Quantification des incertitudes liées aux simulations d'écoulement dans un réservoir pétrolier à l'aide de surfaces de réponse non linéaires

Emmanuel Fetel. ( 2007 )
Institut National Polytechnique de Lorraine

Abstract

Uncertainty Quantification i a key step in any subsurface modeling pipeline. Available data are limited, sparse and heterogeneous, therefore the geomodel itself is an approximation of the reality. This requires to provide tools to access these uncertainties at a minimal cost (memory, time, etc.). In this context, flow simulation adds its own burden : it is time and memory consuming and the results are difficult to interpret due to the complex relationship between reservoir uncertain parameters and production variables. This work develops several methodologies based on response surfaces to analyze this relationship : - construction of response surfaces using the discrete smooth interpolation and the dual kriging approach - analysis of non-linear response surfaces using variance based sensitivity analysis and Bayesian inversion of production history; - integration of fast flow simulation results within the response surface construction step ; - handling of stochastic uncertain parameters characterized by a random effect on reservoir production which cannot be included in a classic deterministic framework. The selection of these methodologies is based on the following considerations : (1) the number of uncertain parameters to handle is usually large, (2) data may be noisy, especially if some parameters has been estimated unimportant and neglected and (3) the complex relationship between uncertain parameters and reservoir production.

Download / Links

BibTeX Reference

@PHDTHESIS{Fetel2007a,
    author = { Fetel, Emmanuel },
     title = { Quantification des incertitudes liées aux simulations d'écoulement dans un réservoir pétrolier à l'aide de surfaces de réponse non linéaires },
      year = { 2007 },
    school = { Institut National Polytechnique de Lorraine },
  abstract = { Uncertainty Quantification i a key step in any subsurface modeling pipeline. Available data are limited, sparse and heterogeneous,  therefore the geomodel itself is an approximation of the reality. This requires to provide tools to access these uncertainties at a minimal cost (memory, time, etc.). In this context, flow simulation adds its own burden : it is time and memory consuming and the results are difficult to interpret due to the complex relationship between reservoir uncertain parameters and production variables. This work develops several methodologies based on response surfaces to analyze this relationship : 
- construction of response surfaces using the discrete smooth interpolation and the dual kriging approach
- analysis of non-linear response surfaces using variance based sensitivity analysis and Bayesian inversion of production history; 
- integration of fast flow simulation results within the response surface construction step ;
- handling of stochastic uncertain parameters characterized by a random effect on reservoir production which cannot be included in a classic deterministic framework.
The selection of these methodologies is based on the following considerations : (1) the number of uncertain parameters to handle is usually large, (2) data may be noisy, especially if some parameters has been estimated unimportant and neglected and (3) the complex relationship between uncertain parameters and reservoir production. }
}