Neural network training data selection using memory reduced cluster analysis for field model development

• US 20040133531A1
• Filed: 03/21/2003
• Published: 07/08/2004
• Est. Priority Date: 01/06/2003
• Status: Active Grant

• Current Assignee: Halliburton Energy Services Incorporated (Halliburton Company)
• Original Assignee: Halliburton Energy Services Incorporated (Halliburton Company)
• Inventors: Wiener, Jacky M., Hamid, Syed, Quirein, John A., Chen, Dingding, Smith, Harry D. Jr., Grable, Jeffery L.
• Granted Patent: US 8,374,974 B2
• Time in Patent Office: Days
• US Class Current: 706/8
• CPC Class Codes: G01V 11/00 Prospecting or detecting by methods combining techniques covered by two or more of main groups G01V1/00 - G01V9/00
• CPC Class Codes: G06F 18/10 Pre-processing; Data cleansing
• CPC Class Codes: G06N 3/08 Learning methods

Abstract

A system and method for selecting a training data set from a set of multidimensional geophysical input data samples for training a model to predict target data. The input data may be data sets produced by a pulsed neutron logging tool at multiple depth points in a cases well. Target data may be responses of an open hole logging tool. The input data is divided into clusters. Actual target data from the training well is linked to the clusters. The linked clusters are analyzed for variance, etc. and fuzzy inference is used to select a portion of each cluster to include in a training set. The reduced set is used to train a model, such as an artificial neural network. The trained model may then be used to produce synthetic open hole logs in response to inputs of cased hole log data.

Claims

1. A method for producing a training data set from a set of multidimensional geophysical input data samples for training a model to predict target data, comprising:

dividing a set of geophysical input data samples into a plurality of first subsets of input data samples,

dividing each of the first subsets into a plurality of first clusters,

generating a first set of prototypes each representing one of the first clusters, and

dividing the first set of prototypes into a plurality of second clusters.

2. A method according to claim 1, further comprising generating a second set of prototypes each representing one of the second clusters.

3. A method according to claim 1, further comprising assigning each of the geophysical input data samples to one of the second clusters.

4. A method according to claim 3, further comprising linking actual target data to each of the geophysical data samples.

5. A method according to claim 4, further comprising performing a statistical analysis of data samples and target data within each cluster.

6. A method according to claim 5, further comprising processing the results of the statistical analysis with fuzzy inference to assign a percentage to each of the second clusters.

7. A method according to claim 6, further comprising selecting the assigned percentage of data samples and target data from each cluster to form a training data set.

8. A method according to claim 4, further comprising generating a plot of the data samples and target data for each cluster.

9. A method according to claim 8, further comprising visually inspecting each plot.

10. A method according to claim 9, further comprising selecting data samples and target data to be included in or excluded from the training set.

11. A method according to claim 5, wherein the statistical analysis includes determining one or more of cluster size, data dispersion ratio, and zone to which the data samples relate.

12. A method according to claim 1, wherein the geophysical data comprises output readings from a pulsed neutron logging tool in at least one cased borehole.

13. A method according to claim 1, wherein the target data comprises logging measurements taken in an open borehole.

14. A method according to claim 13, wherein the target data comprises measurements representing one or more of neutron porosity, formation density and deep resistivity.

15. A method according to claim 1, wherein the model is an artificial neural network adapted to predict target data in response to geophysical input data samples.

16. A method for producing a training data set from a set of multidimensional geophysical input data samples for training a model to predict target data, comprising:

dividing multidimensional geophysical input data samples into a set of clusters, linking each multidimensional geophysical input data sample with corresponding target data, and

performing an analysis of the input samples and target data in each cluster.

17. A method according to claim 16, further comprising selecting a percentage of data samples and corresponding target data from each cluster based on results of the analysis.

18. A method according to claim 17, further comprising combining the data selected from each cluster to form a training data set.

19. A method according to claim 16, further comprising generating a plot of the data samples and target data for each cluster.

