摘要: 这是我在看书中的正式推导前自己的推导过程,只推导了二阶系统,比较简单,可以跳过本文直接看正式证明。正规的推导见 特征模型(3)证明全系数之和等于1。
专栏全部文章见 基于特征模型的全系数自适应控制。
使用二阶特征模型
y
(
k
+
1
)
=
f
1
(
k
)
y
(
k
)
+
f
2
(
k
)
y
(
k
−
1
)
+
g
0
(
k
)
u
(
k
)
y(k+1)=f_1(k)y(k)+f_{2}(k)y(k-1)+g_0(k)u(k)
y(k+1)=f1(k)y(k)+f2(k)y(k−1)+g0(k)u(k)
对于二阶连续系统
y
¨
+
a
y
˙
+
b
y
=
c
u
\ddot y+a\dot y+by=cu
y¨+ay˙+by=cu
写成可控标准型为
{
x
˙
1
=
x
2
x
˙
2
=
−
b
x
1
+
a
x
2
+
c
u
y
=
x
1
\begin{cases} \dot x_1=x_2 \\ \dot x_2=-bx_1+ax_2+cu \\ y=x_1 \end{cases}
⎩
⎨
⎧x˙1=x2x˙2=−bx1+ax2+cuy=x1
欧拉法离散化得到
x
1
(
k
+
1
)
=
x
1
(
k
)
+
x
2
(
k
)
T
x
2
(
k
+
1
)
=
x
2
(
k
)
+
(
−
b
x
1
(
k
)
−
a
x
2
(
k
)
+
c
u
(
k
)
)
T
(1)
\begin{aligned} x_1(k+1) =& x_1(k)+x_2(k)T\tag{1} \\ x_2(k+1) =& x_2(k)+(-bx_1(k)-ax_2(k)+cu(k))T \\ \end{aligned}
x1(k+1)=x2(k+1)=x1(k)+x2(k)Tx2(k)+(−bx1(k)−ax2(k)+cu(k))T(1)
下面想办法消去
x
2
x_2
x2
x
2
(
k
+
1
)
=
x
1
(
k
+
2
)
−
x
1
(
k
+
1
)
T
=
(
1
−
a
T
)
x
2
(
k
)
−
b
T
x
1
(
k
)
+
c
T
u
(
k
)
=
x
1
(
k
+
1
)
−
x
1
(
k
)
T
(
1
−
a
T
)
−
b
T
x
1
(
k
)
+
c
T
u
(
k
)
\begin{aligned} x_2(k+1) =& \frac{x_1(k+2)-x_1(k+1)}{T} \\ =& (1-aT)x_2(k)-bTx_1(k)+cTu(k) \\ =& \frac{x_1(k+1)-x_1(k)}{T}(1-aT)-bTx_1(k)+cTu(k) \end{aligned}
x2(k+1)===Tx1(k+2)−x1(k+1)(1−aT)x2(k)−bTx1(k)+cTu(k)Tx1(k+1)−x1(k)(1−aT)−bTx1(k)+cTu(k)
得到
x
1
(
k
+
2
)
−
x
1
(
k
+
1
)
=
(
x
1
(
k
+
1
)
−
x
1
(
k
)
)
(
1
−
a
T
)
+
(
−
b
T
x
1
(
k
)
+
c
T
u
(
k
)
)
T
x
1
(
k
+
2
)
=
(
2
−
a
T
)
x
1
(
k
+
1
)
−
(
1
−
a
T
+
b
T
2
)
x
1
(
k
)
+
c
T
2
u
(
k
)
(2)
\begin{aligned} x_1(k+2)-x_1(k+1) =& (x_1(k+1)-x_1(k))(1-aT)+(-bTx_1(k)+cTu(k))T \\ x_1(k+2) =& (2-aT)x_1(k+1)-(1-aT+bT^2)x_1(k)+cT^2u(k)\tag{2} \end{aligned}
x1(k+2)−x1(k+1)=x1(k+2)=(x1(k+1)−x1(k))(1−aT)+(−bTx1(k)+cTu(k))T(2−aT)x1(k+1)−(1−aT+bT2)x1(k)+cT2u(k)(2)
重写为差分方程
y
(
k
)
=
f
1
y
(
k
−
1
)
+
f
2
y
(
k
−
2
)
+
g
u
(
k
−
2
)
y(k)=f_1y(k-1)+f_2y(k-2)+gu(k-2)
