표제지
목차
ABSTRACT 12
제1장 서론 14
제2장 이론적배경 17
제1절 머신러닝(Machine Learning) 17
1. 앙상블(Ensemble) 18
2. 랜덤 포레스트(random forest) 19
3. One-Class SVM 21
4. Support Vector Data Description 22
제3장 음파 검층 로그 예측을 위한 머신러닝 모델 구축 연구 24
제1절 대상 데이터 24
제2절 데이터 전처리(Data preprocessing) 26
1. 상관관계 분석을 통한 입력 변수 선정 26
2. 이상치 제거 (outlier removing) 29
제3절 랜덤 포레스트 기법을 이용한 예측 모델 생성 33
1. 15/9-F-1A 예측 모델 결과 34
2. 15/9-F-1B 예측 모델 결과 37
3. 15/9-F-11A 예측 모델 결과 40
제4절 SVDD 기법을 이용한 예측 결과 분석 45
1. 15/9-F-1A 예측 모델 결과 46
2. 15/9-F-11A 예측 모델 결과 66
3. 15/9-F-1B 예측 모델 결과 85
제4장 미측정된 음파 검층 예측 105
1. 15/9-F-1C 예측 결과 105
2. 15/9-F-11B 예측 결과 109
제5장 결론 113
참고문헌 115
Table 3.1. Common log type list 25
Table 3.2. Input and output variable for model construction 28
Table 3.3. Correlation coefficient of 15/9-F-11A and 15/9-F-1A after removing outlier 32
Table 3.4. Result of prediction about total models 43
Table 3.5. Comparison result of RF_1B_1A_model for g value change 53
Table 3.6. Relative difference result of RF_11A_1A_model for g value change 59
Table 3.7. Relative difference result of RF_1B+11A_1A_model for g value change 64
Table 3.8. Comparison result of RF_1B_11A_model for g value change 72
Table 3.9. Relative difference result of RF_1A_11A_model for g value change 78
Table 3.10. Relative difference result of RF_1A+1B_11A_model for g value change 83
Table 3.11. Relative difference result of RF_1A+11A_1B_model for g value change 92
Table 3.12. Relative difference result of RF_1A_1B_model for g value change 97
Table 3.13. Relative difference result of RF_11A_1B_model for g value change 102
Table 3.14. Relative error of total models for g value change 104
Figure 2.1. Example scheme of Random Forest. 20
Figure 2.2. Data description trained on a banana shaped data set. 23
Figure 3.1. Location of Volve field in the North Sea. 24
Figure 3.2. Feature correlation coefficient using Pearson Correlation. 27
Figure 3.3. Cross plot between 15/9-F-11A and 15/9-F-1A. 30
Figure 3.4. Cross plot between 15/9-F-11A and 15/9-F-1A after removing outlier. 31
Figure 3.5. (a) Training, (b) validation using 15/9-F-1B data and (c)... 34
Figure 3.6. (a) Training, (b) validation using 15/9-F-11A data and (c) prediction... 35
Figure 3.7. (a) Training, (b) validation using 15/9-F-1B+11A data and (c)... 36
Figure 3.8. (a) Training, (b) validation using 15/9-F-1A data and (c) prediction... 37
Figure 3.9. (a) Training, (b) validation using 15/9-F-11A data and (c) prediction... 38
Figure 3.10. (a) Training, (b) validation using 15/9-F-1A+11A data and (c)... 39
Figure 3.11. (a) Training, (b) validation using 15/9-F-1A data and (c) prediction... 40
Figure 3.12. (a) Training, (b) validation using 15/9-F-1B data and (c) prediction... 41
Figure 3.13. (a) Training, (b) validation using 15/9-F-1A+1B data and (c)... 42
Figure 3.14. Prediction result of RF_1B_1A_inBND and RF_1B_1A_outBND model... 47
Figure 3.15. Result of PCA of RF_1B_1A_model, inBND_model and... 48
Figure 3.16. Prediction result of RF_1B_1A_inBND and RF_1B_1A_outBND model... 49
Figure 3.17. Prediction result of RF_1B_1A_inBND and RF_1B_1A_outBND model... 50
Figure 3.18. Comparison result of RF_1B_1A_model for g value change 51
Figure 3.19. Histogram of RF_1B_1A_inBND_model error and... 52
Figure 3.20. Well log of input data and predicted DT of RF_1B_1A_model for g value change. 55
Figure 3.21. Prediction result of 15/9-F-1A using 15/9-F-11A training model... 56
Figure 3.22. Prediction result of 15/9-F-1A using 15/9-F-11A training model for g=5 57
