|
碩士論文公告區 |
||
| 年度： | 90 | |
| 姓名： | 陳明澤(Ming-Ze Chen) | |
| 論文題目(中)： | 用類神經網路建立以SPT、CPT及Vs為主之臨界液化曲線 | |
| 論文題目(英)： | Use of artificial neural network to obtain the critical limit state curves based on SPT, CPT, and Vs data | |
| 摘要(中)： | ||
| 評估液化潛能最終目的，即由計算結果加以分析，試圖畫出能夠區分液化及非液化區域之臨界曲線。而土壤液化現象卻為一非常複雜之土壤行為，傳統液化潛能評估較常用之「簡易經驗評估法」，因評估時缺乏考慮土壤液化之間高維度相關參數影響，且繪製臨界狀態曲線皆依靠大量經驗判斷，往往有過於保守之現象。因此，有必要去尋找一種可以解決高維度非線性連結關係，且能避免人為因素影響之工具。 近年來廣泛應用在模擬大地工程非線性問題之類神經網路(ANN)就具備上述要求，其概念即藉由模擬生物腦神經細胞可以快速處理高維度問題能力，經由網路輸入與輸出關係建立訓練模式，當網路收斂時再擷取各神經元相關影響權重，並建立巨集模式，未來便可將新的參數輸入網路中，即能預測輸出值結果。 本研究即採用學習精度高、演算方法簡單且適用於土壤液化分類預測問題之倒傳遞類神經網路模式，對SPT-N、CPT-qc、VS參數為主之資料點進行網路訓練，再配合國內中部地區土壤具有高細粒料含量及高反覆剪應力比之特性，分別建立不同狀態之臨界液化曲線，並利用誤判率及決定係數(R2)檢驗網路準確性，以提供設計者在評估液化潛能時有更為合理之分析模式。結果顯示，以370組SPT、389組CPT及217組Vs資料點所建立之臨界液化曲線，在判斷現地液化表徵方面，乃較傳統簡易經驗曲線能夠有效地反應現地液化情形。 |
||
| 摘要(英)： | ||
| The objective for evaluating liquefaction potential of soils is to use a simple mechanism to define a limit state curve, i.e., separating liquefaction cases form non-liquefaction cases. Because little consideration was given to possible interactions among many soil factors, simplified methods were used to obtain the limit state curves all relied heavily on engineering judgment. However, these a way for highly non-linear and rational approach to defining limit state curve is needed. Artificial neural network (ANN) has the foregoing requisitions which generally apply the non-linear system to geotechnical engineering. The neural network model with a set of multidimensional connection weights and biases. This model, a successfully trained network, can easily be implemented in a spreadsheet, and for a given set of input data, the occurrence of liquefaction and/or non-liquefaction can be predicted. Back-propagation neural network model has a high degree of accuracy in classifying and predicting the occurrence of liquefaction and/or non-liquefaction. Therefore, in order to find the limit state curves, this model is used to train and test the database of SPT-N, CPT-qc, and Vs. Data obtained in Taiwan after 921 Chi-Chi earthquake, which had high fines content and cyclic stress ratio, were used to establish the boundary curves for high fines content. Probability of prediction and correlation coefficient (R 2) were used to examine the correctness of the ANN model. The results show that the obtained critical limit state curves are able to predict the actual performance pretty well. |
||
| 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-0828102-145954 | |