Conventional flow-based two-phase upscaling for simulating the waterflooding process requires the calculations of upscaled two-phase parameters for each coarse interface or block. The whole procedure can be greatly time-consuming especially for large-scale reservoir models.
To address this problem, flow-based two-phase upscaling techniques are combined with machine learning algorithms, in which the flow-based two-phase upscaling is needed only for a small fraction of coarse interfaces (or blocks), while the upscaled two-phase parameters for the rest of the coarse interfaces (or blocks) are directly provided by the machine learning algorithms instead of performing upscaling computation on each coarse interfaces (or blocks).
The new two-phase upscaling workflow was tested for generic (left to right) flow problems using a 2D large-scale model. We observed similar accuracy for results using the machine learning assisted workflow compared with the results using full flow-based upscaling. And significant speedup (nearly 70) is achieved.
The workflow developed in this work is one of the pioneering work in combining machine learning algorithm with the time-consuming flow-based two-phase upscaling method. It is a valuable addition to the existing multiscale techniques for subsurface flow simulation.
[1]
Louis J. Durlofsky,et al.
Ensemble-Level Upscaling for Efficient Estimation of Fine-Scale Production Statistics
,
2008
.
[2]
L. Durlofsky,et al.
Statistical assignment of upscaled flow functions for an ensemble of geological models
,
2011
.
[3]
L. Durlofsky,et al.
A coupled local-global upscaling approach for simulating flow in highly heterogeneous formations
,
2003
.
[4]
Louis J. Durlofsky,et al.
Efficient Incorporation of Global Effects in Upscaled Models of Two-Phase Flow and Transport in Heterogeneous Formations
,
2006,
Multiscale Model. Simul..
[5]
Yanji Wang,et al.
The Application of Machine Learning Algorithm in Relative Permeability Upscaling for Oil-Water System
,
2021,
Day 10 Thu, April 01, 2021.