Professional Documents
Culture Documents
ADVANCED FORMATION
EVALUATION
Day 4: Machine learning in Petrophysics
1
Machine Learning Techniques
1. Neural Networks
3. Genetic Algorithm
4. Fuzzy Logic
2
Machine learning Concept
3
Machine Learning Techniques
1. Neural Networks
3. Genetic Algorithm
4. Fuzzy Logic
4
Neural Networks Concept
5
Neural Networks Concept
An Artificial Neuron
(Processing Element in computer)
6
Neural Networks Concept
Architecture is designed
COMMON ACTIVATION FUNCTIONS
to solve for each problem
Multi-layer feed-forward
Supervised Learning
Unsupervised Learning
8
Neural Network Example (Supervised Learning)
Random weights
ANN architecture
population
(unit function)
(unit function)
9
Neural Network Example (Supervised Learning)
Random weights
ANN architecture
population
10
Neural Network Example (Supervised Learning)
Random weights
ANN architecture
population
11
Neural Network Example (Supervised Learning)
If
Then
12
Neural Network Example (Supervised Learning)
Next, how much does the output of o1 change with respect to its total net input?
Finally, how much does the total net input of o1 change with respect to w5?
To decrease the error, we then subtract this value from the current weight (optionally
multiplied by some learning rate, eta , which we’ll set to 0.5):
13
Neural Network Example (Supervised Learning)
We perform the actual updates in the neural network after we have the new weights
leading into the hidden layer neurons (ie, we use the original weights, not the updated
weights, when we continue the backpropagation algorithm)
Next, we’ll continue the backwards pass by calculating new values for w1, w2, w3,
w4
Finally, we’ve updated all of our weights! When we feed forward the 0.05 and 0.1
inputs originally, the error on the network was 0.298371109. After this first round of
backpropagation, the total error is now down to 0.291027924. It might not seem like
much, but after repeating this process 10,000 times, for example, the error plummets
to 0.0000351085. At this point, when we feed forward 0.05 and 0.1, the two outputs
neurons generate 0.015912196 (vs 0.01 target) and 0.984065734 (vs 0.99 target).
15
Common Back Propagation Algorithms
16
Neural Networks Tool in IP
17
Neural Networks – Permeability Prediction
18
Neural Networks – Permeability Prediction
19
Neural Networks – Rock types
20
Neural Networks – Rock types
21
Neural Networks – Rock types
22
Flow Zone Indicator Prediction
23
Rock Type Prediction
24
Rock Type Prediction
Which backpropagation
algorithm is perfect in
this case?
25
Neural Networks Limit
Training data should be covered all the variety of the data (possible rock
types). I.e. If training for 4 rock types but the formation has 5 rock types
then the prediction will be in-completed (RT5 will be recognized as RT1-
RT4).
Data should be good, “garbage in garbage out”. Neural networks can
not eliminate the garbage by itself (intelligent but not smart!). Therefore
data preparation is important.
The first weighs set is randomly populated. Therefore the learning can
be difference for the same data set and same algorithm.
Efficiency of learning algorithms are variable depending on how
complicate of the training data, how structure of the ANN (normal or
deep learning). Selection of the algorithms is still based on try-and-error
basis -> need further investigation?
ANN tool in IP is not deep learning and only 1 simple learning algorithm
The training is not always successful, it can be over-trained. Therefore
selection of testing data is very important, normally 10% data should be
used for testing.
26
Machine Learning Techniques
1. Neural Networks
3. Genetic Algorithm
4. Fuzzy Logic
27
Self Organizing Maps (SOM)
28
Self Organizing Maps
Lattice (map) 1. Take a set of input vectors (data) and present it to a map grid of 2 dimensions
Neuron = 2. The nodes (neurons) in the map each have a weighting value for each of the
Clusters Neuron Neuron different input vectors (i.e. color columns)
3. Training process:
Outputs (m) a) The weights in each node are initially random values
Neuron Neuron b) “Best matching unit” is calculated by the Euclidean
distance between each node’s weight vector and the
Neuron
chosen input vector as:
Neuron
Weight (wj)
SOM distortion (độ méo) where V is the current input vector and W is the node’s weight
vector. The node where this Distance calculates to the lowest
Input vectors
GR TNPH RHON DTCO value is the “Best Matching Unit”.
Pythagoras’ Theorem c) Once the winning node is determined its weight vectors are
Best Matching Unit (BMU) adjusted to become more similar to the input vector by the
σ: radius of the neighborhood
following equation:
Default in IP
Clustering Methods
1. Minimum distance between all objects in clusters—the distance from Z to C is the minimum of the distances (A to C
and B to C).
