Robust Control via Sequential Semidefinite Programming

From MaRDI portal





A new nonlinear optimization technique is proposed to solve non-convex problems arising in robust control design, where a control law ensuring stability and performance is sought for a given uncertain linear system. Such nonconvex optimization problems can be formulated in various ways (minimization of a concave objective function under linear matrix inequality (LMI) constraints; rank-constrained LMI; bilinear matrix inequalities (BMI); LMI with additional nonlinear algebraic constraints) and various techniques have been proposed in the technical literature to address these problems (branch-and-bound algorithms; conic complementarity linearization; augmented Lagrangian methods). Another, still different technique is proposed in this paper. The proposed numerical method solves LMI problems with additional nonlinear algebraic constraints. It is an extension of sequential quadratic programming to the cone of positive semidefinite matrices, referred to as sequential semidefinite programming. Minimization at each step of the second-order Taylor expansion of the Lagrangian of the problem amounts to solving an LMI problem instead of a quadratic problem. Theoretical proofs of local super-linear and quadratic convergence of the algorithm are given. Using two standard, yet meaningful, small-size test examples (robust control of a flexible actuator and robust autopilot of a missile) it is shown that the proposed algorithm performs favorably compared to augmented Lagrangian methods or concave programming. The bottleneck of the algorithm resides in the limitation of available general-purpose semidefinite programming solvers, currently limited to a few thousand design variables.




Cited in
(72)


Describes a project that uses

Uses Software






This page was built for publication: Robust Control via Sequential Semidefinite Programming

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4537811)