Robustness of stochastic-parameter control and estimation schemes

The effects on closed-loop system stability of modeling errors resulting from overestimating or underestimating the statistical characteristics in linear discrete-time stochastic parameter systems are investigated. Both infinite and moving horizon controllers designed using erroneous parameter characteristics, which can preserve the mean-square and almost sure boundedness of the close-loop system state, are considered. Mean-square, optimal linear, unbiased one-step predictors are shown to possess stability robustness even when designed with the wrong parameter characteristics