Novel aggregation algorithm for large-scale systems control

Abstract

We propose in this paper a novel aggregation algorithm for linear quadratic control problems. The basic idea of our algorithm is to decompose a system into two subsystems and reduce the subsystem which contains only non-dominant eigenvalues in a way that slow-decaying terms in the variables associated with this subsystem could be well preserved. The advantage of the new algorithm will be clearly shown in numeric examples where ninth-order problems are reduced to second-, third- and fourth-order ones which yield suboptimal controls very close to those obtained from full-order Riccati equations.