Methods for downhole waveform tracking and sonic labeling

• US 6,625,541 B1
• Filed: 06/12/2000
• Issued: 09/23/2003
• Est. Priority Date: 06/12/2000
• Status: Expired due to Term

• Current Assignee: Schlumberger Technology Corporation (SLB)
• Original Assignee: Schlumberger Technology Corporation (SLB)
• Inventors: Shenoy, Ramachandra Ganesh, Valero, Henri-Pierre
• Primary Examiner(s): McElheny, Jr., Donald E.
• Application Number:
  US09/591,405
• Time in Patent Office: 1,198 Days
• Field of Search: 702/11, 702/12, 702/13, 702/6, 702/14, 702/17, 703/6, 703/10, 367/73, 367/69
• US Class Current: 702/6
• CPC Class Codes: G01V 1/288 Event detection in seismic signals, e.g. microseismics G01V1/36 takes precedence
• CPC Class Codes: G01V 1/48 Processing data

Abstract

Methods for downhole waveform tracking and sonic labeling employ “tracking algorithms” and Bayesian analysis to classify STC waveforms. More particularly, according to the tracking part of the invention, a probability model is built to distinguish true “arrivals” (e.g. compressional, shear, etc.) from “false alarms” (e.g. noise) before the arrivals are classified. The probability model maps the “continuity” of tracks (slowness/time over depth) and is used to determine whether sequences of measurements are sufficiently “continuous” to be classified as tracks. The probability model is used to evaluate the likelihood of the data in various possible classifications (hypotheses). Prior and posterior probabilities are constructed for each track using the tracking algorithm. The posterior probabilities are computed sequentially and recursively, updating the probabilities after each measurement frame at depth k is acquired. The classification part of the invention determines the hypotheses with the maximum “posterior probability” according to Bayes'"'"' Theorem.

Claims

1. A method for downhole waveform tracking and sonic labeling (classification), said method comprising:

a) logging waveform arrival data at a plurality of depths k;

b) mapping the continuity of the data tracks as a function of slowness or time over depth;

c) building a probability model of the likely classification of the tracks based on said mapping;

d) evaluating the probability model with Bayesian analysis to determine the most likely classifications of the tracks.

2. A method according to claim 1, wherein:

said step of logging waveform arrival data includes recording sonic measurements C_k at each depth and recording the number N_k of recorded measurements at each depth.

3. A method according to claim 2, wherein:

said step of building a probability model includes building a model based on C_k and building a separate model based on N_k.

4. A method according to claim 3, wherein:

said step of building a model based on N_k includes, for each depth,

i) ranking tracks in increasing order of slowness,

ii) creating a first matrix of tracks x sonic measurements, and

iii) assigning a probability to each cell in the first matrix.

5. A method according to claim 4, wherein:

said step of assigning a probability to each cell in the first matrix is based on the total number N_k of recorded measurements, the number 0 of those measurements which appear to be a part of observed tracks and the number N_k−

0 of false alarms.

6. A method according to claim 5, wherein:

the number N_k−

0 of false alarms is modeled as a Poisson distribution with a known mean.

7. A method according to claim 4, wherein:

said step of building a model based on N_k includes, for each depth after a first depth,

i) creating a second matrix of previous depth tracks x current depth tracks, and

ii) assigning a probability to each cell in the second matrix.

8. A method according to claim 7, wherein:

said step of assigning a probability to each cell in the second matrix is based on classifications, track assignments, and the probabilities in the first matrix.

9. A method according to claim 7, wherein:

said step of building a model based on N_k includes building a single model of hypotheses and data likelihood based on the first and second matrices.

10. A method according to claim 3, wherein:

said step of building a model based on C_k includes building a model based on the coordinates of each slowness-time peak derived from the waveforms.

11. A method according to claim 10, wherein:

said step of building a model based on C_k includes filtering slowness-time coordinates through a Kalman filter.

