Source-independent full waveform inversion of seismic data

• US 6,999,880 B2
• Filed: 03/11/2004
• Issued: 02/14/2006
• Est. Priority Date: 03/18/2003
• Status: Expired due to Fees

• Current Assignee: Regents of the University of California (University of California)
• Original Assignee: Regents of the University of California (University of California)
• Inventors: Lee, Ki Ha
• Primary Examiner(s): MCELHENY JR, DONALD E
• Application Number:
  US10/799,131
• Publication Number:
  US 20040186667A1
• Time in Patent Office: 705 Days
• Field of Search: 702/6, 702 11- 14, 702/17, 703/10, 367/73
• US Class Current: 702/14
• CPC Class Codes: G01V 1/30 Analysis G01V1/50 takes precedence

Abstract

A set of seismic trace data is collected in an input data set that is first Fourier transformed in its entirety into the frequency domain. A normalized wavefield is obtained for each trace of the input data set in the frequency domain. Normalization is done with respect to the frequency response of a reference trace selected from the set of seismic trace data. The normalized wavefield is source independent, complex, and dimensionless. The normalized wavefield is shown to be uniquely defined as the normalized impulse response, provided that a certain condition is met for the source. This property allows construction of the inversion algorithm disclosed herein, without any source or source coupling information. The algorithm minimizes the error between data normalized wavefield and the model normalized wavefield. The methodology is applicable to any 3-D seismic problem, and damping may be easily included in the process.

Claims

1. An iterative method of normalized waveform invasion to obtain a model parameter describing one or more physical properties of a medium, the method comprising the steps of:

a) inputting a time domain measurement data set D_ji^d(t), (j=1~NG;

i=1~NS),

where d denotes a data tensor, NG is the number of receivers and NS is the number of sources;

b) means for minimizing a model parameter m below an error bound in the process of creating a normalized waveform inversion, wherein said minimizing step further comprises;

i) Fourier transforming the time domain measurement data set D_ji^d(t) to create a measurement spectral data set D_ji^d(ω

);

ii) normalizing the measurement spectral data set D_ji^d(ω

) to create a normalized data wavefield T_ji^d(ω

);

iii) modeling a medium by iterating the model parameter m describing one or more physical properties of the medium;

by

(1) minimizing a weighted error, between the normalized data wavefield T_ji^d(ω

) and the normalized modeled wavefield T_ji^m(ω

) of the response of the medium, to a level below the error bound;

(2) outputting the iterated model parameter m corresponding to the weighted error below the error bound, as a minimized model parameter.

2. The method of claim 1, wherein within the modeling step the physical property is selected from the group comprising:

a velocity of sound propagation, density, permeability, porosity, resistivity, permittivity, Young'"'"'s modulus, a component of the tensorial stress-strain properties, shear modulus, amount of water present, and amount of oil present.

3. The method of claim 1, wherein within the modeling step the error bound divided by an initial error E₀ is selected from the group consisting of:

below 10<sup>−

5</sup>, below 10⁶, below 10<sup>−

7</sup>, and below 10<sup>−

8</sup>.

5. The method of claim 1 wherein the modeled normalized wavefield T_ji^m(ω

) is independent of the input waveform applied to any of the NS source locations.

6. The method of claim 1 further comprising the step of:

a) storing the minimized model parameter m in a computer-readable medium.

7. The method of claim 1 further comprising the steps of:

a) displaying the minimized model parameter m as a graphical representation of one or more of the physical properties of the medium measured,

b) storing an output image represented by the minimized model parameter m in a computer-readable medium.

8. The method of claim 1 further comprising the step of;

a) storing as a computer program in at least one computer-readable medium,

b) said steps of claim 1.

9. The method of claim 1 wherein said weighted error results from a weighing matrix W_d that is an inverse of the standard deviation of the normalized data wavefield T_ji^d(ω

).

10. The method of normalized waveform inversion of claim 1 wherein said minimizing step weighted error φ

(m) is calculated by a root mean square of the weighing matrix W₆ multiplied by the difference between the normalized data wavefield T_ji^d(ω

) and the normalized modeled wavefield T_ji^m(ω

), or φ

(m)=∥

W_d(T_ji^d(ω

)−

T_ji^d(ω

)∥

² summed over all frequencies, i sources, and j receivers.

