You are on page 1of 56

'

$
An Optimization Technique for Estimating
Velocities and Fracture Orientation in
Orthorombic Media
by
Debora Cores
Jose G. Meza

&

Universidad Simon Bolvar

ISPM2000 - ATLANTA
August 2000

'

$
An Optimization Technique for Estimating
Velocities and Fracture Orientation in
Orthorombic Media
by
Debora Cores
Jose G. Meza

&

Universidad Simon Bolvar

III Jornadas IBO


Intevep, Noviembre 2000

'

$
The Influence of the Seismic Parameter
Acquisition on the Optimization Problem for
Estimating Velocities and Fracture Orientation
by
Debora Cores
Jose G. Meza

&

Universidad Simon Bolvar

Optimization 2001
Aveiro-Portugal, July 2001

'

$
OUTLINE

An orthorombic media (OM)


The Reflection Tomography Problem in OM
Historical Overview
Discretized Problem
Numerical Approach

&

Numerical Results
Conclusions

'
&

Anisotropy

V1 6= V2

The velocity does not changes


with the wave propagation direction.

The velocity changes with the wave propagation direction.

'

$
Anisotropy: An Orthorombic Media (OM)

An orthorombic media is an
anisotropic stratified medium with
vertical fractures.

&

'

PROBLEM: Reflection Tomography (SRT) Problem in OM

Minimize

1
2

kTr ; T (V )k22

T : IRm ! IRn travel time function, T = (T1 (V ) T2 (V ) : : :  Tn (V )) where,


Ti (V ) =

dli
V
(
x
y
z
)
Rayi

Tr 2 IRn real travel time


vector.

V 2 IRm is the velocity

&

vector in OM.

n is the number of layers.

'

Historical Overview

ISOTROPIC MEDIA :
Normal Equations:
Gauss Seidel with Successive Overrelaxation: T. Bishop et al, 1985
Levenberg and Marquardt Method
with SVD descomposition : Lines and
Treitel, 1985: S. Chiu el al, 1986; T.
Zhu and L. Brown, 1987; Farra and
Madariaga, 1998.

&

Low Storage Opt. Techniques:

ANISOTROPIC MEDIA:
In 2D elliptical anisotropic medium
Michelena et al. 1994.
In 2D medium with a sinusoidal aproximation of the velocity: Toshiki et al 1995.
In

3D

Transversally

medium : Grechka, 1995.

Spectral Gradient Method: Castillo,


Cores and Raydan , 2000.

anisotropic

'

DISCRETIZED PROBLEM: Ellipsoidal Aproximation

Contreras et al, 1997


1

V2
j

where

= V1z2 cos2 (1 ) + Vx12 cos2 (2 ) sin2 (1 ) + V1y2 sin2 (2 ) sin2 (1 )
j

j = P SV SH

correspond to the different

&

wave propagation modes.

'

DISCRETIZED PROBLEM: Travel tiem Function

i j ) pair reflecting in

The travel time function for a ray corresponding to the (

the layer ,

P
Tijk (X Y ) = 2k+1
h=2

&

r (xijk ;xijk )
h

h;1
2
vxh

ijk ;yijk )2
(yh
h;1

vy2h

ijk ;zijk )2
(zh
h;1

vz2h

X = (x1  x2  : : :  x2n+1 )
Y = (y1  y2  : : :  y2n+1 )
Z = (z1  z2  : : :  z2n+1 )
Vx = (vx1  vx2  : : :  vx2n+1 )
Vy = (vy1  vy2  : : :  vy2n+1 )
Vx = (vz1  vz2  : : :  vz2n+1 )
zi = fi (xi  yi ):
10

'

$
DISCRETIZED PROBLEM

Consider any symmetry axes, (Group angle 6= Ray angle)

&

11

'

DISCRETIZED PROBLEM

Azimutal Rotation

&
0
@;

cos( )
sin( )
0

Polar Rotation

sin( )

cos( )

1
A

0
@

cos( )

; sin()

cos( )

sin( )

12

1
A

'

GENERAL DISCRETIZED PROBLEM

P
P
P
ns
nr
n (T ; T (S ))
Tr ; T (S = i
rijk
ijk
j
k

Min
s:t: L  S  U
1
k
2

)k2
2

=1

=1

=1

T : <m ! <10n  S = (Vx  Vy  Vz   ) 2 <10n ,


Vx = (vx1  vx2  : : :  vx2n+1 )
Vy = (vy1  vy2  : : :  vy2n+1 )
Vx = (vz1  vz2  : : :  vz2n+1 )
 = (1  2  : : :  2n+1 )
 = (1  2  : : :  2n+1 )
L = (l1  l2  : : :  l10n)
U = (u1  u2  : : :  u10n ).

where,

&

13

'

$
GENERAL DISCRETIZED PROBLEM

v

!

