Upload
dung-tien-hua
View
215
Download
0
Embed Size (px)
Citation preview
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 1/132
Developing Intelligent Synthetic Logs:
Application to Upper Devonian Units in PA
Luisa F. Rolon
Thesis submitted to the
College of Engineering and Mineral Resources
at West Virginia University
in partial fulfillment of the requirements
for the degree of
Master of Science in
Petroleum & Natural Gas Engineering
Committee:
Professor Samuel Ameri, Chair
Dr. Shahab Mohaghegh
Dr. Razi Gaskari
Dr. Daniel Della-Giustina
Department of Petroleum and Natural Gas Engineering
Morgantown, West Virginia
2004
Keywords: Reservoir, Well logs, Upper Devonian, Pennsylvania, Artificial Neural
Network
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 2/132
ABSTRACT
Developing Intelligent Synthetic Logs: Application to Upper Devonian Units in PA
Luisa F. Rolon
A methodology to generate synthetic wireline logs is presented. Synthetic logs
can help to analyze the reservoir properties in areas where the set of logs that arenecessary, are absent or incomplete. The approach presented involves the use of Artificial
Neural Networks, as the main tool, in conjunction with data obtained from conventional
wireline logs. Implementation of this approach aims to reduce operation costs tocompanies.
Development of the neural network model was completed using a General
Regression Neural Network, and four wells that included gamma ray, density, neutron,and resistivity logs. Synthetic logs were generated through two different exercises.
Exercise one involved four wells for development and training of the network.Subsequently verification was carried out using each of the wells that were used to train
the network. The second exercise used three wells for training and development of thenetwork. A fourth well that was not used during training and calibration, was selected for
verification. Three combinations of inputs/outputs were chosen to train the network. In
combination “A” the resistivity log was the output and density, gamma ray, and neutronlogs, and the coordinates and depths (XYZ) the inputs. In combination “B” the density
log was output and the resistivity, the gamma ray, and the neutron logs, and XYZ were
the inputs, and in combination “C” the neutron log was the output and the resistivity, thegamma ray, and the density logs, and XYZ were the inputs.
After development of the neural network model, synthetic logs with a reasonabledegree of accuracy were generated. Results indicate that the best performance was
obtained for combination “A” of inputs and outputs, then for combination “C”, and
finally for combination “B”. In addition, it was determined that accuracy of syntheticlogs is favored by interpolation of data.
As an important conclusion, it was demonstrated that quality of the data plays a
very important role in developing a neural network model. A recommendation for future
works is to do a very careful quality control of the data before a neural network model is built. Conversely it was concluded that lithologic heterogeneities in the reservoir do not
affect performance of a neural network model in generation of synthetic logs.
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 3/132
Acknowledgments
First I thank God for giving me the capability and the courage to finish my thesis and
complete my MS in Petroleum and Natural Gas Engineering.
I don’t have the words to express my thanks and appreciation to Dr. Sam Ameri, chair of
my committee, who brought me to the Petroleum and Natural Gas Engineering
Department and encouraged me to complete my master's degree. I wouldn’t be ableaccomplish this without his support and advise, not only during this work but also
throughout the time I spent at the department.
I want to extend my sincere appreciation and gratitude to my research advisor Dr. Shahab
Mohaghegh, for introducing me to the fascinating area of Neural Networks, for his
friendship, and for his continuous guidance, encouragement, support and patiencethroughout this work.
Special thanks to Dr. Razi Gaskari and Dr. Daniel Della-Guistina for being in my
committee and for the enriching contributions and comments to this work.
My deepest gratitude to Mr. Richard Goings and Dominion Exploration and Production,
Inc. for providing me the data I used for this study. Without this valuable information thiswork couldn't have be completed.
Thanks to my colleagues and friends Nikola Maricic, Emre Artun, Jalal Jalali, BoydHuls, Miguel Tovar and Clay Blaylock, for keeping my morale high and for helping me
when I needed them. Thanks a lot guys, you are the best!
Special thanks to Beverly Matheny for her assistance and enthusiasm during every
semester spent in the department.
Finally, I want to dedicate this thesis to my parents Silvio and Dilma and my husband
Bret, who always gave me their support and encouragement, and most importantly, their
love. It kept me going during some of the difficult moments of this work. Thank you somuch, I love you.
iii
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 4/132
TABLE OF CONTENTS
1. INTRODUCTION ……………………………………………………………………..1
2. BACKGROUND ……………………………………………………………………5
2.1. Geological Setting ……………………………………………………………………..5
2.1.1. Description of the Murrysville & 100 Foot Sands ……………………………62.1.2. Description of the Gordon Sandstone ………………………………………...7
2.1.3. Description of the 2nd
Bradford Sand and the Speechley Sands ……………...7
2.1.4. Structural Geology ……………………………………………………………8
2.2. Well logs fundamentals ……………………………………………………………92.2.1. Gamma-ray log ……………………………………………………………….9
2.2.2. Porosity logs ………………………………………………………………...12
2.2.3. Spontaneous potential (SP) log ……………………………………………..142.2.4. Resistivity log ……………………………………………………………….16
2.2.5. Sonic log …………………………………………………………………….16
2.3. Fundamentals on Artificial Neural Networks ……………………………………172.3.1.Mechanics of neural networks ……………………………………………….192.3.2. Learning mechanisms for Artificial Neural Networks ………………………21
2.3.3. Training the network ………………………………………………………...24
2.3.4. General Regression Neural Network ………………………………………..25
3. LITERATURE REVIEW …………………………………………………………….274. METHODOLOGY …………………………………………………………………...28
4.1. Data Preparation …………………………………………………………………30
4.2. Neural network model development ……………………………………………..31
4.2.1. Exercise 1: Four Wells Combined …………………………………………..31
4.2.2. Exercise 2: Three Wells Combined, one well out …………………………...335. RESULTS …………………………………………………………………………….34
5.1. First Attempt: Data set from Buffalo Valley Field (New Mexico) ………………36
5.2. Southern Pennsylvania Logs: Upper Zone (1000’ to 2000’) …………………….485.2.1. Exercise 1 ……………………………………………………………………48
5.2.2. Exercise 2 ……………………………………………………………………49
5.3. Southern Pennsylvania Logs: Lower Zone (2500’ to 3500’) …………………….50
5.3.1. Exercise 1 ……………………………………………………………………505.3.2. Exercise 2 ……………………………………………………………………50
6. DISCUSSION ……………………………………………………………………….101
7. CONCLUSIONS ……………………………………………………………………106
REFERENCES ………………………………………………………………………...108
APPENDIX A: Calibration results (Southern Pennsylvania data set) ………………...110
APPENDIX B: Training results (Southern Pennsylvania data set) ……………………111
APPENDIX C: Neural network model results (Buffalo Valley Field data set) ………..112
iv
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 5/132
LIST OF FIGURES
Page
Figure 1. Location of the study area and wells analyzed (157, 168, 169,
174). A-A’ represents the line of the cross-section.
2
Figure 2. Gamma Ray correlation of 2nd Bradford (distributary channel –deep).
3
Figure 3. Gamma Ray correlation of Murrysville / 100 Foot Sands
(Braided Stream – Shallow).
4
Figure 4. Cross-section of the Acadian clastic wedge, showing the
relationship of the Price-Rockwell and Catskill delta complexes.
8
Figure 5. Structural map of the area, showing Structural Axes inSouthwestern PA.
8
Figure 6. Gamma ray and sonic logs from the Alberta basin, and their
response to different lithologies.
11
Figure 7. Gamma ray emission spectra of K-40, uranium, and thorium
series.
11
Figure 8. Relationship between gamma-ray deflection and proportion of
shale.
15
Figure 9. Example of SP and resistivity logs from the Alberta basin. 15
Figure 10. Sketch of a biological neuron. 17
Figure 11. Structure of a simple artificial neuron. 18
Figure 12. Architecture of a multilayer perceptron. 22
Figure 13. Supervised and Unsupervised scheme. 23
Figure 14. Training and verification curve. 25
Figure 15. Segment of the matrix prepared for well 157. 30
Figure 16. Cartoon to show distribution of wells used for training /testing
and Production dataset through exercise 1.
32
Figure 17. Different combinations of inputs/ outputs used fordevelopment of neural network model.
33
Figure 18. Combinations of wells for training, testing and verification
used in exercise 2.
35
Figure 19. Location of the Buffalo Valley Field area and the wells used in
the first attempt.
37
Figure 20. .tif files and the resulting Gamma Ray curve afterdigitalization. Well 219 Buffalo Valley Field
38
Figure 21. .tif files and the resulting Density curve after digitalization.
Well 219 Buffalo Valley Field
39
Figure 22. .tif files and the resulting Resistivity curve after digitalization.Well 219 Buffalo Valley Field
40
Figure 23. .tif files and the resulting Neutron Porosity curve after
digitalization. Well 219 Buffalo Valley Field
41
Figure 24. .tif files and the resulting Gamma Ray curve after
digitalization. Well 665 Buffalo Valley Field
42
v
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 6/132
Page
Figure 25. .tif files and the resulting Density curve after digitalization.
Well 665 Buffalo Valley Field
43
Figure 26. .tif files and the resulting Resistivity curve after digitalization.
Well 665 Buffalo Valley Field
44
Figure 27. .tif files and the resulting Neutron Porosity curve after
digitalization. Well 665 Buffalo Valley Field
45
Figure 28. Synthetic resistivity log Well 157, southern Pennsylvania area(lower zone). Exercise 1 and combination A of inputs and outputs.
53
Figure 29. Synthetic resistivity log Well 168, southern Pennsylvania area
(lower zone). Exercise 1 and combination A of inputs and outputs.
54
Figure 30. Synthetic resistivity log Well 169, southern Pennsylvania area
(lower zone). Exercise 1 and combination A of inputs and outputs.
55
Figure 31. Synthetic resistivity log Well 174, southern Pennsylvania area(lower zone). Exercise 1 and combination A of inputs and outputs.
56
Figure 32. Synthetic density log Well 157, southern Pennsylvania area(lower zone). Exercise 1 and combination B of inputs and outputs. 57
Figure 33. Synthetic density log Well 168, southern Pennsylvania area(lower zone). Exercise 1 and combination B of inputs and outputs.
58
Figure 34. Synthetic density log Well 169, southern Pennsylvania area
(lower zone). Exercise 1 and combination B of inputs and outputs.
59
Figure 35. Synthetic density log Well 174, southern Pennsylvania area
(lower zone). Exercise 1 and combination B of inputs and outputs.
60
Figure 36. Synthetic neutron log Well 157, southern Pennsylvania area(lower zone). Exercise 1 and combination C of inputs and outputs.
61
Figure 37. Synthetic neutron log Well 168, southern Pennsylvania area(lower zone). Exercise 1 and combination C of inputs and outputs.
62
Figure 38. Synthetic neutron log Well 169, southern Pennsylvania area
(lower zone). Exercise 1 and combination C of inputs and outputs.
63
Figure 39. Synthetic neutron log Well 174, southern Pennsylvania area
(lower zone). Exercise 1 and combination C of inputs and outputs.
64
Figure 40. Synthetic resistivity log Well 157, southern Pennsylvania area
(lower zone). Exercise 2 and combination A of inputs and outputs.
65
Figure 41. Synthetic resistivity log Well 168, southern Pennsylvania area
(lower zone). Exercise 2 and combination A of inputs and outputs.
66
Figure 42. Synthetic resistivity log Well 169, southern Pennsylvania area(lower zone). Exercise 2 and combination A of inputs and outputs.
67
Figure 43. Synthetic resistivity log Well 174, southern Pennsylvania area
(lower zone). Exercise 2 and combination A of inputs and outputs.
68
Figure 44. Synthetic density log Well 157, southern Pennsylvania area
(lower zone). Exercise 2 and combination B of inputs and outputs.
69
Figure 45 Synthetic density log Well 168, southern Pennsylvania area
(lower zone). Exercise 2 and combination B of inputs and outputs.
