| 研究生: |
簡少禹 Shao-Yu Chien |
|---|---|
| 論文名稱: |
含物理約束之長短型記憶神經網絡模型於結構物動力反應預測之開發與應用 |
| 指導教授: |
陳鵬宇
Peng-Yu Chen |
| 口試委員: | |
| 學位類別: |
碩士 Master |
| 系所名稱: |
工學院 - 土木工程學系 Department of Civil Engineering |
| 論文出版年: | 2024 |
| 畢業學年度: | 112 |
| 語文別: | 中文 |
| 論文頁數: | 246 |
| 中文關鍵詞: | 深度學習 、長短期記憶模型(LSTM) 、物理約束 、MATLAB 、OpenSees 、結構動力反應預測 |
| 外文關鍵詞: | Deep Learning, Long Short-Term Memory (LSTM), Physical Constraints, MATLAB, OpenSees, Structural Dynamic Response Prediction |
| 相關次數: | 點閱:18 下載:0 |
| 分享至: |
| 查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報 |
監測結構物受地震力作用下之歷時響應為評估結構物健康狀況的重要依據,目前國內外多於結構物內各樓層安裝加速度傳感器以監測該樓層之地震響應,但其於實際應用上易受成本與施工之限制。鑒於我國地震頻繁,更應著重結構物之健康監測,因此,本研究嘗試使用含物理約束之長短型記憶神經網絡模型,透過使用地表加速度即可對房屋結構受震反應進行預測,期望以此提高監測之效益以及提升民眾加裝監測儀器之意願。
本研究使用長短期記憶模型(Long Short Term Memory networks, LSTM)並加入物理約束,利用結構動力學原理來降低大量訓練數據之需求,同時考量在地震下之運動狀態,以提高模型預測之合理性與可靠性。其中,訓練與預測之數據來自MATLAB、OpenSees有限元素軟體之模擬數據與美國Center for Engineering Strong Motion Data收集之一棟六層樓鋼筋混凝土建築的地震紀錄,此兩種類型之數據提供了模擬與真實響應之比較,而預測結果顯示,使用含物理約束之模型在模擬數據集上的預測表現優於傳統模型,但在真實數據集上,因結構物運動反應較為複雜,物理約束之改善效果有限。本研究展示了多種示例,並對預測結果進行詳細分析與評估,期望為結構物受震反應預測提供可靠之參考依據。
Monitoring the time-history response of structures under seismic forces is crucial for assessing their health. Currently, accelerometers are installed on various floors within structures both domestically and internationally to monitor seismic responses. However, practical applications are often limited by costs and construction complexity. Given the frequent seismic activity in our country, greater emphasis should be placed on structural health monitoring. Therefore, this study attempts to use a physically constrained Long Short-Term Memory (LSTM) neural network model to predict the seismic response of buildings using only ground acceleration, aiming to improve monitoring efficiency and increase public willingness to install monitoring devices.
This study employs LSTM models with physical constraints, utilizing principles of structural dynamics to reduce the need for extensive training data while considering motion states under seismic conditions. This approach aims to enhance the model's prediction rationality and reliability. The training and prediction data come from simulated data using the MATLAB, OpenSees finite element software and seismic records of a six-story reinforced concrete building collected by the Center for Engineering Strong Motion Data (CESMD). These two types of data provide a basis for comparing simulated and actual responses. The prediction results show that the physically constrained model outperforms traditional models on simulated datasets. However, on real datasets, due to the more complex motion response of actual structures, the improvement effect of physical constraints is limited. This study presents various examples, conducts detailed analysis and evaluation of the prediction results, and aims to provide reliable references for predicting the seismic response of structures.
[1] Ning, C., Xie, Y., & Sun, L. (2023). LSTM, WaveNet, and 2D CNN for nonlinear time history prediction of seismic responses. Engineering Structures, 286, 116083.
[2] Zhang, R., Liu, Y., & Sun, H. (2020). Physics-informed multi-LSTM networks for metamodeling of nonlinear structures. Computer Methods in Applied Mechanics and Engineering, 369, 113226.
[3] Eshkevari, S. S., Takáč, M., Pakzad, S. N., & Jahani, M. (2021). DynNet: Physics-based neural architecture design for nonlinear structural response modeling and prediction. Engineering Structures, 229, 111582.
[4] CESMD Information for Strong-Motion Station San Bernardino - 6-Story Hotel CGS - CSMIP Station 23287. (n.d.). CESMD. https://www.strongmotioncenter.org/cgi-bin/CESMD/Multiplesearch1_DM2.pl?event_name=&magmin=&magmax=&byear=&eyear=&country=Any&state=Any&stn_ident=&network=CE&sta_number=23287&type=Any&Material=Any&Height=&siteclass=Any&vs30min=&vs30max=&accmin=&accmax=&hdistmin=&hdistmax=
[5] Mazzoni, S., McKenna, F., Scott, M. H., & Fenves, G. L. (2006). OpenSees command language manual. Pacific Earthquake Engineering Research (PEER) Center, 264(1), 137-158.
[6] 黃炫文 許博策 盧煉元 蕭輔沛 葉士瑋. (2021, November 5). 開放式地震工程模擬系統(OpenSees)於含塑鉸非線性結構之分析應用. 國家地震工程研究中心. https://www.ncree.narl.org.tw/accomplishment/technicalreports/page/12113
[7] Mazzoni, S., McKenna, F., Scott, M. H., & Fenves, G. L. (2003). OpenSees Example Manual. PEER, Berkeley: University of California, 2003: 52.
[8] 內政部營建署. (2022). 建築物混凝土結構設計規範 (112年版). 內政部營建署. https://glrs.moi.gov.tw/LawContent.aspx?id=GL000236
[9] Zhang, R., Liu, Y., & Sun, H. (2020). Physics-guided convolutional neural network (PhyCNN) for data-driven seismic response modeling. Engineering Structures, 215, 110704.
[10] PyTorch開發團隊. (2019). PyTorch . https://pytorch.org/