2. Maximum distance between all objects in clusters—the distance from Z to C is the maximum of the distances (A to C
and B to C).
3. Average distance between merged clusters—the distance from Z to C is the average distance of all objects that would
be within the cluster formed by merging clusters and C.
4. Average distance between all objects in clusters—the distance from Z to C is the average distance of objects within
cluster Z to objects within cluster C.
5. Minimize the within-cluster sum of squares distance— clusters are formed so as to minimize the increase in the within-
cluster sums of squares. The distance between the two clusters is the increase in these sums of squares if the two
clusters were merged.
30
31
Grouping Clusters Methods
• Clustering method is for grouping the different log lithology into different RT or Facies clusters.
• Hierarchical clustering computes the distances between all nodes and then merging the two closest nodes together to form
a grouping.
• This process is repeated where the distance to all other groups is computed and the two closest groups can then be merged
again.
• This continues until only one group of all nodes remains. The module can output a dendogram which shows the order in
which the nodes and groups were merged.
• The good separation, the good clustering method.
Maximum distance between all objects in clusters Minimize the within-cluster sum of squares distance => The best one
FC-4
Facies Cluster (FC) 1 FC-6
FC-2 FC-5
FC-3
Log Lithology Cluster
Log Lithology Cluster
LTC
LTC
LTC
LTC
LTC
LTC
LTC
LTC
LTC
LTC
Average distance between merged clusters Dendogram Average distance between all objects in clusters
Distance of Clusters
31
32
Define Number of Clusters Group
noise. 5
32
33
Evaluate a SOM
1. SOM distortion calculation as a training quality. The lower this number, the less 'distorted' the SOM training is and therefore the better the
quality of that training
- The distortion number is used to select the SOM geometries (square, Hexagonal and Spherical)
- This number around 1.0 is probably fine
33
SOM in IP
34
SOM in IP
35
SOM in IP
36
SOM in IP
37
SOM in IP
38
39
Cores Description Summary
LITHOFACIES
TMC#6
TMC#8
TMC#11
39
40
SOM Models Definition
TMC#6U
Constrained by – TMC#6L
TMC#6 Core
TMC#7 –
Constrained by TMC#9
TMC#8 Core
TMC#10
Constrained by
TMC#11 Core
–
TMC#13
40
41
SOM Model in TMC#6U & TMC#6L
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Number of Clusters
41
42
SOM Map TMC#6
42
43
Cross Plots
Multi-Curve Crossplot
700
GR (gAPI)
GR (gAPI)
GR (gAPI)
GR (gAPI)
600 80 80 80 80
500
60 60 60 60
400
300 40 40 40 40
200
100 20 20 20 20
0
0 50 100 150 0 0 0 0
-0.1 0 0.1 0.2 0.3 0.4 2.8 2.6 2.4 2.2 2 40 60 80 100 120 140 120 140 160 180
GR (gA PI) TNPH (V /V ) RHON (G/C3) DTCO (US/F) DTSM (US/F)
2,400
0.4 2,200 0.4 0.4 0.4
0.35 2,000 0.35 0.35 0.35
0.3 1,800 0.3 0.3 0.3
0.25 1,600 0.25 0.25 0.25
Frequency
TNPH (V/V)
TNPH (V/V)
TNPH (V/V)
TNPH (V/V)
0.2 1,400 0.2 0.2 0.2