!

!
kX u
ijk
ijk
ijk
u
xh
yh
zh
t
Tijk (S ) =
+
+
+
2

+1

h=2

vxh

vyh

vzh

v

!

!

!
u
n
ijk
ijk
ijk
X u
t xh + yh + zh
2

2 +1

&

h=2n+2;k

vxh

v yh

14

vzh

'

$
GENERAL DISCRETIZED PROBLEM

ijk

xh =

Dxh cos(h ) cos(h ) + Dyh cos(h ) sin(h ) ; Dzh sin(h )


ijk

ijk

zh =

&

yh =

;Dxh sin(h ) + Dyh cos(h )

Dxh sin(h ) cos(h) ; Dyh sin(h) sin(h ) + Dzh cos(h )


ijk
Dxh = xijk
;
x
h;1
h
Dyh = yhijk ; yhijk;1
Dzh = zhijk ; zhijk;1

15

'

NUMERICAL APPROACH

An optimization scheme for solving

MinimizekTr ; T (S )k2
s:t:
LSU
that satisfies the following conditions:
Only function and gradient evaluations (first order information) are
required

&

rf (S ) = JfT (S )(T (S ) ; Tr )

Global convergence
Fast local convergence
Low computational cost and storage
Box constraints
16

'

$
NUMERICAL APROACH

Low cost and storage orthorombic ray tracing algorithm (Loreto and
Cores, 1998-1999)

Minimize T (X Y Z )
The Spectral Proyected Gradient Method (SPG) (Birgin, Martinez and
Raydan, 1999) to solve,

&

Minimize kTr ; T (S )k2


s:t:

LSU

17

'

$
NUMERICAL APPROACH

Spectral Projected Gradient Method (SPG)


Step 1: If

kP (Sk ; rf (Sk )) ; Sk k  tol, then Stop.

Step 2: Nonmonotone Line-search

 = k
Step 2.2: Set S+ = P (Sk ; rf (Sk ))
T
Step 2.3: If f (S+ )  max0j kM ;1 f (Sk;j ) +  (S+ ; Sk ) rf (Sk ) then
k = , Sk+1 = S+ , Wk = Sk+1 ; Sk , yk = rf (Sk+1 ) ; rf (Sk ), go

Step 2.1: Set

to Step 3.

 2 1  2 ] go to Step 2.2
T
Step 3: bk = Wk yk
If bk  0, k+1 = max ,
a
T
else ak = Wk Wk and k+1 = minfmax  maxfmin  b k gg
k

&
else,

18

'

NUMERICAL RESULTS

We consider the following two synthetic models.

MODEL 2

MODEL 1
6

3
1

&
2

0
4

0
4

0
0

2
3

Model 1

Model 2

19

'

$
NUMERICAL RESULTS

The distribution of the sources and recievers was made in:


Squared Mesh: a

a  b squared

Squared Mesh

Radial Mesh

3.5

2.5

1.5

0.5

&
0

0.5

1.5

2.5

3.5

Radial Mesh: a circle of radious .

4
4

20

'

NUMERICAL RESULTS

L U )T be the lower and upper bounds of the velocities and

Let the vector (

fracture orientation angles respectively.

i = 1 : : :  2  n + 1
l(i) = 0:002  Vx (i)  500 = u(i)
l(i) = 0:002  Vy (i)  500 = u(i)
l(i) = 0:002  Vz (i)  500 = u(i)
l(i) = 0  (i)  90 = u(i)
l(i) = 0  (i)  90 = u(i)
Constrained Case: For i = 1 : : :  2  n + 1
l(i) = 0:2  Vx (i)  5 = u(i)
l(i) = 0:2  Vy (i)  5 = u(i)
l(i) = 0:2  Vz (i)  5 = u(i)
l(i) = 10  (i)  30 = u(i)
l(i) = 2  (i)  9 = u(i)
Unconstrained Case: For

&

21

'

$
NUMERICAL RESULTS

Initial Iterates:

Vx0 = Vy0 = Vz0 = (3 4 5 5 4 3)T


0 =
0 = (10 10 10 10 10 10)T
Note: The velocities are measuared in Km
Angles are measuared in degrees.