20. A method according to claim 19, wherein the step of performing an analysis comprises visually inspecting each plot.

21. A method according to claim 16, wherein the step of performing an analysis comprises calculating the dispersion ratio of the data samples and target data for each cluster.

22. A method according to claim 21, further comprising processing the results of the analysis with fuzzy inference to assign a percentage to each of the second clusters.

23. A method according to claim 16, wherein the input data comprises output readings from a pulsed neutron logging tool.

24. A method according to claim 23, wherein the output readings are taken in a cased borehole.

25. A method according to claim 16, wherein the target data comprises logging measurements taken in an open borehole.

26. A method according to claim 16, wherein the target data comprises measurements representing one or more of neutron porosity, formation density and deep resistivity.

27. A method for predicting open borehole logging measurements from actual cased borehole logging measurements, comprising:

collecting open hole logging measurements in a borehole,

collecting cased borehole logging measurements in the borehole,

dividing the cased borehole logging measurements into a set of clusters,

linking each cased borehole logging measurement with corresponding open hole logging measurements,

performing an analysis of the cased borehole logging measurements and corresponding open hole logging measurements for each cluster,

selecting a percentage of the cased borehole logging measurements and corresponding open hole logging measurements from each cluster based on results of the analyses,

training a predictive model with the selected measurements, and

using the trained predictive model to predict open hole logging measurements in response to cased borehole logging measurements.

28. A method according to claim 27, wherein the step of performing an analysis comprises:

plotting the cased borehole logging measurements and corresponding open hole logging measurements for each cluster,

visually inspecting each plot, and

selecting data from each cluster based on the visual inspection.

29. A method according to claim 27, wherein the step of performing an analysis comprises performing a statistical analysis of the cased borehole logging measurements and corresponding open hole logging measurements within each cluster.

30. A method according to claim 29, further comprising processing the results of the statistical analysis with fuzzy inference to assign a percentage to each of the second clusters.

31. A method according to claim 27, wherein the step of dividing the cased borehole logging measurements into a set of clusters comprises:

dividing the cased borehole logging measurements into a plurality of first subsets,

dividing each of the first subsets into a plurality of first clusters,

generating a first set of prototypes each representing one of the first clusters, and

dividing the first set of prototypes into a plurality of second clusters.

32. A method according to claim 31, further comprising assigning each of the cased borehole logging measurements to one of the second clusters.

33. A method according to claim 27, wherein the predictive model is an artificial neural network.

34. A method according to claim 31, wherein the cased borehole logging measurements are outputs of a pulsed neutron logging tool.

35. A method according to claim 27, wherein the open borehole logging measurements comprise measurements representing one or more of neutron porosity, formation density and deep resistivity.

36. A method for predicting open borehole geophysical measurements from actual cased borehole geophysical measurements, comprising:

collecting open hole geophysical measurements in a borehole,

collecting cased borehole geophysical measurements in the borehole,

selecting a percentage of the cased borehole measurements and corresponding open hole measurements as a training data set,

training a predictive model with the selected measurements, and

using the trained predictive model to predict open hole geophysical measurements in response to cased borehole geophysical measurements.

37. A method for predicting cased borehole geophysical measurements from actual open borehole geophysical measurements, comprising:

collecting open hole geophysical measurements in a borehole,

collecting cased borehole geophysical measurements in the borehole,

selecting a percentage of the open hole measurements and corresponding cased borehole measurements as a training data set,

training a predictive model with the selected measurements, and

using the trained predictive model to predict cased hole geophysical measurements in response to open borehole geophysical measurements.

38. A method for producing a synthetic log of at least one geophysical parameter for a well, comprising:

collecting a first log of a plurality of geophysical parameters, including the at least one geophysical parameter, in a first well, the log comprising a plurality of multidimensional data samples,

dividing the data samples into a set of clusters based on the geophysical parameters other than the at least one geophysical parameter,

selecting data from each cluster,

training a predictive model with the selected data,

collecting a second log of the plurality of geophysical parameters, excluding the at least one geophysical parameter, in a second well, and

inputting the second log to the predictive model to produce a synthetic log of the at least one geophysical parameter for the second well.

