現地試驗(SPT及CPT)評估土層剪力波速-基因演算法最佳化類神經網路

Abstract

本研究分別以標準貫入試驗(SPT)及圓錐貫入 試驗(CPT)所獲得之土壤參數，建立評估土層剪力 波速之類神經網路模式，並以基因演算法最佳化 類神經網路之資料結構、隱藏層數、神經元數及 訓練次數等相關參數。個案分析結果顯示，本研 究所建立之基因演算法最佳化類神經網路 (Genetic Algorithms Optimizing Neural Networks, GAONN)評估結果，與現地震測試驗之剪力波速 一致性較高，且優於傳統之迴歸經驗公式。

Using optimizing artificial neural networks (ANN) with genetic algorithms (GAs), the shear wave velocity of soil stratum may be accessed by the parameters obtained from standard penetration tests (SPT) and electronic cone penetration tests (CPT) respectively. The number of hidden layers, neurons, epochs of ANN and the data applying to that are obtained by GAs.To verify the methodology, shear wave velocities evaluated by ANN may be confirmed by compared with the results of in-situ seismic tests, like SCPT or Cross-hole. In this study, shear wave velocities predicted by ANN correspond to the results of in-situ seismic tests and are more reliable than those obtained by traditional regression method.