2023 年 143 巻 3 号 p. 297-304
This paper considers the VRFT (Virtual Reference Feedback Tuning), which is a data-driven controller parameter tuning method using one-shot input-output data, and aims to improve control performance in the case of using noisy data. The paper focuses on pre-filter designs for performance improvement on model matching properties, and proposes a kernel regularization method to mitigate the influence of the noise in the collected data. Furthermore, this paper shows how to determine the hyperparameters in the kernel as well. Finally, the effectiveness of the proposed method is demonstrated using numerical simulations.
J-STAGEがリニューアルされました! https://www.jstage.jst.go.jp/browse/-char/ja/