|
碩士論文公告區 |
||
| 年度： | 92 | |
| 姓名： | 楊月娥(Yueh-E Yang) | |
| 論文題目(中)： | 應用人工智慧技巧於結構動力參數識別 | |
| 論文題目(英)： | Application of Artificial Intelligence Techniques to Structural Dynamic Parameter Identification | |
| 摘要(中)： | ||
| 台灣由於受到歐亞板塊和菲律賓海板塊之擠壓，因此地震活動十分頻繁，平均每年大小地震超過四千個，其中有感地震可達二百餘個。而現今地震預測科技未臻成熟，無法精確預測地震的發生時間、地點和規模大小，故只能根據規範，將耐震設計理念融入建築物設計中；然而結構物可能因施工方式不當或是於使用期間內遭受地震力、風力等外力之反覆作用，使得材料強度因降伏或疲乏而折減時，導致結構系統參數設計值有所改變，結構物是否仍然保有當初設計時之強度便不得而知，因此發展了各種系統識別方法，以求得結構系統受震後之動力行為與系統參數的變化。 本研究首先利用單層類神經網路進行動力系統參數識別，分別建立單自由度線性系統與多自由度線性系統模型，其中地震紀錄分為二種，一種為利用數值模擬方式產生之人造地震紀錄，另一則為真實地震紀錄，二者量測反應紀錄皆由數值模擬方式產生。於利用單層類神經網路進行識別系統權重值時，共採用了二種方法，一為批次演算法，另一為改良式批次演算法，分別找出該系統最佳之權重值，以求得系統之估測反應，並計算量測值和估測值之誤差指數以評估識別之結果；接著再進行識別含雜訊之輸出入資料的系統權重值，以探討其識別結果。 另外，結合類神經網路與基因演算法於結構動力參數識別，首先應用前述之單層類神經網路架構做為系統動力模式，再藉由基因演算法搜索符合該系統之權重值，然後將權重值代入網路架構以取代解微分方程式求得系統反應，最後再與量測反應做比較；同時亦利用該方法應用於進行識別含雜訊之輸出入資料的系統權重值，以驗證該方法應用於單自由度線性系統及多自由度線性系統識別之可行性。 |
||
| 摘要(英)： | ||
| Located at the active arc-continent collision region between the Luzon arc of the Philippine Sea plate and the Eurasian plate, Taiwan is subject to frequent earthquakes. This continual seismic activity which caused great damages to structures resulted in more emphases on the earthquake resistant design of buildings. Structure properties may be deteriorated and degraded with time in an unexpected way due to randomness in the environment and loadings over its lifetime. In particular, when a structure is exposed to strong earthquake, the properties of the structure may be changed and its behavior after an earthquake can be different from that before the earthquake. In order to realize the dynamic behavior of structural systems, we can determine the dynamic models and parameters by system identification techniques. In this study, the single layer neural network is first employed to identify the system parameters of both the SDOF system and the MDOF system. There are two kinds of earthquakes to be used as the system input, one is the artificial earthquake, and the other is the real earthquake. The associated system response is then calculated assuming the system parameters. In addition to the traditional batch mode algorithm, the improved batch algorithm is also used as the training algorithm, while performing the single layer neural network to the above identification problems. The validity and the efficiency of the proposed algorithms are explored by comparing the results of the predicted response with the measured response for both the SDOF system and the MDOF system with or without noise contamination. Furthermore, the advantages of the genetic algorithm and the above single layer neural network are combined to yield a new identification technique. The network topology is employed to replace the procedure for solving the governing (differential) equation when GA is used to identify the system parameters of both the SDOF system and the MDOF system with or without noise contamination. The comparison is made between the predicted acceleration and the measured one for each case. |
||
| 檔案： | 沒有相關檔案 | |
| 相關連結： | http://ethesys.lib.cyut.edu.tw/ETD-db/ETD-search-c/view_etd?URN=etd-0822104-082740 | |