基於全球資料庫的飽和黏土滲透係數估算

在大地工程領域中，有許多參數行為不易取得，滲透係數就是其中一個例子。為此前人發展出許多經驗式想透過簡單取得的參數來對滲透係數進行推估，然而經驗式往往會受到土壤種類和地區的影響，導致推估值的不確定性過大，且考量滲透係數的高變異性，難以藉由少數點的數值來準確的推估完整行為。而近年來數據分析和機器學習在各領域的應用上皆有一定的突破，因此本研究希望藉此來嘗試對飽和黏土的滲透係數和其完整行為進行推估，並能有效降低推估上的不確定性。首先會先藉由回顧前人文獻，了解與滲透係數有關的參數並加以蒐集，上述流程用以建立資料庫;在資料庫建立完成後，會依照前面文獻中提出的經驗式對資料庫進行簡單的檢核，也藉此觀察資料趨勢。接著將會利用層級貝氏模型，並搭配使用Johnson分布系統、吉普森取樣及貝氏分析中的共軛條件，去學習參數間的相關性且同時捕捉資料庫中的群體(場址)行為，並搭配目標群體(場址)中有限的已知資料來推估未知資料的分佈。從結果來看，觀察到層級貝氏模型隨著已知資訊的增加能有效地降低推估滲透係數的不確定性，在後續若搭配可靠度設計來使用能更加精準並有效降低成本。

關鍵字

數據分析；層級貝氏模型；飽和黏土；滲透係數；資料庫；吉普森取樣法； Johnson分布系統

並列摘要

In geotechnical engineering, many behaviors of parameters are difficult to get, the permeability coefficient is one example. For this reason, many researchers several contributed empirical formulas because they would like to estimate the permeability coefficient by simple parameters; however, empirical formulas are usually influenced by soil types and region, resulting in large uncertainties in estimation, and considering the high variability of the permeability, it is not easy to estimate the complete behavior accurately from only a few points. In recent years, there have been some breakthroughs in the application of data analysis and machine learning in various fields. Therefore, this study is an attempt to estimate the permeability coefficient of saturated clay and its complete behavior, and to effectively reduce the uncertainty in the estimation. The first step is to understand the parameters related to the permeability coefficient by reviewing the previous literature and collecting them. The above process is used to build the database; after the database is built, a simple verification of the database is carried out according to the empirical formula proposed in the previous literature, and the data trend is observed. Then, we will use the hierarchical Bayesian model with the Johnson distribution system, Gibbs sampler, and the conjugate conditions in Bayesian analysis to learn the correlation between parameters and catch the group (site) behavior in the database at the same time, and estimate the distribution of unknown data with the limited known data in the target group (site). From the results, it is observed that the hierarchical Bayesian model can effectively reduce the uncertainty of the estimated the permeability coefficient as the known information increases, which can be more accurate and effectively reduce the cost when used with the reliability design later on.

並列關鍵字

hierarchical Bayesian model ； saturated clay ； permeability coefficient ； database ； Gibbs sampler ； Johnson distribution system

參考文獻

Al-Tabbaa, A., and Wood, D. M. (1987). ”Some measurements of the permeability of kaolin.” Géotechnique, 37(4), 499-514.

Google Scholar

Berilgen, S., Berilgen, M., and Ozaydin, I. (2006). “Compression and permeability relationships in high water content clays.” Applied Clay Science, 31(3-4), 249-261.