70
vi
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 7/132
Page
Figure 46. Synthetic density log Well 169, southern Pennsylvania area
(lower zone). Exercise 2 and combination B of inputs and outputs.
71
Figure 47. Synthetic density log Well 174, southern Pennsylvania area
(lower zone). Exercise 2 and combination B of inputs and outputs.
72
Figure 48. Synthetic neutron log Well 157, southern Pennsylvania area
(lower zone). Exercise 2 and combination C of inputs and outputs.
73
Figure 49. Synthetic neutron log Well 168, southern Pennsylvania area(lower zone). Exercise 2 and combination C of inputs and outputs.
74
Figure 50. Synthetic neutron log Well 169, southern Pennsylvania area
(lower zone). Exercise 2 and combination C of inputs and outputs.
75
Figure 51. Synthetic neutron log Well 174, southern Pennsylvania area
(lower zone). Exercise 2 and combination C of inputs and outputs.
76
Figure 52. Synthetic resistivity log Well 157, southern Pennsylvania area(lower zone). Exercise 1 and combination A of inputs and outputs.
77
Figure 53. Synthetic resistivity log Well 168, southern Pennsylvania area(lower zone). Exercise 1 and combination A of inputs and outputs. 78
Figure 54. Synthetic resistivity log Well 169, southern Pennsylvania area(lower zone). Exercise 1 and combination A of inputs and outputs.
79
Figure 55. Synthetic resistivity log Well 174, southern Pennsylvania area
(lower zone). Exercise 1 and combination A of inputs and outputs.
80
Figure 56. Synthetic density log Well 157, southern Pennsylvania area
(lower zone). Exercise 1 and combination B of inputs and outputs.
81
Figure 57. Synthetic density log Well 168, southern Pennsylvania area(lower zone). Exercise 1 and combination B of inputs and outputs.
82
Figure 58. Synthetic density log Well 169, southern Pennsylvania area(lower zone). Exercise 1 and combination B of inputs and outputs.
83
Figure 59. Synthetic density log Well 174, southern Pennsylvania area
(lower zone). Exercise 1 and combination B of inputs and outputs.
84
Figure 60. Synthetic neutron log Well 157, southern Pennsylvania area
(lower zone). Exercise 1 and combination C of inputs and outputs.
85
Figure 61. Synthetic neutron log Well 168, southern Pennsylvania area
(lower zone). Exercise 1 and combination C of inputs and outputs.
86
Figure 62. Synthetic neutron log Well 169, southern Pennsylvania area
(lower zone). Exercise 1 and combination C of inputs and outputs.
87
Figure 63. Synthetic neutron log Well 174, southern Pennsylvania area(lower zone). Exercise 1 and combination C of inputs and outputs.
88
Figure 64. Synthetic resistivity log Well 157, southern Pennsylvania area
(lower zone). Exercise 2 and combination A of inputs and outputs.
89
Figure 65. Synthetic resistivity log Well 168, southern Pennsylvania area
(lower zone). Exercise 2 and combination A of inputs and outputs.
90
vii
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 8/132
Page
Figure 66. Synthetic resistivity log Well 169, southern Pennsylvania area
(lower zone). Exercise 2 and combination A of inputs and outputs.
91
Figure 67. Synthetic resistivity log Well 174, southern Pennsylvania area
(lower zone). Exercise 2 and combination A of inputs and outputs.
92
Figure 68. Synthetic density log Well 157, southern Pennsylvania area
(lower zone). Exercise 2 and combination B of inputs and outputs.
93
Figure 69. Synthetic density log Well 168, southern Pennsylvania area(lower zone). Exercise 2 and combination B of inputs and outputs.
94
Figure 70. Synthetic density log Well 169, southern Pennsylvania area
(lower zone). Exercise 2 and combination B of inputs and outputs.
95
Figure 71. Synthetic density log Well 174, southern Pennsylvania area
(lower zone). Exercise 2 and combination B of inputs and outputs.
96
Figure 72. Synthetic neutron log Well 157, southern Pennsylvania area(lower zone). Exercise 2 and combination C of inputs and outputs.
97
Figure 73. Synthetic neutron log Well 168, southern Pennsylvania area(lower zone). Exercise 2 and combination C of inputs and outputs. 98
Figure 74. Synthetic neutron log Well 169, southern Pennsylvania area(lower zone). Exercise 2 and combination C of inputs and outputs.
99
Figure 75. Synthetic neutron log Well 174, southern Pennsylvania area
(lower zone). Exercise 2 and combination C of inputs and outputs.
100
Figure 76. R 2 values obtained for the upper zone of the Southern
Pennsylvania data set, through exercise 1.
102
Figure 77. R 2 values obtained for the lower zone of the Southern
Pennsylvania data set, through exercise 1.
103
Figure 78. R 2 values obtained for the upper zone of the Southern
Pennsylvania data set, through exercise 1.
103
Figure 79. R 2 values obtained for the lower zone of the Southern
Pennsylvania data set, through exercise 1.
104
Figure 80. R 2 values obtained for the upper zone of the Southern
Pennsylvania data set, through exercise 2.
104
Figure 81. R 2 values obtained for the lower zone of the Southern
Pennsylvania data set, through exercise 2.
105
viii
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 9/132
LIST OF TABLES
Page Table 1. Common values of matrix density for different type of rocks. 13
Table 2. Common activation functions. 20
Table 3. Results for exercise 1 and combinations A, B, and C of inputs
and outputs. Buffalo Valley field data set.
46
Table 4. Results for exercise 2 and combinations A, B, and C of inputs
and outputs. Buffalo Valley field data set.
47
Table 5. Results for exercise 1 and combinations A, B, and C of inputsand outputs. Southern Pennsylvania area data set (upper zone). 48
Table 6. Results for exercise 2 and combinations A, B, and C of inputs
and outputs. Southern Pennsylvania area data set (upper zone).
49
Table 7. Results for exercise 1 and combinations A, B, and C of inputs
and outputs. Southern Pennsylvania area data set (lower zone).
51
Table 8. Results for exercise 2 and combinations A, B, and C of inputs
and outputs. Southern Pennsylvania area data set (lower zone).
52
ix
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 10/132
1. INTRODUCTION
Well logging, a geophysical technique that has been in use for almost one century, plays
an essential role in the determination of production potential in hydrocarbon reservoirs.
The well logging process involves lowering a number of instruments into a borehole,
which measures formation properties as function of depth. The collected data
measurements broadly fall into three categories: electrical, nuclear and acoustic. After
measurement, the log analyst interprets the data from the log in order to determine the
petrophysical parameters of the well. Therefore, logs constitute a very important tool in
the process of exploration and development of any petroleum field.
However, for economical reasons, companies do not always posses all the logs that are
required to determine reservoir characteristics.
This study presents a methodology that can help to solve the aforementioned problem by
generating synthetic wireline logs for those locations where the set of logs that are
necessary to analyze the reservoir properties, are absent or are not complete. Data used
were provided by Dominion E&P. The study area is located in southern Pennsylvania
(figure 1). The approach presented involves the use of artificial neural networks, as the
main tool, in conjunction with data obtained from conventional wireline logs.
During the last few years, artificial neural networks, a multi-dimensional-interpolation
system, have proven to be an efficient tool for prediction of reservoir properties. Actual
data are used to train the network; then the system can develop the best correlation
scheme between different properties of the reservoir and predict the same properties at
points where they are unknown.
1
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 11/132
AAr r mmssttr r oonngg CCoo..
CCr r oossss--SSeeccttii
A
174
SSoouutthhwweesstteer r nn PPeennnnssyyllvvaanniiaa,,
AAr r mmssttr r oonngg CCoo..
Figure 1. Location of the study area and wells analyzed (157, 168, 169, 174). A-A’ represents the line of the cross
2
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 12/132
SW (A) N
174 168 169
2550
250
2600
2650
2700
2750
2800
2850
3100
3050
3150
2900
3000
2950
3350
3300
3250
3200
3650
3600
3550
350
3450
3400
2502550
2600
2650
2700
2750
2800
2850
3100
3050
3150
2900
3000
2950
3350
3300
3250
3200
3550
350
3450
3400
2400
2450
250
2600Speechley
2550
2650
2700
2750
2800
2850
3100
3050
3150
2900
3000
2950
3350
3300
3250
3200
3550
350
3450
3400
2400
2450
250
2550
2600
2650
2700
2750
2800
2850
3100
3050
3150
2900
3000
2950
3350
3300
3250
3200
3400
34502nd Bradfor
350
2400
2450
2350
Figure 2. Gamma Ray correlation of 2nd Bradford (distributary channel – deep)
3
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 13/132
SW (A)
1200
1250
1300
1350
1600
1650
1400
1450
1850
1800
1750
1700
1150
1100
1050
200
1950
1900
1200
1250
1300
1350
1600
1650
1400
1450
1750
1700
1150
1100
1050
20
1950
1900
1200
1250
1300
1350
1600
1550
1650
1400
1500
1450
1850
1800
1750
1700
2150
2100
2050
200
1950
1900
1200
1250
1300
1350
1600
1550
1650
1400
1500
1450
1850
1800
1750
1700
1150
1100
1050
200
1950
1900
169168174
1800
1850
Gordon
1500
1550100 Foo
1500
1550Murrysville
Figure 3. Gamma Ray correlation of Murrysville / 100 Foot Sands (Braided Stream – Shallow)
4
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 14/132
The sequence of rock recorded through the set of logs used here belong to the Upper
Devonian of southern Pennsylvania (the Venango and Bradford plays), which have been
producing natural gas in the area since the late 1800s. The names of the units involved
from bottom to top are the following: 2nd
Bradford, Speechley, Gordon, 100 Foot, and
Murrysville (figures 2 and 3).
This work demonstrates that by following the methodology presented, synthetic logs can
be generated with a reasonable degree of accuracy. The intention of the technique used
here is not to eliminate well logging in a field but it is only meant to become a tool for
reducing costs for companies whenever logging proves to be insufficient and/or difficult
to obtain. This technique in addition, can provide a guide for quality control during the
logging process, by prediction of the response of the log before the log is acquired.
2. BACKGROUND
2.1. Geological Setting
The sequence of rocks involved in this study were deposited during the Devonian period,
when through the assemblage of the Laurussian continent.
This tectonic event, known in Appalachian area as the Acadian orogeny, led to
establishing of several river systems and a large prograding alluvial plain during Late
Devonian time. The mentioned sedimentary complexes are known as the Catskill and the
Price-Rockwell delta complexes (Boswell and Donaldson, 1988; Kammer and Bjerstedt,
1988).
The Catskill delta complex is composed of marine, transitional, and terrestrial clastic
rocks (Boswell and Donaldson, 1988). The lower portion, consisting from bottom to top
5
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 15/132
of the 2nd
Bradford, and the Speechley intervals (names assigned by local drillers), was
deposited in a variety of deltaic environments, related to the delta front. These units
constitute a sedimentary wedge that has been named the Bradford Play. The upper
portion, known as the Venango Play, comprises from bottom to top the Gordon, the 100
Foot and the Murrysville intervals (names assigned by local drillers). The environments
in this interval range from barrier-island complex in the Gordon Sandstone (McBride,
2004), to a complex of fluvial-braided river deposits in the 100 Foot and the Murreysville
zones. Figure 4 shows the Acadian wedge, and the relationship between the Bradford
and the Venango plays, as well as the relationship between the Catskill and the Price-
Rockwell deltas.
2.1.1. Description of the Murrysville & 100 Foot Sands
The Murrysville and 100 Foot sands are generally fine-to-coarse grained, quartz
cemented sandstones and conglomerates. The two formations, where developed, are
generally separated by a shale 30 to 40 foot shale interval. However, as with channel
deposits, this shale can often be scoured into by the younger Murrysville braided sands,
making the division between the two difficult to discern. For both units, much of the
primary porosity is intergranular with secondary porosity developed by mineral
dissolution and replacement, or from fracture porosity in areas with structural control.
