Automatic Wave Equation Migration Velocity Analysis

Download Report

Transcript Automatic Wave Equation Migration Velocity Analysis

TRIP Annual meeting
Differential Semblance Optimization
for Common Azimuth Migration
Alexandre KHOURY
1
Context of the project
•Prestack Wave Equation depth migration
•Wavefield extrapolation method
•Automating the velocity estimation loop (time-consuming)
2
Motivation of the project
•Encouraging results in 2D for Shot-Record migration (Peng Shen, TRIP 2005)
•Efficiency of the Common Azimuth Migration in 3D
enables sparse acquisition in one
direction
very economic algorithm
•Goal of the project:
Implement DSO for Common Azimuth Migration in 3D after a 2D validation
3
Common Azimuth Migration
•Wavefield extrapolation in depth: “survey sinking” in the DSR equation
h
M
d (m, h,  )
I (m, h)
p(m, h,  )
Subsurface offset
•Variable used for Velocity Analysis : Subsurface offset
4
Subsurface Offset
S
M
R
S
R
M
h’
M'
M'
R’=S’
h'  0
R’
S’
h'  0
•For true velocity
P(M , h' , z, t  0)  0
P(M ' ,0, z, t  0)  0
•For wrong velocity
P(M , h' , z, t  0)  0
P(M ' ,0, z, t  0)  0
5
Example: two reflectors data set
6
True velocity common image gather
Offset gather at x=1000 m
7
Example: two reflectors data set
One gather at midpoint x=1000m
c  ctrue
c  ctrue
8
Differential Semblance Optimization
•From I (m, h) we define the objective function :
1
J (c)  h.I
2
For
c  ctrue
c  ctrue
2
1
  | h.I (m, h) |2
2 h,m
h  0  I (m, h)  0
J 0

h  0  I (m, h)  0
h  0  I (m, h)  0
J 0

h  0  I (m, h)  0
•Criteria for determining the true velocity !
9
Differential Semblance Optimization
•Plot of the objective function with respect to the velocity
c=ctrue
10
Gradient calculation
•The objective function :
•Gradient calculation :
J (c ) 
1
h.I
2
2
 J (c),c  h.I , h.I 
 c, (
I * 2
) h I 
c
I * 2
J (c )  ( ) h I
c
•Adjoint-state calculation (Lions, 1971):
code operator (
11
I *
)
c
Migration: Structure of the Common Azimuth Migration
• DSR equation:
1
2
2

[(
k

k
)

(
k

k
)
]
2
m
h
m
h
x
x
y
y
cs
4
Wavefield at depth z
kz 
2
H1
Phase-Shift
.eikz 0z
cr 2
1
 [(kmx  khx ) 2  (kmy  khy ) 2 ]
4
in the Fourier domain
i(
.e
2
1 1 2
  ) iz
cs cr c0
H2
Lens-Correction
H3
General Screen Propagator or FFD
in the space domain
in the space domain
Imaging condition
Wavefield at depth z+z
Image at depth z+z
12
Algorithm of the gradient calculation
B
Gradient at depth
z+z
Wavefield pz
MIGRATION
H
p
c
H
H-1
H*,B*
Gradient at depth
z+2z
Wavefield pz+z
H
H-1
H*,B*
Adjoint variables
propagation Dp, Dc
Wavefield pz+2z
13
Algorithm of the gradient calculation
Velocity representation on a B-spline grid:
B-Spline transformation
Fine grid
B-Spline grid
LBFGS
Optimizer
Adjoint B-Spline transformation
Gradient calculation
respect to B-Spline grid
Gradient calculation
respect to Fine Grid
14
Several critical points
- Avoid wrap-around in the subsurface offset domain
-Avoid artifacts propagation by tapering the data
-Constrain the optimization to keep the velocity in a specified
range
-Careful choice of migration parameters for the accuracy of
the gradient (not necessarily for the migration)
15
Several critical points
- Avoid wrap-around in the subsurface offset domain
-Avoid artifacts propagation by tapering the data
-Constrain the optimization to keep the velocity in a specified
range
-Careful choice of migration parameters for the accuracy of
the gradient (not necessarily for the migration)
16
Wrap-around in the subsurface offset domain
h
For wrong velocity
Image
Gather
17
Wrap-around in the subsurface offset domain
Effect of padding and split-spread for wrong velocity
h
Image
Gather
18
Several critical points
- Avoid wrap-around in the subsurface offset domain
-Avoid artifacts propagation by tapering the data
-Constrain the optimization to keep the velocity in a specified
range
-Careful choice of migration parameters for the accuracy of
the gradient (not necessarily for the migration)
19
Artifacts propagation
Necessity to taper the data on both offset and midpoint axes and in time
20
Several critical points
- Avoid wrap-around in the subsurface offset domain
-Avoid artifacts propagation by tapering the data
-Constrain the optimization to keep the velocity in a specified
range
-Careful choice of migration parameters for the accuracy of
the gradient (not necessarily for the migration)
21
Several critical points
- Avoid wrap-around in the subsurface offset domain
-Avoid artifacts propagation by tapering the data
-Constrain the optimization to keep the velocity in a specified
range
-Careful choice of migration parameters for the accuracy of
the gradient (not necessarily for the migration)
22
Differential Semblance Optimization
•Tests on different data sets:
-Test on flat reflectors with a constant background
velocity
-Test on the top of a salt model
-Test on a 4-Reflectors model
23
Differential Semblance Optimization
•Tests on different data sets:
-Test on flat reflectors with a constant background
velocity
-Test on the top of a salt model
-Test on a 4-Reflectors model
24
Differential Semblance Optimization
Start of the optimization:
V=2300
Image
Gather
25
Differential Semblance Optimization
10 iterations: Right position
Image
Gather
26
Differential Semblance Optimization
•Tests on different data sets:
-Test on flat reflectors with a constant background
velocity
-Test on the top of a salt model
-Test on a 4-Reflectors model
27
Differential Semblance Optimization
Top of salt : image
x=5000
28
Differential Semblance Optimization
Top of salt : one gather
29
Differential Semblance Optimization
Plot of  | h.I ( x, h) |
function h
2
: localization of the energy of the objective
30
Differential Semblance Optimization
•Tests on different data sets:
-Test on flat reflectors with a constant background
velocity
-Test on the top of a salt model
-Test on a 4-Reflectors model
31
Differential Semblance Optimization
True velocity
32
Differential Semblance Optimization
Starting velocity
33
Differential Semblance Optimization
Starting image
34
Differential Semblance Optimization
Optimized image
35
Differential Semblance Optimization
True image
36
Differential Semblance Optimization
Optimized velocity
37
Conclusion
•Migration is critical and has to be artifacts free.
•Is the DSR Migration precise enough for
optimization of complex models ?
•Can we deal with complex velocity model ?
•Next: test on the Marmousi data set and on a 3D data set.
38
Acknowledgment
• Prof. William W. Symes
• Total E&P
• Dr. Peng Shen, Dr Henri Calandra, Dr Paul Williamson
39