Prediction error based controller switching policies

Based on observations of the past inputs and outputs of an unknown system Σ, a countable set of predictors O_a, α ∈ A, is used to predict the system output sequence. Using the resultant prediction errors a decision rule selects a parameter α ∈ A at each time k such that the sequence of selections converges to a predictor with desirable properties. At each time k the selected parameter determines which controller from a family of predesigned controllers Γ_α, α ∈ A, is connected into feedback with the system. We show that for certain MIMO linear systems the resultant closed loop controlled system is stable and can asymptotically track an exogenous reference input