Cascade affine constant recursive algorithm for model-based control

Abstract Paper tackles the trade-off between slow parameter adaptation and parameter variance of recursive least square estimation (rLSE) after a system change in the identification of a time-variant system. This paper proposes cascade affine constant (CAC) estimation for linear systems, which uses rLSE estimated parameters as an apriori knowledge for affine constant estimation, which can be estimated faster with lower variance because of its simple structure. In this configuration, rLSE uses a slower forgetting rate for more accurate dynamics estimation, while affine constant is used to react faster to changes in the system. The comparison is done in the setting with predictive functional control as its performance is impacted greatly by the quality of model parameters and highly correlated with the quality of model parameters. The metrics of the developed method are overall better than other methods.