&

Stopping Criterium :

kP (Sk ; rf (Sk )) ; Sk k2  10;6


Sun Station Ultra 10
M=8 in the SPG Method

22

'

Model 1 (P-Wave), Unconstrained Case


Squared Mesh, ns=2, nr=15

&

Radial Mesh, ns=5, nr=16

Vx

Vxa

Vy

Vya

Vz

Vza

Vxa

Vya

Vza

1.5

1.35

1.7

1.69

1.9

2.12

1.28

1.7

2.23

2.97

2.3

2.29

2.5

1.68

2.56

2.3

1.95

2.8

2.79

3.3

3.3

3.45

2.8

2.86

2.8

2.8

3.3

3.3

3.45

2.8

2.86

2.97

2.3

2.3

2.5

1.68

2.65

2.29

1.94

1.5

1.35

1.7

1.7

1.9

2.1

1.28

1.69

2.23

Squared Mesh, ns=2, nr=15

Radial Mesh, ns=5, nr=16

(Aprox.)

(Aprox.)

(Aprox.)

(Aprox.)

34.43

20

20

39.7

19.99

48.88

15

14.91

68.52

14.99

4.47

25

23.62

58.25

24.98

4.38

25

23.28

58.25

24.98

48.84

15

14.91

68.52

14.99

35.24

20

20

39.71

19.99

23

'

Model 1 (P-Wave), Constrained Case


Squared Mesh, ns=2, nr=15

Radial Mesh, ns=5, nr=16

Vx

Vxa

Vy

Vya

Vz

Vza

Vxa

Vya

Vza

1.5

1.49

1.7

1.7

1.9

1.91

1.49

1.69

1.91

1.99

2.3

2.29

2.5

2.5

1.99

2.3

2.5

2.8

2.8

3.3

3.3

2.99

2.8

3.3

2.8

2.8

3.3

3.3

2.99

2.79

3.3

1.99

2.3

2.29

2.5

2.5

1.99

2.29

2.51

1.5

1.49

1.7

1.7

1.9

1.91

1.49

1.7

1.91

Squared Mesh, ns=2, nr=15

&

Radial Mesh, ns=5, nr=16

(Aprox.)

(Aprox.)

(Aprox.)

(Aprox.)

8.95

20

19.99

8.97

19.99

8.68

15

14.91

8.79

14.99

5.81

25

23.42

5.78

24.95

5.81

25

23.42

5.78

24.98

8.68

15

14.91

8.79

14.99

8.95

20

19.99

8.97

19.99

24

'
&

NUMERICAL RESULTS

25

$
%

'
&

NUMERICAL RESULTS

26

$
%

'

Model 1 (P-S Wave), Unconstrained Case


Squared Mesh, ns=2, nr=15

&

Radial Mesh, ns=5, nr=16

Vx

Vxa

Vy

Vya

Vz

Vza

Vxa

Vya

Vza

1.5

1.92

1.7

1.66

1.9

1.24

1.14

1.59

2.12

1.95

2.3

3.13

2.5

2.37

2.77

2.11

1.58

2.85

2.8

2.85

3.3

3.19

3.7

2.85

2.46

2.7

2.83

2.9

2.85

3.1

3.21

3.69

2.84

2.48

1.8

1.81

2.17

2.3

2.45

2.76

2.2

1.69

1.3

1.9

1.6

1.64

1.8

1.46

1.28

1.69

2.15

Squared Mesh, ns=2, nr=15

Radial Mesh, ns=5, nr=16

(Aprox.)

(Aprox.)

(Aprox.)

(Aprox.)

71.82

20

19.99

34.49

19.99

14.64

15

14.89

53.22

14.99

6.78

25

48.34

18.61

6.78

25

1.74

46.94

18.55

10.44

15

15.03

48.78

14.99

64.52

20

19.97

40.57

19.99

27

'

Model 1 (P-S Wave), Constrained Case


Squared Mesh, ns=2, nr=15

&

Radial Mesh, ns=5, nr=16

Vx

Vxa

Vy

Vya

Vz

Vza

Vxa

Vya

Vza

1.5

1.42

1.7

1.64

1.9

1.84

1.29

1.59

1.83

1.89

2.3

2.14

2.5

2.38

1.98

2.22

2.39

2.86

2.8

2.84

3.3

3.17

2.85

2.85

3.19

2.7

2.84

2.9

2.86

3.1

3.23

2.85

2.84

3.21

1.8

1.89

2.16

2.3

2.42

1.81

2.09

2.4

1.3

1.38

1.6

1.66

1.8

1.87

1.49

1.7

1.87

Squared Mesh, ns=2, nr=15

Radial Mesh, ns=5, nr=16

(Aprox.)

