computational challenges for finding big oil by seismic inversion
DESCRIPTION
Computational Challenges for Finding Big Oil by Seismic Inversion. Motivation for Better Seismic Imaging Strategy. Jack Buckskin. ¼ billion $$$ well. 35,055 Feet. Kaskida Tiber. Motivation for Better Seismic Imaging Strategy Oil Well Blowouts. - PowerPoint PPT PresentationTRANSCRIPT
Computational Challenges Computational Challenges for Finding Big Oil by for Finding Big Oil by
Seismic InversionSeismic Inversion
JackJackBuckskinBuckskin
KaskidaKaskidaTiberTiber
35,055 Feet
Motivation for Better Seismic Imaging StrategyMotivation for Better Seismic Imaging Strategy
¼ billion $$$ well¼ billion $$$ well
Motivation for Better Seismic Imaging StrategyMotivation for Better Seismic Imaging StrategyOil Well BlowoutsOil Well Blowouts
OverpressureZone
Motivation for Better Seismic Imaging StrategyMotivation for Better Seismic Imaging StrategyOil Well BlowoutsOil Well Blowouts
= Low Seismic Velocity Zone
Motivation for Better Seismic Imaging StrategyMotivation for Better Seismic Imaging StrategyMud VolcanoesMud Volcanoes
6.3 km2
13 people killed 30,000 people displacedMay 29, 2006
• Computational Challenge Seismic InversionComputational Challenge Seismic Inversion
OutlineOutline
• Full waveform InversionFull waveform Inversion
• Multisource InversionMultisource Inversion
Given: Given: dd = L= LmmSeismic Inverse ProblemSeismic Inverse Problem
Find: Find: m(x,y,z)m(x,y,z)
Soln: min || LSoln: min || Lmm--dd || ||22
mm = [L L] L = [L L] L ddTT TT-1-1
L L ddTT
migrationmigration
waveformwaveforminversioninversion
Given: Given: dd = L= LmmComputational ChallengesComputational Challenges
Find:Find:mm = [L L] L = [L L] L ddTT TT-1-1
20x20x10 km3
dx=1 m
# time steps ~ 10# time steps ~ 1044
# shots > 10# shots > 1044
m > 10m > 10 unknown velocity valuesunknown velocity values
10101515Total =Total =
d > 10d > 1077
1313wordswords
• Computational Challenge Seismic InversionComputational Challenge Seismic Inversion
OutlineOutline
• Full waveform InversionFull waveform Inversion
• Multisource InversionMultisource Inversion
Multisource Migration:Multisource Migration: mmmigmig=L=LTTdd
Forward Model:Forward Model:
m =[Lm =[LTTL]L]-1-1LLTTddMultisrc-Least FWI:Multisrc-Least FWI:
Multisource Encoded FWIMultisource Encoded FWI
m’ = m - Lm’ = m - LTT[Lm - d][Lm - d]
f ~ [Lf ~ [LTTL]L]-1-1
ff Steepest DescentSteepest Descent
PreconditionedPreconditioned
Nd +Nd =[Nd +Nd =[NL +NL ]mL +NL ]m11 222211 2211 11 22
multisource preconditionermultisource preconditioner
Multiscale Waveform TomographyMultiscale Waveform TomographyMultiscale Waveform TomographyMultiscale Waveform Tomography
1. Collect data d(x,t)1. Collect data d(x,t)
2. Generate synthetic data d(x,t) by FD method2. Generate synthetic data d(x,t) by FD methodsynsyn..
3. Adjust v(x,z) until ||d(x,t)-d(x,t) || minimized by CG.3. Adjust v(x,z) until ||d(x,t)-d(x,t) || minimized by CG.synsyn.. 22
4. To prevent getting stuck in local minima:4. To prevent getting stuck in local minima: a). Invert early arrivals initiallya). Invert early arrivals initially
mute
7
b). Use multiscale: low freq. high freq.b). Use multiscale: low freq. high freq.
