Top Banner
A Deep-Learning-Based Geological Parameterization Method for History Matching Yimin Liu Wenyue Sun Louis J. Durlofsky SESAAI Meeting March 30, 2018
39

A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Aug 13, 2021

Download

Documents

dariahiddleston
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

A Deep-Learning-Based Geological Parameterization Method

for History Matching

Yimin Liu Wenyue Sun Louis J. Durlofsky

SESAAI Meeting

March 30, 2018

Page 2: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Outlineq Background

q Deep-learning-based neural style transfer

q Convolutional neural network PCA (CNN-PCA) for geological Parameterization

q Parameterization results

q History matching results

q Conclusions and future work

2

Page 3: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

History Matching

3

Page 4: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

History Matching Problem

πšπ«π π¦π’π§π’Ž

(

𝟏𝟐 𝒅 π’Ž βˆ’ 𝒅𝐨𝐛𝐬 𝑻𝐢234(𝒅(π’Ž) βˆ’ 𝒅𝐨𝐛𝐬)

+𝟏𝟐 π’Žβˆ’ π’Ž8 𝑻𝐢934(π’Ž βˆ’π’Ž8 )

:

4

q Decision variable: π’Ž - model parameters

q Challenges:

Β§ π’Ž can be high dimensional

Β§ π’Ž should preserve geology

q Solution: map π’Ž onto lower dimensional space

Log permeability

Page 5: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Reparameterization for History Matching

5

q Map π’Ž to a new variable 𝝃

q Favorable properties:

Β§ dim 𝝃 β‰ͺ dim(π’Ž)

Β§ 𝝃 is uncorrelated

Β§ π’Ž@ preserves geological realism

π’Ž β‰ˆ π’Ž@ = 𝒇(𝝃)

πšπ«π π¦π’π§πƒ

(

𝟏𝟐 𝒅 𝝃 βˆ’ 𝒅𝐨𝐛𝐬 𝑻𝐢234(𝒅(𝝃) βˆ’ 𝒅𝐨𝐛𝐬)

+𝟏𝟐 𝝃 βˆ’ 𝝃D

𝑻𝐢𝝃34(𝝃 βˆ’ 𝝃D)

:

q Optimization variable: 𝝃

Page 6: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Principal Component Analysis (PCA)

q PCA: Oliver (1996), Sarma et al. (2006)

Ø Generate 𝑁F realizations using geostatistical algorithm

Ø Perform SVD and reduce dimension

Ø Generate new realization:

6

π’Žπ©πœπš = π‘Όπ’πœ¦π’πŸ/πŸπƒπ’ +π’Ž8

l << Nπ‘ͺ (Nπ‘ͺ:#ofgridblocks)

𝒀 =𝟏

𝑁F βˆ’ 𝟏[π’ŽπŸ βˆ’ π’Ž8 ,π’ŽπŸ βˆ’ π’Ž8 ,… ,π’Ž _Μ‚ βˆ’ π’Ž8 ]

𝒀 = π‘Όπœ¦πŸ/πŸπ‘½π‘» β‰ˆ π‘Όπ’πœ¦π’πŸ/πŸπ‘½π’π‘»

𝝃𝒍~𝑡(𝟎, 𝑰)

Page 7: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

PCA Representation for New Realizationsq Works well when π’Ž follows Gaussian distribution

7

PCA 𝒍 = 70SGEMS Nπ‘ͺ = 3600

π’Žπ©πœπš = π‘Όπ’πœ¦π’πŸ/πŸπƒπ’ +π’Ž8 𝝃𝒍~𝑡(𝟎, 𝑰)

Non

-Gau

ssia

nG

auss

ian

PCA 𝒍 = 70SGEMS Nπ‘ͺ = 3600

Page 8: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Optimization-based PCA (O-PCA)q Optimization-based PCA: Vo and Durlofsky (2014, 2015)

Ø Formulate PCA as an optimization problem with regularizationØ Objective: minimize difference to π’Žπ©πœπš and original histogram Ø Essentially post-process π’Žπ©πœπš with point-wise mapping

8

O-PCA 𝒍 = 70SGEMS Nπ‘ͺ = 3600

π’Žfghi= πšπ«π π¦π’π§π’™

π’Žghi(𝝃𝒍) βˆ’ 𝒙 𝟐𝟐+ γ𝒙𝑻(𝟏 βˆ’ 𝒙) π‘₯m ∈ [π‘₯o, π‘₯p]

(figures from Vo and Durlofsky, 2014)

Page 9: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Limitations of O-PCAq Underlying PCA honors only two-point correlations

q O-PCA point-wise mapping honors single point statistics

q Difficult to preserve multiple point statistics

9

SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2

Unconditional Realizations

Page 10: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Neural Style Transfer

