Method for more rapidly producing the representative stochastic model of a heterogeneous underground reservoir defined by uncertain static and dynamic data

1) A method for forming a stochastic numerical model of Gaussian or Gaussian-related type, representative of the distribution of a physical quantity in a porous heterogeneous medium, adjusted in relation to dynamic data, characteristic of the displacement of fluids in the medium, and local static data measured at a certain number of measuring points along wells through the medium, and involving a certain uncertainty margin, wherein the model is optimized by means of an iterative deformation process comprising forming, on each iteration, a combined realization obtained by linear combination on the one hand of an initial realization, best representing a part of the medium, and of at least a second independent realization of the same stochastic model, and an objective function measuring the difference between real dynamic data and the dynamic data simulated by means of a flow simulator is minimized, for the combined realization, by adjustment of the combination coefficients, the iterative adjustment process being continued until an optimum realization of the stochastic model is obtained, characterized in that it comprises:

• a) transforming the local static data into punctual pseudo-data in accordance with probability laws and a spatial variability model, and
• b) adjusting the pseudo-data while respecting the probability laws from which they result, and the spatial variability model, by means of an iterative process wherein a first Gaussian white noise associated with the set of pseudo-data is combined with a second Gaussian white noise.