Global sensitivity analysis of stochastic computer models with joint metamodels

Amandine Marrel (1) , Bertrand Iooss (2, 3) , Sébastien da Veiga (1, 2) , Mathieu Ribatet (4)

1 IFPEN - IFP Energies nouvelles
2 GdR MASCOT-NUM - Méthodes d'Analyse Stochastique des Codes et Traitements Numériques
3 EDF R&D MRI - Management des Risques Industriels
4 I3M - Institut de Mathématiques et de Modélisation de Montpellier

The global sensitivity analysis method used to quantify the influence of uncertain input variables on the variability in numerical model responses can be applicable to deterministic computer codes. Deterministic means here that the same set of input variables gives always the same output value. This paper proposes a global sensitivity analysis methodology for stochastic computer codes, for which the result of each code run is itself random. The framework of the joint modeling of the mean and dispersion of heteroscedastic data is used. To deal with the complexity of computer experiment outputs, nonparametric joint models are discussed and a new Gaussian process-based joint model is proposed. The relevance of these models is analyzed based upon two case studies. Results show that the joint modeling approach yields accurate sensitivity index estimations even when heteroscedasticity is strong.

Mots clés: Computer experiment; Generalized additive model; Gaussian process; Joint modeling; Sobol indices; Uncertainty

Domaines: Statistiques [math.ST]; Théorie [stat.TH]

Amandine Marrel, Bertrand Iooss, Sébastien da Veiga, Mathieu Ribatet. Global sensitivity analysis of stochastic computer models with joint metamodels. 2010. ⟨hal-00525489v1⟩