Linear quadratic bumpless transfer (Q1576496): Difference between revisions

The authors consider the problem of bumpless transfer in the linear quadratic context. A methodical procedure is proposed which, apart from assuming that the controllers are finite dimensional, linear, and time-invariant, makes few additional assumptions on the controller. In this way, this method of bumpless transfer is applicable to most general types of problems, and requires no ad hoc procedures for choosing \(D\)-matrices. Formulae for a full-information gain which minimizes a certain cost function are derived. These formulae incorporate two weighting matrices to add flexibility to the design, and they can also be used when the controllers in question are non-invertible. The results are derived for both continuous and discrete time domains. One degree of freedom (DOF) and 2 DOF controller switching formulae are developed, although the authors note that the 1 DOF can be considered as a special case of the 2 DOF case. The most detailed derivation is given for the 1 DOF continuous time case, to convey the main ideas of the scheme. Due to the similarity between the discrete and continuous derivations, the discrete formulae are derived. When extended to the infinite horizon, the formulae depend only on known matrices and the solution to a single Riccati equation. Finally, it is shown that the formulae given here reduce to the Hanus conditioned transfer formulae under certain circumstances.