Computation of simplified models for estimator and controller design

Most existing model reduction methods produce simplified models which can be used in simulation studies, but are not, in general, suitable for estimator or controller design. Exceptions are the techniques in [1] and [2], which synthesise reduced models by minimising an expected negative log-likelihood function, and which in addition produce reduced order predictors and filters respectively. In this paper we minimise a slightly modified performance index in order to obtain a reduced order model and estimator. In addition, a reduction technique is proposed which produces a reduced order model together with a reduced order dynamic controller, which can be implemented on the full system. For both techniques, an efficient method for computing the performance index and its derivatives is given.