Escolar Documentos
Profissional Documentos
Cultura Documentos
pubs.acs.org/EF
Hybrid Multiscale Modeling through Direct Substitution of PoreScale Models into Near-Well Reservoir Simulators
Tie Sun, Yashar Mehmani, and Matthew T. Balho*
Department of Petroleum and Geosystems Engineering, The University of Texas at Austin, 1 University Station, C0300, Austin, Texas
78712-0228, United States
ABSTRACT: A hybrid, multiscale reservoir simulator has been developed that directly substitutes pore-scale models for
continuum-scale, nite-dierence grids. Over 7500 pore scale models (75 million pores) are inserted into a 1 m2 pore-scale
region around a producing well, which is coupled to an outer, ne-scale nite-dierence model with 10 000 cell-centered grids.
This computationally demanding problem is possible because the domain is decomposed into manageable subdomains
(thousands of pore-scale models and one continuum region) and then coupled together using nite-element mortars to ensure
continuity between the subdomains. Here, ow is single-phase, single-component, and steady-state, but the potential of the
simulator for more complex, near-well ows is demonstrated. A priori, direct upscaling and a posteriori, global upscaling are both
performed, and it is shown that the direct approach results in signicant inaccuracies in the pressure and ow elds but the global
upscaling approach results in a good approximation.
1. INTRODUCTION
Sub-surface simulators are used to model ow and transport of
many applications in porous media, including hydrocarbon
recovery, carbon sequestration, groundwater remediation, and
nuclear waste storage. These simulators are modeled at the
continuum scale using semi-empirical equations (e.g., Darcys
law) because of the large scales (103104 m) of interest in the
sub-surface. The models require macroscopic inputs, such as
permeability, porosity, and relative permeability curves. In
general, these values must be estimated from the geological data
or measured from laboratory experiments.
Recent improvements in pore-scale (104102 m) modeling suggest that direct numerical simulation through the pore
space has reached a level of accuracy that they could be used as
surrogates for experiments.15 This is largely a result of
improved imaging capabilities and better description and
solution of the fundamental ow equations in the pore space.
Pore-scale approaches include direct computational uid
dynamics modeling using the nite-element method (FEM)
or Lattice Boltzmann,6 level-set approaches,7 and network
modeling.8,9 Network models require some approximations to
the ow equations but have the advantage of modeling at
slightly larger scales (103102 m).
Despite improvements in pore-scale modeling, questions
remain as to the ability to extract macroscopic parameters from
a stand-alone, small-scale model. Even if the model is large
enough to be considered a representative elementary volume
(REV), inaccuracies can arise because the imposed boundary
conditions do not reect ow and transport in surrounding
media.1013 Ideally, ow and transport could be modeled in a
single, large pore-scale model that spans 102100 m and
millions of pores, but this is computationally impractical. As a
result, accurate multiscale methods are sought that upscale
pore-level ow behavior to the continuum scale. A lofty goal
would involve multiscale simulation via direct substitution of
pore-scale models for ne grids in strategic parts of a reservoir
(e.g., near wells). Such a hybrid (pore-to-continuum) model
2012 American Chemical Society
Article
2. MATHEMATICAL APPROACH
2.1. Multiscale Simulator Description and Modeling
Approach. Figure 1 depicts the geometry of the near-well, hybrid,
and multiscale simulator, which is areal and pseudo-three-dimensional
(3D) [the pore-scale models are 3D, but the continuum blocks are
only 1 grid block in thickness and therefore two-dimensional (2D)].
Although the geometry is Cartesian and square, the wellbore geometry
is a multisided polygon, so that it approximates a circle with a radius of
10 cm. The pore-scale region extends to a radius of 0.5 m, and the
continuum region extends an additional 0.5 m. The pore-scale region
includes 7548 pore-scale, network models, each with dimensions of 1
1 1 cm. Therefore, the pore-scale region covers an area of 0.75 m2.
Although 7548 unique network models could have been used, for
simplicity, 40 dierent networks were repeated randomly in the
domain. All but one of the networks were mapped from computergenerated sphere packings using a Delaunay tessellation,23 and the last
network was mapped from a real sandstone,24 imaged using X-ray
microtomography. This sandstone network added heterogeneity to the
problem. The network models ranged from 400 to 40 000 pores, and
permeability ranged from 1 107 to 9 105 cm2. In all, the porescale region in the simulations contains 3075 million pores. The
Ndof
pi =
jj
j=1
(1)
Article
Kx , ij =
K y , ij =
solution for steady-state, radial ow around a well. The wellknown solution is given as
P(r ) = Pw + (P0 Pw )
(2)
qy , ijy
Apy , ij
(4)
qx , ijx
Apx , ij
log(r /rw )
log(r0/rw )
(3)
where qx,ij and px,ij are the ow and pressure drop along the x
direction extracted from the multiscale model for the ijth network
model. The boundary pressures (used to compute p) on the network
were computed by taking a weighted average of the nodal pressures on
the mortar interface grid. In contrast to direct upscaling, where Kx and
Ky are estimated a priori for the 40 repeated network models, 7548 sets
of Kx and Ky are estimated from the a posteriori, global upscaling
approach.
