Data-driven controller tuning based on unfalsified control for sensitivity minimization

By achieving low sensitivity, it is possible to suppress influence of perturbation in a plant and disturbances. In modelbased controller syntheses, the sensitivity function is minimized using a mathematical model of a plant while internal stability of the closed-loop system is maintained. However, there are some difficult cases where a precise mathematical model of the plant cannot be obtained. Moreover, if there is a large error between a plant model and an actual plant, the designed controller might not show desired performance, at the worst case, the system would be destabilized. This paper focuses on data-driven controller tuning methods because they does not need system identification and design controller by using only input-output data. This paper proposes a design method to achieve sensitivity minimization by using only input-output data. © 2017 The Institute of Electrical Engineers of Japan.