You are on page 1of 34

Universidad Simón Bolívar

Un nuevo enfoque para trazar rayos


en 3D y otros problemas inversos en
geofísica
Debora Cores Carrera
cores@cesma.usb.ve
debcores@hotmail.com
Caracas-Venezuela
Universidad Simón Bolívar

OUTLINE

The Ray tracing problem (RT)


The Inverse problem approach (IP)
Ellipsoidal velocity

A general travel time equation


Brief historical overview
The optimization Solver
Adavantages of the solver
Numerical Results for RT and IP
Full waveform inversion
Conclusions
Universidad Simón Bolívar
The Ray Tracing Problem
Xr 2 n 1
dl l
Minimize T ( x, y , z ) 
X v( x, y, z) 

X r i
Xs
s
i  2 v i

v ( x , y , z ) is the group velocity and dl is the differential


along the ray l2

The number of
layers is given
by n
Universidad Simón Bolívar
Tomography Inverse problem
1
Minimize f (v )  || Tobs  T (v ) ||2
2

2
l  vu

l 2j
l 5j
T ( v )  (T1 (v ),...Tnr (v )) T

2 n 1 j
l
T j (v )  
j
l 3 l 4j i

i 2 vi

f (v )  J T (v )T (T (v )  Tobs )
Universidad Simón Bolívar

Ellipsoidal Velocity
Solving the eigenvalue problem: (G(1,2 )  WI )U  0
Where 1 and 2 are the polar and azimuthal phase angles.
3
Gik  C 
ijkl j l
j ,l 1

 1   cos(2 ) sin( 1 )
 2   sin( 2 ) sin( 1 )

 3   sin( 1 )
Universidad Simón Bolívar

Ellipsoidal Velocity
Approximating the eigenvalues of the Christoffel equation
and using the Byun Transformation, Contreras et al. in
1997 obtained an ellipsoidal group velocity:
1 1 1
2
 2
cos 2
(1i )  2
cos 2
(2i ) sin 2
(1i ) 
(vi ) (( vZ , j )i ) (( vNMO , j )i )
[ X ;Z ]

1
2
sin 2
(2i ) sin 2
(1i )
(( vNMO[Y ;Z ] , j )i )

vi is the group velocity in the layer delimited by


interfaces i-1 and i. (vNMO[ X ;Z ] , j )i is the i-th component of
the normal move out velocity in the symmetry plane [X,Z]
with wave propagation mode j=P,SV or SH .
Universidad Simón Bolívar

Ellipsoidal velocity
fi ( xi , yi )  fi 1 ( xi 1, yi 1 ) ( xi  xi 1 )2  ( yi  yi 1 )2
cos(1i )  sin( 1i ) 
li li

yi  yi 1
sin( 2i ) 
( xi  xi 1 ) 2  ( yi  yi 1 ) 2

xi  xi 1
cos(2i ) 
( xi  xi 1 )2  ( yi  yi 1 )2
Universidad Simón Bolívar

Ellipsoidal Velocity
Then, for j=P,SV,SH and i=2,…,2n+1

zi xi yi2


2 2
1 1
 2
 2
 2
vi li (( vZ , j )i ) (( vNMO[ XZ ] , j )i ) (( vNMO[YZ ] , j )i )

where,
yi  yi  yi 1
zi  zi  zi 1

xi  xi  xi 1
Universidad Simón Bolívar

A More general ellipsoidal velocity


The distance segment (xi ' , yi ' , zi ' ) between two
consecutive points at interfaces i-1 and i,
(xi ' , yi ' , zi ' )  Rp Ra (xi , yi , zi )
T T

 cos i 0  sin i 
 
Rp   0 1 0 
 sin  0 cos i 
 i

 cos i sin  i 0
 
Ra    sin  i cos i 0
 0 1 
 0
Universidad Simón Bolívar

A general travel time equation


2 n 1
( zi ' )2 ( xi ' )2 ( yi ' )2
TXXs r ( X , Y , Z )  
i 2
2
 2

(( vZ , j )i ) (( vNMO[ XZ ] , j )i ) (( vNMO[YZ ] , j )i )2

where,
xi '  xi cosi cos i  yi sin i cosi  zi sin i

yi '  xi sin  i  yi cos i


zi '  xi cos i sin i  yi sin i sin  i  zi cosi
Universidad Simón Bolívar
Brief Historical Overview

Ray Tracing Approaches

Solving Differential Equations Solving Optimization Problems

Especially in the 70’s More recently

•P.L. Jacson (1970) •Um and Thurber (1987)


•H. Jacob (1970) •Prothero et al. (1988)
•R.L. Wesson (1970-1971) •Mao and Stuard (1997)
•Julian and Gubbins (1970-1971) •Cores et al. (2000)
•Pereyra et al. (1980)
Universidad Simón Bolívar
Brief Historical Overview
Inverse tomography Approaches
Reconstruction Techniques Damped Gauss Newton

