Computational Challenges for Subsurface Flow Modeling and...
Transcript of Computational Challenges for Subsurface Flow Modeling and...
![Page 1: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/1.jpg)
1
Computational Challenges for Subsurface Flow Modeling and
Optimization
Louis J. Durlofsky
Department of Energy Resources EngineeringStanford University
![Page 2: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/2.jpg)
2
• Multiscale geology
• Unconventional resources and complex recovery processes increasingly important
− Tar sands, oil shales, shale gas, coalbed methane
−Recovery / CO2 storage: multiphysics & multiscale
• Use of computational optimization and data assimilation procedures; multiple objective optimization
• Uncertainty assessment, risk management
Computational Issues for Oil / Gas Production and CO2 Sequestration
![Page 3: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/3.jpg)
3
Darcy Flow through Porous Formations
• Darcy velocity u, volumetric flow rate q (u = q /A)
p = p1 p = p2
rock permeability k, fluid viscosity μ, porosity φL
bulk
pore ,VV
dxdpku =−= φ
μ
• In multiple dimensions, p∇−= kuμ1
![Page 4: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/4.jpg)
4
cj
jijjj
jijj mySyt
=⎟⎠⎞
⎜⎝⎛⋅∇+⎟
⎠⎞
⎜⎝⎛
∂∂
∑∑ uρφρ nc equations
)(
jj
rjj p
Sk∇−= ku
μ
Governing Equations for Subsurface Flow(Multicomponent, Multiphase, Darcy Scale)
• Component mass balances for component i:
• Darcy’s laws for phase j :
• Additional (problem specific) effects: energy equation, geomechanics, chemical reactions, …
• Equilibrium relationships: K = yi,vapor / yi,oleic
![Page 5: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/5.jpg)
5(photo by Eric Flodin)
Multiscale Geology & Reservoir Flow
• Reservoir: 5 km × 5 km × 100 m
• Grid blocks: 5 m × 5 m × 0.2 m
• 1000 × 1000 × 500 = 0.5 × 109 blocks
![Page 6: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/6.jpg)
6
Slip band sets
Deformed host rock
Fault rock
Orthogonal SBs
(from Ahmadov et al., 2007)
Fault Zone Features5 cm
![Page 7: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/7.jpg)
7(from Ahmadov et al., 2007)
Thin Section of Slip Bands
• Low φ and k in slip bands; presence of open fractures may have strong impact on flow
1 mm blue indicates φ fracture
![Page 8: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/8.jpg)
8
Upscaling Approaches
• Upscaling (computational homogenization) methods provide averaged or coarse-grained results
Fine grid Coarse grid
( ) wqSftS ~)( −=⋅∇+∂∂ uφ
( ) ~)( tqpS =∇⋅⋅∇ kλ ( ) ~)( **t
cc qpS =∇⋅⋅∇ kλ
( ) wcc
c
qSftS ~)( ** −=⋅∇+∂∂ uφ
coarse functions (k*, φ*, λ*, f* ) generally precomputed
![Page 9: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/9.jpg)
Multiscale Finite Element / Volume Methods
• Construct (local) basis functions to capture subgrid k(x)
• Reconstruct fine-scale u for transport calculations
• FE (Hou, Wu, Efendiev, …), FV (Jenny, Lee, Tchelepi, Lunati, …),MFE (Arbogast, Aarnes, Krogstad, Lie, Juanes, Chen …)
(figures from Hesse, Mallison, Tchelepi)
MsFVM
basis functions
jip ,
1,1 −− jip
jip ,1−
1,1 +− jip
1,1 −+ jip
jip ,1+
1,1 ++ jip
1, −jip
1, +jip
)1( )2(
)4( )3(
primal & dual grids
![Page 10: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/10.jpg)
10
Reduced-Order Modeling Procedures
• ROM approaches require “training” run using full-order model [g(x)=0] and basis construction
• “Snapshot” matrix X= [x1, x2, … xN]; SVD of Xprovides basis matrix Φ (x=Φz)
• Basic POD:
• Trajectory piecewise linearization (TPWL):
( ) gΦδJΦΦgJδ Tr
T −=→−=
ΦJΦJ 11 ++ = iTir
( )1 1
11 11 1 11( ) ( )
i ii i i n i
r i in
r r
n+ +−+ + + ++
+
⎡ ⎤⎛ ⎞ ⎛ ⎞∂ ∂= − − + −⎢ ⎥⎜ ⎟ ⎜ ⎟∂ ∂⎢ ⎥⎝ ⎠ ⎝ ⎠⎣ ⎦
g gz J z u ux
z zu
![Page 11: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/11.jpg)
11
Oil from Unconventional Sources
tar sands oil shales
• Alberta tar sands: ~170 × 109 bbl in reserves
• US oil shales: ~2 × 1012 bbl resource (Green River Formation, UT, CO, WY)
