LAUSR

Abstract Virtual Reference Feedback Tuning (VRFT) is a direct data-driven control design method employed to tune a controller’s parameters aiming to achieve a prescribed closed-loop performance. Its primary formulation leads to a biased estimate in the presence of noise, so an instrumental variable (IV) alternative has been proposed and this alternative has been favoured whenever the noise level is significant. Even though VRFT thus formulated has been very successful, the bias reduction through the IV approach comes at the cost of an important increase in the variance of the parameters’ estimate. In this work we propose a different solution for the parameters estimation in VRFT which reduces bias without increasing the variance — the Constrained Total Least Squares (CTLS). The effectiveness of the proposed solution is illustrated by three case studies, showing that the mean square error of the parameters’ estimate is smaller when compared to previously proposed solutions and, most importantly, that the closed-loop performance is significantly better.