Robust stabilization of uncertain linear systems (Q2704008)

The author considers the control problem for a linear dynamical system described by the equation NEWLINE\[NEWLINE\dot x = [A + F_1\Omega_1(t)E_1] x + [B + F_2\Omega_2(t)E_2] u,\tag{1}NEWLINE\]NEWLINE where \(x \in\mathbb R^m\) is a state, \(u\in\mathbb R^k\) is a control, \(A\), \(B\), \(F_i\) and \(E_i\) are given matrices of the corresponding sizes, and \(\Omega_i(t)\) are measurable matrix functions corresponding to the unknown parameters and satisfying the conditions \((2) \Omega_i(t)^T\Omega_i(t)\leq I\), \(i=1,2\), for all \(t\); here \(I\) is the identity matrix. It is assumed that for the initial state \(x(0) = x_0\) prescribed for (1) and for any admissible control, the resultant Cauchy problem has a unique solution defined on the interval \((0,+\infty)\). The problem is to find robust control laws in the form of a linear state feedback \((3) u=-\Theta^Tx\) such that, by virtue of system (1), (3), some function \(V(x) = x^TPx\) with a symmetric positive definite matrix \(P\) satisfies the inequality \(\dot V < -x^TQx - u^TRu\) for all \(x\) and \(t\), and for all admissible functions \(\Omega_i(t)\) satisfying conditions (2). Here \(Q = Q^T > 0\) and \(R = R^T > 0\) are given matrices. It is shown that the set of the parameters \(\Theta\) of robust control laws (3) as well as necessary and sufficient conditions for the existence of such laws is determined by a two-parameter family of matrix inequalities. In particular, the minimax control law in a certain differential game whose equation does not contain unknown parameters and whose objective functional belongs to a certain class is a robust control law. Thus, by applying the solution of inverse variational minimax control problems one can find conditions, directly expressed via the linear feedback parameters and not requiring the solution of the above-mentioned inequalities, under which a given feedback corresponds to a robust control law.