Quantile sensitivity measures based on subset simulation importance sampling

lSubset simulation importance sampling is applied to solving quantile sensitivity measures.lDifference between quantile sensitivity measures and other sensitivity measures such as variance-based global sensitivity measures, failure-probability-based global sensitivity measures, and moment independent measures is discussed.lResults of numerical and practical examples show the accuracy and efficiency of the subset simulation importance sampling.lIt is shown that the subset simulation importance sampling is more efficient than the conventional Monte Carlo simulation. Global sensitivity measures based on quantiles of the output are an efficient tool in measuring the effect of input variables for problems in which α−th quantiles are the functions of interest and for identification of inputs which are the most important in achieving the specific values of the model output. Previously proposed methods for numerical estimation of such measures are costly and not practically feasible in cases in which the quantile level α is very small or high. It is shown that the subset simulation importance sampling (SS-IS) method previously applied for solving small failure probability problems can be efficiently used for estimating quantile global sensitivity measures (QGSM). Considered test cases and engineering examples show that the proposed SS-IS method is more efficient than the previously proposed Monte Carlo method.