非線性迴歸應用於大區域之地下水參數優選

Abstract

本研究的目的在於利用非線性迴歸參數優選模式,建立區域性大尺度及多層地下水數值 模擬時的參數優選技術,以提高數值模擬之精度.一般對於大尺度的問題,參數維度太大以 致計算量太高,一直是參數優選時常面臨的問題,本研究除了應用傳統上的參數化來降低維 度外,為避免過度簡化原來系統,更進一步充份利用問題本身的結構,將一多層次大區域的 地下水流系統,分解成數個子系統,再個別進行優選,以使計算量更有效的降低,研究中之參 數優選部份乃採用美國地質調查局(U.S.G.S)發展之地下水參數優選程式(MODFLOWP),並將 其應用在濁水溪沖積扇實例模擬之參數優選上,經參數優選後之數值模式不但能準確模擬 地下水流況,其抽水量的優選結果亦可作為濁水溪沖積扇地下水管理的參考. The purpose of this study is to develop a parameter identification techniqu efor large scale groundwater simulation basing on non-linear regression method .In general, the reguirement of large amount of computational effort is always the problem for large scale parameters optimization. Besides the traditionalp arameterization, this research develop model decomposition technique to effec- tively reduce the computational dimensionof parameters to be optimized. Thepro cedure, basing on the model structure, decomposed a large scale multi-layersgr oundwater system into several subsystems. Each subsystem consists only threela yers and the optimization computation is focused on that. The results is thenm odified by the cross layer mass balance. The program of MODFLOWP is used to pe rform the parameter optimization for each subsystem, which is developed byUnit ed States Geological Survey. This procedure is applied on the Cho-Shui Riverfa n to verify the capability of the technique. The results demonstrates that the optimized parameters can accurately reproduce the observed data. The optimized pumping rate is also a valuable reference for groundwater management of Cho-Sh uiRiver.