Figure 3.23. Comparison result of RF_11A_1A_model for g value change. 58
Figure 3.24. Well log of input data and predicted DT of RF_11A_1A_model for g value change. 60
Figure 3.25. Prediction result of 15/9-F-1A using 15/9-F-1B and 15/9-F-11A... 61
Figure 3.26. Prediction result of 15/9-F-1A using 15/9-F-1B and 15/9-F-11A... 62
Figure 3.27. Comparison result of RF_1B+11A_1A_model for g value change. 63
Figure 3.28. Well log of input data and predicted DT of RF_1B+11A_1A_model for g value change. 65
Figure 3.29. Prediction result of RF_1B_11A_inBND and RF_1B_11A_outBND... 66
Figure 3.30. Result of PCA of RF_1B_11A_model, inBND_model and... 67
Figure 3.31. Prediction result of RF_1B_11A_inBND and RF_1B_11A_outBND... 68
Figure 3.32. Prediction result of RF_1B_11A_inBND and RF_1B_11A_outBND... 69
Figure 3.33. Comparison result of RF_1B_11A_model for g value change. 70
Figure 3.34. Histogram of RF_1B_11A_inBND_model error and... 71
Figure 3.35. Well log of input data and predicted DT of RF_1B_11A_model for g value change 74
Figure 3.36. Prediction result of 15/9-F-11A using 15/9-F-1A training model for g=1 75
Figure 3.37. Prediction result of 15/9-F-11A using 15/9-F-1A training model for g=5 76
Figure 3.38. Comparison result of RF_1A_11A_model for g value change. 77
Figure 3.39. Well log of input data and predicted DT of RF_1A_11A_model for g value change 79
Figure 3.40. Prediction result of 15/9-F-11A using 15/9-F-1A and 15/9-F-1B... 80
Figure 3.41. Prediction result of 15/9-F-11A using 15/9-F-1A and 15/9-F-1B... 81
Figure 3.42. Comparison result of RF_1A+1B_11A_model for g value change. 82
Figure 3.43. Well log of input data and predicted DT of RF_1A+1B_11A_model for g value change 84
Figure 3.44. Prediction result of 15/9-F-1B using 15/9-F-1A and 15/9-F-11A... 86
Figure 3.45. Result of PCA of RF_1B_11A_model, inBND_model and... 87
Figure 3.46. Prediction result of 15/9-F-1B using 15/9-F-1A and 15/9-F-11A... 88
Figure 3.47. Prediction result of 15/9-F-1B using 15/9-F-1A and 15/9-F-11A... 89
Figure 3.48. Comparison result of RF_1A+11A_1B_model for g value change. 90
Figure 3.49. histogram of RF_1A+11A_1B_inBND_model error and... 91
Figure 3.50. Well log of input data and predicted DT of RF_1A+11A_1B_model for g value change 93
Figure 3.51. Prediction result of 15/9-F-1B using 15/9-F-1A training model for g=1 94
Figure 3.52. Prediction result of 15/9-F-1B using 15/9-F-1A training model for g=5 95
Figure 3.53. Comparison result of RF_1A_1B_model for g value change 96
Figure 3.54. Well log of input data and predicted DT of RF_1A_1B_model for g value change 98
Figure 3.55. Prediction result of 15/9-F-1B using 15/9-F-11A training model for g=1 99
Figure 3.56. Prediction result of 15/9-F-1B using 15/9-F-11A training model for g=5 100
Figure 3.57. Comparison result of RF_11A_1B_model for g value change. 101
Figure 3.58. Well log of input data and predicted DT of RF_11A_1B_model for g value change 103
Figure 4.1. Training and validation result of RF_1A+1B+11A_1C_model. 105
Figure 4.2. Tsne result of RF_1A+1B+11A_1C_model for g value change 106
Figure 4.3. Well log of input data and predicted DT of RF_1A+1B+11A_1C_model for g value change. 108
Figure 4.4. Training and validation result of RF_1A+1B+11A_11B_model 109
Figure 4.5. Tsne result of RF_1A+1B+11A_11B_model for g value change 110
Figure 4.6. Well log of input data and predicted DT of RF_1A+1B+11A_11B_model for g value change 112