(Aprox.)

(Aprox.)

(Aprox.)

8.08

20

19.89

5.28

20

8.96

15

14.97

8.62

14.98

5.48

25

10

5.42

13.02

5.45

25

10.25

5.94

14.91

8.99

15

15.28

6.62

14.99

8.41

20

19.91

8.76

19.99

28

'
&

NUMERICAL RESULTS

29

$
%

'
&

NUMERICAL RESULTS

30

$
%

'

Model 2 (P Wave), Unconstrained Case


Squared Mesh, ns=2, nr=15

&

Radial Mesh, ns=5, nr=16

Vx

Vxa

Vy

Vya

Vz

Vza

Vxa

Vya

Vza

1.5

1.9

1.7

1.7

1.9

1.5

1.49

1.7

1.91

2.6

2.3

2.08

2.5

2.07

2.01

2.29

2.48

3.96

2.8

2.82

3.3

2.62

3.46

2.81

2.96

3.97

2.8

2.89

3.3

2.69

3.36

2.8

2.95

2.82

2.3

2.49

2.5

1.47

2.01

2.31

2.5

1.5

1.88

1.7

1.69

1.9

1.49

1.49

1.69

1.9

Squared Mesh, ns=2, nr=15

Radial Mesh, ns=5, nr=16

(Aprox.)

(Aprox.)

(Aprox.)

(Aprox.)

90

20

19.8

6.7

19.99

38.69

15

3.3

12.83

48.46

25

45.68

62.77

36.31

36.77

25

49.29

57.28

34.31

71.88

15

3.13

12.88

90

20

20.15

6.73

20

31

'

Model 2 (P Wave), Constrained Case


Squared Mesh, ns=2, nr=15

Radial Mesh, ns=5, nr=16

Vx

Vxa

Vy

Vya

Vz

Vza

Vxa

Vya

Vza

1.5

1.5

1.7

1.69

1.9

1.89

1.5

1.7

1.89

1.96

2.3

2.27

2.5

2.46

2.29

2.5

3.03

2.8

2.81

3.3

3.33

2.99

2.79

3.3

2.98

2.8

2.79

3.3

3.27

2.99

2.8

3.3

2.04

2.3

2.33

2.5

2.53

1.99

2.29

2.5

1.5

1.49

1.7

1.7

1.9

1.9

1.5

1.69

1.89

Squared Mesh, ns=2, nr=15

&

Radial Mesh, ns=5, nr=16

(Aprox.)

(Aprox.)

(Aprox.)

(Aprox.)

5.33

20

20.06

4.66

20

5.83

15

14.54

7.56

15.48

5.99

25

23.11

5.94

23.69

5.99

25

21.95

5.99

24.29

6.24

15

14.22

7.44

15.18

5.26

20

19.94

4.66

19.99

32

'
&

NUMERICAL RESULTS

33

$
%

'
&

NUMERICAL RESULTS

34

$
%

'

Model 2 (P-S Wave), Unconstrained Case


Squared Mesh, ns=2, nr=15

Vx

Vxa

Vy

Vya

Vz

Vza

1.5

1.92

1.7

1.71

1.9

2.39

2.3

1.87

3.73

2.8

2.7

3.63

1.8
1.3

&

Radial Mesh, ns=5, nr=16

Vxa

Vya

Vza

1.503

1.502

1.69

1.89

2.5

1.92

1.76

2.02

2.25

3.02

3.3

2.29

3.51

2.89

2.96

2.9

2.75

3.1

2.71

2.91

2.81

2.76

2.89

2.44

2.3

1.35

1.96

2.28

2.69

1.77

1.6

1.59

1.8

1.303

1.302

1.59

1.79

Squared Mesh, ns=2, nr=15

Radial Mesh, ns=5, nr=16

(Aprox.)

(Aprox.)

(Aprox.)

(Aprox.)

