Propagation of epistemic uncertainties using dempster-shafer theory in bepu evaluation - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2018

Propagation of epistemic uncertainties using dempster-shafer theory in bepu evaluation

Résumé

BEPU evaluation is generally based on computer simulators such as thermal-hydraulic system codes with expansive computational cost. Furthermore, two types of uncertainties are present in the BEPU evaluation the aleatory uncertainty which describes the natural variability of random events and the epistemic uncertainty due to lack of knowledge. The former is usually modelled by probability theory where some conditions on data quantity and quality must be satisfied for probability density functions (pdfs) fitting. However, while the epistemic uncertainty is taken into account, with less and imprecise available data (i.e. parameters of physical model and correlations present in the computer simulator), the use of probabilistic methods on uncertainty modelling and propagation cannot be justified. The Dempster-Shafer Theory (DST) of evidence provides an adapted framework for representing the parameters with epistemic uncertainty, when it is not possible to build a coherent probabilistic model from the available knowledge. In this theory, instead of pdfs, the parameters are modeled by focal sets with associated degrees of belief. Input uncertainty modeled by DST can be propagated within a BEPU evaluation by mapping input focal sets to the output space. The main problem is how to control the computation cost because the mapped results are obtained by finding the optimal values of the output in each input focal set. In this paper, we propose a new scheme to propagate epistemic uncertainties modelled by DST through a time-consuming computer simulator. Besides classical Monte Carlo method using Cartesian product input mass construction method, we propose a novel procedure using vacuous dimension extension and mass combination rules (e.g. Dempster's combination rule for independent input variables ) after mapping the input focal sets into the output space. This method significantly reduces the number of input focal sets. The obtained output function has less focal sets, but larger range. Thus, a complete coverage of the output focal set space can be provided. Epistemic uncertainty is estimated globally in a conservative point of view. As a result of a trade-off between calculation cost, estimation accuracy and the quantity of details on epistemic uncertainty presented, our proposal enables propagation of epistemic uncertainties modeled by DST with a very limited computation budget making possible its practical use within a BEPU evaluation.
Fichier principal
Vignette du fichier
201700003799.pdf (855.52 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-02415473 , version 1 (17-12-2019)

Identifiants

  • HAL Id : hal-02415473 , version 1

Citer

M. Marques, H. Ynuhui, A. Marrel. Propagation of epistemic uncertainties using dempster-shafer theory in bepu evaluation. ANS Best Estimate Plus Uncertainty International Conference (BEPU 2018), May 2018, Lucca, Italy. ⟨hal-02415473⟩

Collections

CEA DEN
40 Consultations
202 Téléchargements

Partager

Gmail Facebook X LinkedIn More