0 km0 km 20 km20 km
0 km0 km
6 km6 km 3 km/s3 km/s
6 km/s6 km/s
Boonyasiriwat et al., 2009, TLEBoonyasiriwat et al., 2009, TLE
3 km/s3 km/s
6 km/s6 km/s
Initial modelInitial model
5 Hz5 Hz
10 Hz10 Hz
20 Hz20 Hz
Waveform TomogramsWaveform Tomograms
3 km/s3 km/s
6 km/s6 km/s
3 km/s3 km/s
6 km/s6 km/s
3 km/s3 km/s
6 km/s6 km/s
0 km0 km
6 km6 km
0 km0 km
6 km6 km
0 km0 km
6 km6 km
0 km0 km
0 km0 km 20 km20 km
6 km6 km
Low-pass FilteringLow-pass Filtering
18
Offset (km)
Tim
e (s)
(a) Original CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
Offset (km)
Tim
e (s)
(b) 5-Hz CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
Offset (km)Tim
e (s)
(c) 10-Hz CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
(b) 0-15 Hz CSG (c) 0-25 Hz CSG
Dynamic Early-Arrival Muting WindowDynamic Early-Arrival Muting Window
19
Offset (km)
Tim
e (s)
(a) Original CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
Offset (km)
Tim
e (s)
(b) 5-Hz CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
Offset (km)
Tim
e (s)
(c) 10-Hz CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
0-15 Hz CSG
Offset (km)
Tim
e (s)
(a) Original CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
Offset (km)
Tim
e (s)
(b) 5-Hz CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
Offset (km)
Tim
e (s)
(c) 10-Hz CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
0-25 Hz CSG
Window = 1 s Window = 1 s
19
Offset (km)
Tim
e (s)
(a) Original CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
Offset (km)
Tim
e (s)
(b) 5-Hz CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
Offset (km)
Tim
e (s)
(c) 10-Hz CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
0-15 Hz CSG
Offset (km)
Tim
e (s)
(a) Original CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
Offset (km)
Tim
e (s)
(b) 5-Hz CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
Offset (km)
Tim
e (s)
(c) 10-Hz CSG
0 2 4
0
0.5
1
1.5
2
2.5
3
3.5
4
0-25 Hz CSG
Window = 2 s Window = 2 s
Dynamic Early-Arrival Muting WindowDynamic Early-Arrival Muting Window
2000 20202.52.5
00
Dep
th (
km)
Dep
th (
km)
X (km)X (km)
Traveltime TomogramTraveltime Tomogram
15001500
30003000
Vel
ocity
(m
/s)
Vel
ocity
(m
/s)
Waveform TomogramWaveform Tomogram
2.52.5
00
Dep
th (
km)
Dep
th (
km)
ResultsResults
2100 2020
2.52.5
00
Dep
th (
km)
Dep
th (
km)
X (km)X (km)
Waveform TomogramWaveform Tomogram
15001500
30003000
Vel
ocity
(m
/s)
Vel
ocity
(m
/s)
2.52.5
00
Dep
th (
km)
Dep
th (
km)
Vertical Derivative of Waveform TomogramVertical Derivative of Waveform Tomogram
Kirchhoff Migration ImagesKirchhoff Migration Images
22
Kirchhoff Migration ImagesKirchhoff Migration Images
22
• Computational Challenge Seismic InversionComputational Challenge Seismic Inversion
OutlineOutline
• Full waveform InversionFull waveform Inversion
• Multisource InversionMultisource Inversion
1980
Multisource SeismicMultisource SeismicImagingImaging
vs
copper
VLIW
Superscalar
RISC
1970 1990 2010
1
100
100000
10
1000
10000
Aluminum
Year
202020001980
CPU Speed vs Year
FWI Problem & Possible Soln.FWI Problem & Possible Soln.
• Problem:Problem: FWI computationally costly FWI computationally costly
• Solution:Solution: Multisource Encoded FWI Multisource Encoded FWI
Preconditioning speeds up by factor 2-3Preconditioning speeds up by factor 2-3
Iterative encoding reduces crosstalkIterative encoding reduces crosstalk
Multisource Migration:Multisource Migration: mmmigmig=L=LTTdd
Forward Model:Forward Model:
Multisource Phase Encoded ImagingMultisource Phase Encoded Imaging
d +d +dd =[ =[L +L +LL ]m ]m11 222211
LL{dd{
=[=[L +L +LL ]( ](dd + + dd ) ) 11 222211
TT TT
= = L d +L d +L dL d + + 11 222211
TT TT
LL dd + +L L dd22 112211
Crosstalk noiseCrosstalk noiseStandard migrationStandard migration
TT TT
m = m +(k+1) (k)
Multi-Source Waveform Inversion StrategyMulti-Source Waveform Inversion Strategy(Ge Zhan) (Ge Zhan)
Generate multisource field data with known time shift
Generate synthetic multisource data with known time shift from estimated
velocity model
Multisource deblurring filter
Using multiscale, multisource CG to update the velocity model with
regularization
Initial velocity model
144 shot gathers144 shot gathers
3D SEG Overthrust Model(1089 CSGs)
15 km
3.5 km
15 km
3.5 km
Dynamic QMC TomogramDynamic QMC Tomogram (99 CSGs/supergather)(99 CSGs/supergather)
Static QMC TomogramStatic QMC Tomogram(99 CSGs/supergather)(99 CSGs/supergather)
15 km
Dynamic Polarity TomogramDynamic Polarity Tomogram(1089 CSGs/supergather)(1089 CSGs/supergather)
Numerical ResultsNumerical Results
Multisource FWI SummaryMultisource FWI Summary(We need faster migration algorithms & better velocity models)(We need faster migration algorithms & better velocity models)
IO 1 vs 1/20
Cost 1 vs 1/20 or better
Resolution dx 1 vs 1
Sig/MultsSig ?