12. A method according to claim 11, wherein:

said step of building a model based on C_k includes correlating slowness across depth as the output of an ARMA filter with known coefficients driven by white Gaussian noise.

13. A method according to claim 11, wherein:

said step of building a model based on C_k includes modeling each slowness-time pair as a two-dimensional vector.

14. A method according to claim 13, wherein:

said step of filtering slowness-time coordinates through a Kalman filter includes filtering the slownesses through a one-dimensional Kalman filter and filtering the times through a separate one-dimensional Kalman filter.

15. A method according to claim 1, wherein:

said step of evaluating the probability model with Bayesian analysis uses only the data in the current depth for classification to compute posterior probability.

16. A method according to claim 1, wherein:

said step of evaluating the probability model with Bayesian analysis incorporates information provided by all data up to and including the current depth.

17. A method according to claim 1, wherein:

said step of evaluating the probability model with Bayesian analysis incorporates information from a fixed number of prior depths and a fixed number of subsequent depths relative to the current depth.

18. A method according to claim 1, wherein:

said step of evaluating the probability model with Bayesian analysis assumes that the probabilities of classifications at depth k, given the probabilities of classifications at depth k−

1 and data from depths 1 through k−

1, depends only on the probabilities of classifications at depth k−

1.

19. A method according to claim 1, wherein:

said step of evaluating the probability model with Bayesian analysis assumes that the data likelihood at the current depth depends on the previous data only through the probable classifications at the current depth.

20. A method for downhole waveform tracking, said method comprising:

a) logging waveform arrival data at a plurality of depths k; and

b) mapping the continuity of the data as a function of depth utilizing a probability model of the likelihood that data at one depth represents the same waveform arrival as data at other depths.

21. A method according to claim 20, wherein:

said step of logging waveform arrival data includes recording measurements C_k at each depth and recording the number N_k of recorded measurements at each depth.

22. A method according to claim 21, wherein:

said step of utilizing a probability model includes building a model based on C_k and a separate model based on N_k.

23. A method according to claim 22, wherein:

the waveforms are sonic waveforms, and said step of building a model based on N_k includes, for each depth,

i) ranking tracks in increasing order of slowness,

ii) creating a first matrix of tracks x sonic measurements, and

iii) assigning a probability to each cell in the first matrix.

24. A method according to claim 23, wherein:

said step of assigning a probability to each cell in the first matrix is based on the total number N_k of recorded measurements, the number 0 of those measurements which appear to be a part of observed tracks and the number N_k−

0 of false alarms.

25. A method according to claim 24, wherein:

the number N_k−

0 of false alarms is modeled as a Poisson distribution with a known mean.

26. A method according to claim 23, wherein:

said step of building a model based on N_k includes, for each depth after a first depth,

i) creating a second matrix of previous depth tracks x current depth tracks, and

ii) assigning a probability to each cell in the second matrix.

27. A method according to claim 26, wherein:

said step of assigning a probability to each cell in the second matrix is based on track assignments and the probabilities in the first matrix.

28. A method according to claim 26, wherein:

said step of building a model based on N_k includes building a single model of hypotheses and data likelihood based on the first and second matrices.

29. A method according to claim 22, wherein:

the waveforms are sonic waveforms; and

said step of building a model based on C_k includes building a model based on the coordinates of each slowness-time peak derived from the waveforms.

30. A method according to claim 29, wherein:

said step of building a model based on C_k includes filtering slowness-time coordinates through a Kalman filter.

31. A method according to claim 30, wherein:

said step of building a model based on C_k includes correlating slowness across depth as the output of an ARMA filter with known coefficients driven by white Gaussian noise.

32. A method according to claim 30, wherein:

said step of building a model based on C_k includes modeling each slowness-time pair as a two-dimensional vector.

33. A method according to claim 30, wherein:

said step of filtering slowness-time coordinates through a Kalman filter includes filtering the slownesses through a one-dimensional Kalman filter and filtering the times through a separate one-dimensional Kalman filter.