88.92

20

18.4

20

32.38

15

18.36

22.73

66.92

25

22.55

89.16

13.05

28.39

25

36.24

86.88

40.32

74.75

15

14.57

21.16

90

20

21.22

1.07

19.98

35

'

Model 2 (P-S Wave), Constrained Case


Squared Mesh, ns=2, nr=15

&

Radial Mesh, ns=5, nr=16

Vx

Vxa

Vy

Vya

Vz

Vza

Vxa

Vya

Vza

1.5

1.5

1.7

1.69

1.9

1.86

1.49

1.69

1.903

1.91

2.3

2.22

2.5

2.37

1.81

2.06

2.21

2.86

2.8

2.85

3.3

3.21

3.01

2.81

3.55

2.7

2.83

2.9

2.85

3.1

3.19

2.73

2.88

2.91

1.8

1.89

2.07

2.3

2.42

2.01

2.27

2.59

1.3

1.29

1.6

1.61

1.8

1.84

1.29

1.6

1.81

Squared Mesh, ns=2, nr=15

Radial Mesh, ns=5, nr=16

(Aprox.)

(Aprox.)

(Aprox.)

(Aprox.)

5.84

20

21.49

8.46

20.1

7.36

15

15.46

4.93

13.46

25

20.31

25.13

5.78

25

19.14

20.43

7.47

15

14.29

4.47

12

5.01

20

18.94

7.64

19.86

36

'
&

NUMERICAL RESULTS

37

$
%

'
&

NUMERICAL RESULTS

38

$
%

'

NUMERICAL RESULTS
# rays vs. velocity error
0.1

# rays vs. polar error

# rays vs. azimutal error

7.5

4
S

0.08
S

0.06

6.5
6

2
R

5.5

0.04

5
0.02
0

4.5
R
0

1000

2000

R
0

# rays vs. cputime

&

80

20
R
0

1000

2000

700

3500

600

3000

500

2500

400

200

1000

100
0

1000

39

2000

300

1500

500

1000

# rays vs. # backtracking

4000

2000

40

2000

# rays vs. # iteration

100

60

1000

2000

1000

2000

'

CONCLUSIONS

We solve the velocity and fracture orientation inversion problem in OM


using the Spectral Projected Gradient Method (SPG) and an ellipsoidal
approximation of the velocity.
This is a highly nonlinear problem that has many solutions, so
regularization of the problem is required.
The SPG method obtain good precision for the velocities estimates using
a relative small number of rays and no regularization.
To obtain a good estimate of the azimuthal and polar angle vectors

&

regularity is essential.
To get a constrained region (regularity ) is not a difficult task in seismic
since the maximum and minimum values of the velocities in the medium
is know a priori.
40

'

A better estimate of the azimuthal angle vector can be obtained if there


are rays in all different azimuths (For example, using Radial Mesh).

None of the Mesh distributions used in this work give enough information
for obtaining a good estimate of the polar angle vector.
The problem in obtaining a better estimate of the polar angle vector is not
the optimization scheme used, but 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, obviously this imply an increase
in the cpu-time.

&

Also, increasing the number of rays, the number of iterations and number
of back-trackings may be reduced.
On the other hand, the error in the polar angle vector increases even if
the number of rays increase, since the seismic data distribution is not the
adequated for estimating the polar angle vector.
41

'

The ray tracing takes most CPU time required for the inversion, so a
parallel low cost ray tracing will reduce the CPU time.

&

42

'

NUMERICAL RESULTS

TEST 1: Corresponding to Model 1

L = (l1  : : :  l30 )T and U = (u1  : : :  u30 )T where


li = 0:002 i = 1 : : :  18 li = 0 i = 19 : : :  30
ui = 500 i = 1 : : :  18 ui = 5 i = 19 : : :  30
nls = 3, ns = 6, nlr = 4 and nr = 28
Initial Velocities

Vx0

Vy0

Vz0

VxR

Real Velocities

VyR

VzR

Vxa

1.5

1.7

1.9

2.3

&

Approximated Velocities

Vya

Vxa

1.4999973

1.7019851

1.9019638

2.5

1.9984656

2.3011494

2.5005751

2.8

3.3

2.9998747

2.7991592

3.3009248

2.8

3.3

2.9997968

2.8008410

3.2994352

2.3

2.5

2.0011699

2.2988378

2.4999144

1.5

1.7

1.9

1.5062839

1.6980178

1.8985726

43

'

