| Summary: | 碩士 === 國立交通大學 === 土木工程系所 === 107 === The main purpose of this study is to implement electric resistivity technique as a non-destructive geophysical method to investigate the groundwater table level of the unconfined aquifer. Current literatures revealed that the most adopted electric resistivity technique in groundwater table (GWT) investigation is the two-dimensional electric resistivity tomography (2D-ERT) method. However, the inversion scheme used in 2D-ERT may induce various uncertainties during the GWT determination, and the value discrepancy between the inverted apparent resistivity and actual resistivity may be large and unrealistic. On the other hand, the one-dimensional vertical electrical sounding (1D-VES) allows the variations of thickness and resistivity values of each assumed layers, which forms an advantage for this method in the GWT investigation. This study focused at the feasibility assessment of VES in the GWT level monitoring and proposed appropriate inversion parameters according to the study outcomes.
This study tackled this issue using numerical simulation, by first defining the scope for a case, where the simulated model consist of a homogeneous layer with GWT at 2.5m. The GWT depth detection performances of both the 2D-ERT and 1D-VES are evaluated through numerical simulation, in order to verify the potential advantage of 1D-VES in this specific task. Upon confirming the forte of 1D-VES, multi-layered strata was next simulated for application testing. However, it was found that for scenarios with numerous local minimums, the resultant outcome from 1D-VES is highly dependent on the given initial parameters, if the commonly used 1D-VES linear inversion approach is adopted for the GWT detection. This led to the difficulty in GWT interface determination and impacted the robustness of 1D-VES. Based on this premise, this study pivoted to a global non-linear inversion algorithm to minimize impact arising from local minimums in conventional 1D-VES.
This study adopted simulated annealing optimization approach as the global optimization algorithm in the GWT layer detection. This is performed through forward-modelling the electrical resistivity data and process the simulated data using simulated annealing in Matlab. Four soils with different capillary degree was simulated in this study and appropriate inversion parameters (such as initial guess and search range) were obtained through thorough examination. From the numerical simulation, the impact of capillary phenomenon towards the inverted outcome increases as the GWT becomes shallower. This study therefore sets the simulated GWT level as 2.5m deep for all four simulated soil models. This study attempted to increase the amount of inverted layers in order to mimic the soil resistivity change due to capillary effect. Theoretically, as the inverted layer increases, the capillary phenomenon from above the GWT should be better simulated and resulted in lower error. However, the optimization stability of simulated annealing approach was severely impacted as the amount of inverted layer and inverted parameters increases. Based on the simulated outcome, 3 layers were found as the optimal number of inverted layers, which would reduce inversion error effectively and enhance the outcome stability concurrently. Since simulated annealing approach is computationally expensive and the increment of inverted layers may render this algorithm unstable, this study recommended that the computation enhancement this optimization algorithm could be further improved and other global optimization method could also be evaluated.
|
|---|