11. The method of claim 1 wherein said normalization step is further comprised of the steps of:

a) choosing one of the measurement spectral data set D_ji^d(ω

) at one of the measurement locations, such as D_1i^d(ω

),

(1) said chosen measurement spectral data set D_1i^d(ω

) known as a reference measurement;

ii) dividing the measurement spectral data set D_1i^d(ω

) amplitude at a particular frequency in the measurement spectral data set D_ji^d(ω

),

iii) by the amplitude of the reference measurement D_1i^d(ω

) at the same frequency,

b) to create a normalized data wavefield T_ji^d(ω

).

12. The method of claim 1 wherein the error bound of the minimizing step is calculated by first assigning an initial error E₀, then dividing each subsequent iterated error bound by the initial error E₀.

14. The method of claim 1 further comprising the step of;

a) storing as a computer program in at least one computer-readable medium,

i) said steps of claim 1.

17. The method of claim 1 further comprising the step of;

a) storing as a computer program in at least one computer-readable medium, said steps of claim 1.

18. An article of manufacture for normalized waveform inversion using a computer, said article comprising:

a computer readable medium comprising instructions for a computer to execute, said execution comprising the steps of claim 1.

4. An iterative method of using normalized waveform inversion to obtain a model parameter describing one or more physical properties of a medium, the method comprising the steps of:

a) inputting one or more each of NS source and NG measurement spatial locations;

b) measuring time domain data at each of the NG measurement locations resulting from an input waveform at one of the NS source locations propagating through a medium,

i) for each of the NS source locations,

ii) thereby forming a time domain measurement data set D_ji^d(t), (j=1~NG;

i=1~NS), where d denotes a data tensor;

c) Fourier transforming the time domain measurement data set D_ji^d(t) to create a measurement spectral data set D_ji^d(ω

) having frequency and amplitude information for each of the NG measurement locations;

d) normalizing the measurement spectral data set D_ji^d(ω

) to create a normalized data wavefield T_ji^d(ω

);

e) modeling the medium using an iterated model parameter m describing one or more physical properties of the medium,

(1) the NS source and the NG measurement spatial locations used as respective model input and model response spatial locations contained within the model of the medium,

(2) and initializing the iterated model parameter in with corresponding one or more known bulk properties or the medium being modeled,

ii) said modeling step comprising;

iii) creating a measurement model by;

(1) applying a delta function source collocated with the i^th NS source,

(2) modeling the response at the NG measurement locations, using a time domain modeling method, to create a synthetic medium response at the j^th receiver due to the i^th source, P_ji^m(t), m denoting the model response,

(3) repeating the applying and modeling steps at each of the NS source locations and NG measurement locations until the measurement model is full, and

iv) Fourier transforming the model response P_ji^m(t), to obtain a frequency domain synthetic response P_ji^m(ω

);

v) forming a normalized modeled wavefield using the frequency domain synthetic response T_ji^m(ω

)=P_ji^m(ω

)[P_1i^m(ω

)]<sup>−

1</sup>; and

f) minimizing a weighted error, between the normalized data wavefield T_ji^d(ω

) and the normalized modeled wavefield T_ji^m(ω

) of the response of the medium, to a level below an error bound,

i) said weighted error met by using the iterated model parameter m, known as the a minimized model parameter m.

13. An iterative method of using normalized waveform inversion to obtain a model parameter describing one or more physical properties of a medium, the method comprising the steps of:

a) inputting one or more each of NS source and NG measurement spatial locations;

b) measuring time domain data at each of the NG measurement locations resulting from an input waveform at one of the NS source locations propagating through a medium,

i) for each of the 195 source locations,

ii) thereby forming a time domain measurement data set D_ji^d(t), (j=1~NG;

i=1~NT), where d denotes a data tensor;

c) Fourier transforming the time domain measurement data set D_ji^d(t) to create a measurement spectral data set D_ji^d(ω

) having frequency and amplitude information for each of the NG measurement locations;

d) normalizing the measurement spectral data set D_ji^d(ω

) to create a normalized data wavefield T_ji^d(ω

);

e) modeling the medium using an iterated model parameter m describing one or more physical properties of the medium,

the NS source and the NG measurement spatial locations used as respective model input and model response spatial locations contained within the model of the medium,

and initializing the iterated model parameter m with corresponding one or more known bulk properties of the medium being modeled,

 

said modeling step comprising;

i) creating a measurement model by;

(1) applying a delta function source collocated with the i^th NS source,

(2) modeling the response at the NG measurement locations, using a time domain modeling method, to create a synthetic medium response at the j^th receiver due to the i^th source, P_ji^m(t), m denoting the model response,

(3) repeating the applying and modeling steps at each of the NS source locations and NG measurement locations until the measurement model is full, and

ii) Fourier transforming the model response P_ji^m(t), to obtain a frequency domain synthetic response P_ji^m(t);

iii) forming a normalized modeled wavefield using the frequency domain synthetic response T_ji^m(ω

)=P_ji^m(ω

)[P_1i^m(ω

)]<sup>−

1</sup>; and

f) minimizing a weighted error, between the normalized data wavefield T_ji^d(ω

) and the normalized modeled wavefield T_ji^m(ω

) of the response of the medium, to a level below an error bound,

g) said weighted error met by using the iterated model parameter m, known as the a minimized model parameter m.

