Académique Documents
Professionnel Documents
Culture Documents
3D seismic
3D seismic+production
Life-‐time
Life-‐time
4D seismic+production
P2
m
Proxy flow Response r
Structure model complexity
Rock p
Fluid Time
stochastic
1 1
0.8
pi 0.8
pj
0.6
CDF
0.6
CDF
cdf
cdf
0.4 0.4
0.2 0.2
0
0 0.2 0.4 0.6 0.8 1
0
2 2.5 3 3.5 4
Dim. Reduction
KvKh watExp
Kv/Kh Corey Water Exp classification
Generalized sensitivity
any parameter, any response
y A measure of sensitivity is the L1 norm difference
between a class-‐conditional and marginal cdfs
pi = corey exponent pj = TI ; pi = corey exponent
TI|krwMax - class # 1
1 1
FǷ(pi ) FǷ(pi | p j , ck )
0.8 0.8
cdf
0.6 0.6
cdf
cdf
0.4 0.4 FǷ(pi |ck )
0.2 0.2
FǷ(pi |ck )
0 0
0 0.2 0.4 0.6 0.8 1 TI1 TI3 TI8 TI9 TI10 TI13
x TI
y A measure of interaction sensitivity is the L1 norm
difference between a conditional class-‐conditional and
conditional cdfs
Application
to WCA watExp
Training Image
Max Water rel perm
krwMax
Water Corey exponent TI
Kv/Kh
KvKh
Res. Oil Sat.
SOWCR
Sensitive
NotSensitive
TI watExp TI|krwMax
watExp|krwMax
Wat exp
krwMax SOWCR|KvKh
KvKh|TI
Krw Max TI watExp|TI
KvKh|krwMax
KvKh
KvKh
krwMax|KvKh
TI|KvKh
SOW CR
SOWCR
Sensitive KvKh|watExp
NotSensitive
TI|watExp
0 0.5 1 1.5 2 2.5
'r krwMax|TI
krwMax|SOWCR
'Kv / Kh SOWCR|krwMax Sensitive
watExp|KvKh NotSensitive
Interaction is asymmetric 0 0.2 0.4 0.6 0.8 1
geological
scenario
uncertainty:
3 training images
TI1: 50% TI2: 25% TI3: 25%
Production
Data:
Water rate/well
Two modeling questions
Well 3 Well 4
Time/Days
Trying to falsify TIs with data
represent data in lower dimensions using
multi-‐dimensional scaling
Eigencomponent 2
Production data
TI1 responses
TI2 responses
TI3 responses
Eigencomponent 1
MDS: distance = difference in water rate response for all wells
9 dimensions = 99% of variance
f (Data | TIk )
Kernel density estimation in 9D
for TI1 for TI2 for TI3
f data|TIk P TIk
P TIk|Data #
¦ f data|TI P TI
k k k
Streamline geometry
at final time step
Example of region geometry
History match results for all TIs
CPU: Average of 24 flow simulations/model
A few history matches
From TI2 From TI3
600
P10
400
200
P50
0 P90
2000 2100 2200 2300
Time, days
Comparison
Runs/
P(TI1|D) P(TI2|D) P(TI3|D)
model
Rejection
3% 33% 64% 250
Sampler
MPS
Boolean
Surface based
Process based
The missing link
y Geological interpretation: attempting to
understand the genesis and process of past
deposition
?
y Geostatistics: attempting to model the
geometries currently present with a practical
application in mind
Two challenges
1. What methodology bridges this gap?
2. If so, how to bridge this gap?
Limitation of covariances
data model
1 2 3
0 10 20 30 40 0 10 20 30 40
Training images
From Boolean From high resolution seismic
Geostatistical model
High performance
Training image
Geostatistical model