Bias-variance trade-off issues in robust controller design using statistical confidence bounds

Abstract Robust control aims to account for model uncertainty in design. Traditional methods for robust control typically assume knowledge of hard bounds on the system frequency response. However, this does not match well with system identification procedures which typically yield statistical confidence bounds on the estimated model. This paper explores a new procedure for obtaining a better match between robust control and system identification by using statistical confidence bounds for robust control design. It is shown that by changing the nominal design it is possible to reduce the overall variability from an a-priori specified desired performance. The proposed procedure is particularly straightforward and leads to a closed form solution for the final robust controller.