C.J. Ferreira
Evaluation of Regions of Influence for Dimensionality Reduction in Emulation of Production Data
Ferreira, C.J.; Avansi, G.; Vernon, I.; Schiozer, D.J.; Goldstein, M.
Authors
G. Avansi
Professor Ian Vernon i.r.vernon@durham.ac.uk
Professor
D.J. Schiozer
Professor Michael Goldstein michael.goldstein@durham.ac.uk
Professor
Abstract
Oil and gas companies use reservoir simulation models for production forecasting and for business and technical decisions at the various stages of field management. The size and complexity of the reservoirs often requires reservoir models with a high resolution (number of grid blocks) to improve the reservoir behaviour prediction. As a consequence, simulation time becomes a limiting factor for routine workflows such as probabilistic history-matching, production optimization or uncertainty quantification, which requires a higher number of reservoir simulations. One possible solution to this problem is to use Bayesian statistic techniques known as emulation to substitute the simulator in parts of the workflow. An emulator is an approximate representation of a complex physical model; it is usually several orders of magnitude faster to evaluate than simulation, hence facilitating previously intractable calculations because of its speed. However, the challenge to incorporate spatial attributes, such as geostatistical realizations, as inputs remains. It is unfeasible to consider the reservoir spatial property value from each grid cell as a single input, so it is necessary to perform a dimensionality reduction to handle spatial properties as inputs in the emulation process. The use of region of influence is a way to deal with a high-dimensional model in an emulation setting and reduce the spatial properties space. Therefore, we evaluate different types of region of influence during the dimensionality reduction process to emulate production data of a complex numerical model. The regions of influence evaluated were defined using: streamlines, producer-injector pairs, Voronoi based on injection wells and Voronoi based on production wells. The dimensionality method considered were Principal Variables and Stepwise AIC. Our goal is to present and discuss alternatives to treat the high-dimensional input space, i.e., spatial reservoir properties instead of multipliers, to build effective emulators for production history data to use in oil industry workflows, which typically are time-consuming.
Citation
Ferreira, C., Avansi, G., Vernon, I., Schiozer, D., & Goldstein, M. (2018). Evaluation of Regions of Influence for Dimensionality Reduction in Emulation of Production Data. . https://doi.org/10.3997/2214-4609.201802145
Conference Name | ECMOR XVI - 16th European Conference on the Mathematics of Oil Recovery |
---|---|
Conference Location | Barcenola, Spain |
Publication Date | 2018-09 |
Deposit Date | May 16, 2019 |
Volume | 2018 |
DOI | https://doi.org/10.3997/2214-4609.201802145 |
You might also like
FLAMINGO: calibrating large cosmological hydrodynamical simulations with machine learning.
(2023)
Journal Article
First observation of 28O
(2023)
Journal Article
Gaussian Process Models for Well Placement Optimisation
(2022)
Conference Proceeding
A Bayesian Computer Model Analysis of Robust Bayesian Analyses
(2022)
Journal Article
Downloadable Citations
About Durham Research Online (DRO)
Administrator e-mail: dro.admin@durham.ac.uk
This application uses the following open-source libraries:
SheetJS Community Edition
Apache License Version 2.0 (http://www.apache.org/licenses/)
PDF.js
Apache License Version 2.0 (http://www.apache.org/licenses/)
Font Awesome
SIL OFL 1.1 (http://scripts.sil.org/OFL)
MIT License (http://opensource.org/licenses/mit-license.html)
CC BY 3.0 ( http://creativecommons.org/licenses/by/3.0/)
Powered by Worktribe © 2024
Advanced Search