10

Content Image

Style Image

Output Image

q Recent work in deep learning and computer visionq Gatys et al. (2016), Johnson et al. (2016)q Enables transfer of photo into artistic style

(images from Johnson: github.com/jcjohnson/fast-neural-style)

Page 11: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Neural Style Transfer

11

q Recent work in deep learning and computer visionq Gatys et al. (2016), Johnson et al. (2016)q Enables transfer of photo into artistic style

Content ImagePCA Model

Style Image - Training Image

Output ImagePost-processed model

Page 12: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Neural Style Transfer Algorithm

12

q O: output image, C: content image, S: style image

q 𝐿hfrstrs 𝑂, 𝐢 : difference between output and content image

q 𝐿vswxt 𝑂, 𝑆 : difference between output and style image

q Output image preservesØ Content (objects) in the content image

Ø Style (color, texture) in the style image

Ø Object and texture are characterized by multipoint statistics

𝑂 = argmin|

{𝐿hfrstrs 𝑂, 𝐢 + 𝛾𝐿vswxt 𝑂, 𝑆 }Content Loss Style Loss

Page 13: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Analogy to O-PCAq Neural style transfer algorithm

q O-PCA

𝑂 = argmin|

𝐿hfrstrs 𝑂,𝐢 + 𝛾𝐿vswxt 𝑂, 𝑆Content Loss Style Loss

13

π’Žfghi= argmin𝒙

π’Žghi βˆ’ 𝒙 𝟐𝟐 + 𝛾𝒙𝑻(𝟏 βˆ’ 𝒙)

Style LossContent Loss

q O-PCA losses based on point-wise difference

q Neural style transfer based on high-level features extracted from deep convolutional neural network (CNN)

Page 14: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Artificial Neural Networkq Nonlinear function 𝑦 = 𝑔(π‘₯) with multiple layers of neurons

q Linear function: π‘ŠοΏ½β„Žo34 + 𝑏q Nonlinear activation function 𝑓 οΏ½ , e.g., ReLU, sigmoid

14

β„Žo = 𝑓(π‘ŠοΏ½β„Žo34 + 𝑏)

q Convolutional neural networks (CNN):Ø Convolutional layersØ Suitable for image input

Page 15: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Neural Style Transfer Algorithm

15

𝐿hοΏ½οΏ½οΏ½οΏ½οΏ½οΏ½ 𝑂, 𝐢 =οΏ½1

𝑁o𝐷o𝐹o 𝑂 βˆ’ 𝐹o 𝐢 οΏ½

οΏ½

o

𝐿vswxt 𝑂, 𝑆 = οΏ½1𝑁xοΏ½ 𝐺o 𝑂 βˆ’ 𝐺o 𝑆 οΏ½

οΏ½

o

q Limitations of neural style transfer algorithm: Ø Need to solve optimization onlineØ Derivatives of the output image w.r.t input images not clear

Page 16: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Fast Neural Style Transfer Algorithm

q Train a transform net, Johnson et al. (2016)Ø Hour glass shape deep CNN with same input and output size

q Same loss function, optimize parameters in the transform net

16

Page 17: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Fast Neural Style Transfer Algorithmq Construct PCA with 1000 SGeMS realizationsq Train the model transform net on 3000 random PCA modelsq Training takes 3 minutes on 1 GPU (NVIDIA Telsa K80)q Final step: threshold at 0.5

17

Page 18: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Outlineq Background

q Deep-learning-based neural style transfer

q Convolutional neural network PCA (CNN-PCA) for geological parameterization

q Parameterization results

q History matching results

q Conclusions and future work

18

Page 19: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Unconditional Binary System

19

q Binary facies model:

q Training image size: 250 x 250

q Model size: 60 x 60, 𝑁� = 3600

q No hard data

q Goal: low-dimensional representation

q Reduced dimension: 𝑙 = 70

Training image

One model realization

Page 20: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Unconditional Binary System

20

PCA Real. 1

PCA Real. 2

O-PCA Real. 1

O-PCA Real. 2 CNN-PCA Real. 2

CNN-PCA Real. 1

Page 21: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Conditional Binary System

21

q Binary facies model

q Training image size: 250 x 250

q Model size: 60 x 60, 𝑁� = 3600

q Hard data at 16 well locations

q Reduced dimension: 𝑙 = 70

q 200 new random realizations

q Additional hard-data loss

Training image

One model realization

argmin|

{𝐿hfrstrs 𝑂, 𝐢 + 𝛾4𝐿vswxt 𝑂,𝑆

+𝛾�𝐿�isi(𝑂)}

Page 22: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Conditional Binary System

