Robust Reference Signal Self-Organizing Control based on Deep Reinforcement Learning

カテゴリ: 論文誌(論文単位)

グループ名: 【D】産業応用部門（英文）

発行日: 2022/11/01

タイトル(英語): Robust Reference Signal Self-Organizing Control based on Deep Reinforcement Learning

著者名: Hiromichi Iwasaki (Graduate School of Science and Technology, Tokai University), Atsushi Okuyama (Department of Precision Engineering, Tokai University)

著者名(英語): Hiromichi Iwasaki (Graduate School of Science and Technology, Tokai University), Atsushi Okuyama (Department of Precision Engineering, Tokai University)

キーワード: adaptive control，control theory based on machine learning，reference signal self-organization，deep reinforcement learning，track-following control，inverted pendulum with inertia rotor

要約(英語): In general, the effects of modeling errors, parameter variations, and external disturbances of the controlled object degrade tracking control performance. To address these problems, we have developed a control method based on deep reinforcement learning. Accordingly, a reference signal self-organizing control system based on a deep deterministic policy gradient (DDPG) is proposed, which is an extension of an existing control system using DDPG. In a previous study, we confirmed the realization of the swing-up and stabilizing motions of an inverted pendulum using the proposed control system(8). However, the addition of a new ability to the system could not be verified in that study. Thus, in this work, we aim to verify whether a new function can be added to the proposed control system. By performing a control simulation, we verified whether the proposed system can achieve robustness by using the inverted pendulum with an inertia rotor. A control simulation of the system was performed by adding noise into the system, and its control performance was investigated to confirm the robustness of the system. The simulation results indicate that the pendulum could not be inverted using an un-retrained control system. However, it was confirmed to have been inverted and stabilized by the swing-up and stabilizing control using a retrained control system. Moreover, the retrained control system could effectively function under the effect of noise with an accuracy close to that of a noise-free state. Therefore, we confirmed that the addition of robustness can be realized in the proposed control system.

本誌: IEEJ Journal of Industry Applications Vol.11 No.6 （2022）

本誌掲載ページ: 737-743 p

原稿種別: 論文／英語

電子版へのリンク: https://www.jstage.jst.go.jp/article/ieejjia/11/6/11_21005735/_article/-char/ja/