ARARX模型的辨识算法.doc

Homework Recursive extended least squares identification In this section, I focus on the following ARMAX model where {} and {} are respectively input and output series, is white noise, and are some polynomials with respect to the backward operator : , , and for t ≤ 0. I use the residual based recursive extened least squares algorithm. The procedure of MATLAB: clc;clear all; t=500; for k=1:t u(k)=1; end v=randn(1,t)/100; y(1)=v(1);y(2)=1.5*y(1)+0.5*u(1)+v(2)-v(1); y(3)=1.5*y(2)-0.7*y(1)+0.5*u(1)+u(2)+v(3)-v(2)+0.2*v(1); for k=4:t y(k)=1.5*y(k-1)-0.7*y(k-2)+u(k-1)+0.5*u(k-2)+v(k)-v(k-1)+0.2*v(k-2); end s=zeros(6,t); q=zeros(6,t); p=100000*eye(6); v1=zeros(1,t); v1(1)=y(1); q(1:6,2)=[-y(1) 0 u(1) 0 v1(1) 0]; L=p*q(1:6,2)/(1+q(1:6,2)*p*q(1:6,2)); s(1:6,2)=L*y(2); v1(2)=y(2)-q(1:6,2)*s(1:6,2); for k=3:t q(1:6,k)=[-y(k-1) -y(k-2) u(k-1) u(k-2) v1(k-1) v1(k-2)]; L=p*q(1:6,k)/(1+q(1:6,k)*p*q(1:6,k)); s(1:6,k)=s(1:6,k-1)+L*(y(k)-q(1:6,k)*s(1:6,k-1)); p=(eye(6)-L*q(1:6,k))*p; v1(k)=y(k)-q(1:6,k)*s(1:6,k); end a1=-1.5*ones(1,t);a2=0.7*ones(1,t);b1=ones(1,t); b2=0.5*ones(1,t);d1=-ones(1,t);d2=0.2*ones(1,t); i=1:1:t; figure(1) subplot(311); plot(i,a1,b,i,s(1,1:t),r);legend(blue--给定参数a1 red--估计参数a11); subplot(312); plot(i,a2,b,i,s(2,1:t),r);legend(blue--给定参数a2 red--估计参数a22); subplot(313); plot(i,b1,b,i,s(3,1:t),r);legend(blue--给定参数b1 red--估计参数b11); figure(2) subplot(311); plot(i,b2,b,i,s(4,1:t),r);legend(blue--给定参数b2 red--估计参数b22); subplot(312); plot(i,d1,b,i,s(5,1:t),r);legend(blue--给定参数d1 red--估计参数d11); subplot(313); plot(i,d2,b,i,s(6,1:t),r);legend(blue--给定参数d2 red--估计参数d22); e1=zeros(1,t);e2=zeros(1,t);e3=zeros(1,t);e4=zeros(1,t);e5=zeros(1,t);e6=zeros(1,t); for k=1:t e1(k)=norm(s(1,k)+1.5,2)/norm(-1.5,2); e2(k)=norm(s(2,k)-0.7,2)/norm(0.7,2); e3(k)=norm(s(3,k)-1,2)/norm(1,2); e4(k)=norm(s(4,k)-0.5,2)/norm(0.5,2); e5(k)=norm(s(5,k)+1,2)/norm(-1,2); e6(k)=norm(s(6,k)-0.2,2)/norm(0.2,2); end figure(3) subpl