x
˙
=
A
x
+
B
w
w
+
B
u
u
+
B
q
p
q
=
C
q
x
+
D
q
w
w
+
D
q
u
u
+
D
q
p
p
u
=
K
x
{\displaystyle {\begin{aligned}{\dot {x}}&=Ax+B_{w}w+B_{u}u+B_{q}p\\q&=C_{q}x+D_{qw}w+D_{qu}u+D_{qp}p\\u&=Kx\end{aligned}}}
Where:
x
∈
R
n
w
∈
R
m
u
∈
R
k
{\displaystyle {\begin{aligned}x&\in R^{n}\\w&\in R^{m}\\u&\in R^{k}\\\end{aligned}}}
In case of norm-bound uncertainty, we have:
p
=
Δ
(
t
)
q
‖
Δ
‖
≤
1
{\displaystyle {\begin{aligned}p&=\Delta (t)q\\&\|\Delta \|\leq 1\end{aligned}}}
The matrices
A
∈
R
n
×
n
;
B
w
∈
R
n
×
m
;
B
u
∈
R
n
×
k
;
B
p
∈
R
n
×
N
p
;
K
∈
R
k
×
n
{\displaystyle A\in R^{n\times n};\;B_{w}\in R^{n\times m};\;B_{u}\in R^{n\times k};B_{p}\in R^{n\times N_{p}};K\in R^{k\times n}}
.
C
q
∈
R
N
q
×
n
;
D
q
u
∈
R
N
q
×
k
D
q
w
∈
R
N
q
×
m
D
q
p
∈
R
N
q
×
N
p
{\displaystyle C_{q}\in R^{N_{q}\times n};\;D_{qu}\in R^{N_{q}\times k}D_{qw}\in R^{N_{q}\times m}D_{qp}\in R^{N_{q}\times N_{p}}}
.
The reachable set can be defined:
R
S
=
{
x
(
T
)
|
u
=
K
x
;
x
(
0
)
=
0
;
T
≥
0
;
∫
0
T
w
T
w
d
t
<
1
}
{\displaystyle {\begin{aligned}RS&=\{x(T)|u=Kx;\;\;x(0)=0;\;\;T\geq 0;\;\;\int _{0}^{T}w^{T}wdt<1\}\\\end{aligned}}}
The elipsoid
E
=
{
ε
∈
R
n
|
ε
T
Q
ε
≤
1
}
⊇
R
S
{\displaystyle E=\{\varepsilon \in R^{n}|\varepsilon ^{T}Q\varepsilon \leq 1\}\supseteq RS}
The Optimization Problem
edit
The following optimization problem should be solved:
Find
μ
>
0
:
Y
[
Q
A
T
+
A
Q
+
B
u
Y
+
Y
T
B
u
T
+
B
w
B
w
T
+
μ
B
p
B
p
T
(
C
q
Q
+
D
q
u
Y
)
T
C
q
Q
+
D
q
u
Y
−
μ
I
]
<
0
K
=
Y
Q
−
1
{\displaystyle {\begin{aligned}{\text{Find}}\;&\mu >0:Y\\&{\begin{bmatrix}QA^{T}+AQ+B_{u}Y+Y^{T}B_{u}^{T}+B_{w}B_{w}^{T}+\mu B_{p}B_{p}^{T}&(C_{q}Q+D_{qu}Y)^{T}\\C_{q}Q+D_{qu}Y&-\mu I\end{bmatrix}}<0\\&K=YQ^{-1}\end{aligned}}}
Or
Find
μ
>
0
:
σ
[
Q
A
T
+
A
Q
+
σ
B
u
B
u
T
+
B
w
B
w
T
+
μ
B
p
B
p
T
(
C
q
Q
−
σ
D
q
u
B
u
T
+
μ
D
q
p
B
p
T
)
T
C
q
Q
−
σ
D
q
u
B
u
T
+
μ
D
q
p
B
p
T
−
μ
(
I
−
D
q
p
D
q
p
T
)
]
<
0
K
=
−
σ
2
B
u
T
Q
−
1
{\displaystyle {\begin{aligned}{\text{Find}}\;&\mu >0:\sigma \\&{\begin{bmatrix}QA^{T}+AQ+\sigma B_{u}B_{u}^{T}+B_{w}B_{w}^{T}+\mu B_{p}B_{p}^{T}&(C_{q}Q-\sigma D_{qu}B_{u}^{T}+\mu D_{qp}B_{p}^{T})^{T}\\C_{q}Q-\sigma D_{qu}B_{u}^{T}+\mu D_{qp}B_{p}^{T}&-\mu (I-D_{qp}D_{qp}^{T})\end{bmatrix}}<0\\&K=-{\frac {\sigma }{2}}B_{u}^{T}Q^{-1}\end{aligned}}}
This LMI allows us to investigate stability for the robust control problem in the case of polytopic uncertainty and gives on the controller for this case
[1] - Matlab implementation using the YALMIP framework and Mosek solver
A list of references documenting and validating the LMI.