Log correlation is often difficult in this interval due to rapid sediment changes within the
braided river environment. Often there is no clearly defined channel, but rather multiple
channels that are laterally discontinuous due to the high sediment load combined with
intermittent water movement, mainly during flooding events. Sand/conglomerate
channel lenses that interfinger with shale from overbank deposits often characterize the
6
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 16/132
two units. Variation in gamma ray, density, and resistivity values taken from well logs
can vary greatly from location to location, even in wells that are close offsets,
approximately 1000 ft apart (figure 2).
2.1.2. Description of the Gordon Sandstone
The Gordon sandstone is composed of a series of conglomerate, sandstone, and shales.
The sandstones of this facies are fine- to medium-grained, well sorted, well rounded, and
have good porosity and permeability. Log porosity has be measured up to 25% and
permeabilities have been measured up to 250 mD (McBride, 2004). The conglomerate
and coarse-grained sandstone of the upper upper shoreface and the foreshore tend to have
more cement and therefore a lower porosity and permeability (McBride, 2004). Gamma
ray, density, and resistivity values are more or less constant, although changes in
thickness occur from location to location (figure 2).
2.1.3. Description of the 2nd
Bradford Sand and the Speechley Sands
The 2nd
Bradford and the Speechley sands are quartz and feldspar cemented sandstone-
and-siltstone reservoirs. In general, these reservoirs have similar characteristics in all the
logs used, with little discontinuity among wells. There is no discernable water leg in
much of the area, and the dominant trapping mechanism is mostly structural with some
influence from stratigraphic changes. Gamma ray, density, and resistivity values are
relatively constant among wells in the study area. The Speechley interval is considerably
finer-grained than the 2nd Bradford (figure 3).
7
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 17/132
Figure 4. Cross-section of the Acadian clastic wedge, showing the relationship of thePrice-Rockwell and Catskill delta complexes (adopted from Boswell et al, 1996).
MurrysvilleAnticline
Figure 5. Structural map of the area, showing Structural Axes in Southwestern PA
8
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 18/132
2.1.4. Structural Geology
All of the wells in the study area fall within the Valley and Ridge Province of
Southwestern Pennsylvania. More specifically, the four wells analyzed in this work, are
located along the crest of the Murrysville anticline, a southwest / northeast trending
structural feature, which constitutes one of many low amplitude folds throughout
Southwestern Pennsylvania (figure 5). The observed structural features are
Pennsylvanian in age and formed during the Alleghenian orogeny, a tectonic event
correlated with the collision of the North American Plate with other continental plates, in
the late Paleozoic (Middle Ordovician to Permian).
2.2. Well logs fundamentals
2.2.1. Gamma-ray log
This log (figure 6) measures the natural gamma-ray emission of the various layers
penetrated in the well, a property related to their content of radiogenic isotopes of
potassium, uranium and thorium.
The tool may detect gamma ray energies of less than 0.5 to more than 2.5 millivolts.
Figure 7 shows the individual emission spectra for thorium, uranium, and potassium.
These elements (particularly potassium) are common in clay minerals and some
evaporites. In terrigenous clastic successions the log reflects the “cleanness”(lack of
clays) or “shaliness” (high radioactivities) (see figures 6 and 8) on the API scale of the
rock, averaged over an interval of depth. Because of this property, gamma-ray log
patterns mimic vertical sand-content or carbonate-content trends
9
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 19/132
It must be emphasized that the gamma-ray reading is not a function of grain size or
carbonate content, only the proportion of radioactive elements, which may be related to
the proportion of shale content. For example, clay free sandstones or conglomerates with
any mix of sand and pebble-clast sizes generally give similar responses, and lime
mudstone gives the same response as grainstone (Cant, 1992). The concentration of
radioactive elements in shale increases with compaction, so the shale line should be
readjusted if a thick section is being studied.
The Gamma Ray (GR) tool emits gamma rays into sedimentary formations to an average
penetration of approximately one foot. When gamma rays pass through a formation, they
experience successive Compton-scattering collisions with formation atoms (Cant, 1992).
These collisions cause gamma rays to lose energy (Bassiouni, 1994). Subsequently,
formation atoms absorb this energy through the photoelectric effect.
The amount of absorption is a function of the formation density. Therefore, the
radioactivity level shown on the GR log will be different for two formations that have the
same amount of radioactive material per unit volume but with different densities. The
lower density formation will appear to be more radioactive.
Cased hole use is an important application of the GR Log because the tool can be run in
completion and work-over operations. The GR log also can be used as a substitute for the
SP log in cased holes or open holes where the SP resolution is poor. When run with
casing collar locator logs, the GR also allows for a very accurate positioning of
perforating guns. GR logs are also used in locating source beds as well as in the
interpretation of depositional environment (Ameri, 2004).
10
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 20/132
Figure 6. Gamma ray and sonic logs from the Alberta basin, and their response
to different lithologies (adopted from Cant, 1992).
Figure 7. Gamma ray emission spectra of K-40, uranium, and thorium series
(adopted from Bassiouni, 1994)
11
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 21/132
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 22/132
Rock Type Matrix density (g/cm3)
Sand or Sandstone 2.65
Limestone 2.71
Dolomite 2.87
Anhydrite 2.98
Table 1. Common values of matrix density for different type of rocks.
The formula for calculating density porosity is :
)(
)(
f ma
bma
den ρ ρ
ρ ρ φ
−−
=
whereden
φ is density derived porosity,ma
ρ is the matrix density,b
ρ is the bulk density,
and f
ρ is the average density of the fluids in pore spaces.
Density of formation water ranges from 0.95 g/cc to 1.10 g/cc approximately depending
on temperature, pressure and salinity. Average density of oil is slightly lower than these
values and varies over an equally wide range. According to Bassiouni (1994), the
investigation ratio of the tool is shallow, therefore it investigates the invaded zone and
f ρ can be expressed by:
h xomf xo f )S 1( S ρ ρ ρ −+=
,
wheremf
ρ is the mud-filtrate density, S xo is the mud filtrate saturation in the invaded
zone, and h ρ is the invaded zone hydrocarbon density. In water-bearing zones where S xo
is equal to 1,h
ρ can be assumed to be equal tomf
ρ . For gas-bearing formations, it is
13
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 23/132
reasonable to assume thatmf
ρ is equal to 1.00. Therefore for practical reasons, 1.00 can
be used as a general value for the term f
ρ (Bassiouni, 1994).
2.2.2.2. Neutron Log
The neutron, a particle constituent of the atom, exhibits a high penetrating potential
because its lack of electric charge. Because of its penetration power, the neutron plays an
important role in well logging applications.
Because hydrogen is responsible for most of the slowing-down effect, measuring the
concentration of epithermal neutrons indicates the hydrogen concentration in the
material. In shale-free, water-bearing formations, the hydrogen concentration reflects the
porosity and lithology.
The neutron log measures the hydrogen concentration or hydrogen index in the rock. The
tool emits neutrons of a known energy level, and measures the energy of neutrons
reflected from the rock. Because energy is lost most easily to particles of similar mass,
the hydrogen concentration can be delineated.
2.2.3. Spontaneous potential (SP) log
This log records the electric potential between an electrode pulled up the hole and a
reference electrode at the surface. This potential exists because of electrochemical
differences between the waters within the formation and the drilling mud, and because of
ionic selection effects in shales (the surfaces of clay minerals selectively allow passage of
cations compared to anions). The potential is measured in millivolts relative to shale the
line (figure 9).
14
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 24/132
Figure 8. Relationship between gamma-ray deflection and proportion of shale
(adopted from Cant, 1992).
Figure 9. Example of SP and resistivity logs from the Alberta basin, (adopted from
Cant, 1992).
15
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 25/132
In shaly sections the maximum SP response to the right from normal, depending on the
salinity of the drilling mud. The best test of the reliability of the SP log in determining
lithology is to calibrate the log against cores and cuttings.
SP log is useful for detection of permeable beds, location of bed boundaries for
correlation purposes, determination of formation water resistivity, and indication of
formation shaliness.
2.2.4. Resistivity log
This log records the resistance of interstitial fluids to the flow of an electric current, either
transmitted directly to the rock through an electrode, or magnetically induced deeper into
the formation from the hole, as it is the case with induction logs (figure 9). The term
“deep” refers to horizontal distance from the well bore. Varying the length of the tool and
focusing the induced current measure resistivities at different depths into the rock.
Several resistivity and induction curves are commonly shown on the same track (figure
9).
Resistivity logs are used for evaluation of fluids within formations. They can also be used
for identification of coals (high resistance), thin limestones in shale (high resistance) and
bentonites (low resistance). In turn, for wells where few types of logs have been run, the
resistivity log may be useful for picking tops and bottoms of formations, and for
correlating between wells. Freshwater-saturated porous rocks have high resistivities, so
the log can be used in these cases to separate shales from porous media (sandstones and
carbonates).
2.2.5. Sonic log
This log (figure 6) measures the velocity of sound waves in rock (Ameri, 2004).
16
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 26/132
Velocity depends on 1) lithology 2) amount of interconnected pore space, and 3) type of
fluid in the pores.
The log is useful for delineating beds of low-velocity material such as coal (figure 6) or
poorly cemented sandstones, as well as high-velocity material such as tightly cemented
sandstones and carbonates or igneous basement. Sonic logs are also important in
understanding and calibrating seismic lines.
2.3. Fundamentals on Artificial Neural Networks
An artificial neural network is a computing parallel scheme based on the biological
neural system configuration (Mohaghegh, et al., 2002; Poulton, 2002, Faucett, 1994,
White et al., 1995). A neuron in the brain (figure 10) is a unique piece of equipment that
consists of three types of components called dendrites, cell body or soma, and axon.
Dendrites are the sensitive part of neuron that receive signal from other neuron. Soma
calculates and sums the signals and transmitted to other cells through axon. The
biological neuron carries information and transfers to other neuron in a chain of
networks.
Soma
Axon
Dendrite ofanother Neuron
Synaptic
Gap
Dendrite
Axon of
another Neuron
Soma
Axon
Dendrite ofanother Neuron
Synaptic
Gap
Dendrite
Axon of
another Neuron
Figure 10. Sketch of a biological neuron (adopted and modified from
Faucett, 1994).
17
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 27/132
The process of growing and learning in the human brain is one of creating neural
connections through an associative process based on patterns received from the senses.
The artificial neuron imitates the functions of the three components of the biological
neuron and their unique process of learning.
The process of learning in an artificial neural network is similar to the process that is
performed in the human brain. Artificial neural networks are systems mathematically
designed to receive, process, and transmit information. Information processing occurs at
the neurons. The simple neuron (figure 11) consists of an input layer, activation function,
and output layer. The input layer receives signals from the external environment (or other
neuron). The activation function is the internal neuron that calculates and sum the input
signals.
weights
Figure 11. Structure of a simple artificial neuron (adopted and modified from
Faucett, 1994)
These signals are then transmitted to an output layer and retransmitted. The input layer,
activation function, and output layer in artificial neuron are similar to the function of
18
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 28/132
dendrites, soma, and axon in the biological neuron. Still, as seen in figure 11, an artificial
network has many inputs but only one output.
2.3.1.Mechanics of Neural Networks
Artificial neural networks are generalizations of mathematical models of human
cognition (Faucett, 1994). They function based on the following assumptions:
Information processing occurs at many simple elements (neurons).−
−
−
−
Signals are passed between neurons over connection links.
Each connection link has an associated weight, which, in a typical neural net,
multiplies the transmitted signal.
Each neuron applies an activation function to its net input to determine its output
signal.
Assume we have n input units, X i ,…,X n with input signals x1 ,…,xn. When the network
receive the signals ( xi) from input units ( X i), the net input to output ( y_in j) is calculated by
summing the weighted input signals as follows:
∑=
n
1iiji
w x
The matrix multiplication method for calculating the net input is shown in the equation
below:
∑=
=n
1iiji j
w xin _ y
where wij is the connection weights of input unit xi and output unit y j.
The network output ( y j) is calculated using the activation function f(x). In which y j = f(x),
where x is y_in j. The computed weight from the training is stored and will become the
19
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 29/132
information for the future application. Table 2 shows the most common types of
activation functions.
