類神經網路於入滲池最佳化設計之應用

Abstract

地下水為一寶貴之水資源，可經由適當之經營管理以求獲得最大之預期效應。部份地區由於過渡抽取地下水，造成海水入侵、地盤下陷等傷害，可利用各種「人工補注」方法，以作為地下水之貯蓄補充。 本研究主要目的乃針對一已知地質參數之空曠平地，以不同大小、不同入滲池數量組合，以及細顆粒淤積對入滲池補注之影響作探討，並分析於實際營運時，空池清底頻率對長期補注量之影響。研究步驟可分為兩大主題，一為類神經網路之訓練，另一為優選模式之建立。本研究中，類神經網路的學習對象為2DFEMFAT模式，其目的除節省模式的計算量外，可經由類神經網路解析式推求補注量對各入滲池幾何變數之敏感度；至於參數優選模式則分別以固定時間內之最大地下水補注量，與固定時間內之最大總用水效益為目的，可求得各自的最佳入滲池幾何參數組合與最佳空池清底頻率。一般均認為淤泥沈積速率變大與淤泥水力傳導係數變小需要較快之清池頻率，然而本研究優選結果顯示過大之淤泥沈積速率與過小之淤泥水力傳導係數則未必符合前述之結果。

Artificial recharge can be an important component in a regional groundwater management. Among the feasible alternatives on groundwater artificial recharge, recharge pond is the most popular filed practice. Nevertheless, there are very few studies on the optimal design and operation on recharge pond. Base on above highlight, this research is to investigate the general guideline related to the optimal design and operation policy on the recharge pond. The Genetic Algorithms (GAs), Artificial Neural Network (ANN) and numerical simulation are applied to study the problem. A synthetic procedure based on the 2D-FEMAT is performed to simulate the recharge rates of a three-dimension recharge pond. Hundreds of simulation results are then generates by varying the pond dimensions, which are the training data of the ANN. Finally, the optimal dimension and operation policy are calculate by the GAs and trained ANN. The scheme is then applied to solve the optimal dimension and operation policy for several hypothesis cases including one pond, two ponds or four ponds. The decay of infiltration rates caused by the sedimentation of fine particles is also investigated. The numerical study demonstrate that the permeability of the sediment and the sedimentation rates significantly determine the decay rates of infiltration and affect the size of recharge ponds and operation policy. The study also illustrates that the proposed scheme can obtain the optimal dimension and operation policy under steady state assumptions and is a valuable planning tool for the design and evaluation of groundwater recharge ponds.