In this paper, we derive copula-based and empirical dependency models (DMs) for simulating non-independent variables, and then propose a new way for determining the distribution of the model outputs conditional on every subset of inputs. Our approach relies on equivalent representations of the model outputs using DMs, and an algorithm is provided for selecting the representations that are necessary and sufficient for deriving the distribution of the model outputs given each subset of inputs. In sensitivity analysis, the selected representations allow for assessing the main, total and interactions effects of each subset of inputs. We then introduce the first-order and total indices of every subset of inputs with the former index less than the latter. Consistent estimators of such indices are provided, including their asymptotic distributions. Analytical and numerical results are provided using single and multivariate response models.
翻译:在本文中,我们为模拟非独立变量而得出基于千兆瓦和经验的依赖模型,然后提出新的方法,以确定模型产出的分布情况,但须视投入的每个子集而定。我们的方法依赖对模型产出使用模式的同等表述,并提供一种算法,用于选择对计算每一子集投入的模型产出的分布而言必要和足够的表述。在敏感性分析中,选定的表达方式允许评估每一子集投入的主要、总和和互动影响。然后我们采用前一子投入的第一阶和总指数,而前一组投入的指数比后一组指数要少。提供了这些指数的一致估计,包括非关键分布。使用单一和多变量反应模型提供了分析和数字结果。