B}_KC_2 \end{array}\right] +(*)^T %\nonumber\\&& +\alpha \left[\begin{array}{cc} R & I \\ I & S \end{array}\right]<0}\\ \left[\begin{array}{cc} -r \left[\begin{array}{cc} R & I \\ I & S \end{array}\right] & \left[\begin{array}{cc} AR+B_2{\cal C}_K & A+B_2D_KC_2\\ {\cal A}_K & SA+{\cal B}_KC_2 \end{array}\right] \\ (*)^T & -r \left[\begin{array}{cc} R & I \\ I & S \end{array}\right] \end{array}\right] <0\\ \left[\begin{array}{cc} \sin\theta & \cos\theta \\ -\cos\theta & \sin\theta \end{array}\right]\otimes \left[\begin{array}{cc} AR+B_2{\cal C}_K & A+B_2D_KC_2\\ {\cal A}_K & SA+{\cal B}_KC_2 \end{array}\right]+ (*)^T <0 \end{array} }

ちなみに、コントローラK_0に積分器

\displaystyle{(13)\quad\dot{x}_I=r-y}

を接続して、制御対象に対するコントローラとして次を得ます。

\displaystyle{(14)\quad  \boxed{K: \left\{\begin{array}{l} \left[\begin{array}{c} \dot{x}_K \\ \dot{x}_I \end{array}\right]= \left[\begin{array}{cc} A_K & B_{K2} \\ 0 & 0 \end{array}\right] \left[\begin{array}{c} x_K \\ x_I \end{array}\right] + \left[\begin{array}{cc} B_{K1} & 0\\ -1& 1 \end{array}\right] \left[\begin{array}{c} y \\ r \end{array}\right] \\ u= \left[\begin{array}{cc} C_K & D_{K2} \end{array}\right] \left[\begin{array}{c} x_K \\ x_I \end{array}\right] + \left[\begin{array}{cc} D_{K1} & 0 \end{array}\right] \left[\begin{array}{c} y \\ r \end{array}\right] \end{array}\right.} }

[2] 上の制御系設計は混合感度問題として定式化されていることを示すために、次式に注目します。

\displaystyle{(15)\quad \hat{P}_{CL}(s)= \left[\begin{array}{c} \hat{G}_{y_{11}}(s)\\ \hat{G}_{y_{12}}(s) \end{array}\right]= \left[\begin{array}{c} \hat{W}_S(s)\hat{S}(s)\\ \hat{W}_T(s)\hat{T}(s) \end{array}\right] }

ただし

\displaystyle{(16)\quad \hat{G}_{y_{11}}(s) =\omega_I\frac{\frac{1}{s}}{1+\hat{G}(s)\hat{K}_0(s)\frac{1}{s}} =\underbrace{\frac{\omega_I}{s}}_{\hat{W}_S(s)} \underbrace{\frac{1}{1+\hat{G}(s)\hat{K}(s)}}_{\hat{S}(s)} }
\displaystyle{(17)\quad \hat{G}_{y_{12}}(s) =\omega_Ds\frac{\hat{G}(s)\hat{K}_0(s)\frac{1}{s}}{1+\hat{G}(s)\hat{K}_0(s)\frac{1}{s}} =\underbrace{\omega_Ds}_{\hat{W}_T(s)} \underbrace{\frac{\hat{G}(s)\hat{K}(s)}{1+\hat{G}(s)\hat{K}(s)}}_{\hat{T}(s)} }

したがって、

\displaystyle{(18)\quad ||\hat{P}_{CL}(j\omega)||_2=\sigma_1(\hat{P}_{CL}(j\omega)) =\sqrt{|\hat{W}_S(j\omega)\hat{S}(j\omega)|^2+|\hat{W}_T(j\omega)\hat{T}(j\omega)|^2} }

に注意して

\displaystyle{(19)\quad \sup_{\omega\in{\rm\bf R}}||\hat{P}_{CL}(j\omega)||_2<\gamma }

を最小化することは

\displaystyle{(20)\quad \forall\omega: |\hat{W}_S(j\omega)\hat{S}(j\omega)|^2+|\hat{W}_T(j\omega)\hat{T}(j\omega)|^2<\gamma^2 }