Neural networks can be divided into three architectures, namely single layer, multilayer
network and competitive layer. The number of layers in a net is defined based on the
number of interconnected weights in the neuron. A single layer network consists only of
one layer of connected weights, whereas, multilayer networks consist of more than one
layer of connection weights. The network also consists of an additional layer called a
hidden layer.
Function Definition Identity x Logistic
X e−+1
1
Hyperbolic
Exponential
Softmax
Unit sum
Square root
Sine sin( x)
Table 2. Common activation functions.
Multilayer networks can solve more complicated problems than those solved by a single
layer network. Both networks are also called “feed forward” networks where the signal
flows from the input units to the output units in a forward direction. For example, a
recurrent network is a feedback network with a closed-loop signal from the unit back to
itself.
20
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 30/132
Figure 12 shows the architecture of a simple network, a multilayer perceptron (MPL),
where the basic parts are the neurons (also known as nodes), layers and connection
weights. Input and output layers are trained with user supplied data. The hidden layer
performs a mapping from the input to the output layer. The architecture of different
networks is defined by the connection strategy between neurons and layers (Poulton
2002).
2.3.2. Learning mechanisms for Artificial Neural Networks
Human beings are capable of learning with the aid of a teacher or on their own. The first
case is considered supervised learning, the latter is unsupervised learning. In supervised
learning, a teacher supplies both the material to be learned and corrects the student when
the response to the material is incorrect. In unsupervised learning, the student receives the
material to be learned and has to drawn his/her own conclusion as to what the material
means. The most basic division of artificial neural networks is whether they learn in a
supervised or unsupervised mode. In supervised mode, common in most applications, the
user supplies a set of patterns that the neural network should learn. These patterns are
then associated with responses dealing with a classification or an estimation of a
parameter value. In the unsupervised mode, the network is supplied with the set of
patterns to learn, but it is not supplied with a prior classification or parameter value
(Poulton, 2002).
While the architecture is important in defining how the network will function, the specific
learning algorithm used to train the network defines how it will learn. The learning
algorithm involves the input signals, the weights, the activation function (table 2), and the
21
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 31/132
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 32/132
Network, which uses the Backpropagation-learning algorithm. Backpropagation
algorithm is one of the well-known algorithms in neural networks. Backpropagation
algorithm has been popularized in 1980s as a euphemism for generalized delta
rule. Backpropagation of errors or generalized delta rule is a decent method to
minimize the total squared error of the output computed by the net (Faucett, 1994).
2.3.2.2. Unsupervised Learning
Unsupervised learning (figure 13) method is not given any target value. A desired output
of the network is unknown. During training the network performs some kind of data
compression such as dimensionality reduction or clustering. The network learns the
distribution of patterns and makes a classification of that pattern where, similar patterns
are assigned to the same output cluster. Kohonen network is the best example of
unsupervised learning network. According to Sarle (1997) Kohonen network refers to
three types of networks that are Vector Quantization, Self-Organizing Map and Learning
Vector Quantization
Figure 13. Supervised and Unsupervised scheme (adopted and modified from Jaeger,
2002)
outout
ii
out
i
Mod
unknoIn utCorre
:out
Unsupervised B.
Mod
:Teach
A. Supervised
23
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 33/132
2.3.3. Training the network
Training the network is time consuming. It usually learns after several epochs, depending
on how large the network is. Thus, a large network requires more training time than a
small one. Basically, the network is trained for several epochs and stopped after
reaching the maximum epoch. For the same reason minimum error tolerance is
used provided that the differences between network output and known outcome is
less than the specified value (see for example Pofahl et al., 1998). We could also
stop the training after the network meet certain stopping criteria.
During training the network might learn too much. This problem is referred to as
overfitting. Overfitting is a critical problem in most all-standard neural networks
architectures. Furthermore, neural networks and other artificial intelligence machine
learning models are prone to overfitting (Lawrence et al., 1997). One of the solutions is
early stopping (Sarle, 1995), but this approach needs more critical intention, as this
problem is harder than expected (Lawrence et al., 1997). The stopping criterion is also
another issue to consider in preventing overfitting (Prechelt, 1998). To crack this problem
during training, a validation set is used instead of a training data set. After a few epochs,
the network is tested with the verification data. The training is stopped as soon as the
error on verification set increases rapidly higher than the last time it was checked
(Prechelt, 1998). Figure 14 shows that the training should stop at time t when verification
error starts to increase.
24
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 34/132
E r r o r
verification
training
Timet
2.3.4. General Regression Neural Network
Figure 14. Training and verification curve (adopted from Prechelt, 1998)
The general regression neural network (GRNN) is Donald Specht's term for a neural
network invented by him in 1990 (Specht, 1991). The general regression neural network
(GRNN) is a memory-based network that provides estimates of continuous variables and
converges to the underlying regression surface. GRNNs are based on the estimation of
probability density functions, feature fast training times and can model non-linear
functions.
The GRNN is a one-pass learning algorithm with a highly parallel structure. It is that,
even with sparse data in a multidimensional measurement space, the algorithm provides
smooth transitions from one observed value to another. The algorithmic form can be used
for any regression problem in which an assumption of linearity is not justified.
GRNN can be thought as a normalized RBF (Radial Basis Functions) network in which
there is a hidden unit centered at every training case. These RBF units are usually
probability density functions such as the Gaussian. The only weights that need to be
learned are the widths of the RBF units. These widths are called "smoothing parameters.
The main drawback of GRNN is that it suffers badly from the curse of dimensionality.
25
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 35/132
GRNN cannot ignore irrelevant inputs without major modifications to the basic
algorithm. So GRNN is not likely to be the top choice if there are more than 5 or 6 no
redundant inputs.
The regression of a dependent variable, Y, on an independent variable, X, is the
computation of the most probable value of Y for each value of X based on a finite
number of possibly noisy measurements of X and the associated values of Y. The
variables X and Y are usually vectors.
In order to implement system identification, it is usually necessary to assume some
functional form. In the case of linear regression, for example, the output Y is assumed to
be a linear function of the input, and the unknown parameters, ai, are linear coefficients.
The procedure presented in Donald F. Specht’s article (Specht, 1991) does not need to
assume a specific functional form.
A Euclidean distance is estimated between an input vector and the weights, which are
then rescaled by the spreading factor. The radial basis output is then the exponential of
the negatively weighted distance.
The GRNN equation is:
∑
∑
=
=
−
−
=n
1i2
2
i
n
1i2
2
i
i
2
Dexp
2
DexpY
) X ( Y
σ
σ
The estimate Y(X) can be visualized as a weighted average of all of the observed values,
Yi, where each observed value is weighted exponentially according to its Euclidian
distance from X. Y(X) is simply the sum of Gaussian distributions centered at each
training sample. However the sum is not limited to being Gaussian.
26
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 36/132
In this theory, σ is the smoothing factor and can be seen as the spread of the Gaussian
bell.
3. LITERATURE REVIEW
Artificial neural networks have been broad used in reservoir characterization because
their ability to extract nonlinear relationships between a sparse set of data (Banchs and
Michelena, 2002). The most common architecture used is the backpropagation artificial
neural network (BPANN). Studies in this subject have used wireline measurement logs,
and seismic attributes to predict reservoir properties such as effective porosity, fluid
saturation and rock permeability, and to define lithofacies and predict log responses, i.e.
generation of synthetic logs. In all cases it was demonstrated that ANN are powerful tool
for recognition-pattern, system identification, and prediction of any variable in the future
with a better correlation coefficient (R 2) over traditional statistical analysis like linear
regression.
Mohaghegh, et al. (1998), describes a methodology developed to generate synthetic
Magnetic Resonance Imaging logs using conventional well logs such as Spontaneous
Potential,Gamma Ray, Caliper, and Resistivity for four wells located in East Texas, Gulf
of Mexico, Utah, and New Mexico. The methodology incorporates a backpropagation
artificial neural network as its main tool. The synthetic Magnetic Resonance Imaging
(MRI) logs were generated with a high degree of accuracy even when the model
developed used data not employed during model development.
Mohaghegh, et al. (1999), Present an approach that involves neural-network-design
software, for low cost/ high effectiveness log analysis in a field scale. The cost reduction
27
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 37/132
is achieved by analyzing only a group of the wells in the field. The intelligent software
tool is built to learn and reproduce the analyzing capabilities of the engineer on the
remaining wells. As part of this study, logs that were missed in several wells and that
were necessary for analysis were generated. The tool used for this procedure was also a
backpropagation neural network
Bhuiyan (2001), developed a backpropagation neural network to generate synthetic
magnetic resonance imaging logs (MRI) in order to provide information about reservoir
characteristics of the Cotton Valley formation. In this work, data preparation previous to
network training, involved fuzzy logic for group well logs together based on similarity
criteria of the reservoir formation and to identify the most influential logs for a well.
Tonn (2002), uses seismic attributes, density and sonic logs to train an ANN in order to
predict the GR response of the Athabasca oil sands in western Canada, in order to solve
the reservoir properties and therefore chose the best location to place injection and
production wells for a steam injection program.
4. METHODOLOGY
The main objective of the study was to develop a systematic approach, that uses an
artificial neural network model, with the aim of generate synthetic wireline logs, from
other conventional wireline logs. Developing of this technique intends to generate
synthetic curves of any nonexistent log at any specific location, which can be necessary
to measure reservoir characteristics such as effective porosity, fluid saturation and rock
permeability. Implementation of this approach will reduce operation costs to companies,
28
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 38/132
by avoiding acquisition of new logs than now will be able to be obtained from already
acquired data.
The methodology applied in this work is based on the approaches presented by Bhuiyan,
(2001), and Mohaghegh et al., (1999).
In order to generate the synthetic logs, an artificial-neural-network-design software,
NeuroShell®2, previously developed by Ward Systems Group®, was used to find the
best model that could be applied to the data set used for this project.
NeuroShell®2 includes a set of procedures for building and executing a complete neural
network application. The software gives users the ability to create and execute a variety
of neural network architectures through different modules that allow the user to input
data, prepare the data for training, built the neural network, apply the model to new data
and examine the outputs. Outputs can be evaluated in terms of R-squared (R 2). R
2is the
relative predictive power of a model. R 2
is a descriptive measure between 0 and 1. The
closer it is to one, the better your model is. R-squared is defined as:
YY
2
SS
SSE 1 R −= , where
∑ −= 2 ) y y( SSE
∑ −= 2
YY ) y y( SS
= y actual value
= y the predicted value of , and y
= y the mean of the values. y
29
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 39/132
R 2 values can be interpreted, as indicators of how good are the results produced by the
network, although they are not the ultimate measure, but is the user who finally decides if
the network is working properly or not.
4.1. Data Preparation
The first step of the data preparation was to identify the depth of the pay zone/s. Five
units are producing in the study area: the Murrysville, the 100 foot, the Gordon, the
Speechely, and the 2nd
Bradford formations, which were included in this study.
Taken in account the sedimentary characteristics of the aforementioned formations, the
studied interval can be divided in two segments with different lithologic and petrophysic
characteristics. In addition breaking of the logs allowed better visualization of the actual
logs and the logs generated by the network. These two intervals are from 1000 to 2000
feet and from 2500 to 3500 feet.
The second step was to prepare a matrix in a spreadsheet to be imported to NeuroShell®.
The matrix for each well contained the well name, the depths, the longitude, the latitude,
and the values of the resistivity (RILD), density (DEN), gamma ray (GRGC), and neutron
(DNND) logs. Figure 15 is an example of the arrangement used to prepare the matrices.
ID DEPTH LAT LONG RILD DEN NPRL GRGC DNND
157 2000 40.5859 79.4719 32.87 2.70 8.79 144.66 15775.31
157 2000 40.5859 79.4719 31.73 2.71 9.08 145.10 15718.19
157 1999 40.5859 79.4719 30.91 2.71 9.38 142.85 15628.33
157 1999 40.5859 79.4719 30.82 2.71 9.58 141.16 15647.24
157 1998 40.5859 79.4719 31.57 2.71 9.61 142.10 15765.67
157 1998 40.5859 79.4719 32.53 2.69 9.43 142.63 15928.85
Figure 15. Segment of the matrix prepared for well 157.