15. An article of manufacture for normalized waveform inversion using a computer, said article comprising:

a computer readable medium comprising instructions for a computer to execute, said execution comprising the steps of;

a) inputting one or more each of NS source and NG measurement spatial locations;

b) measuring time domain data at each of the NG measurement locations resulting from an input waveform at one of the NS source locations propagating through a medium,

i) for each of the NS source locations,

ii) thereby forming a time domain measurement data set D_ji^d(t), (j=1~NG;

i=1~NS), where d denotes a data tensor;

c) Fourier transforming the time domain measurement data set D_ji^d(t) to create a measurement spectral data set D_ji^d(ω

) having frequency and amplitude information for each of the NG measurement locations;

d) normalizing the measurement spectral data set D_ji^d(ω

) to create a normalized data wavefield T_ji^d(ω

);

e) modeling the medium using an iterated model parameter m describing one or more physical properties of the medium,

the ATS source and the NG measurement spatial locations used as respective model input and model response spatial locations contained within the model of the medium.

and initializing the iterated model parameter m with corresponding one or more known bulk properties of the medium being modeled,

 

said modeling step comprising;

i) creating a measurement model by;

(1) applying a delta function source collocated with the i^th NS source,

(2) modeling the response at the NG measurement locations, using a time domain modeling method, to create a synthetic medium response at the j^th receiver due to the i^th source, P_ji^m(ω

), m denoting the model response,

(3) repeating the applying and modeling steps at each of the NS source locations and NG measurement locations until the measurement model is full, and

ii) Fourier transforming the model response P_ji^m(ω

), to obtain a frequency domain synthetic response P_ji^m(ω

);

iii) forming a normalized modeled wavefield using the frequency domain synthetic response T_ji^m(ω

)=P_ji^m(ω

)[P_1i^m(ω

)]<sup>−

1</sup>; and

f) minimizing a weighted error, between the normalized data wavefield T_ji^d(ω

) and the normalized modeled wavefield T_ji^m(ω

) of the response of the medium, to a level below an error bound,

i) said weighted error met by using the iterated model parameter m, known as the a minimized model parameter m.

16. An iterative method of using normalized waveform inversion to obtain a model parameter describing one or more physical properties of a medium, the method comprising the steps of:

a) inputting one or more each of NS source and NG measurement spatial locations;

b) measuring time domain data at each of the NG measurement locations resulting from an input waveform at one of the NS source locations propagating through a medium,

i) for each of the NS source locations,

ii) thereby forming a time domain measurement data set D_ji^d(ω

), (j=1~NG;

i=1~NS) where d denotes a data tensor;

c) Fourier transforming the time domain measurement data set D_ji^d(t) to create a measurement spectral data set D_ji^d(ω

) having frequency and amplitude information for each of the NG measurement locations;

d) normalizing the measurement spectral data set D_ji^d(ω

) to create a normalized data wavefield T_ji^d(ω

);

e) modeling the medium using an iterated model parameter m describing one or more physical properties of the medium.

the NS source and the NG measurement spatial locations used as respective model input and model response spatial locations contained within the model of the medium,

and initializing the iterated model parameter m in with corresponding one or more known bulk properties of the medium being modeled,

 

said modeling step comprising;

i) creating a measurement model by;

(1) applying a delta function source collocated with the i^th source,

(2) modeling the response at the NG measurement locations, using a frequency domain modeling method, to create a frequency domain synthetic response at the j^th receiver due to the i^th source, P_ji^m(ω

), m denoting the model response,

(3) repeating the applying and modeling steps at each of the NS source locations and NG measurement locations until the frequency domain synthetic response is full, and

ii) forming a normalized modeled wavefield using the frequency domain synthetic response T_ji^m(ω

)=P_ji^m(ω

)[P_1i^m(ω

)]<sup>−

1</sup>; and

f) minimizing a weighted error, between the normalized data wavefield T_ji^d(ω

) and the normalized modeled wavefield T_ji^m(ω

) of the response of the medium, to a level below an error bound,

i) said weighted error met by using the iterated model parameter m, known as the a minimized model parameter m.