literature presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf ·...
TRANSCRIPT
![Page 1: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/1.jpg)
Literature Presentation
Jenny Tjan
Technical University Delft
27 February 2018
![Page 2: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/2.jpg)
Introduction
2 / 23
![Page 3: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/3.jpg)
Structure
1. Introduction
2. Problem
3. Iterative Numerical Methods
4. Proper Orthogonal Decomposition (POD)
5. Reservoir Simulation
6. Research Question
7. Summary
3 / 23
![Page 4: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/4.jpg)
Problem
Ax = b (1)
Properties of A
I Large
I SPD (symmetric positive definite)
I Sparse
Condition number
κ2(A) =λmax(A)
λmin(A)(2)
4 / 23
![Page 5: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/5.jpg)
Iterative Numerical Methods
Ax = b (3)
5 / 23
![Page 6: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/6.jpg)
Conjugate Gradient
Convergence
‖x− xk‖ A ≤ 2‖x− x0‖ A
(√κ2(A)− 1√κ2(A) + 1
)k
. (4)
6 / 23
![Page 7: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/7.jpg)
Conjugate Gradient
Figure: Eigenvalues of A
√κ2(A)− 1√κ2(A) + 1
≈ 0.5971 (5)
7 / 23
![Page 8: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/8.jpg)
Preconditioner
M−1Ax = M−1b, (6)
M: preconditioner
Requirements for M
I SPD
I M−12 exists and symmetric
8 / 23
![Page 9: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/9.jpg)
Preconditioner
Figure: Eigenvalues of M−1A
√κ2(M−1A)− 1√κ2(M−1A) + 1
≈ 0.2663 (7)
9 / 23
![Page 10: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/10.jpg)
Deflated Method
PAx̂ = Pb (8)
P: Deflation matrixZ: Deflation-subspace matrix
(a) Eigenvalues of PA (b) Eigenvalues of A
10 / 23
![Page 11: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/11.jpg)
Proper Orthogonal Decomposition (POD)
Known solutions:Axi = bi (9)
X = [x1 . . . xm] (10)
Correlation matrix:
R =1
mXX> (11)
Eigenvaluesλ1 > λ2 > . . . > λm (12)
11 / 23
![Page 12: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/12.jpg)
Proper Orthogonal Decomposition (POD)
Correlation matrix:
R =1
mXX> (13)
Eigenvaluesλ1 > λ2 > . . . > λm (14)
max1≤l≤m
l∑i=1
λi (R)
m∑i=1
λi (R)
≤ α (15)
where 0 < α ≤ 1
Basis matrix:Ψ :=
[ψ1 . . . ψl
](16)
Basis matrix as deflation-subspace matrix Z
12 / 23
![Page 13: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/13.jpg)
Deflated Preconditioner Conjugate Gradient
M−1PAx̂ = M−1Pb (17)
P: Deflation matrixZ: Deflation-subspace matrixM: Preconditioner
13 / 23
![Page 14: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/14.jpg)
Deflated Preconditioner Conjugate Gradient
Figure: Eigenvalues of M−1PA
κ2(M−1PA) ≤ κ2(M−1A) (18)
14 / 23
![Page 15: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/15.jpg)
ROM-based Preconditioner
M−1rom = M + Q(1− AM) (19)
Q: Correction matrixZ: Deflation-subspace matrix
Figure: Eigenvalues of M−1romA
15 / 23
![Page 16: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/16.jpg)
Porous Media
16 / 23
![Page 17: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/17.jpg)
Reservoir Simulation
Mass conversation law + Darcy’s velocity =
ctρϕ∂p
∂t−∇
(ρ
K
µ(∇p − ρg∇d)
)= ρq (20)
ρ fluid density ϕ rock porosityK rock permeability µ fluid viscosityg gravity constant d reservoir depthq source term ct total compressibility
p pressure (unknown)
17 / 23
![Page 18: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/18.jpg)
Incompressible Model
Fluid density and rock porosity does not depend on pressure+ more assumptions
− 1
µ∇(K∇p) = q. (21)
Discretization:
Tp = q (22)
18 / 23
![Page 19: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/19.jpg)
Constant Compressible Model
Fluid density depends on pressure+ more assumptions
ϕ∂ρ(p)
∂t− ρ0µ∇(K∇p) = ρ(p)q. (23)
Discretization:
Vρ(pk+1)− ρ(pk)
∆tk+ Tpk+1 = q(pk+1). (24)
19 / 23
![Page 20: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/20.jpg)
Newton-Raphson
1. Initial: p0, ε
2. While∣∣pk+1 − pk
∣∣ > ε
3. Solve Jf(pk)δp = −f(pk)
4. Update pk+1 = pk + δp
5. k = k + 1
20 / 23
![Page 21: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/21.jpg)
Research Question
DPCG ROM-based prec
1. Complexity ? ?2. Memory ? ?3. Convergence ? ?4. Iterations ? ?5. Error ? ?
21 / 23
![Page 22: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/22.jpg)
Summary
1. Introduction
2. Problem
3. Iterative Numerical Methods
4. Proper Orthogonal Decomposition (POD)
5. Reservoir Simulation
6. Research Question
7. Summary
22 / 23
![Page 23: Literature Presentationta.twi.tudelft.nl/nw/users/vuik/numanal/tjan_presentation1.pdf · 3.Iterative Numerical Methods 4.Proper Orthogonal Decomposition (POD) 5.Reservoir Simulation](https://reader034.vdocuments.net/reader034/viewer/2022051918/6009f34c00524e49dd3102aa/html5/thumbnails/23.jpg)
Literature Presentation
Jenny Tjan
Technical University Delft
27 February 2018