30
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 40/132
4.2. Neural network model development
Development of the neural network model was completed using four wells that included
gamma ray, density, neutron, and resistivity logs. Different training algorithms were
attempted until the best results in terms of R 2 and matching of the synthetic logs
generated by the network versus the actual logs were achieved. The algorithm that has
been most frequently used in previous publications similar to this study, is the
backpropagation (Mohaghegh et al., 1998, Mohaghegh et al., 1999, Bhuiyan, 2001),
however, this work found that the best results were obtained using a General Regression
Neural Network (GRNN).
The network consists of three layers: an input layer made up of 7 neurons, a hidden layer
made up of 7000 neurons, and finally an output layer consisting of only one neuron. The
smooth factor applied was always 0.122 obtained by default from the NeuroShell.
Training, calibration and verification were carried out through two different exercises that
are described as follows:
4.2.1. Exercise 1: Four Wells Combined
In this exercise the entire set of data, consisting of 4 wells, was used during development
and training of the network and then each one of these wells were used to verify the
trained network (figure 16).
The data brought into the network as inputs/outputs were the locations of the wells (in
terms of latitude and longitude), Depths, Deep Induction (RILD) log values, Density
(DEN) log values, Gamma Ray (GRGC) log values, and Neutron (DNND) log values.
31
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 41/132
174
169
168
157
Verification
wells
174
169
168
157
Trainingand
calibrationwells
Figure 16. Cartoon to show distribution of wells used for training /testing and Production dataset through exercise 1.
Combinations of different inputs/outputs were chosen to train the network (figure 17); at
each combination one of the logs aforementioned was predicted from the other
information. In combination “A” the resistivity log was used as an actual output while the
density, the gamma ray, the neutron, and the coordinates and depths (XYZ) were used as
inputs, in combination “B” the density log was used as an actual output while the
resistivity, the gamma ray, the neutron, and XYZ were used as inputs, and in combination
“C” the neutron log was used as an actual output while the resistivity, the gamma ray, the
density, and XYZ were used as inputs. The percentages used for training, calibration and
verification were 80%, 15%, and 5% respectively. There were in total 3 combinations
used for exercise 1.
32
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 42/132
Figure 17. Different combinations of inputs/ outputs used for development of neural network model
Combination C
Combination B
Combination A
XYZ = Coordinates
and Depths
NEU = Neutron
GR = Gamma Ray
DEN = Density
RES = Resistivity
= Inputs
= Actual Output
I
A
III
Z
Y
X
XYZNEURE S GR DEN
AI
IIIA I
Z
Y
X
XYZNEURES GR DEN
III
Z
Y
X
XYZNEURE S GR DEN
AI
4.2.2. Exercise 2: Three Wells Combined, one well out
Differing from exercise 1, this exercise used only three wells for training and
development of the network while the fourth well, never used during training and
calibration, was selected to generate synthetic logs out of the other three wells
33
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 43/132
(verification). Since the verification data set in this exercise consisted of a data set never
used during training, the model was developed only with training and calibration data
sets. The percentages used were distributed 85% for training and 15% for calibration.
Therefore, wells 157, 168, and 169 where combined to generate logs in well 174; wells
157, 168, and 174 where combined to generate logs in well 169; wells 157, 174, and 169
where combined to generate logs in well 168; and wells 174, 169, and 168 where
combined to generate logs in well 157.
Figure 18 represents the combinations of wells used through this exercise (there were in
total four possible combinations). Combinations of inputs/output used in exercise 1 were
repeated for this exercise.
5. RESULTS
As mentioned before, the neural network used in this work to generate the synthetic logs
involved General Regression architecture. During training the network uses a data set
consisting of inputs and outputs. For calibration the data set consists of a similar number
of inputs and outputs, but in this case they are used to validate the network by verifying
how well the network is performing on data that were never seen before during the
training process. In this fashion, the partially constructed network is checked at certain
intervals of training by applying the calibration data set. Finally the verification set is
used to prove the ability of the network to provide accurate results on the unseen data.
Therefore, the values of R 2 obtained for each of the dataset mentioned, reflect the
performance of the network during training, calibration and verification.
34
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 44/132
Verification
well
169
174
157
Training
and
calibration
wells
168
169
Verification
well
174
168
157
Training
and
calibration
wells
157
Verification
well
168
169
174
Training
and
calibration
wells
174
Verification
well
169
168
157
Training
and
calibration
wells
Figure 18. Combinations of wells for training, testing and verification used in exercise 2.
Although the most important criteria to determine if the network is capable of generating
logs with a certain degree of accuracy is indeed the degree of matching between the plots
of the actual logs with the plots of logs generated by the network. Independently of the
R 2
values obtained, it was observed that the best matching between actual and synthetic
generated logs was obtained when high values of R 2 (higher than 0.7) were obtained.
Oppositely, when poor values of R 2 were obtained, matching was also poor.
35
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 45/132
The following are the results obtained from the neural network models developed for this
work. Models were developed throughout the exercises discussed above and using
different combinations of inputs and outputs (combinations A, B, and C). As mentioned
before, both exercises were completed for the interval from 1000 to 2000 feet and from
2500 to 3500 feet.
Performance of the model is showed in terms of values of the R-squared (R 2), the
coefficient of determination (r 2), and the coefficient of correlation obtained for the
training (TRN), calibration (TST) and verification (PRO) data sets. Log plots are also
included, in order to let the reader visualize the degree of matching between the actual
logs and the logs generated by the network.
5.1. First Attempt: Data set from Buffalo Valley Field (New Mexico)
A first attempt to generate synthetic well-logs was done using a set of logs obtained at the
web page of the New Mexico Energy, Minerals and Natural Resources Department.
(http://www.emnrd.state.nm.us). The data is part of the data used for a project that
intends to characterize the Morrow Formation at Buffalo Valley field, from well logs and
seismic data, using an artificial neural network.
The log data set analyzed consisted of resistivity, gamma ray, density, and neutron-
porosity logs. Logs were originally obtained as images of the hard copies, and were
available in .tif format, therefore they had to be digitized in order to be converted to a
format compatible with excel (.las format). Figure 19 shows the location of the wells and
figures 20 to 27 show the quality of the original .tif files and the resulting curves after
digitalization.
36
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 46/132
The combinations of inputs and outputs that were used to analyze this set of data
corresponds with the same used for exercise 1 and 2, however, for this case, the neutron
porosity log was used instead of the neutron log, since this last curve was absent in all the
wells. Development of the network models was carried out according with the
methodology steps discussed in exercise 1 and 2.
665
754219
1 mile
321
MEXICO
OKLAHOMA
TEXAS
NEW MEXICO
Figure 19. Location of the Buffalo Valley Field area and the wells used in the first attempt.
37
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 47/132
8 0 0 0
8 1 0 0
8 2 0 0
8 3 0 0
1 0 0 1 5 0 2 0 0
.tif file Digitized log
Gamma Ray (API units)
Figure 20. .tif files and the resulting Gamma Ray curve after digitalization. Well 219 Buffalo Valley Field
38
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 48/132
8 0 0 0
8 1 0 0
8 2 0 0
8 3 0 0
2 . 0 2 . 5 3 . 0
.tif file Digitized log
Density (g/ccm)
Figure 21. .tif files and the resulting Density curve after digitalization. Well 219 Buffalo Valley Field
39
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 49/132
8 0 0 0
8 1 0 0
8 2 0 0
8 3 0 0
- 2 0 0 0 2 0 0 4 0 0
.tif file Digitized log
Resistivity (ohm-m)
Figure 22. .tif files and the resulting Resistivity curve after digitalization. Well 219 Buffalo Valley Field
40
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 50/132
8000
8100
8200
8300
-0.050.150.35
.tif file Digitized log
Neutron Porosity
Figure 23. .tif files and the resulting Neutron Porosity curve after digitalization. Well 219 Buffalo Valley Field
41
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 51/132
8000
8100
8200
8300
0 20 40 60 8 0 100
.tif file Digitized log
Gamma Ray (API units)
Figure 24. .tif files and the resulting Gamma Ray curve after digitalization. Well 665 Buffalo Valley Field
42
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 52/132
8000
8100
8200
8300
2. 0 2 . 2 2 . 4 2 . 6 2 . 8
.tif file Digitized log
Density (g/ccm)
Figure 25. .tif files and the resulting Density curve after digitalization. Well 665 Buffalo Valley Field
43
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 53/132
.tif file Digitized log
Resistivity (ohm-m)
8000
8100
8200
8300
0 10 00 20 00 3000
Figure 26. .tif files and the resulting Resistivity curve after digitalization. Well 665 Buffalo Valley Field
44
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 54/132
8000
8100
8200
8300
-0.100.000.100.200.30
.tif file Digitized log
Neutron Porosity
Figure 27. .tif files and the resulting Neutron Porosity curve after digitalization. Well 665 Buffalo Valley Field
45
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 55/132
R 2 values obtained for the verification data set, out of the Buffalo Valley field data, were
fairly good, when the neural network model was developed using four wells for training
and calibration, and one of the wells used during training for verification (exercise 1)
Results are shown in table 3.
COMBINATION A
Inputs: Density, Gamma Ray, Neutron Porosity, XYZ
Outputs: Resistivity
Data Set R 2
TRN 0.9295
TST 0.9188
PRO 0.9124PRO well 219 0.766
PRO well 321 0.9342
PRO well 665 0.7281
PRO well 754 0.8128
COMBINATION B
Inputs: Resistivity, Gamma Ray, Neutron Porosity, XYZ
Outputs: Density
Data Set R 2
TRN 0.9708
TST 0.961
PRO 0.9341
PRO well 219 0.6229
PRO well 321 0.895
PRO well 665 0.735
PRO well 754 0.8199
COMBINATION C
Inputs: Resistivity, Density, Gamma Ray, XYZ
Outputs: Neutron Porosity
Data Set R 2
TRN 0.803
TST 0.769
PRO 0.7173
PRO well 219 0.7974
PRO well 321 0.9387
PRO well 665 0.8552
PRO well 754 0.5801
Table 3. Results for exercise 1 and combinations A, B, and C of inputs and outputs. Buffalo Valley field data set.
46
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 56/132
Conversely, when the network was trained using three wells and was verified using a
fourth well that has not been used during training (exercise 2), extremely poor R 2
values
were observed. Table 4 shows the results obtained throughout exercise 2, for
combinations A, B, and C of inputs and outputs.
COMBINATION A COMBINATION B COMBINATION C
Training wells: 219, 321, 665 Training wells: 219, 321, 665 Training wells: 219, 321, 665
Verification well: 754 Verification well: 754 Verification well: 754
Data Set R 2 Data Set R 2 Data Set R 2
TRN 0.9555 TRN 0.9715 TRN 0.8664
TST 0.9377 TST 0.9603 TST 0.8254
PRO well 754 -0.4601 PRO well 754 -0.1129 PRO well 754 -0.1082Training wells: 219, 321, 754 Training wells: 219, 321, 754 Training wells: 219, 321, 754
Verification well: 665 Verification well: 665 Verification well: 665
Data Set R 2 Data Set R 2 Data Set R 2
TRN 0.9619 TRN 0.9858 TRN 0.8059
TST 0.9627 TST 0.98 TST 0.803
PRO well 665 0.3422 PRO well 665 -1.4643 PRO well 665 0.2685
Training wells: 219, 754, 665 Training wells: 219, 754, 665 Training wells: 219, 754, 665
Verification well: 321 Verification well: 321 Verification well: 321
Data Set R 2 Data Set R 2 Data Set R 2
TRN 0.9459 TRN 0.8674 TRN 0.9504
TST 0.9249 TST 0.801 TST 0.8546
PRO well 321 -11.5431 PRO well 321 -2.6582 PRO well 321 -0.2056
Training wells: 754, 665, 321 Training wells: 754, 665, 321 Training wells: 754, 665, 321
Verification well: 219 Verification well: 219 Verification well: 219
Data Set R 2 Data Set R
2Data Set R
2
TRN 0.7571 TRN 0.9688 TRN 0.8401
TST 0.7162 TST 0.9766 TST 0.807
PRO well 219 -139.6099 PRO well 219 -0.9922 PRO well 219 -1.3319
Table 4. Results for exercise 2 and combinations A, B, and C of inputs and outputs. Buffalo Valley field data set.
