既存制御の操作量活用による強化学習の効率性向上技術

カテゴリ: 部門大会

論文No: GS5-4

グループ名: 【C】2022年電気学会電子・情報・システム部門大会

発行日: 2022/08/24

タイトル(英語): Improving the Efficiency of Reinforcement Learning by Utilizing the Operation of Existing Controller

著者名: 秦 洋（東芝インフラシステムズ）,高野 俊也（東芝インフラシステムズ）

著者名(英語): Yang Qin (TOSHIBA Infrastructure Systems & Solution Cooperation),Toshiya Takano (TOSHIBA Infrastructure Systems & Solution Cooperation)

キーワード: 強化学習|PI制御|協調制御|学習効率性|溶存酸素濃度制御|reinforcement learning|PI controller|collaborative control|training efficiency|Dissolved Oxygen control

要約(日本語): Reinforcement learning (RL) methods have achieved impressive results on a wide range of control problems. However, they usually require an extensive amount of training data in order to converge to a meaningful solution, which largely prohibits their usage and reliability for complex control tasks in the real world. To tackle this issue, we suggest a novel collaborative control method that combines the reinforcement learning and existing controller, e.g., Proportional Integral (PI) controller, to improve the training efficiency of reinforcement learning. Specifically, we evaluate the proposal on a Dissolved Oxygen (DO) control problem in a simulated wastewater treatment system. Evaluation results demonstrate that the training time is greatly reduced greatly through the method.