NUMERICAL RESULTS

TEST 1: Corresponding to Model 1

L = (l1  : : :  l30 )T and U = (u1  : : :  u30 )T where


li = 0:002 i = 1 : : :  18 li = 0 i = 19 : : :  30
ui = 500 i = 1 : : :  18 ui = 5 i = 19 : : :  30
nls = 3, ns = 6, nlr = 4 and nr = 28
Initial Angles

0

&

Real Angles

R

Approximated Angles

a

1.5322834

0.0033316

1.3463286

0.0049373

1.4478746

0.0000000

1.4356391

0.3929935

1.3585243

0.0341731

1.5861549

0.0000000

CPU ; time = 22:82min, iter = 507

and

44

line ; searches = 126

'

NUMERICAL RESULTS

TEST 2: Corresponding to Model 1

L = (l1  : : :  l30 )T and U = (u1  : : :  u30 )T where


li = 0:002 i = 1 : : :  18 li = 0 i = 19 : : :  30
ui = 6 i = 1 : : :  18 ui = 40 i = 19 : : :  30
nls = 3, ns = 6, nlr = 4 and nr = 28
Initial Velocities

Vx0

Vy0

Vz0

VxR

3.5

3.5

Real Velocities

VyR

VzR

Vxa

1.5

1.7

1.9

3.5

2.7

2.5

3.5

3.5

&

Approximated Velocities

Vya

Vxa

1.4867472

1.6086736

1.7609429

2.9

2.7073193

2.4508121

2.7925423

2.8

3.3

3.0203938

2.7488306

3.1746943

2.9

2.7

3.2

3.0147983

2.7494752

3.1800087

3.5

2.6

2.4

2.8

2.6841448

2.4514979

2.8105226

1.4

1.6

1.8

1.5616237

1.6898496

1.7545015

45

'

NUMERICAL RESULTS

TEST 2: Corresponding to Model 1

L = (l1  : : :  l30 )T and U = (u1  : : :  u30 )T where


li = 0:002 i = 1 : : :  18 li = 0 i = 19 : : :  30
ui = 6 i = 1 : : :  18 ui = 40 i = 19 : : :  30
nls = 3, ns = 6, nlr = 4 and nr = 28
Initial Angles

0

&

Real Angles

R

Approximated Angles

a

30

10

1.4881100

9.9471689

30

10

5.3222734

10.2593893

30

10

0.3525908

9.8410988

30

10

1.2738337

9.7745561

30

10

5.3366231

9.8324231

30

10

4.7458023

10.0201751

CPU ; time = 41:9min, iter = 379

and

46

line ; searches = 93

'

NUMERICAL RESULTS

TEST 3: Corresponding to Model 1

L = (l1  : : :  l30 )T and U = (u1  : : :  u30 )T where


li = 0:2 i = 1 : : :  18 li = 20 i = 19 : : :  24 li = 5 i = 25 : : :  30
ui = 6 i = 1 : : :  18 ui = 40 i = 19 : : :  24 ui = 20 i = 25 : : :  30
nls = 4, ns = 20, nlr = 6 and nr = 66
Initial Velocities

Real Velocitie s

Vx0

Vy0

Vz0

VxR

VyR

VzR

Vxa

1.5

1.7

1.9

3.5

3.5

3.5

2.7

2.5

3.5

3.5

&

Approximated Velocities

Vya

Vxa

1.4376400

1.7011965

1.9290055

2.9

2.6773830

2.4507081

2.8187457

2.8

3.3

2.9590742

2.7493184

3.2417599

2.9

2.7

3.2

2.9503655

2.7497884

3.2461665

3.5

2.6

2.4

2.8

2.6758592

2.4530501

2.8238717

1.4

1.6

1.8

1.3586700

1.5958847

1.9018278

47

'

NUMERICAL RESULTS

TEST 3: Corresponding to Model 1

L = (l1  : : :  l30 )T and U = (u1  : : :  u30 )T where


li = 0:2 i = 1 : : :  18 li = 20 i = 19 : : :  24 li = 5 i = 25 : : :  30
ui = 6 i = 1 : : :  18 ui = 40 i = 19 : : :  24 ui = 20 i = 25 : : :  30
nls = 4, ns = 20, nlr = 6 and nr = 66
Initial Angles

