Abstract There is an increasing interest in utilizing surrogate models to perform the global sensitivity analysis (GSA) for the uncertainty quantification since they allow to quantify efficiently the relative importance of each input parameter with relatively small computational costs. This paper aims at investigating the feasibility of applying this method to a stochastic soil-tunnel system in seismic conditions. An advanced technique named sparse polynomial chaos expansions (SPCE) is introduced to build a surrogate model that allows performing a GSA combined with the Sobol’ indices. The accuracy and efficiency of the SPCE-GSA method are validated by comparison with the true numerical predictions and Monte Carlo simulations. Parametric sensitivity analyses are performed for a wide range of the soil shear wave velocities, ground motion intensities, probability distribution types, and coefficients of variation, assuming a nonlinear soil behavior. The influences of the sampling size, polynomial degree, and ground motion characteristics on the variability in the sensitivity indices are evaluated. The results show that the soil shear wave velocity and modulus reduction factor are the two variables that significantly affect the seismic deformations of tunnels.

中文翻译：

---

使用稀疏多项式混沌展开的概率隧道地震变形的全局灵敏度分析

摘要 由于替代模型允许以相对较小的计算成本有效地量化每个输入参数的相对重要性，因此对利用替代模型执行全局灵敏度分析 (GSA) 进行不确定性量化的兴趣越来越大。本文旨在研究将该方法应用于地震条件下的随机土隧道系统的可行性。引入了一种名为稀疏多项式混沌扩展 (SPCE) 的高级技术来构建代理模型，该模型允许执行与 Sobol 指数相结合的 GSA。SPCE-GSA 方法的准确性和效率通过与真实数值预测和蒙特卡罗模拟的比较得到验证。对大范围的土壤剪切波速度、地面运动强度、概率分布类型和变异系数，假设土壤行为是非线性的。评估了采样大小、多项式次数和地震动特性对灵敏度指标变异性的影响。结果表明，土体剪切波速度和模量折减因子是影响隧道地震变形的两个重要变量。