利用類神經網路推估非等向性含水層參數之研究

Abstract

含水層特性參數的調查是瞭解地下水流通特性的重要工作，含水層的空間變異性之大雖廣為認知，然目前利用抽水試驗的方式求得簡單的地下水有效水文參數，諸如有效流通係數、有效水力傳導係數、儲水係數等等。本研究利用修正型ANN模式、修正型PCA-ANN模式結合Papadopoulos解析解運用於地下水文有效參數推估以提高精確性，其所推估有效參數近似於幾何平均值。另外，再利用曲線法、直線法所推估參數代入Papdopulous 解析解運用於濁水溪沖積扇之現地抽水試驗(溪洲站)。由結果顯示，修正型ANN模式及修正型PCA-ANN模式所推求目標函數誤差值較小，原因為類神經網路模擬抽水試驗資料與井函數建立關係以優選最佳參數值，並建立一套推定含水層參數(儲水係數、流通係數)之自動程序，提高推求參數之精確性；而曲線法與直線法之目標函數較大原因，圖形套配過程中容易造成人為誤差以致目標函數提高。當水文地質參數推估確定後，本研究在區域尺度中，利用MODFLOW-96模式模擬地下水流空間分佈初步優選研究區域最佳地下水可開發水量區域，並希爾法以安全出水量驗證。優選後之最佳地下水可開發水量區域，並進行現地抽水試驗，依據洩降水位資訊利用修正型ANN模式推求滲透性方向性及大小，並利用滲透性較佳水文地質之方向性進一步優選抽水井位置最佳配置，將可提高抽水率，以提供地下水管理之參考。

Obtaining reasonable hydrological parameters is a key challenge in groundwater modeling. The analysis of the temporal evolution of pump-induced drawdown is a common approach to estimate the effective transmissivity and storage coefficients in the heterogeneous aquifer. In this study, we compared modified ANN model, modified PCA-ANN model cooperated with Papadopoulos solution, and applied them to the estimation of effective aquifer parameters with better precision. Results show that modified ANN model and modified PCA-ANN model can efficiently obtain the most representative (or effective) aquifer parameters wich are close to geometric means. In-situ time-drawdown from Shi-Chou station on the Choushui River alluvial fan, Taiwan, is further adopted to test the applicability and reliability of the proposed methods, as well as provide a basis for comparison with the Straight-line method (SLM) and the Type-curve method (TCM). Results show that both of the modified methods give better estimation than SLM and TCM in terms of RRMSE. Besides, model uncertainties could be easily induced with the application of the TCM or SLM for not using a systematic approach to optimize the parameters of well functions. In order to diminish the uncertainties, relationships between well function and simulated drawdown are established by modified ANN and modified PCA-ANN to automatically optimize those parameters. Aforementioned methods are further applied to optimize the location of pumping wells in a local-scaled numerical model, MODFLOW-96. The optimal zone for exploiting groundwater could be initially decided by modeling the groundwater flow distribution, and validated by Hill method. Pumping tests are carried out in the optimal zone to generate time-drawdown data for modified ANN method to estimate the value and direction of transmissivities. Optimization of the location of the pumping wells is conducted in the direction with major transmissivities in order to increase pumping efficiency. This study expects to develop a mechanism of parameters optimization for groundwater management.