0

&

Real Angles

R

Approximated Angles

a

30

10

37.5290024

9.9980061

30

10

22.9054416

10.0189242

30

10

29.4423860

9.9919281

30

10

28.9710453

9.9914708

30

10

23.3144158

10.0187249

30

10

33.8245728

9.9980831

CPU ; time = 14 hours, iter = 2633

and

48

line ; searches = 491

'

NUMERICAL RESULTS

TEST 4: Corresponding to Model 2

L = (l1  : : :  l50 )T and U = (u1  : : :  u50 )T where


li = 0:2 i = 1 : : :  30 li = 0 i = 31 : : :  50
ui = 6 i = 1 : : :  30 ui = 40 i = 31 : : :  40 ui = 5 i = 41 : : :  50
nls = 1, ns = 2, nlr = 2 and nr = 15
Vx0 Vy0 Vz0 VxR VyR VzR
Vxa
Vya
Vxa
2.5

2.6

1.5

1.7

1.9

1.5161705

1.7044085

1.8036717

3.5

3.2

3.6

2.7

2.5

2.9

2.6627301

2.4650854

2.8372655

3.8

4.1

2.8

3.3

2.9680257

2.7636329

3.2310795

4.1

4.3

4.5

3.3

3.5

3.6

3.1759822

3.4724621

3.6971851

4.3

4.5

4.5

3.5

3.6

3.8

3.7108556

3.7538805

3.5634800

4.3

4.5

4.5

3.4

3.5

3.7

3.8196233

3.5490186

3.5673048

4.1

4.3

4.5

3.2

3.4

3.5

3.1410118

3.4148884

3.5903418

3.8

4.1

2.9

2.7

3.2

2.9575074

2.7362643

3.2367196

3.5

3.2

3.6

2.6

2.4

2.8

2.6577356

2.4378346

2.8389060

2.5

2.6

1.4

1.6

1.8

1.4447071

1.5913248

1.8147302

&

49

'

NUMERICAL RESULTS

TEST 4: Corresponding to Model 2

L = (l1  : : :  l50 )T and U = (u1  : : :  u50 )T where


li = 0:2 i = 1 : : :  30 li = 0 i = 31 : : :  50
ui = 6 i = 1 : : :  30 ui = 40 i = 31 : : :  40 ui = 5 i = 41 : : :  50
nls = 1, ns = 2, nlr = 2 and nr = 15
a
0 0 R R
a

&

23

30

26.9826548

0.0023920

23

30

27.8955924

0.0056462

23

30

28.1072325

0.0169888

23

30

32.2204377

0.9686877

23

30

36.8106643

3.5928424

23

30

6.5430180

0.9112516

23

30

38.4896379

1.0487732

23

30

27.2661348

0.0000001

23

30

28.0762195

0.0000000

23

30

22.3459052

0.0065288

CPU ; time = 2:17 hours, iter = 2264


50

and

line ; searches = 529

'

NUMERICAL RESULTS

TEST 5: Corresponding to Model 2

L = (l1  : : :  l50 )T and U = (u1  : : :  u50 )T where


li = 0:2 i = 1 : : :  30 li = 0 i = 31 : : :  50
ui = 6 i = 1 : : :  30 ui = 40 i = 31 : : :  40 ui = 5 i = 41 : : :  50
nls = 2, ns = 6, nlr = 4 and nr = 28
Vx0 Vy0 Vz0 VxR VyR VzR
Vxa
Vya
Vxa
2.5

2.6

1.5

1.7

1.9

1.5146239

1.6834034

1.8664965

3.5

3.2

3.6

2.7

2.5

2.9

2.6520520

2.4716170

2.8821383

3.8

4.1

2.8

3.3

2.9540754

2.7565136

3.2766908

4.1

4.3

4.5

3.3

3.5

3.6

3.2934095

3.5050076

3.6028292

4.3

4.5

4.5

3.5

3.6

3.8

3.7115094

3.6230909

3.5623813

4.3

4.5

4.5

3.4

3.5

3.7

3.7822892

3.5025558

3.5610915

4.1

4.3

4.5

3.2

3.4

3.5

3.2050763

3.3926163

3.4969652

3.8

4.1

2.9

2.7

3.2

2.9172908

2.7421486

3.2543635

3.5

3.2

3.6

2.6

2.4

2.8

2.6143580

2.4293056

2.8551777

2.5

2.6

1.4

1.6

1.8

1.3901383

1.6163345

1.8183832

&

51

'