•Dines and Lytle •Bishop et al. (1985)


(1979)
•Chiu et al. (1986)
•Ivansson (1983)
•Zhu and Brown (1987)
•Lines and Treitel
•Farra and Madariaga
(1984)
(1988)

Conjugate Gradient type methods

Pica et al. (1990)


•Michelena et al. (1993)
Universidad Simón Bolívar

The Optimization Approach used for solving both Problems

The Projected Spectral Gradient (PSG)


Method (Raydan et al. (2000))

Considered a low cost •Local Storage requirements


and storage technique as •Few floating point operations
any of the extensions of per iteration
conjugate gradient
methods (Polak-Ribiere, •Fast Local Convergence
Hestenes-Stiefel) for a •Do not require to solve a
nonlinear optimization linear system of equation per
problem. iteration
Universidad Simón Bolívar
Projected Spectral Gradient (PSG) Min f (x)
Method s. t. l  xu
1. Given x0   n , 0   and M  0
2. If || P( x k  g k )  xk || 0 , stop
3. Compute d k  P( xk   k g k )  xk and set :
4. If f ( xk 1 )  max f ( xk  j )    gkT dk , then
k  , xk 1  xk  k dk , yk  gk 1  gk , sk  xk 1  xk go to step 5
skT sk
5.  k 1  T
sk y k

Where: P is the projection on   {x  n / l  x  u} and g k  f ( xk )


Universidad Simón Bolívar
Advantages of the Optimization Approach
1. The projection over  is simple and has low computational
cost
2. The objective function does not decrease monotonicaly because
of step lenght and the non monotone line search (step 4).
Implying less function evaluations to converge from any initial
point (Global convergence).
3. The step size  k is not the classical choice for the steepest
descent method. It speeds up the convergence of the PSG
method.
4. The PSG method is related to the Quasi Newton methods. It can
be view as a two point method.
5. The PSG method is competitive and many times out performs
the extensions of CG methods (CONMIN and PR+)
6. The method converge to the global minimun if we have an
stratified and dipped model with constant velocity between
layers
Universidad Simón Bolívar
Numerical Results for Ray Tracing

5 layer synthetic model where P-S converted waves velocities are considered
Universidad Simón Bolívar
Numerical Results for Ray Tracing

1. 157 recievers and 3


sources randomly
genereted at the
surface.
2. The average CPU
time for 1 shot is 3 s
(from different initial
rays).
3. Convergence to the
global minimum is
obtained.

5 layer synthetic model where P-S converted wave velocities are considered
Universidad Simón Bolívar
Numerical Results for Ray Tracing
1. 157 recievers and 5 sources
randomly generated at the
surface.
2. Lateral heterogeneous
model :
v( x, y )  ax  by  c,
a  (0,1.7,1.5,1.3,0.8,0.8,1.3,1.5,1.7)T ,
b  (0,1,1,1,1,1,1,1,1)T ,
c  (0,800,700,500,150,150,500,700,800)T

3. We can not guarantee


convergence to the global
minumum.
4. The average CPU time for
the first shot was 50 s (from
different initial rays).

4 layer synthetic lateral heterogeneous model of complex stratigraphy


Universidad Simón Bolívar
Numerical Results for Ray Tracing
We consider a 5 layer ellipsoidal anisotropic medium,where the velocities are
given by the formula:

1 1 ( zi' )2 ( xi' )2 ( yi' )2


 2
 2
 2
,
vi li (( vz , j )i ) (( v[ x , z ], j )i ) (( v[ x , y ], j )i )
xi'  xi cos(i ) cos( i )  yi sin( i ) cos(i )  zi sin( i ),
yi'   xi sin( i )  yi cos( i ),
zi'  xi sin( i ) cos( i )  yi sin( i ) sin( i )  zi cos(i ),

Where  i and i denote the polar and azimuthal rotation angles in the
layer i, and j=P,SV,SH, i=1,2,...,2n+1
If the medium is an stratified or dipped model, the approach converges to a
global minimum
Universidad Simón Bolívar

Numerical Results for Ray Tracing


157 receivers at the surface and 1
source in the origen.