PCA Real. 1

PCA Real. 2

O-PCA Real. 1

O-PCA Real. 2

CNN-PCA Real. 1

CNN-PCA Real. 2

22

q All 200 CNN-PCA models match all hard data

Page 23: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Outlineq Background

q Deep-learning-based neural style transfer

q Convolutional neural network PCA (CNN-PCA) for geological parameterization

q Parameterization results

q History matching results

q Conclusions and future work

23

Page 24: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

History Match Conditional Modelq Oil-water, 60 x 60 grid

q 2 injectors, 2 producers, BHP controlled

q ksand = 2000 md, kmud = 0.02 md

q Data: production and injection rates for 1000 days every 100days

q Number of data: 𝑁� = 80

q Goal: 30 RML posterior models

q Optimizer: PSO-MADS

24

Injector Producer

Page 25: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Permeability Estimation

25

True model

One O-PCA prior model O-PCA posterior model

One CNN-PCA prior model CNN-PCA posterior model

Page 26: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

PROD-1 Water Rate

26

O-PCA Prior models O-PCA Posterior Models

CNN-PCA Prior models CNN-PCA Posterior Models

Page 27: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Infill Well Prediction

infill well locations

q Two infill wells P3, P4q Drilled at 1000 daysq Prediction to 2000 days

27

Page 28: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Permeability Estimation

28

True model

O-PCA #1 O-PCA #2

CNN-PCA #1 CNN-PCA #2

O-PCA #3

CNN-PCA #3

Page 29: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Infill Well PROD-4 Prediction

29

O-PCA Water Rate CNN-PCA Water Rate

24 curves 13 curves

Page 30: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Field Prediction

30

O-PCA Oil Rate CNN-PCA Oil Rate

15 curves 6 curves

Page 31: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Conclusions

q Developed CNN-PCA by combining deep-learning-based neural style transfer algorithm with PCA

q CNN-PCA better preserves channel geometry compared to O-PCA

q CNN-PCA history matching solutions provide more accurate prediction for infill wells

31

Page 32: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Future Work

32

q Extend CNN-PCA to bimodal, three-facies and three dimensional reservoir models

q Apply CNN-PCA with gradient-based history matching

q Implement CNN-PCA treatment with ensemble-based history matching methods

Page 33: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Acknowledgmentsq Stanford CS 231N course

Ø Instructors: Justin Johnson, Serena Yeung, Fei-fei LiØ Course project teammate: Yuanlin Wen (Google)

q Open source code:Ø O-PCA: Hai Xuan Vo (Chevron)Ø Neural style transfer (PyTorch): Hang ZhangØ Fast neural style transfer (PyTorch): Abhishek Kadian

q PSO-MADS: Obi Isebor

q Stanford CEES

33

Page 34: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

34

Thank You!Q & A

Page 35: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Backup Slides

35

Page 36: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Convolutional Layersq Each neuron connects to a local region in previous layerq Convolution: No filters 𝑀o sliding through previous layersq Convolutional layer 𝑁o channels:

Ø Each channel is called a feature map of 𝐷o neuronsØ All channels form a feature matrix 𝐹o ∈ 𝑅^οΏ½Γ— οΏ½

36

Page 37: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Convolutional Neural Networkq Neural network consists of mainly convolutional layersq Nonlinear layers:

Ø Activation layer, e.g., ReLU 𝑓 π‘₯ = max(0,π‘₯)Ø Pooling layer: down sampling to reduce dimension

q Image classification:Ø Input: image of size π»Γ—π‘ŠΓ—3Ø Output: score for predefined image classes

37

VGG-16 Deep CNN (Simonyan and Zisserman, 2015)

Page 38: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

Feature Matrix and Gram Matrix

38

q Content representation: feature matrix 𝐹o ∈ 𝑅^οΏ½Γ— οΏ½

q Style representation: Gram matrix 𝐺o = 𝐹o𝐹oοΏ½/(𝑁o𝐷o)οΏ½

* Images modified from Gatys et al. (2016)

Page 39: A Deep-Learning-Based Geological Parameterization Method ......SGeMS Random Real. 1 O-PCA Random Real. 1 O-PCA Random Real. 2 Unconditional Realizations Neural Style Transfer 10 Content

CNN-PCA for Reparameterizationq Construct PCA with 𝑁F SGeMS realizations

q Post-process π’Žπ©πœπš with fast neural style transfer algorithm

39

π’Žπ©πœπš = π‘Όπ’πœ¦π’πŸ/πŸπƒπ’ +π’Ž8 𝝃𝒍~𝑡(𝟎, 𝑰)