Article
Figure 3. Pressure prole of simulation case II derived from three approaches: (a) multiscale, (b) direct upscaling, and (c) a posteriori, global
upscaling. Note that, in panel a, the points represent pore pressure in the pore-scale models. The pressure unit is Pa.
Figure 4. Flow eld of simulation case II derived from three approaches: (a) multiscale, (b) direct upscaling, and (c) a posteriori, global upscaling.
Units are m3/s.
Figure 5. Pressure prole for simulation case II at four angles around the well, 0, 90, 180, and 270, from the positive x axis extracted from the
three models. The solid line indicates the direct upscaling model; the dashed line indicates the a posteriori, global upscaling model; and the black
points represent the multiscale model. Note that, in this simulation, pressures from all three models closely match each other. The pressure unit is
Pa.
5831
Article
Figure 6. Pressure prole of simulation case III derived from three approaches: (a) multiscale, (b) direct upscaling, and (c) a posteriori, global
upscaling. Note that the direct upscaling model yields more pressure drop in the pore-scale subdomain compared to the multiscale model and global
upscaling model. The pressure unit is Pa.
Figure 7. Flow eld of simulation case III derived from three approaches: (a) multiscale, (b) direct upscaling, and (c) a posteriori, global upscaling.
Units are m3/s.
Figure 8. Flow rate dierence between (a) the multiscale simulator and the direct upscaling approach and (b) the multiscale simulator and the a
posteriori, global upscaling approach in case III. Note that the scale is logarithmic. Units are m3/s.
Article
Figure 9. Pressure prole for simulation case III at four angles around the well, 0, 90, 180, and 270, from the positive x axis extracted from the
three models. The solid line indicates the direct upscaling model; the dashed line indicates the a posteriori, global upscaling model; and the black
points represent the multiscale model. The pressure unit is Pa.
Figure 10. Permeabilities (cm2) of each network obtained from the two upscaling approaches for case III. The red triangles represent direct
upscaling permeability, and the blue dots represent permeabilities obtained from the a posteriori, global upscaling approach.
Article
Figure 11. Histograms for 2 of the 40 networks used in the simulations (one homogeneous, sphere pack network and the other, the sandstone
network) for both the x and y directions. Panels a and b show the permeability histogram for the heterogeneous sandstone network. Panels c and d
show the permeability histogram for a relative homogeneous uniform sphere pack network. The histograms are overlaid with direct upscaled
permeability (indicated by the dark line) and the geometric mean of the globally upscaled permeability (indicated by the gray line).
ACKNOWLEDGMENTS
NOMENCLATURE
K = permeability (L2)
p = pressure (F/L2)
q = ow rate (L3/T)
r = radius (L)
N = number of pores in the pore-scale model
Ndof = number of degrees of freedom on mortar space
= coecients of mortar basis functions
= basis functions used in mortar space
= viscosity (F L2 T1)
REFERENCES
4. CONCLUSION
A hybrid, multiscale sub-surface simulator has been developed
that allows for direct substitution of pore-scale models for ne,
continuum-scale grids. Here, a pore-scale region of over 7500
pore-scale network models was inserted around a well and
coupled to an outer continuum scale region. Continuity of
pressure and ux between subdomains was ensured using niteelement mortars. Pore-level heterogeneity was captured, which
would not be possible using a conventional, continuum
simulator. Although network models were used in this work,
other pore-scale approaches (e.g., Lattice Boltzmann) could be
employed.
A priori, direct upscaling can lead to inaccuracies for
heterogeneous media in pressure and ux elds because the
boundary conditions are arbitrary and do not reect ow in
surrounding media. However, an a posteriori, global approach
showed that upscaled permeabilities were dependent upon the
boundary conditions and resulted in a good approximation to
the multiscale solution, both qualitatively and quantitatively.
The simulator could be extended to model more nonlinear,
dynamic ows that require pore-level modeling. Examples
include the injection of carbon dioxide in sequestration
strategies, non-Darcy ow, or matrix stimulation with acids.
Future work should focus on these time-dependent applications
and optimize computational eciency in a parallel computing
environment.
Article
AUTHOR INFORMATION
Corresponding Author
Article
5836