Investigation on uncertainty and sensitivity analysis of complex systems

Abstract : By means of taylor series expansion, a general analytic formula is derived to characterise the uncertaintypropagation from input variables to the model response,in assuming input independence. By using power-lawand exponential functions, it is shown that the widelyused approximation considering only the first ordercontribution of input uncertainty is sufficiently good onlywhen the input uncertainty is negligible or the underlyingmodel is almost linear. This method is then applied to apower grid system and the eoq model.The method is also extended to correlated case. Withthe extended method, it is straightforward to identify theimportance of input correlations in the model response.This allows one to determine whether or not the inputcorrelations should be considered in practicalapplications. Numerical examples suggest theeffectiveness and validation of our method for generalmodels, as well as specific ones such as thedeterministic hiv model.The method is then compared to Sobol’s one which isimplemented with sampling based strategy. Resultsshow that, compared to our method, it may overvaluethe roles of individual input factors but underestimatethose of their interaction effects when there arenonlinear coupling terms of input factors. A modificationis then introduced, helping understand the differencebetween our method and Sobol’s one.Finally, a numerical model is designed based on avirtual gambling mechanism, regarding the formation ofopinion dynamics. Theoretical analysis is proposed bythe use of one-at-a-time method. Sampling-basedmethod provides a global analysis of output uncertaintyand sensitivity.
Liste complète des métadonnées

Cited literature [271 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-01801799
Contributor : Abes Star <>
Submitted on : Monday, May 28, 2018 - 4:53:45 PM
Last modification on : Monday, February 4, 2019 - 2:38:03 PM
Document(s) archivé(s) le : Wednesday, August 29, 2018 - 3:03:21 PM

File

2017LEMA1021.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-01801799, version 1

Collections

Citation

Yueying Zhu. Investigation on uncertainty and sensitivity analysis of complex systems. General Physics [physics.gen-ph]. Université du Maine, 2017. English. ⟨NNT : 2017LEMA1021⟩. ⟨tel-01801799⟩

Share

Metrics

Record views

108

Files downloads

63