Pick-freeze estimation of sensitivity indices for models with dependent causal processes input

Mathilde Grandjacques 1 Alexandre Janon 2, 3 Olivier Adrot 4 Benoît Delinchant 1, 3
1 G2ELab - Laboratoire de Génie Electrique de Grenoble
2 LM-Orsay - Laboratoire de Mathématiques d'Orsay
3 GdR MASCOT-NUM - Méthodes d'Analyse Stochastique des Codes et Traitements Numériques
4 G-SCOP_GCSP - GCSP
G-SCOP - Laboratoire des sciences pour la conception, l'optimisation et la production
Abstract : This paper address sensibility theory for dynamic models, linking correlated inputs to observed outputs. Usual estimation methods of Sobol indices are based on the fact that the input variables are independent. We present in this paper a method to overpass this constraint for Gaussian processes of high dimension in a time related framework. A general method exists with very weak hypothesis but computations are quite impossible in high dimension. Our proposition leads to a natural generalization of Sobol indices for time dependent, causal and correlated inputs. The method of estimation is a modification of the pick-freeze scheme. After having detailed the scheme for the general Gaussian case we detailed the case of high dimensional autoregressive model, which can be also associated with state models. We then apply the results to the case of a building model.
Type de document :
Communication dans un congrès
11th International Conference on Monte Carlo and Quasi-Monte Carlo Methods in Scientific Computing, MCQMC2014, Apr 2014, Leuven, Belgium. 2014
Liste complète des métadonnées
https://hal.archives-ouvertes.fr/hal-01064160
Contributeur : Sylvie Garcia <>
Soumis le : lundi 15 septembre 2014 - 16:15:49
Dernière modification le : lundi 9 avril 2018 - 12:22:24

Identifiants

  • HAL Id : hal-01064160, version 1

Collections

LM-ORSAY | G-SCOP | G-SCOP_OSP_GCSP | UGA | G-SCOP_OSP | G2ELAB

Citation

Mathilde Grandjacques, Alexandre Janon, Olivier Adrot, Benoît Delinchant. Pick-freeze estimation of sensitivity indices for models with dependent causal processes input. 11th International Conference on Monte Carlo and Quasi-Monte Carlo Methods in Scientific Computing, MCQMC2014, Apr 2014, Leuven, Belgium. 2014. 〈hal-01064160〉

Exporter

BibTeX TEI DC DCterms EndNote

Partager

Métriques

Consultations de la notice

173

Contact

support.ccsd.cnrs.fr
hal.support@ccsd.cnrs.fr
Logo CCSD