| Summary: | 碩士 === 國立臺灣大學 === 土木工程學研究所 === 83 ===
This study develops a methodology for identification of distributed transmissivity in a one-dimensional unsteady heterogeneous groundwater system. This method is a Bayesian estimation method for it incorporate the prior information ( observed transmissivity), which are used to construct parameter statistical structure, and used this spaual structure can estimate the unknown parameter by Kriging method. The posterior information ( observed piezometric head ), which can be to used to modify coefficients of parameter of the simulation model, in an optimized way.
In this study, Kriging, an unbiased minimum variance estimator, is used to reconstruct the transmissivity distribution for the entire flow domain. By assuming a polynomial drift function and an exponential type semivariogram, the number of parameter is greatly reduced. In the simulation model, the exponential scheme is used for caculation. A Quasi-Newton method is implemented for the model to find the best direction of parameter modification. An influence coefficient technique is used to derive the relationship between the model parameter and the piezometric heads.
This model is tested by an artifically generated aquifer. Good results was shown in estimation of transmissivity using this identification model. But, while the number of parametre to be identified is increased, the computer time is also increased. It is found that if the piezometric head observations increased, the proposed approach give a better estimation of the model parameters, and the number of iteration need decreases.
|
|---|
