Updating geological facies models using the Ensemble Kalman filter

Patent metadata

Field	Value
Publication number	US-10545260-B2
Application number	US-201213422070-A
Country	US
Kind code	B2
Filing date	Mar 16, 2012
Priority date	Apr 15, 2011
Publication date	Jan 28, 2020
Grant date	Jan 28, 2020

Abstract

The invention relates to a method for history matching a facies geostatistical model using the ensemble Kalman filter (EnKF) technique. The EnKF is not normally appropriate for discontinuous facies models such as multiple point simulation (MPS). In the method of the invention, an ensemble of realizations are generated and then uniform vectors on which those realizations are based are transformed to Gaussian vectors before applying the EnKF to the Gaussian vectors directly. The updated Gaussian vectors are then transformed back to uniform vectors which are used to update the realizations. The uniform vectors may be vectors on which the realizations are based directly; alternatively each realization may be based on a plurality of uniform vectors linearly combined with combination coefficients. In this case each realization is associated with a uniform vector made up from the combination coefficients, and the combination coefficient vector is then transformed to Gaussian and updated using EnKF.

First claim

The invention claimed is:

1. A computer-based method of history matching a facies geostatistical model of a hydrocarbon reservoir, the method comprising:
creating an ensemble of initial realizations of a facies geostatistical model in which each one of the initial realizations is a function of at least one of a plurality of initial uniform random vectors, the model being a discontinuous multipoint simulation model;
deriving simulated dynamic responses from each initial realization of the ensemble of initial realizations for a given time period;
performing a Gaussian score transform on each initial uniform random vector to create an ensemble of initial Gaussian random vectors;
deriving a correlation matrix correlating each one of the initial Gaussian random vectors and the simulated dynamic responses, and correlating the simulated dynamic responses with each other;
obtaining measured data from a reservoir for the given time period;
using the correlation matrix, applying ensemble Kalman filter, using a computing processor, to the ensemble of initial Gaussian random vectors to update the ensemble of initial Gaussian random vectors to the measured data, thereby creating updated Gaussian vectors;
computing updated uniform vectors corresponding to the updated Gaussian vectors;
deriving from the updated uniform vectors an ensemble of updated realizations; and
generating a channel probability map based on the ensemble of updated realizations,
wherein,
the method preserves statistical data and geological features of the model.

2. The method of claim 1 , wherein each one of the initial realizations is directly based on one of the initial uniform random vectors.

3. The method of claim 1 , wherein each one of the initial realizations is a function of a plurality of further uniform random vectors linearly combined with respective combination coefficients, and wherein the initial uniform vector is composed of said combination coefficients.

4. The method of claim 3 , wherein said further uniform random vectors are fixed prior to being combined.

5. The method of claim 1 , wherein the facies geostatistical model is also a function of a training image.

6. A computer-based method of history matching a facies geostatistical model of a hydrocarbon reservoir based on uniform random vectors, the method comprising:
creating an ensemble of initial realizations of a facies geostatistical model in which each one of the initial realizations is a function of at least one of a plurality of uniform random vectors, the model being a discontinuous multipoint simulation model;
fixing the plurality of uniform random vectors and then linearly combining the plurality of uniform random vectors with respective combination coefficients;
each realization thereby being associated with a uniform vector comprising the respective combination coefficients, whereby an ensemble of initial uniform combination coefficient vectors is provided;
deriving simulated dynamic responses from each initial realization for a given time period;
performing a Gaussian score transform on each initial uniform combination coefficient vector to create an initial Gaussian random vector that includes an ensemble of initial Gaussian combination coefficient vectors;
deriving a correlation matrix correlating each one of the Gaussian combination coefficient vectors and the simulated dynamic responses, and correlating said dynamic responses with each other;
obtaining measured data from a reservoir for the given time period;
using the correlation matrix, applying ensemble Kalman filter, using a computing processor, to the ensemble of initial Gaussian combination coefficient vectors to update the Gaussian combination coefficient vectors to the measured data, thereby creating updated Gaussian vectors;
computing updated uniform vectors corresponding to the updated Gaussian vectors;
deriving from the updated uniform vectors an updated ensemble of realizations; and

generating a channel probability map based on the updated ensemble of realizations, wherein,
the method preserves statistical data and geological features of the model.

7. The method of claim 6 , wherein said model is also a function of a training image.

8. The method of claim 6 , wherein each one of the initial realizations is directly based on one of the initial uniform random vectors.

9. The method of claim 6 , wherein each one of the initial realizations is a function of a plurality of further uniform random vectors linearly combined with respective combination coefficients, and wherein the initial uniform vector is composed of said combination coefficients.

10. The method of claim 9 , wherein said further uniform random vectors are fixed prior to being combined.

11. The method of claim 6 , wherein the facies geostatistical model is also a function of a training image.

Assignees

• Hu Lin Ying
• Zhao Yong
• Liu Yongshe
• Conocophillips Co

Inventors

• Hu Lin Ying
• Zhao Yong
• Liu Yongshe

Classifications

• G01V2210/665

  using geostatistical modeling · CPC title

• G06F2111/10

  Numerical modelling · CPC title

• G01V2210/663

  Modeling production-induced effects · CPC title

• G06F30/20

  Design optimisation, verification or simulation (optimisation, verification or simulation of circuit designs G06F30/30) · CPC title

• G01V99/005 Primary

  Physics · mapped topic