傳統的土壤液化評估法以安全係數的大小為液化潛能的評估標準,當FS<1,則代表土層會液化,反之FS>1,則代表土層不會液化。有鑑於地震所產生的災害風險是一種不確定性之表現,本質是機率的,而非「會」與「不會」二元式簡單判定。因此,若能運用統計技術及機率理論推估液化發生之可能性,這對液化防制工作而言,將有非常大之幫助。 本研究首先蒐集共669筆國內外SPT-N現場液化與非液化案例調查資料,針對目前廣為採用的SPT簡易評估法(包括Seed法、NJRA法、T-Y法及HBF法等四種方法),分別以邏輯迴歸分析、可靠度理論及貝氏定理等方法建構四種不同機率預測模式。其次藉由可考慮預測模式不確定性折減度之訊息理論,評估不同機率模式之可信度及相對權重,以作為評量各模式之可靠度基礎,同時除了比較各模式間的差異性及相對保守程度外,也探討各模式在不同評估法之適用性。分析結果顯示,四種機率模式之可靠度及適用性皆相當優良,其中各模式在Seed法及HBF法之機率預測表現較佳;模式可靠度部分則以邏輯迴歸模式之可信度最高。 綜整研究成果顯示:本研究建構之四種機率評估模式在推求液化機率方面相當方便實用,值得進一步發展及運用。最後進行各不同簡易評估法推估所得安全係數與液化機率對應關係建立之研究,以提供工程界在從事抗液化工程設計與分析上之參考。
In engineering practices, the assessment of liquefaction potential of soils is usually represented by a safety factor, in which FS<1 indicates that soil will liquefy whereas FS>1 indicates that soil will not liquefy. However, the risk of soil liquefaction resulting from earthquakes contains a lot of uncertainties in nature. Therefore, it will be more suitable if statistics and probability theory can be applied to the estimation of liquefaction potential of soils. For this purpose, 669 domestic and international cases of SPT-N liquefaction and non-liquefaction were collected as the data base in this study. The liquefaction evaluation methods selected for study are the commonly used simplified empirical SPT-N methods, such as the Seed, Japan Road Association, Tokimatsu and Yoshimi, and Hyperbolic Function methods. The probability models applied are based on the logistic regression, the reliability-based theory, and the Bayes’ theorem, respectively. In addition, information theory is then employed to evaluate the reliability of each probability model. Meanwhile, the variation and relative conservatism between models were compared, and the applicability of each model in each evaluation method was also investigated. The results showed that the reliability and applicability of the four probability models were excellent, in which the Seed and HBF methods had a better performance. As for the reliability of different probability models, the one based on the logistic regression was the highest.