47
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 57/132
5.2. Southern Pennsylvania Logs: Upper Zone (1000’ to 2000’)
5.2.1. Exercise 1
Table 5 contains the R 2 values and the correlation and determination coefficients obtained
through exercise 1 for the upper interval of the logs belonging to the Southern
Pennsylvania area.
COMBINATION A
Inputs: Density, Gamma Ray, Neutron, XYZ
Outputs: Resistivity
Data Set R 2
TRN 0.9377
TST 0.9412
PRO 0.9207PRO well 157 0.9264
PRO well 168 0.9619
PRO well 169 0.9102
PRO well 174 0.9262
COMBINATION B
Inputs: Resistivity, Gamma Ray, Neutron, XYZ
Outputs: Density
Data Set R 2
TRN 0.8338
TST 0.8225
PRO 0.8099
PRO well 157 0.8126
PRO well 168 0.8668
PRO well 169 0.831
PRO well 174 0.8161
COMBINATION C
Inputs: Resistivity, Density, Gamma Ray, XYZ
Outputs: Neutron
Data Set R 2
TRN 0.942
TST 0.9291
PRO 0.9234
PRO well 157 0.9299
PRO well 168 0.9398
PRO well 169 0.9331
PRO well 174 0.9506
Table 5. Results for exercise 1 and combinations A, B, and C of inputs and outputs. Southern Pennsylvania area data
set (upper zone).
48
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 58/132
Figures 28 to 39 show the relation between the actual logs and the logs generated by the
neural network model during verification. Training and calibration results are included in
the appendix.
5.2.2. Exercise 2
Results obtained through exercise 1 for the upper zone of the logs of Southern
Pennsylvania area, substantially improved in comparison with the results obtained from
Buffalo Valley data set. However, improvements were more significant when exercise 2
was performed. R 2 values rose from negative values to values over 0.8.
COMBINATION A COMBINATION B COMBINATION CTraining wells: 157, 168, 169 Training wells: 157, 168, 169 Training wells: 157, 168, 169
Verification well: 174 Verification well: 174 Verification well: 174
Data Set R 2 Data Set R 2 Data Set R 2
TRN 0.9498 TRN 0.8612 TRN 0.9366
TST 0.9418 TST 0.8301 TST 0.9267
PRO well 174 0.9091 PRO well 174 0.6002 PRO well 174 0.7854
Training wells: 157, 168, 174 Training wells: 157, 168, 174 Training wells: 157, 168, 174
Verification well: 169 Verification well: 169 Verification well: 169
Data Set R 2 Data Set R
2 Data Set R
2
TRN 0.9455 TRN 0.8612 TRN 0.9437
TST 0.9530 TST 0.8301 TST 0.9348
PRO well 169 0.9218 PRO well 169 0.8132 PRO well 169 0.9226
Training wells: 157, 174, 169 Training wells: 157, 174, 169 Training wells: 157, 174, 169
Verification well: 168 Verification well: 168 Verification well: 168
Data Set R 2 Data Set R 2 Data Set R 2
TRN 0.9667 TRN 0.8397 TRN 0.9435
TST 0.9675 TST 0.6957 TST 0.9291
PRO well 168 0.9623 PRO well 168 0.7466 PRO well 168 0.9213
Training wells: 174, 169, 168 Training wells: 174, 169, 168 Training wells: 174, 169, 168
Verification well: 157 Verification well: 157 Verification well: 157
Data Set R
2
Data Set R
2
Data Set R
2
TRN 0.9464 TRN 0.8313 TRN 0.948
TST 0.9555 TST 0.8277 TST 0.9299
PRO well 157 0.9376 PRO well 157 0.6453 PRO well 157 0.8003
Table 6. Results for exercise 2 and combinations A, B, and C of inputs and outputs. Southern
Pennsylvania area data set (upper zone).
49
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 59/132
Despite some wells had R 2 values between 0.6 and 0.7, the synthetic logs generated by
the network during verification still showed a high degree of accuracy (figures 40 to 51).
It is important to mention that the first results for the verification dataset at this interval
were not successful. R 2 values obtained for this first effort were relatively low. The
reason for these poor results was that data initially included a log interval that had been
run at a cased segment, hence, the values recorded were highly anomalous and
consequently they did inject a significant error into the network. Once the data were
cleaned of this error, results improved. The final results obtained after cleaning the data
are summarized in table 6. Plots of actual logs versus logs generated by the neural
network model during training and calibration are included in the appendix.
5.3. Southern Pennsylvania Logs: Lower Zone (2500’ to 3500’)
5.3.1. Exercise 1
Values of R2, and correlation and determination coefficients, obtained for the lower
interval of the logs belonging to the Southern Pennsylvania area through exercise 1 were
good in general (over 0.8). Table 7 summarizes these results. The logs generated during
verification are shown in figures 52 to 63. Training and calibration results are included in
the appendix.
5.3.2. Exercise 2
Results obtained through exercise 2 are summarized in table 8. Figures 64 to 75 show the
logs generated during verification. Training and calibration results are included in the
appendix.
50
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 60/132
COMBINATION A
Inputs: Density, Gamma Ray, Neutron, XYZ Outputs: Resistivity
Data Set R 2
TRN 0.9536
TST 0.9388
PRO 0.9426
PRO well 157 0.9582
PRO well 168 0.955
PRO well 169 0.9199
PRO well 174 0.9568
COMBINATION B
Inputs: Resistivity, Gamma Ray, Neutron, XYZ Outputs: Density
Data Set R 2
TRN 0.8118
TST 0.7946
PRO 0.8336
PRO well 157 0.8229
PRO well 168 0.816
PRO well 169 0.7911
PRO well 174 0.8064
COMBINATION C
Inputs: Resistivity, Density, Gamma Ray, XYZ
Outputs: Neutron
Data Set R 2
TRN 0.9313
TST 0.9133
PRO 0.9311
PRO well 157 0.9087
PRO well 168 0.94
PRO well 169 0.9215
PRO well 174 0.9354
Table 7. Results for exercise 1 and combinations A, B, and C of inputs and outputs. SouthernPennsylvania area data set (lower zone).
51
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 61/132
Combination A Combination B Combination C
Training wells: 157, 168, 169 Training wells: 157, 168, 169 Training wells: 157, 168, 169
Verification well: 174 Verification well: 174 Verification well: 174
Data Set R 2 Data Set R 2 Data Set R 2
TRN 0.9511 TRN 0.9511 TRN 0.9303
TST 0.9352 TST 0.8882 TST 0.9185
PRO well 174 0.8628 PRO well 174 0.5844 PRO well 174 0.8447
Training wells: 157, 168, 174 Training wells: 157, 168, 174 Training wells: 157, 168, 174
Verification well: 169 Verification well: 169 Verification well: 169
Data Set R 2
Data Set R 2 Data Set R
2
TRN 0.9704 TRN 0.8244 TRN 0.9354
TST 0.9413 TST 0.7873 TST 0.9138
PRO well 169 0.8815 PRO well 169 0.6898 PRO well 169 0.8869
Training wells: 157, 174, 169 Training wells: 157, 174, 169 Training wells: 157, 174, 169
Verification well: 168 Verification well: 168 Verification well: 168
Data Set R 2 Data Set R 2 Data Set R 2
TRN 0.9531 TRN 0.8155 TRN 0.9301
TST 0.9416 TST 0.7754 TST 0.9014
PRO well 168 0.8945 PRO well 168 0.76 PRO well 168 0.8811
Training wells: 174, 169, 168 Training wells: 174, 169, 168 Training wells: 174, 169, 168
Verification well: 157 Verification well: 157 Verification well: 157
Data Set R 2 Data Set R 2 Data Set R 2
TRN 0.9671 TRN 0.81 TRN 0.9461
TST 0.9564 TST 0.8103 TST 0.9328
PRO well 157 0.8825 PRO well 157 0.7172 PRO well 157 0.7420
Table 8. Results for exercise 2 and combinations A, B, and C of inputs and outputs. Southern
Pennsylvania area data set (lower zone).
52
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 62/132
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
10 100 1000
Resistivity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 28 . Synthetic resistivity log generated during verification through exercise 1 and
combination A of inputs and outputs. Well 157, southern Pennsylvania area (upper zone).
53
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 63/132
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
10 100 1000
Resistivity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 29 . Synthetic resistivity log generated during verification through exercise 1 and
combination A of inputs and outputs. Well 168, southern Pennsylvania area (upper zone).
54
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 64/132
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
10 100 1000
Resistivity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 30 . Synthetic resistivity log generated during verification through exercise 1 and
combination A of inputs and outputs. Well 169, southern Pennsylvania area (upper zone).
55
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 65/132
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
10 100 1000
Resistivity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 31 . Synthetic resistivity log generated during verification through exercise 1 and
combination A of inputs and outputs. Well 174, southern Pennsylvania area (upper zone).
56
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 66/132
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
2.5 2.6 2.7 2.8
Density (g/ccm)
D e p t h
( f e e t )
Actual
Network
Figure 32 . Synthetic density log generated during verification through exercise 1 and
combination B of inputs and outputs. Well 157, southern Pennsylvania area (upper zone).
57
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 67/132
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
2.5 2.6 2.7 2.8
Density (g/ccm)
D e p t h
( f e e t )
Actual
Network
Figure 33 . Synthetic density log generated during verification through exercise 1 and
combination B of inputs and outputs. Well 168, southern Pennsylvania area (upper zone).
58
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 68/132
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
2.5 2.6 2.7 2.8
Density (g/ccm)
D e p t h
( f e e t )
Actual
Network
Figure 34 . Synthetic density log generated during verification through exercise 1 and
combination B of inputs and outputs. Well 169, southern Pennsylvania area (upper zone).
59
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 69/132
1200
1300
1400
1500
1600
1700
1800
1900
2000
2.5 2.6 2.7 2.8
Density (g/ccm)
D e p t h
( f e e t )
Actual
Network
Figure 35 . Synthetic density log generated during verification through exercise 1 and
combination B of inputs and outputs. Well 174, southern Pennsylvania area (upper zone).
60
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 70/132
1200
1300
1400
1500
1600
1700
1800
1900
2000
15000 16000 17000 18000 19000 20000
Neutron (snu)
D e p t h
( f e e t )
Actual
Network
Figure 36 . Synthetic neutron log generated during verification through exercise 1 and
combination C of inputs and outputs. Well 157, southern Pennsylvania area (upper zone).
61
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 71/132
1200
1300
1400
1500
1600
1700
1800
1900
2000
15000 16000 17000 18000 19000 20000
Neutron (snu)
D e p t h
( f e e t )
Actual
Network
Figure 37 . Synthetic neutron log generated during verification through exercise 1 and
combination C of inputs and outputs. Well 168, southern Pennsylvania area (upper zone).
62
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 72/132
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
15000 16000 17000 18000 19000 20000
Neutron (snu)
D e p t h
( f e e t )
Actual
Network
Figure 38 . Synthetic neutron log generated during verification through exercise 1 and
combination C of inputs and outputs. Well 169, southern Pennsylvania area (upper zone).
63
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 73/132
1200
1300
1400
1500
1600
1700
1800
1900
2000
15000 16000 17000 18000 19000 20000
Neutron (snu)
D e p t h
( f e e t )
Actual
Network
Figure 39 . Synthetic neutron log generated during verification through exercise 1 and
combination C of inputs and outputs. Well 174, southern Pennsylvania area (upper zone).
