左天雄2006-07-252018-07-092006-07-252018-07-092002http://ntur.lib.ntu.edu.tw//handle/246246/2785本研究施行現地連續表面波試驗(CSWT) 分析 頻散曲線及電子震測錐試驗(SCPT) 量測土層剪力 波速。另進行波傳矩陣法之理論頻散曲線分析。以 理論頻散曲線、現地頻散曲線、及SCPT 量測之土 層剪力波速，建立訓練及測試資料為80 %及20 % 之倒傳遞類神經網路，進行現地頻散曲線評估土層 剪力波速分析。網路選用雙層隱藏層及神經元個數 為40 之架構、以亂數決定網路權重初始值、訓練 時期停止誤差值為10-3，習速率在0.1～1 區間、及 訓練次數小於100。 個案研究結果顯示，以倒傳遞類神經網路，進 行頻散曲線映射土層剪力波速分析之結果，具有避 免傳統之建立繁瑣力學模式反算土層剪力波速、分 析結果有一致性及可靠性。The continuous surface wave tests (CSWT) to access dispersion curves and electronic seismic cone penetration tests (SCPT) to evaluate shear wave velocity were performed by this study. Furthermore, the theoretical dispersion curves are developed by matrix wave propagation method. Based on theoretical dispersion curves, field dispersion curves, and shear wave velocity of soil stratum obtained by SCPT, the backpropagation neural networks (BPNN) are built with 80% training data and 20% test data to evaluate shear wave velocity of soil stratum from field dispersion curves. The BPNN used in this study have two hidden layers of 40 neurons. The training BPNN are with initial weights and biases created by random numbers, the goal error of 0.001, the learning rate of 0.1~1, and training cycles less than 100. Case study shows that the results of the mapping of shear wave velocity of soil stratum from field dispersion curves by BPNN may avoid creating complicated mechanical models to do inversion method. Comparisons indicate that the neural network models have consistence and reliability.application/pdf516686 bytesapplication/pdfzh-TW國立臺灣大學土木工程學系暨研究所頻散曲線剪力波速連續表面波試驗(CSWT)電子震測錐試驗(SCPT)倒傳遞類神經網路(BPNN)dispersion curveshear wave velocitycontinuous surface wave test (CSWT)electronic cone penetration test (SCPT)backpropagation neural networks (BPNN)reporthttp://ntur.lib.ntu.edu.tw/bitstream/246246/2785/1/902211E002097.pdf