(vz , p )i  1500  100 * i m / s,


(v[ x , z ], p )i  1200  50 * i m / s,
(v[ x , y ], p )i  1350  80 * i m / s,
(vz , s )i  1400  100 * (n  3  i ) m / s,
(v[ x , z ], s )i  1000  50 * (n  3  i ) m / s,
(v[ y , z ], s )i  1150  80 * (n  3  i ) m / s
for i=2,...,n+1

5 layer synthetic ellipsoidal anisotropic medium


Universidad Simón Bolívar
Numerical Results for the tomography inversion
Universidad Simón Bolívar
Numerical Results for the tomography inversion
Universidad Simón Bolívar
Numerical Results for the tomography inversion

We fixed CPU time and the


grid size (500x500) to observe
the reduction in the gradient
and the residual during that
period of time
Universidad Simón Bolívar
Numerical Results for the tomography inversion

We used a (20x20) Real velocities Initial velocities


grid size to measure
The initial velocities have an error of 50% from the real velocities
the precision of PR+
and PSG

Final velocities (PSG) Final velocities (PR+)


The quality of the solution by the 2 methods are almost the same
Universidad Simón Bolívar
Numerical Results for the tomography inversion

1. SIRT has low computational cost per iteration but requires too many
iterations and therefore consumes more CPU time.
2. PSG, PR+ and CONMIN reach quickly a good precision (10e-03) when
compared to SIRT and Gauss Newton methods.
3. Gauss Newton is fast, in CPU time, for very small size of the grid.
4. The PSG and PR+ methods outperform CONMIN for very large
problems.
5. The PSG method is always slightly faster , in CPU time, than PR+.
Universidad Simón Bolívar

Anisotropic tomography inversion


Bounds on the unknown parameters: For i=2,…,2n+1
0.2km  (vZ , j )i  (vz )i  5km
0.2km  (vNMO[Y ;Z ] , j )i  (v y )i  5km
0.2km  (vNMO[ X ;Z ] , j )i  (vx )i  5km
10   i  30
2  i  9
Stopping criteriun:
P( X k  f ( X k ))  X k 2  1006

M=8 (SPG)
Universidad Simón Bolívar
Anisotropic tomography inversion
Square mesh Radial mesh
Universidad Simón Bolívar

Anisotropic tomography inversion


Square mesh ns=2 nr=5 Radial mesh ns=5 nr=16
ap.
i (vx )i ( v x )ap.i (v y )i (v y )ap.i (vz )i (vz )ap.i ( v x ) i (v y )ap.i (vz )ap.i

2 1.5 1.5 1.7 1.69 1.9 1.86 1.49 1.69 1.9


3 2 1.91 2.3 2.22 2.5 2.37 1.81 2.06 2.21
4 3 2.86 2.8 2.85 3.3 3.21 3.01 2.81 3.55
5 2.7 2.83 2.9 2.85 3.1 3.19 2.73 2.88 2.91
6 1.8 1.89 2 2.07 2.3 2.42 2.01 2.27 2.59
7 1.3 1.29 1.6 1.61 1.8 1.84 1.29 1.6 1.81
Universidad Simón Bolívar

Anisotropic tomography inversion


Square mesh ns=2 nr=15 Radial mesh ns=5 nr=16
i i  ap.
i i  ap.i  ap.i  ap.i

2 5 5.84 20 21.49 8.46 20.1


3 7 7.36 15 15.46 4.93 13.46
4 3 6 25 20.31 2 25.13
5 3 5.78 25 19.14 6 20.43
6 7 7.47 15 14.29 4.47 12
7 5 5.01 20 18.94 7.64 19.86
Universidad Simón Bolívar
Numerical results on the anisotropic tomography
inversion
• This is a highly nonlinear problem that has many solutions, so regularization of the
problem and priori information is required.
• The SPG method gets a good precision for estimating the velocities using small
number of rays.
• The problem for obtaining a better estimate of the polar angle vector is not the
optimization scheme used, it depends on the seismic data acquisition.
• Increasing the number of rays, the error in the velocity vector and in the azimuthal
angle vector can be reduced, but the CPU time increase.
• None of the mesh distribution used here give enough information for obtaining a
good estimate of the polar angle vector. May be the travel time information is not
appropiate for estimating fracture orientation.
Universidad Simón Bolívar

Fullwave inversion
1
Minimize p' ( c ), p( c )
2
lij  cij  uij
Where,
p' (c)  C p 1p(c)
p(c)  p( xr , t; xs )  p( xr , t; xs )obs

p is the pressure wavefield at the velocity matrix c


(computed by a finite difference scheme (Luo
and Schuster 1991) ), and the matrix C p is a
covariance operator.
Universidad Simón Bolívar
Full waveform inversion (for Modified Marmousi model)
Universidad Simón Bolívar

Full wave inversion


Universidad Simón Bolívar

Conclusions
• The PSG method is a simple, global and fast method for
large scale problems (Example: inversion and ray tracing).
• The PSG method reachs quickly to a good precision (For
example 10e-02 or 10e-03).
• The PSG method only requires firts order information.
• The PSG method does not require exhastive line search
which implies less function evaluations per iteration.
• We also used the SPG method for Full waveform inversion,
obtaining very good results.

You might also like