を最小化していることになります。

演習B62…Flipped Classroom
1^\circ 水タンクにおいて、バルブに無駄時間がある場合の次の状態方程式を考えます。

\displaystyle{(21)\quad\underbrace{\frac{d}{dt}(h(t)-h^*)}_{\dot{x}(t)}=\underbrace{-\frac{k}{2S\sqrt{h^*}}}_{a}\underbrace{(h(t)-h^*)}_{x(t)}+\underbrace{\frac{1}{S}}_{b}\underbrace{(q_i(t-L)-q_i^*)}_{u(t-L)}}

これに対するH_\infty制御系を設計する次のプログラムを実行して得られる3枚の図の意味について説明せよ。

MATLAB
%b62.m 
%-----
 clear all, close all
 S=1; k=0.1; hs=1; L=10;
 a=-k/(2*sqrt(hs)*S); b=1/S; c=1;
 wD=L; wI=0.02;
 A=[a 0;-c 0]; B1=[0;1]; B2=[b;0];
 C1=[0 wI;wD*c*a 0]; D11=[0;0]; D12=[0;wD*c*b];
 C2=[c 0;0 1]; D21=[0;0]; D22=[0;0];
%-----
 alpha=0.001; r=0.1; th=pi/3;
 LMIs=of_synlmi6(A,B1,B2,C1,C2,D11,D12,D21,D22,alpha,r,th);
 cobj=zeros(1,decnbr(LMIs)); 
 cobj(1)=1; 
 [cost,xopt]=mincx(LMIs,cobj);  
 gopt=dec2mat(LMIs,xopt,1)
 R=dec2mat(LMIs,xopt,2); 
 S=dec2mat(LMIs,xopt,3);
 ak=dec2mat(LMIs,xopt,4); 
 bk=dec2mat(LMIs,xopt,5); 
 ck=dec2mat(LMIs,xopt,6); 
 dk=dec2mat(LMIs,xopt,7); 
 [u,sd,v]=svd(eye(size(A,1))-R*S); 
 sd=diag(sqrt(1./diag(sd))); 
 Ni=sd*v'; Mti=u*sd; 
 AK=Ni*(ak-S*(A-B2*dk*C2)*R-bk*C2*R-S*B2*ck)*Mti; 
 BK=Ni*(bk-S*B2*dk); 
 CK=(ck-dk*C2*R)*Mti; 
 DK=dk;
%----- 
 pl=eig(A);
 ACL=[A+B2*DK*C2 B2*CK;BK*C2 AK];
 plCL=eig(ACL)
 figure(1)
 dregion(-alpha,0,r,th,r*[-2,1,-1,1]) 
 plot(real(pl),imag(pl),'o',real(plCL),imag(plCL),'x')
%-----  
 ACL=[A+B2*DK*C2 B2*CK;BK*C2 AK];
 BCL=[B1+B2*DK*D21;BK*D21];
 CCL=[C2(1,:) zeros(1,2)];
 AK=[AK BK(:,2);zeros(1,3)];
 BK=[BK(:,1) zeros(2,1); -1 1];
 CK=[CK DK(:,2)];
 DK=[DK(:,1) 0];
%-----   
 w=logspace(-2,1,100); 
 H=exp(-L*i*w); ga_H=20*log10(abs(H));
 ga_DL=20*log10(abs(H-1));
 ga_WT=20*log10(abs(L*i*w));
 figure(2)
 w=logspace(-3,0,100); 
 semilogx(w,ga_DL,w,ga_WT),grid, legend('Delta','WT') 
 %-----  
 ga_WS=20*log10(abs(wI./(i*w))); 
 ga_WT=20*log10(abs(wD*(i*w)));
 G(:,:)=freqresp(ss(a,b,c,[]),w); G=G.'; ga_G=20*log10(abs(G));
 K(:,:)=freqresp(ss(AK,BK(:,1),CK,DK(:,1)),w); K=K.'; ga_K=20*log10(abs(K)); 
 T(:,:)=freqresp(ss(ACL,BCL(:,1),CCL(1,:),[]),w); T=T.'; ga_T=20*log10(abs(T)); 
 ga_S=20*log10(abs(1-T)); 
 figure(3)
 subplot(121),semilogx(w,ga_G+ga_K,w,ga_WS,'b',w,-ga_WT,'r'),grid,legend('L=GK','WS','1/WT')
 subplot(122),semilogx(w,ga_T,'r',w,-ga_WT,'r',w,ga_S,'b',w,-ga_WS,'b'),grid,legend('T','1/WT','S','1/WS')
%-----
function LMIs=of_synlmi6(A,B1,B2,C1,C2,D11,D12,D21,D22,alpha,r,th)
