地下水為臺灣重要之水資源之一，過度使用地下水會引發海水入侵、地層下陷等諸多問題，最常見蓄水方法是使用地表水庫，然而，由於土地徵用、污染問題、蒸發或入滲消耗，過度仰賴地表水庫有時會出現問題，此外，在沿海等相對平坦的地區建設水庫會遇到許多限制。含水層儲蓄回抽 (aquifer storage and recovery，簡稱ASR) 其原理是將從地表收集豐水期的水將其注入地下含水層中進行儲存，以供未來需要時使用的儲水方式。進行鑽探本身費時費力，在所有地點進行顯然不可行，因此人工神經網路(ANN)是一個利用有限的資料推估整個範圍資料的一種方式。本研究的目的為利用ANN進行空間內插找出台灣適合進行ASR的可行場址，將座標與觀測井的直線距離作為人工神經網路的輸入參數，以流通係數數值與地下水水質作為輸出結果。人工神經網路結果顯示在2層隱藏層且每層神經元個數為16時有最好的預測效果 ，最後加入河川分布圖篩選台灣ASR優選場址，以確保廠址周圍具有水源，結果可供政府機構判斷適合ASR的場址。;Groundwater is one of Taiwan′s important water resources. Excessive use of groundwater can lead to various problems such as salt water intrusion and land subsidence. The most common method of water storage is through surface water reservoirs. However, over-reliance on surface water reservoirs can sometimes pose problems due to land acquisition, pollution issues, evaporation, or infiltration. Additionally, constructing reservoirs in relatively flat areas, such as coastal regions, faces many restrictions. Aquifer Storage and Recovery (ASR) is a water storage method that involves collecting water from the surface during periods of abundance and injecting it into underground aquifers for storage, to be used in the future when needed. Drilling itself is time-consuming and labor-intensive, and it is obviously not feasible to conduct it in all locations. Therefore, artificial neural network (ANN) is a way to use limited data to estimate the entire range of data. The purpose of this study is to use ANN for spatial interpolation to identify feasible sites in Taiwan for ASR. The straight-line distance between coordinates and observation well locations is used as input parameters for the artificial neural network, with transmissivity values and groundwater quality as output results. The results from the artificial neural network indicate that the best predictive performance occurs with two hidden layers, each with 16 neurons. Finally, a river distribution map is incorporated to filter and select optimal ASR sites in Taiwan, ensuring that there is a water source around the selected locations. These results can be used by government agencies to determine suitable ASR sites.