y(k)=f1y(k−1)+f2y(k−2)+gu(k−2)
其中
{
f
1
=
2
−
a
T
f
2
=
−
(
1
−
a
T
+
b
T
2
)
g
=
c
T
2
(3)
\begin{cases} f_1=2-aT \\ f_2=-(1-aT+bT^2) \\ g=cT^2 \end{cases} \tag{3}
⎩
⎨
⎧f1=2−aTf2=−(1−aT+bT2)g=cT2(3)
若将式(1)的后向欧拉式改为前向欧拉式
x
1
(
k
+
1
)
=
x
1
(
k
)
+
x
2
(
k
+
1
)
T
x_1(k+1)=x_1(k)+x_2(k+1)T
x1(k+1)=x1(k)+x2(k+1)T
则式(2)变为
x
1
(
k
+
1
)
=
(
2
−
a
T
−
b
T
2
)
x
1
(
k
)
−
(
1
−
a
T
)
x
1
(
k
−
1
)
+
c
T
2
u
(
k
)
x_1(k+1)=(2-aT-bT^2)x_1(k)-(1-aT)x_1(k-1)+cT^2u(k)
x1(k+1)=(2−aT−bT2)x1(k)−(1−aT)x1(k−1)+cT2u(k)
差分方程变为
y
(
k
+
1
)
=
f
1
y
(
k
)
+
f
2
y
(
k
−
1
)
+
g
u
(
k
)
y(k+1)=f_1y(k)+f_2y(k-1)+gu(k)
y(k+1)=f1y(k)+f2y(k−1)+gu(k)
或
y
(
k
)
=
f
1
y
(
k
−
1
)
+
f
2
y
(
k
−
2
)
+
g
u
(
k
−
1
)
y(k)=f_1y(k-1)+f_2y(k-2)+gu(k-1)
y(k)=f1y(k−1)+f2y(k−2)+gu(k−1)
写成特征模型的矩阵形式
y
(
k
)
=
[
f
1
(
k
)
f
2
(
k
)
g
(
k
)
]
[
y
(
k
−
1
)
y
(
k
−
2
)
u
(
k
−
1
)
]
y(k)=\left[\begin{matrix} f_1(k) & f_2(k) & g(k) \end{matrix}\right] \left[\begin{matrix} y(k-1) \\ y(k-2) \\ u(k-1) \end{matrix}\right]
y(k)=[f1(k)f2(k)g(k)]
y(k−1)y(k−2)u(k−1)
使用递推最小二乘辨识参数
K
(
k
)
=
P
(
k
−
1
)
ϕ
(
k
)
1
+
ϕ
T
P
(
k
−
1
)
ϕ
(
k
)
θ
(
k
)
=
θ
(
k
−
1
)
+
K
(
k
)
(
y
(
k
)
−
ϕ
(
k
)
T
θ
(
k
−
1
)
)
P
(
k
)
=
(
I
−
K
(
k
)
ϕ
(
k
)
T
)
P
(
k
−
1
)
\begin{aligned} K(k) =& \frac{P(k-1)\phi(k)}{1+\phi^\mathrm{T}P(k-1)\phi(k)} \\ \theta(k) =& \theta(k-1)+K(k) (y(k)-\phi(k)^{\mathrm{T}} \theta(k-1)) \\ P(k) =& \left(I-K(k)\phi(k)^{\mathrm{T}}\right) P(k-1) \end{aligned}
K(k)=θ(k)=P(k)=1+ϕTP(k−1)ϕ(k)P(k−1)ϕ(k)θ(k−1)+K(k)(y(k)−ϕ(k)Tθ(k−1))(I−K(k)ϕ(k)T)P(k−1)
另外注意到不管哪种推导方式,待辨识的3个参数在采样周期
T
T
T很小时满足
f
1
(
k
)
+
f
2
(
k
)
+
g
(
k
)
=
1
−
b
T
2
+
c
T
2
≈
1
f_1(k)+f_2(k)+g(k)=1-bT^2+cT^2\approx 1
f1(k)+f2(k)+g(k)=1−bT2+cT2≈1
称为全系数之和等于
1
1
1,同时
f
1
f_1
f1、
f
2
f_2
f2和
g
g
g这3个参数都可以根据式(3)大概限制在一定范围内。
特征模型(2)简单二阶系统的证明
于 2023-07-10 16:45:08 首次发布