 [n,m]=size(B2);  [p,n]=size(C2);
 setlmis([]); 
 gam=lmivar(1,[1 0]);  
 [R,xxx,Rdec]=lmivar(1,[n 1]); 
 [S,xxx,Sdec]=lmivar(1,[n 1]); 
 Ak=lmivar(2,[n n]); 
 Bk=lmivar(2,[n p]); 
 Ck=lmivar(2,[m n]); 
 Dk=lmivar(2,[m p]); 
% 
 lmiRS=newlmi;
 lmiterm([lmiRS 1 1 R],A,1,'s');       %#1:R*A'+AR 
 lmiterm([lmiRS 1 1 Ck],B2,1,'s');     %#1:Ck'*B2'+B2*Ck 
 lmiterm([lmiRS 2 1 0],A');            %#1:A' 
 lmiterm([lmiRS 2 1 Ak],1,1);          %#1:Ak 
 lmiterm([lmiRS 2 1 -Dk],C2',B2');     %#1:C2'*Dk'*B2' 
 lmiterm([lmiRS 2 2 S],1,A,'s');       %#1:A'*S+S*A 
 lmiterm([lmiRS 2 2 Bk],1,C2,'s');     %#1:C2'*Bk'+Bk*C2 
 lmiterm([lmiRS 1 3 0],B1);            %#1:B1 
 lmiterm([lmiRS 1 3 Dk],B2,D21);       %#1:B2*Dk*D21 
 lmiterm([lmiRS 2 3 S],1,B1);          %#1:S*B1 
 lmiterm([lmiRS 2 3 Bk],1,D21);        %#1:Bk*D21 
 lmiterm([lmiRS 3 3 gam],-1,1);        %#1:-gam 
 lmiterm([lmiRS 4 1 R],C1,1);          %#1:C1*R 
 lmiterm([lmiRS 4 1 Ck],D12,1);        %#1:D12*Ck 
 lmiterm([lmiRS 4 2 0],C1);            %#1:C1 
 lmiterm([lmiRS 4 2 Dk],D12,C2);       %#1:D12*Dk*C2 
 lmiterm([lmiRS 4 3 0],D11);           %#1:D11 
 lmiterm([lmiRS 4 3 Dk],D12,D21);      %#1:D12*Dk*D21 
 lmiterm([lmiRS 4 4 gam],-1,1);        %#1:-gam 
%-----
 lmiPL1=newlmi; 
 lmiterm([lmiPL1 1 1 R],A,1,'s');      %#2:R*A'+A*R 
 lmiterm([lmiPL1 1 1 Ck],B2,1,'s');    %#2:Ck'*B2'+B2*Ck 
 lmiterm([lmiPL1 2 1 Ak],1,1);         %#2:Ak 
 lmiterm([lmiPL1 1 2 0],A);            %#2:A 
 lmiterm([lmiPL1 1 2 Dk],B2,C2);       %#2:B2*Dk*C2 
 lmiterm([lmiPL1 2 2 S],1,A,'s');      %#2:A'*S+S*A 
 lmiterm([lmiPL1 2 2 Bk],1,C2,'s');    %#2:C2'*Bk'+Bk*C2 
% 
 lmiterm([lmiPL1 1 1 R],2*alpha,1);    %#2:2*alpha*R
 lmiterm([lmiPL1 2 1 0],2*alpha);      %#2:2*alpha*I 
 lmiterm([lmiPL1 2 2 S],2*alpha,1);    %#2:2*alpha*S 
%-----
 lmiPL2=newlmi; 
 lmiterm([lmiPL2 1 1 R],-r,1);         %#3:-r*R 
 lmiterm([lmiPL2 2 1 0],-r);           %#3:-r*I 
 lmiterm([lmiPL2 2 2 S],-r,1);         %#3:-r*S 
% 
 lmiterm([lmiPL2 1 3 R],A,1);          %#3:A*R
 lmiterm([lmiPL2 1 3 Ck],B2,1);        %#3:B2*Ck 
 lmiterm([lmiPL2 2 3 Ak],1,1);         %#3:Ak 
 lmiterm([lmiPL2 1 4 0],A);            %#3:A 
 lmiterm([lmiPL2 1 4 Dk],B2,C2);       %#3:B2*Dk*C2 
 lmiterm([lmiPL2 2 4 S],1,A);          %#3:S*A 
 lmiterm([lmiPL2 2 4 Bk],1,C2);        %#3:Bk*C2 
% 
