Consider an open loop LPV system
\(P\) described by the following set of equations:
$$P:\left\{ \begin{array}{*{20}l} \dot{x}(t) = A(\theta (t))x(t) + B_{1} (\theta (t))w(t) + B_{2} (\theta (t))u(t), \hfill \\ z(t) = C_{1} (\theta (t))x(t) + D_{11} (\theta (t))w(t) + D_{12} (\theta (t))u(t), \hfill \\ y(t) = C_{2} (\theta (t))x(t) + D_{21} (\theta (t))w(t), \hfill \\ \end{array} \right.$$
(31)
where
\(y\) denotes the measured output,
\(z\) is the controlled output,
\(w\) is the reference and the disturbance inputs and
\(u\) is the control inputs. The matrices in (
9) are affine functions of the parameter vector that varies in
polytope \(\varTheta\) with vertices
\(\theta_{1} , \ldots ,\theta_{j}\) that is:
$$\theta (t) \in \varTheta = {\text{conv}}\left\{ {\theta_{1} , \ldots ,\theta_{j} } \right\} \triangleq \left\{ {\sum\limits_{j = 1}^{r} {\alpha_{i} \theta_{j} ,\alpha_{j} \ge 0,\sum\limits_{j = 1}^{r} {\alpha_{j} = 1} } } \right\}$$
The LPV synthesis problem consists in finding a controller
\(K(\theta )\) described by:
$$K(\theta ):\left\{ \begin{aligned} &\dot{x}_{K} (t) = A_{K} (\theta (t))x_{K} (t) + B_{K} (\theta (t))y(t), \hfill \\ & u(t) = C_{K} (\theta (t))x_{K} \hfill \\ \end{aligned} \right.$$
(32)
Such that closed-loop system (
31) (with input
\(w\) and output
\(z\)) is internally stable and the induced
\(L_{2}\) norm of
\(w \to z\) is bounded by a given number
\(\gamma \succ 0\) for all possible parameter trajectories.
$$P_{\text{cl}} :\left[ {\begin{array}{*{20}c} {\dot{\xi }(t)} \\ {z(t)} \\ \end{array} } \right] = \left[ {\begin{array}{*{20}c} {A_{\text{cl}} (\theta (t))} & {B_{\text{cl}} (\theta (t))} \\ {C_{\text{cl}} (\theta (t))} & {D_{\text{cl}} (\theta (t))} \\ \end{array} } \right]\left[ {\begin{array}{*{20}c} {\xi (t)} \\ {w(t)} \\ \end{array} } \right]$$
(33)
The characterization of robust stability and performance for closed-loop system
\(P_{cl}\) (31) is proved by the following theorem:LPV system (31) would have a quadratic stability and gain level if there exists a matrix such that:
$$\left[ {\begin{array}{*{20}c} {A_{\text{cl}}^{T} (\theta )X + XA_{\text{cl}} (\theta )} & {XB_{\text{cl}} (\theta )} & {C_{\text{cl}} (\theta )^{T} } \\ {B_{\text{cl}} (\theta )^{T} X} & { - \gamma I} & {D_{\text{cl}} (\theta )^{T} } \\ {C_{\text{cl}} (\theta )} & {D_{\text{cl}} (\theta )} & { - \gamma I} \\ \end{array} } \right] \prec 0$$
(34)
This implies for synthesis inequalities (
35) that, without loss of generality, we can replace the search over the polytope
\(\varTheta\) by the search over the vertices of this set consequently, condition (
34) can be reduced to a finite set of linear matrix inequalities (LMI).