64
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 74/132
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
10 100 1000
Resistivity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 40 . Synthetic resistivity log generated during verification through exercise 2 and
combination A of inputs and outputs. Well 157, southern Pennsylvania area (upper zone).
65
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 75/132
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
10 100 1000
Resistivity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 41 . Synthetic resistivity log generated during verification through exercise 2 and
combination A of inputs and outputs. Well 168, southern Pennsylvania area (upper zone).
66
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 76/132
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
10 100 1000
Resistivity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 42 . Synthetic resistivity log generated during verification through exercise 2 and
combination A of inputs and outputs. Well 169, southern Pennsylvania area (upper zone).
67
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 77/132
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
10 100 1000
Resistivity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 43 . Synthetic resistivity log generated during verification through exercise 2 and
combination A of inputs and outputs. Well 174, southern Pennsylvania area (upper zone).
68
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 78/132
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
2.4 2.5 2.6 2.7 2.8
Density (g/ccm)
D e p t h
( f e e t )
Actual
Network
37 005 27157
Figure 44 . Synthetic density log generated during verification through exercise 2 and
combination B of inputs and outputs. Well 157, southern Pennsylvania area (upper zone).
69
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 79/132
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
2.4 2.5 2.6 2.7 2.8
Density (g/ccm)
D e p t h
( f e e t )
Actual
Network
Figure 45 . Synthetic density log generated during verification through exercise 2 and
combination B of inputs and outputs. Well 168, southern Pennsylvania area (upper zone).
70
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 80/132
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
2.4 2.5 2.6 2.7 2.8
Density (g/ccm)
D e p t h
( f e e t )
Actual
Network
Figure 46 . Synthetic density log generated during verification through exercise 2 and
combination B of inputs and outputs. Well 169, southern Pennsylvania area (upper zone).
71
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 81/132
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
2.4 2.5 2.6 2.7 2.8
Density (g/ccm)
D e p t h
( f e e t )
Actual
Network
Figure 47 . Synthetic density log generated during verification through exercise 2 and
combination B of inputs and outputs. Well 174, southern Pennsylvania area (upper zone).
72
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 82/132
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
14000 16000 18000 20000 22000
Neutron (snu)
D e p t h
( f e e t )
Actual
Network
Figure 48 . Synthetic neutron log generated during verification through exercise 2 and
combination C of inputs and outputs. Well 157, southern Pennsylvania area (upper zone).
73
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 83/132
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
14000 16000 18000 20000 22000
Neutron (snu)
D e p t h
( f e e t )
Actual
Network
Figure 49 . Synthetic neutron log generated during verification through exercise 2 and
combination C of inputs and outputs. Well 168, southern Pennsylvania area (upper zone).
74
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 84/132
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
14000 16000 18000 20000 22000
Neutron (snu)
D e p t h
( f e e t )
Actual
Network
Figure 50 . Synthetic neutron log generated during verification through exercise 2 and
combination C of inputs and outputs. Well 169, southern Pennsylvania area (upper zone).
75
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 85/132
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
14000 16000 18000 20000 22000
Neutron (snu)
D e p t h
( f e e t )
Actual
Network
Figure 51 . Synthetic neutron log generated during verification through exercise 2 and
combination C of inputs and outputs. Well 174, southern Pennsylvania area (upper zone).
76
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 86/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
10 100
Resistivity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 52. Synthetic resistivity log generated during verification through exercise 1 and
combination A of inputs and outputs. Well 157, southern Pennsylvania area (lower zone).
77
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 87/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
10 100
Resistivity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 53. Synthetic resistivity log generated during verification through exercise 1 and
combination A of inputs and outputs. Well 168, southern Pennsylvania area (lower zone).
78
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 88/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
10 100
Resistivity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 54. Synthetic resistivity log generated during verification through exercise 1 and
combination A of inputs and outputs. Well 169, southern Pennsylvania area (lower zone).
79
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 89/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
10 100
Resistivity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 55. Synthetic resistivity log generated during verification through exercise 1 and
combination A of inputs and outputs. Well 174, southern Pennsylvania area (lower zone).
80
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 90/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
2.5 2.6 2.7 2.8
Density (g/ccm)
D e p t h
( f e e t )
Actual
Network
Figure 56. Synthetic density log generated during verification through exercise 1 and
combination B of inputs and outputs. Well 157, southern Pennsylvania area (lower zone).
81
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 91/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
2.5 2.6 2.7 2.8
Density (g/ccm)
D e p t h
( f e e t )
Actual
Network
Figure 57. Synthetic density log generated during verification through exercise 1 and
combination B of inputs and outputs. Well 168, southern Pennsylvania area (lower zone).
82
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 92/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
2.5 2.6 2.7 2.8
Density (g/ccm)
D e p t h
( f e e t )
Actual
Network
Figure 58. Synthetic density log generated during verification through exercise 1 and
combination B of inputs and outputs. Well 169, southern Pennsylvania area (lower zone).
83
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 93/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
2.5 2.6 2.7 2.8
Density (g/ccm)
D e p t h
( f e e t )
Actual
Network
Figure 59. Synthetic density log generated during verification through exercise 1 and
combination B of inputs and outputs. Well 174, southern Pennsylvania area (lower zone).
84
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 94/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
15000 16000 17000 18000 19000 20000
Neutron (snu)
D e p t h
( f e e t )
Actual
Network
Figure 60. Synthetic neutron log generated during verification through exercise 1 and
combination C of inputs and outputs. Well 157, southern Pennsylvania area (lower zone).
85
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 95/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
15000 16000 17000 18000 19000 20000
Neutron (snu)
D e p t h
( f e e t )
Actual
Network
Figure 61. Synthetic neutron log generated during verification through exercise 1 and
combination C of inputs and outputs. Well 168, southern Pennsylvania area (lower zone).
86
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 96/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
15000 16000 17000 18000 19000 20000
Neutron (snu)
D e p t h
( f e e t )
Actual
Network
Figure 62. Synthetic neutron log generated during verification through exercise 1 and
combination C of inputs and outputs. Well 169, southern Pennsylvania area (lower zone).
87
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 97/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
15000 16000 17000 18000 19000 20000
Neutron (snu)
D e
p t h
( f e e t )
Actual
Network
Figure 63. Synthetic neutron log generated during verification through exercise 1 and
combination C of inputs and outputs. Well 174, southern Pennsylvania area (lower zone).
88
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 98/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
10 100
Resistiv ity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 64. Synthetic resistivity log generated during verification through exercise 2 and
combination A of inputs and outputs. Well 157, southern Pennsylvania area (lower zone).
89
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 99/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
10 100
Resistivity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 65. Synthetic resistivity log generated during verification through exercise 2 and
combination A of inputs and outputs. Well 168, southern Pennsylvania area (lower zone).
90
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 100/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
10 100
Resistivity (ohm-m)
D e p t h
( f e e t )
Actual
Network
Figure 66. Synthetic resistivity log generated during verification through exercise 2 and
combination A of inputs and outputs. Well 169, southern Pennsylvania area (lower zone).
91
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 101/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
10 100
Resistivity (ohm-M)
D e p t h
( f e e t )
Actual
Network
Figure 67. Synthetic resistivity log generated during verification through exercise 2 and
combination A of inputs and outputs. Well 174, southern Pennsylvania area (lower zone).
92
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 102/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
2.5 2.55 2.6 2.65 2.7 2.75 2.8
Density (ccm)
D e p t h
( f e e t )
Actual
Network
Figure 68. Synthetic density log generated during verification through exercise 2 and
combination B of inputs and outputs. Well 157, southern Pennsylvania area (lower zone).
93
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 103/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
2.5 2.6 2.7 2.8
Density (gr/ccm)
D e p t h
( f e e t )
Actual
Network
Figure 69. Synthetic density log generated during verification through exercise 2 and
combination B of inputs and outputs. Well 168, southern Pennsylvania area (lower zone).
94
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 104/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
2.5 2.6 2.7 2.8
Density (gr/ccm)
D e p t h
( f e e t )
Actual
Network
Figure 70. Synthetic density log generated during verification through exercise 2 and
combination B of inputs and outputs. Well 169, southern Pennsylvania area (lower zone).
95
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 105/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
2.5 2.6 2.7 2.8
Density (gr/ccm)
D e p t h
( f e e t )
Actual
Network
Figure 71. Synthetic density log generated during verification through exercise 2 and
combination B of inputs and outputs. Well 174, southern Pennsylvania area (lower zone).
96
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 106/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
14000 16000 18000 20000 22000
DNND (snu)
D e p t h
( f e e t )
Actual
Network
Figure 72. Synthetic neutron log generated during verification through exercise 2 and
combination C of inputs and outputs. Well 157, southern Pennsylvania area (lower zone).
97
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 107/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
14000 16000 18000 20000 22000
DNND (snu)
D e p t h
( f e e t )
Actual
Network
Figure 73. Synthetic neutron log generated during verification through exercise 2 and
combination C of inputs and outputs. Well 168, southern Pennsylvania area (lower zone).
98
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 108/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
14000 16000 18000 20000 22000
DNND (snu)
D e p t h
( f e e t )
Actual
Network
Figure 74. Synthetic neutron log generated during verification through exercise 2 and
combination C of inputs and outputs. Well 169, southern Pennsylvania area (lower zone).
99
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 109/132
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
14000 16000 18000 20000 22000
DNND (snu)
D e p t h
( f e e t )
Actual
Network
Figure 75. Synthetic neutron log generated during verification through exercise 2 and
combination C of inputs and outputs. Well 174, southern Pennsylvania area (lower zone).
100
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 110/132
6. DISCUSSION
Results discussed in chapter 5, for exercises 1 and 2, indicate that the best neural network
model performance was obtained in general for combination “A”. Combination “C” was
ranked second and combination “B” represents the lowest performance (figures 76, 77,
78, and 79).
These results indicate that the most predictable was the resistivity log and the less
predictable was the density log. The inferior response of the neural network model to
combination “B” is reflected in the way that the network captures the deflections of the
log. The high peaks of the log where high contrasts of density are present are captured
with high accuracy, conversely, small changes cannot be captured accurately but in these
cases the network averaged the values. An explanation for the lower degree of
predictability of the density log is due to radioactive fluctuations (relative to the cesium
source) during the logging operation. Hence, radiations can take different ways at each
time and log response can vary from place to place.
For exercise 2 the best results, were obtained, for wells number 168 and 169 (figures 80
and 81 ). The reason is probably because these wells are located at the middle of the cross
section A-A’, so the neural network model can interpolate information from adjacent
wells. An explanation for anomalous results obtained for combination “A” at the upper
zone, could not be resolved by this study. However, it is important to be mention that,
generally speaking, geology in the study area is simple. Therefore, it is possible that for
areas with more complex geology, this condition could change, and the location of the
wells in terms of geometry, could not have any relation with the performance of the
neural network.
101
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 111/132
A very interesting point to discuss in this study is the fact that poor results were obtained
because of the quality of the data, as a consequence of digitalization of the logs in
original .tif format. These poor results are indicative of the high degree of sensitivity that
neural network models have to the quality of the data. It is highly recommended for
future studies involving generation of intelligent synthetic logs, to perform a strict quality
control of data prior to building the neural network model, especially if the logs that will
be used as inputs in the model are not directly obtained from the borehole, but have been
digitized. It was demonstrated that once it was realized that data from Buffalo Valley had
low quality as a result of the imprecise digitizing, and they were replaced by data from
Southern Pennsylvania, the effectiveness of the neural network models in predicting logs
for exercises 1 and 2 improved considerably. Furthermore, before data from the upper
zone of the Pennsylvanian data set were cleaned, the results were poor, once the
anomalous data were detected and cleaned, results improved noticeably.
Finally, it is important to mention that the main reason why logs of the Pennsylvanian
area were split into two intervals was because it was desired to determine if heterogeneity
of the rocks influenced the performance of the network model. It was demonstrated for
all models built in this study that heterogeneities in lithology of the reservoir have low
influence for the networks since R 2 values obtained for both zones were very similar to
each other.