pictures from http://www.technologyreview.com/NanoTech/wtr_16059,318,p1.html (left), S. Graham (right)
![Page 12: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/12.jpg)
In-situ Upgrading of Oil Shale
(Picture from www.shell.com)
kerogen(s) → oil + gas
(via several reactions)
19
![Page 13: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/13.jpg)
In-situ Upgrading Simulations
Fan, Durlofsky & Tchelepi (2009)
Temperature
Kerogen concentration
Temperature
Kerogen concentration
20
![Page 14: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/14.jpg)
Geological Sequestration of CO2
19(from Benson, 2007)
![Page 15: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/15.jpg)
15(from Benson, 2007)
Geologic Storage of Carbon
Trapping Mechanisms
• Structural
• Residual
• Solubility
• Mineral
![Page 16: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/16.jpg)
16
Optimization of Well Placement and Settings
log10(k)
0
4
2
• Determine well locations, orientation and injection schedule to minimize mobile CO2
• Applying global stochastic search (PSO) and local search (GPS, HJ) optimizers
(work with David Cameron)
![Page 17: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/17.jpg)
17
Well 4Well 2 Well 3Well 1
Optimal Injection Strategy (4 periods)
Injection Rate
Well 1 Well 2 Well 3 Well 4
Default
Mobile CO2 (top view, vertically averaged, low hysteresis)
0
15%
Optimized
Optimization Results
![Page 18: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/18.jpg)
18
Optimization of Field Development
• Goal is, e.g., to maximize net present value of multi-well development project
• Complications:– Number of wells must be determined in optimization
– Each well can be of any type (categorical)
– Reservoir geology is uncertain
– Well settings also need to be optimized
– Multiple objectives may be important
– Each function evaluation entails reservoir simulation
![Page 19: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/19.jpg)
19
Some Possible Well Types …
(from TAML, 1999)
![Page 20: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/20.jpg)
20
… Coupled with Multiple Geomodels
Nr = # of realizations (potentially 100s or 1000s)∑
==
rN
ii
rN 1)NPV( 1NPV
![Page 21: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/21.jpg)
21
Retrospective Optimization* for Well Placement under Geological Uncertainty
• Brute force approach: at each iteration evaluate
• RO approach:– Define sequence of sub-problems Pk with increasing Nk
– E.g., for Nr =80, use 4 sub-problems with Nk = (4, 8, 24, 80)
– Optimize ⟨ J ⟩k using any core optimization algorithm
– Initial guess for Pk+1 is solution to Pk
– Early sub-problems faster to evaluate; later sub-problems converge quickly because initial guess is close to optimum
∑=
=rN
ii
r
JN
J1
1
*Chen & Schmeiser, 2001; Wang & Schmeiser, 2008
![Page 22: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/22.jpg)
22
![Page 23: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/23.jpg)
23
![Page 24: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/24.jpg)
24
• 100 PSO iterations × 20 PSO particles × 104 realizations ≈ 200,000 reservoir simulations
Performance of Brute-Force PSO (no RO)
Wang et al. (2011)
![Page 25: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/25.jpg)
25Wang et al. (2011)
![Page 26: Computational Challenges for Subsurface Flow Modeling and ...wiki.siam.org/siag-gs/images/siag-gs/f/f3/Durlofsky.pdf · −Tar sands, oil shales, shale gas, coalbed methane −Recovery](https://reader036.fdocuments.net/reader036/viewer/2022070107/60231374e6a6ed01a0174bc2/html5/thumbnails/26.jpg)
26
Summary
• Computational challenges:
− Multiscale geology and effects on flow, particularly for complex (multiphysics) processes
− Increasing importance of unconventional resources− Optimization of production or CO2 sequestration
• Computational advances could lead to better predictive models, improved recovery, realistic UQ, and could facilitate production of unconventional resources