Machine learning and reinforcement learning for feature analysis of decent solution and optimization of steel structures
| Project/Area Number |
18K18898
|
|---|
| Research Category |
Grant-in-Aid for Challenging Research (Exploratory)
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Medium-sized Section 23:Architecture, building engineering, and related fields
|
|---|
| Research Institution | Kyoto University |
|---|
Principal Investigator |
Ohsaki Makoto 京都大学, 工学研究科, 教授 (40176855)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
寒野 善博 東京大学, 大学院情報理工学系研究科, 教授 (10378812)
木村 俊明 京都大学, 工学研究科, 助教 (60816057)
|
|---|
| Project Period (FY) |
2018-06-29 – 2020-03-31
|
| Project Status |
Completed (Fiscal Year 2019)
|
| Budget Amount *help |
¥5,980,000 (Direct Cost: ¥4,600,000、Indirect Cost: ¥1,380,000)
Fiscal Year 2019: ¥2,210,000 (Direct Cost: ¥1,700,000、Indirect Cost: ¥510,000)
Fiscal Year 2018: ¥3,770,000 (Direct Cost: ¥2,900,000、Indirect Cost: ¥870,000)
|
|---|
| Keywords | 構造最適化 / 機械学習 / 強化学習 / 鋼構造骨組 / 建築骨組 / 構造設計 / 最適化 |
|---|
| Outline of Final Research Achievements |
A machine learning method has been developed for extracting the features related to decent mechanical performances of steel building frames. Using the proposed metod, meta-level knowledges are obtained for locations and cross-sectional properties of members, and decent designs are found with small computational cost. Furthermore, the sequential process of design update is modeled as a Markov decision process and an agent is trained using reinforment learning. As a result, an agent is developed to find designs that satisfies various design requirements. It has also been shown that the optimization method often used machine learning has better performancs for a certain type of optimum design problem than the conventional methods.
|
| Academic Significance and Societal Importance of the Research Achievements |
大規模建築鋼構造骨組の優れた力学的性能につながる部材配置や部材断面の特徴量を，機械学習の手法であるサポートベクターマシンで求める手法を提案し，その手法を用いて，静的地震荷重に対する応答を最小化する問題に対する近似最適解を，少ない計算量で得られることを示した。また，構造設計において部材断面を逐次変更するプロセスをマルコフ決定過程としてモデル化し，強化学習を用いて最適方策を求めることにより，さまざまな実際的条件を考慮した構造設計を行うエージェントを開発した。さらに，交互方向乗数法や次元削減などの機械学習で用いられる手法に基づいて，扱いにくい最適設計問題に対して優れた近似解を得る手法を提案した。
|
Report
(3 results)
Research Products
(20 results)
