2021 Fiscal Year Final Research Report
Data-driven design of controller with ultra-high degree of freedom
Project/Area Number |
19K15017
|
Research Category |
Grant-in-Aid for Early-Career Scientists
|
Allocation Type | Multi-year Fund |
Review Section |
Basic Section 21040:Control and system engineering-related
|
Research Institution | The University of Kitakyushu |
Principal Investigator |
Fujimoto Yusuke 北九州市立大学, 国際環境工学部, 准教授 (60826204)
|
Project Period (FY) |
2019-04-01 – 2022-03-31
|
Keywords | データ駆動制御 / 正則化 |
Outline of Final Research Achievements |
The purpose of this project is to design a controller in a data-driven manner especially from the data observed in one preliminary experiment. Although typical controllers have several (up to dozens of) parameters, such degrees of freedom would be small for control of a complex system. Hence this work focused on data-driven designs of controller with hundreds/thousands of parameters. In more detail, we established some methods to tune parameters of such ultra-high degree of freedom controllers, gave noise reduction methods, and validated their effectiveness through practical experiments. In particular, related to noise reduction methods we developed some methods and made some contributions from the viewpoint of regularization theory.
|
Free Research Field |
制御工学
|
Academic Significance and Societal Importance of the Research Achievements |
制御工学自身は,モータあるいはそれを組み込んだロボットハンド,車両の速度制御など多様かつ産業上も重要な応用を数多く持つ.通常は対象の数理モデルを基に制御器を設計するが,制御対象は年々複雑化しており,モデルを得ること自体が難しい.そこで,データから直接制御器を設計するデータ駆動制御が近年注目されている.一方で既存のデータ駆動制御は高々十数個程度の可調整パラメータからなり,極めて複雑な対象に対しては限定的な性能しか出せない.そこで本研究では数百から数千ものパラメータを持つ制御器をデータから設計した.ただし,パラメータ数が多いとノイズの影響を受けやすい.その影響の低減が本研究の一つの主眼であった.
|