NUMERICAL RESULTS

TEST 5: Corresponding to Model 2

L = (l1  : : :  l50 )T and U = (u1  : : :  u50 )T where


li = 0:2 i = 1 : : :  30 li = 0 i = 31 : : :  50
ui = 6 i = 1 : : :  30 ui = 40 i = 31 : : :  40 ui = 5 i = 41 : : :  50
nls = 2, ns = 6, nlr = 4 and nr = 28
a
0 0 R R
a

&

23

30

32.3391389

0.0000489

23

30

34.1106580

0.0000030

23

30

32.3994191

0.0060426

23

30

28.2734365

0.2313614

23

30

33.7757315

1.8283798

23

30

14.0161269

1.4759419

23

30

31.3560557

0.2281124

23

30

31.2219247

0.0000007

23

30

32.5670208

0.0000002

23

30

26.4797392

0.0000259

CPU ; time = 16 hours, iter = 5661

and

52

line ; searches = 1247

'

NUMERICAL RESULTS

TEST 6: Corresponding to Model 2

L = (l1  : : :  l50 )T and U = (u1  : : :  u50 )T where


li = 0:2 i = 1 : : :  30 li = 0 i = 31 : : :  50
ui = 6 i = 1 : : :  30 ui = 40 i = 31 : : :  40 ui = 5 i = 41 : : :  50
nls = 3, ns = 12, nlr = 5 and nr = 45
Vx0 Vy0 Vz0 VxR VyR VzR
Vxa
Vya
Vxa
2.5

2.6

1.5

1.7

1.9

1.5015321

1.6815115

1.8446112

3.5

3.2

3.6

2.7

2.5

2.9

2.6330562

2.4550646

2.8898359

3.8

4.1

2.8

3.3

2.9731147

2.7550248

3.2525118

4.1

4.3

4.5

3.3

3.5

3.6

3.3050384

3.453214

3.6168122

4.3

4.5

4.5

3.5

3.6

3.8

3.9870890

3.8343516

3.0960452

4.3

4.5

4.5

3.4

3.5

3.7

3.9948479

3.7634267

3.1293398

4.1

4.3

4.5

3.2

3.4

3.5

3.2320777

3.3985567

3.5229264

3.8

4.1

2.9

2.7

3.2

2.9321892

2.7441214

3.2395075

3.5

3.2

3.6

2.6

2.4

2.8

2.6083586

2.4465557

2.8750930

2.5

2.6

1.4

1.6

1.8

1.3929838

1.6172687

1.8505478

&

53

'

NUMERICAL RESULTS

TEST 6: Corresponding to Model 2

L = (l1  : : :  l50 )T and U = (u1  : : :  u50 )T where


li = 0:2 i = 1 : : :  30 li = 0 i = 31 : : :  50
ui = 6 i = 1 : : :  30 ui = 40 i = 31 : : :  40 ui = 5 i = 41 : : :  50
nls = 3, ns = 12, nlr = 5 and nr = 45
a
0 0 R R
a

&

23

30

34.3991447

0.0043882

23

30

35.4810917

0.0013095

23

30

30.9784433

0.0881668

23

30

33.9740601

0.0000000

23

30

33.5488468

3.7197016

23

30

37.6389450

0.3137009

23

30

32.450822

0.0000000

23

30

28.059081

0.087736

23

30

34.2884020

0.0047640

23

30

25.9546206

0.0020617

CPU ; time = 60 hours, iter = 3974

and

54

line ; searches = 962

'

$
NUMERICAL RESULTS

150 rays

840 rays

2700 rays

1.749

0.622

1.27

0.631

0.54

1.25

k( r 
r ) ; ( k 
k )k2

1.63

0.317

0.24

Iterations

2264

5661

3974

Line-searches

529

1247

962

CPU-time

16

60

kSr ; Sk k2
kVr ; Vk k2

&

55

'

CONCLUSIONS

We solve the velocity and fracture orientation inversion problem in OM


using the Spectral Projected Gradient Method (SPG) and an ellipsoidal
approximation of the velocity.
This is a highly nonlinear problem that has many solutions, so
regularization of the problem is required..
The SPG method obtain good precision for the velocities estimates using
a relative small number of rays and no regularization.
To estimate de azimuthal and polar angles regularity is essential.

&

A better estimate of the azimuthal angle can be obtained if there are rays
in all different azimuths (For example using Radial Mesh).
The ray tracing takes most CPU time required for the inversion, so a
parallel low cost ray tracing will reduce the CPU time.
56