污染物於含水層中的傳輸行為常藉由移流-延散方程式描述，其中延散度為重要輸入參數，而推估延散度須先進行現地追蹤劑試驗，分析追蹤劑試驗濃度穿透曲線，並藉由數學模式所產生之標準曲線套配進行參數推估即可求得試驗場址之延散度，傳統標準曲線套配法需花費大量時間及具有曲線擬合不佳等缺點，造成實際應用的難題。本研究提出以類神經網路模式結合二維徑向收斂流場追蹤劑試驗模式建構倒傳遞套配模式(Back Propagation Neural Network Fitting Model, 簡稱BPNFM)提高延散度推估效率及精確度。結果顯示有效孔隙率介於 範圍內網路輸出值與目標輸出值之相對誤差均低於0.9 %；縱向延散度介於 範圍內相對誤差均低於3%；側向延散度介於 範圍內相對誤差均低於0.25 %，各套配模式在其適用範圍內均可獲得良好之輸出精確度。而套配效率上，倒傳遞套配模式可大幅縮短標準曲線套配法套配過程所耗費的時間，因此證實二維徑向收斂流場追蹤劑試驗套配模式可有效率地套配追蹤劑試驗數據，獲致可靠之延散度參數。 The transport process of solute in the aquifer as widely described by advection-dispersion equation (ADE). Dispersivities are the important input parameters for ADE. To obtain those parameters, a general methodology suggested to analyze the breakthrough curves (BTCs) plotted from tracer test. However, the artificial curve fitting causes lots of time consuming and errors. In this study, we try to promote one efficiency and accuracy method to estimate dispersivities. A back propagation neural network fitting model (BPNFM), combing with the neural network model and the two dimensional radially convergent flow tracer test is developed. Results show that the relative errors between target and network output data of effective porosity fall in the region from 0.05 to 0.5 is less than 0.9%; of longitudinal dispersivity distributed from 0.2m to 40m is less than 3%; of transverse dispersivity distributed from 0.033m to 6.427m is less than 0.25%. The consuming time can be reduced significantly by BPNFM, while the predicted parameters fall in the model setting, BPNFM is a accuracy and efficiency method for analyzing the dispersivities of the tracer tests.
