Quadratic adaptive control of normal mixtures

The paper deals with adaptive control of a system described by a mixture of normal regression models with external variables (ARX). A recently proposed method of quasi-Bayes recursive estimation of such models is complemented with optimal quadratic control design of such a mixture using certainty-equivalent combination. The design technology guarantees that an optimal stabilizing controller is found if it exists. Use of mixtures extends substantially practical applicability of adaptive controllers as it allows us to deal with systems having a different linear dynamics in different states/working points.