Stochastic Control for Systems With Faulty Sensors

Abstract

The problem of control of linear discrete-time stochastic systems with faulty sensors is considered. The anomaly sensors are assumed to be modeled by a finite-state Markov chain whose transition probabilities are completely known. A passive type multiple model adaptive control (MMAC) law is developed by applying a new generalized pseudo-Bayes algorithm (GPBA), which is based on an n-step measurement update method. The present and other existing algorithms are compared through some Monte Carlo simulations. It is then shown that, for a case of only measurement noise uncertainty (i.e., a case when the certainty equivalence principle holds), the proposed MMAC has better control performance than MMAC’s based on using other existing GPBA’s.