Switching reconfigurable control based on hidden Markov models

Abstract

A supervisor algorithm for application in switching reconfigurable control of time varying plants is presented. The possible outcomes of the parameters configuring plant dynamics are described by a hidden Markov model whose state, assumed unavailable for direct observation, is estimated from plant i/o data using a nonlinear filtering algorithm. The use of a shared state representation allows a natural fitting of this supervisor to Predictive Reconfigurable Control.