39. A method according to claim 38, further comprising analyzing the data in each cluster.

40. A method according to claim 39, further comprising performing an analysis on data in each cluster.

41. A method according to claim 39, further comprising:

plotting the data in each cluster, and

visually inspecting the data plots.

42. A method according to claim 41, further comprising identifying formation type represented by a cluster.

43. A method for producing a synthetic value of at least one geophysical parameter for a well, comprising:

collecting a first data sample set of a plurality of geophysical parameters, including the at least one geophysical parameter, relating to a first well,

dividing the first data sample set into a set of clusters based on the geophysical parameters other than the at least one geophysical parameter,

selecting data from each cluster,

training a predictive model with the selected data,

collecting a second data sample set of the plurality of geophysical parameters, excluding the at least one geophysical parameter, relating to a second well, and

inputting the second data sample set to the predictive model to produce a synthetic value of the at least one geophysical parameter for the second well.

44. A method according to claim 43, wherein the first data sample set comprises geophysical parameters normally measured by open hole logging and by cased hole logging of the well.

45. A method according to claim 43, wherein the first data sample set comprises seismic data.

46. A method according to claim 43, wherein the first data sample set comprises sidewall core data.

47. A method according to claim 43, wherein the first data sample set comprises data collected by carbon oxygen logging.

48. A method according to claim 43, further comprising:

using the clusters to identify geological facies types, and

training separate predictive models for each facies type.

49. A method of operating a hydrocarbon bearing field, comprising:

drilling a plurality of wells in the hydrocarbon bearing field,

performing open hole logging in a subset of the wells,

performing cased hole logging in substantially all of the wells including the subset of wells,

using open hole logging data and cased hole logging data from the subset of wells to train a predictive model to produce synthetic open hole data in response to inputs of cased hole data, and

using the trained predictive model and cased hole data from the wells to produce synthetic open hole data.

50. A method according to claim 49, further comprising using the synthetic open hole data to plan operations for the wells.

51. A method according to claim 49, wherein the subset of wells comprises less than one-half of the plurality of wells.

52. A method according to claim 49, wherein the subset of wells comprises less than one-fifth of the plurality of wells.

53. Apparatus for producing synthetic values of at least one geophysical parameter for a well, comprising a predictive model trained by:

collecting a first data sample set of a plurality of geophysical parameters, including the at least one geophysical parameter, relating to a first well,

dividing the first data sample set into a set of clusters based on the geophysical parameters other than the at least one geophysical parameter,

selecting data from each cluster, and

training the predictive model to produce a synthetic value of the at least one geophysical parameter in response to inputs of the plurality of geophysical parameters, excluding the at least one geophysical parameter.

54. Apparatus according to claim 53, wherein the predictive model comprises an artificial neural network.

55. Apparatus according to claim 53, wherein the predictive model comprises computer code.

56. A method for producing synthetic geophysical measurements in a well having at least one depth interval in which one or more actual measurements cannot be accurately taken, comprising:

collecting at least one log of a plurality of geophysical measurements in a borehole, the at least one log having missing or defective measurements of at least one parameter in at least one depth interval,

selecting a training data set comprising at least a portion of the plurality of geophysical measurements from depth intervals other than the at least one depth interval,

training a predictive model with the training data set, and using the trained predictive model to produce synthetic values of the missing or defective measurements of the at least one parameter in response to inputs comprising at least a portion of the geophysical measurements taken in the at least one depth interval.

57. A method according to claim 56, wherein the at least one log comprises an open hole log and the at least one parameter is measured by the open hole log.

58. A method according to claim 57, wherein the at least one log comprises a cased hole log.

59. A method according to claim 58, wherein the training data set comprises parameters measured by both the open hole log and the cased hole log.