102
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 112/132
Exercise 1 - Upper Zone (1000' to 2000')
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
TRN TST PRO PROwell 157
PROwell 168
PROwell 169
PROwell 174
R - s q u a r e d
Combination A
Combination B
Combination C
Figure 76. R 2 values obtained for the upper zone of the Southern Pennsylvania data set, through exercise 1.
Exercise 1 - Lower Zone (2500' to 3500')
0.0000
0.1000
0.2000
0.3000
0.4000
0.5000
0.6000
0.7000
0.8000
0.9000
1.0000
TRN TST PRO PRO
well
157
PRO
well
168
PRO
well
169
PRO
well
174
R - s q u a r e d
Combination A
Combination B
Combination C
Figure 77. R 2 values obtained for the lower zone of the Southern Pennsylvania data set, through exercise 1.
103
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 113/132
Exercise 2 - Upper Zone (1000' to 2000')
0.0000
0.1000
0.2000
0.3000
0.4000
0.5000
0.6000
0.7000
0.8000
0.9000
1.0000
PRO well 157 PRO well 168 PRO well 169 PRO well 174
R - s q u a r e d
Combination A
Combination B
Combination C
Figure 78. R 2 values obtained for the upper zone of the Southern Pennsylvania data set, through exercise 1.
Exercise 2 - Lower Zone (2500' to 3500')
0.0000
0.1000
0.2000
0.3000
0.4000
0.5000
0.6000
0.7000
0.8000
0.9000
1.0000
PRO well 157 PRO well 168 PRO well 169 PRO well 174
R - s q u a r e d
Combination A
Combination B
Combination C
Figure 79. R 2 values obtained for the lower zone of the Southern Pennsylvania data set, through exercise 1.
104
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 114/132
Exercise 2 - Upper Zone (1000' - 2000')
0.0000
0.1000
0.2000
0.3000
0.4000
0.5000
0.6000
0.7000
0.8000
0.9000
1.0000
Combination A Combination B Combination C
R - s q u a r e d Well 157
Well 168
Well 169
Well 174
Figure 80. R 2 values obtained for the upper zone of the Southern Pennsylvania data set, through exercise 2.
Exercise 2 - Lower Zone
0.0000
0.1000
0.2000
0.3000
0.4000
0.5000
0.6000
0.7000
0.8000
0.9000
1.0000
Combination A Combination B Combination C
R - s q u a r e d Well 157
Well 168
Well 169
Well 174
Figure 81. R
2 values obtained for the lower zone of the Southern Pennsylvania data set, through exercise 2.
105
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 115/132
7. CONCLUSIONS
This work demonstrates that generation of synthetic logs with a reasonable degree of
accuracy is possible by using a neural network model and following the methodology
herein described.
Three neural network models to predict resistivity, density, and neutron logs were built
through exercises 1 and 2, as well as using different combinations of inputs and outputs,
namely combination “A” to predict resistivity logs, combination “B” to predict density
logs, and combination “C” to predict neutron logs.
Results indicate that the best performance was obtained for combination “A” of inputs
and outputs, then for combination “C”, and finally for combination “B”. Therefore
performance for combination “A” indicates that the resistivity log was the most
predictable log. On the other hand performance in combination “B” demonstrates that
density log was the least predictable as a consequence of radioactive fluctuations of the
Cs source during the logging operation.
Results demonstrate as well that in areas where geology is simple, as is the case of the
study of this work, accuracy of synthetic logs may be favored by interpolation of data.
Therefore, the best results were obtained for wells located in the central part of the cross
section studied. This condition could change in areas where geology presents more
complexities.
106
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 116/132
It was demonstrated that in studies that involve generation of intelligent synthetic logs
quality of data plays a very important role during the development of the neural network
model. Quality of logs could be defective when logs are not available in digital format
and have to be digitized. Therefore it is highly recommended for future works to perform
a very careful quality control of the data before a neural network model is built.
It was also demonstrated that lithologic heterogeneities in the reservoir do not
significantly affect performance of a neural network model in the generation of synthetic
logs.
107
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 117/132
REFERENCES
Ameri, S., 2004. Notes for the Advanced formation Evaluation course. West Virginia
University, Morgantown, West Virginia.
Amizadeh, F., 2001. Soft computing information applications for the fusion and analysis
of large data sets: Fact Inc./dGB-USA. University of Berkeley, California.Banchs, R. and Michelena, R., 2002. From 3D seismic attributes to pseudo-well-log
volumes using neural networks: Practical considerations. The Leading Edge, Vol
21, No.10 , pp 996-1001.
Bassiouni, Z., 1994. Theory, measurement, and interpretation of well logs. SPE textbook
series, vol. 4, 372 p.
Bhuiyan, M., 2001. An Intelligent System’s Approach to Reservoir Characterization inCotton Valley. M.S. Thesis, West Virginia University, Morgantown, West
Virginia.
Boswell, R. M., and Donaldson, A. C., 1988. Depositional architecture of the Catskilldelta complex; central Appalachian basin, U.S., in Canadian Society of
PetroleumGeologists Memoir 14, pp. 65-84. Boswell, R.M., Heim, L.R., Wrightstone, G.R., and Donaldson, A.C., 1996. Play Dvs:
Upper Devonian Venango sandstones and siltstones in Roen, J.B. and Walker,B.J. (Eds), The Atlas of Major Appalachian Gas Plays, Publication V-25, West
Virginia Geologicaland Economic Survey, Morgantown, WV. Pp. 63-69.
Cant, D., 1992. Subsurface facies analysis, in Walker, R. and James, N., eds. FaciesModels: response to sea level change. Geological Association of Canada, 454 p.
Faucett, L., 1994. Fundamentals of neural networks. Architectures, Algorithms, and
applications. Prentice Hall, Englewood Cliffs. NJ.
Hebb, D., 1949. The organization of behavior: Wiley
Jaeger, H., 2002. Supervised training of recurrent neural networks. An INDY MathBooster tutorial, AIS, September / October 2002.
Kammer, T.W., and Bjerstedt, T.W., 1988. Genetic stratigraphy and depositional
systems of the Upper Devonian-Lower Mississippian Price-Rockwell deltacomplex in the central Appalachians. USA: Sedimentary Geology, v. 54, p 265-
301.
Lawrence, S., Giles, C. L., and Tsoi, A., 1997. Lessons in Neural Network Training:
Training may be harder than Expected. Proceedings of the Fourteenth NationalConference on Artificial Intelligence, AAAI-97, (pp. 540-545), Menlo Park,
California: AAAI Press.
McBride, P., 2004. Facies Analysis of the Devonian Gordon Stray Sandstone in WestVirginia. M.S. Thesis, West Virginia University, Morgantown, West Virginia.
Mohaghegh, S., Popa, A., Koperna, G., 1999 Reducing the Cost of Field-Scale Log
Analysis Using Virtual Intelligence Techniques. SPE 57454, Proceedings, SPEEastern Regional Conference Oct. 21-22, Charleston, WV.
Mohaghegh, S., Richardson, M., Ameri, S., 1998 Virtual Magnetic Imaging Logs:
Generation of Synthetic MRI Logs from Conventional Well Logs. SPE 51075,Proceedings, SPE Eastern Regional Conference Nov. 9-11, Pittsburgh, PA.
108
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 118/132
Mohaghegh, S., Wilson T., & Toro J., 2002. An intelligent system approach to
reservoir characterization. Technical application. West Virginia University. Norwegian Formation Evaluation Society, 2002 / 2003 Season Stavanger May
Meeting.
Pofahl, W., Walczak, S., Rhone, E., and Izenberg, S., 1998. Use of an Artificial
Neural Network to Predict Length of Stay in Acute Pancreatitis. AmericanSurgeon, Sep98, Vol. 64 Issue 9, (pp: 868 – 872)
Poulton, M., 2002. Neural networks as an intelligence amplification tool: A review of
applications. Geophysics, Vol. 67, No.3. p 979-993.
Prechelt, L., 1998. Early Stopping-but when? Neural Networks: Tricks of the trade, (pp.
55-69). Retrieved March 28, 2002 from World Wide Web:
http://wwwipd.ira.uka.de/~prechelt/Biblio/
Sarle, W., 1997. Neurak Network FAQ, part 1 of 7: Introduction. Periodic posting to the
Usenet newsgroup comp.ai.neural-nets.
URL:http://ftp.sas.com/pub/neural/FAQ.html
Specht, D., 1991. A General Regression Neural Network. IEEE Transactions Of Neural
Networks. Vol 2, No. 6Tonn, T., 2002. Neural network seismic reservoir characterization in a heavy oilreservoir. The Leading Edge, Vol 21, No.3 , pp 309-312.
White, D., Aminiam K., Mohaghegh S., & Esposito P., 1995. The application of ANNfor Zone identification in a complex reservoir: Society of Petroleum Engineers.
SPE Eastern Regional Conference & Exhibiton.
109
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 119/132
110
APPENDIX A: Training results (Southern Pennsylvania data set)
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 120/132
111
0
1000
2000
3000
4000
5000
10 100 1000
Resistivity (ohm-m)
I n d e x
Actual
Network
Training results for combination A of inputs and outputs. Southern Pennsylvania data set, upper zone.
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 121/132
112
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
5500
6000
6500
7000
2.5 2.6 2.7 2.8
Density (g/ccm)
I n d e x
Actual
Network
Training results for combination B of inputs and outputs. Southern Pennsylvania data set, upper zone.
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 122/132
113
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
5500
15000 16000 17000 18000 19000 20000
Neutron (snu)
I n d e x
Actual
Network
Training results for combination C of inputs and outputs. Southern Pennsylvania data set, upper zone.
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 123/132
114
0
1000
2000
3000
4000
5000
6000
10 100
Resistivity (ohm-m)
I n d e x
Actual
Network
Training results for combination A of inputs and outputs. Southern Pennsylvania data set, lower zone.
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 124/132
115
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
5500
6000
6500
2.5 2.6 2.7 2.8
Density (g/ccm)
I n d e x
Actual
Network
Training results for combination B of inputs and outputs. Southern Pennsylvania data set, lower zone.
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 125/132
116
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
5500
6000
15000 16000 17000 18000 19000 20000
Neutron (snu)
I n d e x
Actual
Network
Training results for combination C of inputs and outputs. Southern Pennsylvania data set, lower zone.
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 126/132
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 127/132
118
0
1000
2000
3000
4000
5000
10 100 1000
Resistivity (ohm-m)
I n d e x
Actual
Network
Calibration results for combination A of inputs and outputs. Southern Pennsylvania data set, upper zone.
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 128/132
119
0
100
200
300
400
500
600
700
800
900
1000
1100
2.5 2.6 2.7 2.8
Density (g/ccm)
I n d e x
Actual
Network
Calibration results for combination B of inputs and outputs. Southern Pennsylvania data set, upper zone.
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 129/132
120
0
100
200
300
400
500
600
700
800
900
1000
1100
1200
1300
1400
1500
1600
15000 16000 17000 18000 19000 20000
Neutron (snu)
I n d e x
Actual
Network
Calibration results for combination C of inputs and outputs. Southern Pennsylvania data set, upper zone.
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 130/132
121
0
100
200
300
400
500
600
700
800
900
1000
1100
10 100
Resistivity (ohm-m)
I n d e x
Actual
Network
Calibration results for combination A of inputs and outputs. Southern Pennsylvania data set, lower zone.
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 131/132
122
Calibration results for combination B of inputs and outputs. Southern Pennsylvania data set, lower zone.
0
100
200
300
400
500
600
700
800
900
1000
1100
1200
2.5 2.6 2.7 2.8
Density (g/ccm)
I n d e x
Actual
Network
7/24/2019 Developing Intelligent Synthetic Logs Application to Upper Devonian Units in PA.pdf
http://slidepdf.com/reader/full/developing-intelligent-synthetic-logs-application-to-upper-devonian-units-in 132/132
0
100
200
300
400
500
600
700
15000 16000 17000 18000 19000 20000
Neutron (snu)
I n d e x
Actual
Network