 lmiterm([lmiPL2 3 3 R],-r,1);         %#3:-r*R
 lmiterm([lmiPL2 4 3 0],-r);           %#3:-r*I 
 lmiterm([lmiPL2 4 4 S],-r,1);         %#3:-r*S 
%-----
 sth=sin(th); cth=cos(th);
 lmiPL3=newlmi; 
 lmiterm([lmiPL3 1 1 R],sth*A,1,'s');  %#4:sth*(R*A'+A*R) 
 lmiterm([lmiPL3 1 1 Ck],sth*B2,1,'s');%#4:sth*(Ck'*B2'+B2*Ck) 
 lmiterm([lmiPL3 2 1 Ak],sth,1);       %#4:sth*(Ak) 
 lmiterm([lmiPL3 1 2 0],sth*A);        %#4:sth*(A) 
 lmiterm([lmiPL3 1 2 Dk],sth*B2,C2);   %#4:sth*(B2*Dk*C2) 
 lmiterm([lmiPL3 2 2 S],1,sth*A,'s');  %#4:sth*(A'*S+S*A) 
 lmiterm([lmiPL3 2 2 Bk],1,sth*C2,'s');%#4:sth*(C2'*Bk'+Bk*C2) 
% 
 lmiterm([lmiPL3 1 3 R],cth*A,1);      %#1:cth*(A*R) 
 lmiterm([lmiPL3 1 3 R],1,-cth*A');    %#1:cth*(-R*A')
 lmiterm([lmiPL3 1 3 Ck],cth*B2,1);     %#1:cth*(B*Ck) 
 lmiterm([lmiPL3 1 3 -Ck],-cth*B2',1);  %#1:cth*(-Ck'*B') 
 lmiterm([lmiPL3 2 3 Ak],cth,1);       %#4:cth*(Ak) 
 lmiterm([lmiPL3 1 4 -Ak],-cth,1);     %#4:cth*(-Ak') 
 lmiterm([lmiPL3 1 4 0],A);            %#4:cth*(A) 
 lmiterm([lmiPL3 2 3 0],-A');          %#4:cth*(-A') 
 lmiterm([lmiPL3 1 4 Dk],cth*B2,C2);   %#4:cth*(B2*Dk*C2) 
 lmiterm([lmiPL3 2 3 -Dk],-cth*C2',B2');%#4:cth*(-C2'*Dk'*B2') 
 lmiterm([lmiPL3 2 4 S],1,cth*A);      %#4:cth*(S*A) 
 lmiterm([lmiPL3 2 4 S],-cth*A',1);    %#4:cth*(-A'*S) 
 lmiterm([lmiPL3 2 4 Bk],1,cth*C2);    %#4:cth*(Bk*C2) 
 lmiterm([lmiPL3 2 4 -Bk],-cth*C2',1); %#4:cth*(-C2'*Bk') 
% 
 lmiterm([lmiPL3 3 3 R],sth*A,1,'s');  %#4:sth*(R*A'+A*R)
 lmiterm([lmiPL3 3 3 Ck],sth*B2,1,'s');%#4:sth*(Ck'*B2'+B2*Ck) 
 lmiterm([lmiPL3 4 3 Ak],sth,1);       %#4:sth*(Ak) 
 lmiterm([lmiPL3 3 4 0],sth*A);        %#4:sth*(A) 
 lmiterm([lmiPL3 3 4 Dk],sth*B2,C2);   %#4:sth*(B2*Dk*C2) 
 lmiterm([lmiPL3 4 4 S],1,sth*A,'s');  %#4:sth*(A'*S+S*A) 
 lmiterm([lmiPL3 4 4 Bk],1,sth*C2,'s');%#4:sth*(C2'*Bk'+Bk*C2) 
%-----
 posX=-newlmi;
 lmiterm([posX 1 1 R],1,1);            %#5:R 
 lmiterm([posX 2 1 0],1);              %#5:I 
 lmiterm([posX 2 2 S],1,1);            %#5:S 
%-----
 lmiDk=-newlmi;                        
 lmiterm([lmiDk 1 1 0],1e2);           %#6:1e2 
 lmiterm([lmiDk 2 2 0],1e2);           %#6:1e2 
 lmiterm([lmiDk 2 1 Dk],1,1);          %#6:Dk 
%-----
 lmig=newlmi;                          
 lmiterm([lmig,1,1,gam],1,1);          %#7:gam 
 lmiterm([lmig,1,1,0],-1e3);           %#7:1e3 
 LMIs=getlmis;
end
%-----
%eof