A Riccati-genetic algorithms approach to fixed-structure controller synthesis

A practical approach to the design of controllers with fixed structure (low order, decentralized etc.) that can be tuned via H/sub 2/ or H/sub /spl infin// performance measures is proposed. The design problem is split into a convex subproblem that can involve a large number of decision variables, and a nonconvex subproblem with a small number of decision variables. The former problem can be solved with efficient Riccati solvers, while the latter one is solved using genetic algorithms. The proposed method is flexible and can be used for different H/sub 2/ or H/sub /spl infin// performance or robustness measures. In this paper low-order robust H/sub 2/ design and low-order decentralized mixed sensitivity design are presented. Application of these methods to a benchmark problem and a large scale industrial problem demonstrates that the approach is numerically efficient and leads to performance comparable or superior to that of previously published methods.