Comparison of Surrogate Models for Extreme Quantile Estimation in the Context of EMC Risk Analysis - Archive ouverte HAL Access content directly
Conference Papers Year :

Comparison of Surrogate Models for Extreme Quantile Estimation in the Context of EMC Risk Analysis

(1, 2) , (1) , (2) , (3)
1
2
3

Abstract

Various EMC problems may be studied from numerical simulations involving 3D Maxwell equation solvers. However, the EMC risk analysis, either from a susceptibility or emissivity point of view requires various configurations of coupling paths described by important sets of unknown or uncertain parameters. The use of surrogate models is very relevant to speed up the risk analysis process. More specifically, values at risk corresponding to extreme values of relevant fields, currents or voltages are often the most important information with regard to a possible EMC risk. Specific methods such as controlled stratification provide a way to sample the input space of random variables in an efficient way to estimate extreme quantiles of a distribution. However, it requires a simple (i.e. fast calculation time) companion model. This companion model has to be correlated to the reference model in a specific sense. Building a surrogate model would be a possible way. This paper discusses various surrogate models and provides some conclusions about their ability to provide either a direct estimation of extreme quantiles of the response of interest, or a companion model for the controlled stratification method.
Fichier principal
Vignette du fichier
sapporo_template_v10.pdf (694.86 Ko) Télécharger le fichier
Origin : Files produced by the author(s)
Loading...

Dates and versions

hal-02314034 , version 1 (11-10-2019)

Identifiers

  • HAL Id : hal-02314034 , version 1

Cite

T. Houret, Philippe Besnier, S Vauchamp, Philippe Pouliguen. Comparison of Surrogate Models for Extreme Quantile Estimation in the Context of EMC Risk Analysis. Asia-Pacific EMC Symposium, Jun 2019, Sapporo, Japan. ⟨hal-02314034⟩
37 View
95 Download

Share

Gmail Facebook Twitter LinkedIn More