Stnd. FWI Multsrc. FWIStnd. FWI Multsrc. FWI
Multisource FWI SummaryMultisource FWI Summary(We need faster migration algorithms & better velocity models)(We need faster migration algorithms & better velocity models)
Future: Multisource MVA, Interpolation, Future: Multisource MVA, Interpolation, Field Data, Migration Filtering, LSM Field Data, Migration Filtering, LSM
Research GoalsResearch GoalsG.T. Schuster (Columbia Univ.,G.T. Schuster (Columbia Univ., 1984)1984)
Seismic Interferometry: VSP, SSP, OBSSeismic Interferometry: VSP, SSP, OBS
Multisource+Preconditioned RTM+MVA+Inversion+Modeling: Multisource+Preconditioned RTM+MVA+Inversion+Modeling:
TTI 3D RTM, GPU: TTI 3D RTM, GPU: Stoffa+CSIM, UUtah K. Johnson SCI, PSU, KAUSTStoffa+CSIM, UUtah K. Johnson SCI, PSU, KAUST
ShaheenShaheen
CorneaCornea
Multisource S/N RatioMultisource S/N Ratio
# geophones/CSG# geophones/CSG
# CSGs# CSGs
L [d + d +.. ]1 221
d +d T d , d 2211
L [d + d + … ]1 2
T , …. +….
Multisrc. Migration vs Standard Migration
# iterations# iterations
Iterative Multisrc. Migration vs Standard Migration
vs
vs
MSMSS-1
M~~
# geophones/CSG# geophones/CSG # CSGs# CSGs
MSMI
Crosstalk TermCrosstalk Term
Time Statics
Time+Amplitude Statics
QM Statics
LL dd + +L L dd22 112211
TT TT
SummarySummary
Time Statics
Time+Amplitude Statics
QM Statics
1. Multisource crosstalk term analyzed analytically1. Multisource crosstalk term analyzed analytically
2. Crosstalk decreases with increasing 2. Crosstalk decreases with increasing , randomness, , randomness, dimension, iteration #, and decreasing depthdimension, iteration #, and decreasing depth
3. Crosstalk decrease can now be tuned3. Crosstalk decrease can now be tuned
4. Some detailed analysis and testing needed to refine 4. Some detailed analysis and testing needed to refine predictions.predictions.
LL dd + +L L dd22 112211
TT TT
• Fast Multisource Least Squares Fast Multisource Least Squares Kirchhoff Mig.Kirchhoff Mig.
• Multisource Waveform Inversion (Ge Zhan)Multisource Waveform Inversion (Ge Zhan)
Multisource TechnologyMultisource Technology
0Z
k(m
)3
0 X (km) 16
The Marmousi2 Model
The area in the white box is used for S/N calculation.
0 X (km) 16
0Z
k(m
)3
0Z
(k
m)
3
0 X (km) 16
Conventional Source: KM vs LSM (50 iterations)
LSM (100x)
KM (1x)
0 X (km) 16
0Z
k(m
)3
0Z
(k
m)
3
0 X (km) 16
200-source Supergather: KM vs LSM (300 its.)
LSM (33x)
KM (1/200x)
S/N
0
1 I300
S/N =7
The S/N of MLSM image grows as the square root of the number of iterations.
MI
• Fast Multisource Least Squares Migration ( Dai)Fast Multisource Least Squares Migration ( Dai)
• Multisource Waveform Inversion (Boonyasiriwat)Multisource Waveform Inversion (Boonyasiriwat)
Multisource TechnologyMultisource Technology
Comparing CIGsComparing CIGs
23
Comparing CIGsComparing CIGs
24
CIG from Traveltime Tomogram CIG from Waveform Tomogram
Comparing CIGsComparing CIGs
25
Comparing CIGsComparing CIGs
26
CIG from Traveltime Tomogram CIG from Waveform Tomogram
Comparing CIGsComparing CIGs
27
Comparing CIGsComparing CIGs
28
CIG from Traveltime Tomogram CIG from Waveform Tomogram
17
Data Pre-ProcessingData Pre-Processing
3D-to-2D conversion3D-to-2D conversion
Attenuation compensationAttenuation compensation
Random noise removalRandom noise removal
17
Source Wavelet EstimationSource Wavelet Estimation
Pick the water-bottomPick the water-bottom
Stack along the water-bottom to obtain an estimate ofStack along the water-bottom to obtain an estimate ofsource waveletsource wavelet
Generate a stacked sectionGenerate a stacked section
In some cases, source wavelet inversion can be used.In some cases, source wavelet inversion can be used.
17
Gradient Computation and InversionGradient Computation and Inversion
Multiscale inversion: low to high frequencyMultiscale inversion: low to high frequency
Dynamic early-arrival muting windowDynamic early-arrival muting window
Normalize both observed and calculated data within the sameNormalize both observed and calculated data within the sameshotshot
Quadratic line search method (Nocedal and Wright, 2006)Quadratic line search method (Nocedal and Wright, 2006)A cubic line search can also be used.A cubic line search can also be used.