2023 Fiscal Year Final Research Report
Development of an AI-Technology-Based Control Method with Stability Guarantees
| Project/Area Number |
21K14178
|
|---|
| 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 | Gunma University |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2021-04-01 – 2024-03-31
|
| Keywords | 分散制御 / 強化学習 / レトロフィット制御 |
|---|
| Outline of Final Research Achievements |
A novel control method that combines reinforcement learning with retrofit control, a technique used when only part of the system model is known, has been proposed. Specifically, a method for learning control laws while retaining the rectifier, a distinctive structure of retrofit control, has been investigated. This approach enables adaptation to environmental changes through reinforcement learning while theoretically ensuring system stability. The effectiveness of the proposed method has been validated through numerical simulations on a simplified power system model and simple real-world experiments.
|
| Free Research Field |
システム制御工学
|
| Academic Significance and Societal Importance of the Research Achievements |
一般に，強化学習は環境の変化に適応する制御手法として知られているが，制御系の最も基本的で重要な性質の安定性を理論的に保証することが困難であった．本研究では制御を行う前のシステムが安定であるという事前情報を活用し，レトロフィット制御理論を応用することにより，強化学習による制御系の安定性を理論的に保証することができた．これにより，産業界などでの実用化の際に，学習途中での機器の暴走などを防止することが可能となる．
|