0.15 1,200 0.15 0.15 0.15
0.1 1,000 0.1 0.1 0.1
800
0.05 0.05 0.05 0.05
600
0 0 0 0
400
-0.05 -0.05 -0.05 -0.05
200
-0.1 -0.1 -0.1 -0.1
0
-0.15 0 0.2 0.4 -0.15 -0.15 -0.15
0 50 100 150 2.8 2.6 2.4 2.2 2 40 60 80 100 120 140 120 140 160 180
GR (gA PI) TNPH (V /V ) RHON (G/C3) DTCO (US/F) DTSM (US/F)
2 2 1,100 2 2
1,000
2.1 2.1 2.1 2.1
900
2.2 2.2 2.2 2.2
800
2.3 2.3 2.3 2.3
RHON (G/C3)
RHON (G/C3)
RHON (G/C3)
RHON (G/C3)
700
Frequency
2.4 2.4 600 2.4 2.4
2.5 2.5 500 2.5 2.5
400
2.6 2.6 2.6 2.6
300
2.7 2.7 2.7 2.7
200
2.8 2.8 100 2.8 2.8
DTCO (US/F)
DTCO (US/F)
DTCO (US/F)
Frequency
100 100 100 100
90 90 90 2,000 90
80 80 80 1,500 80
70 70 70 1,000 70
60 60 60 60
500
50 50 50 50
0
40 40 40 40 60 80 100 120 140 40
0 50 100 150 -0.1 0 0.1 0.2 0.3 0.4 2.8 2.6 2.4 2.2 2 120 140 160 180
GR (gA PI) TNPH (V /V ) RHON (G/C3) DTCO (US/F) DTSM (US/F)
1,600
190 190 190 190
1,400
180 180 180 180
1,200
170 170 170 170
DTSM (US/F)
DTSM (US/F)
DTSM (US/F)
DTSM (US/F)
Frequency
160 160 160 160 1,000
0 1 2 3 4 5 43 6
44
Star Plots
44
45
Rock Types in TMC#6 Core
Calcite cemented
Lower-shoreface
SST 4 main facies in TMC#6
3
Arg. Bioturbated
Lower-shoreface
SST
2
Arg. Bioturbated
Lower-shoreface
Mudstone
Arg. Bioturbated
Lower-shoreface
SST
1
Bioturbated
Mid-shoreface
SST
4 45
46
Rock Types in TMC#6 Core
46
47
Rock Types in TMC#6
SVW-1X SV-1XST SV-1X SV-4P SV-3P SV-5P SV-1P SV-2P
47
48
SOM Model in TMC#7 - TMC#9
4.9
4.8
4 main facies
4.7 Possible 7 facies
4.6
4.4
4.3
4.2
4.1
3.9
3.8
3.7
3.6
3.5
3.4
3.3
3.2
3.1
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Number of Clusters
48
49
SOM Map TMC#7 – TMC#9
49
50
Cross Plots
Multi-Curve Crossplot
GR (gAPI)
GR (gAPI)
GR (gAPI)
GR (gAPI)
80 80 80 80
600
60 60 60 60
400
40 40 40 40
200
20 20 20 20
0
0 50 100 150 0 0 0 0
-0.1 0 0.1 0.2 0.3 0.4 2.8 2.6 2.4 2.2 2 40 60 80 100 120 140 120 140 160 180
GR (gA PI) TNPH (V /V ) RHON (G/C3) DTCO (US/F) DTSM (US/F)
2,500
0.4 0.4 0.4 0.4
0.35 0.35 0.35 0.35
0.3 2,000 0.3 0.3 0.3
0.25 0.25 0.25 0.25
Frequency
TNPH (V/V)
TNPH (V/V)
TNPH (V/V)
TNPH (V/V)
0.2 1,500 0.2 0.2 0.2
0.15 0.15 0.15 0.15
0.1 1,000 0.1 0.1 0.1
0.05 0.05 0.05 0.05
0 500 0 0 0
-0.05 -0.05 -0.05 -0.05
-0.1 -0.1 -0.1 -0.1
0
-0.15 0 0.2 0.4 -0.15 -0.15 -0.15
0 50 100 150 2.8 2.6 2.4 2.2 2 40 60 80 100 120 140 120 140 160 180
GR (gA PI) TNPH (V /V ) RHON (G/C3) DTCO (US/F) DTSM (US/F)
2 2 1,200 2 2
1,100
2.1 2.1 2.1 2.1
1,000
2.2 2.2 900 2.2 2.2
RHON (G/C3)
RHON (G/C3)
RHON (G/C3)
Frequency
700
2.4 2.4 2.4 2.4
600
2.5 2.5 500 2.5 2.5
DTCO (US/F)
DTCO (US/F)
DTCO (US/F)
Frequency
100 100 100 2,500 100
90 90 90 90
2,000
80 80 80 80
1,500
70 70 70 70
1,000
60 60 60 60
500
50 50 50 50
0
40 40 40 40 60 80 100 120 140 40
0 50 100 150 -0.1 0 0.1 0.2 0.3 0.4 2.8 2.6 2.4 2.2 2 120 140 160 180
GR (gA PI) TNPH (V /V ) RHON (G/C3) DTCO (US/F) DTSM (US/F)
1,800
190 190 190 190
1,600
180 180 180 180
1,400
170 170 170 170
1,200
DTSM (US/F)
DTSM (US/F)
DTSM (US/F)
DTSM (US/F)
Frequency
160 160 160 160
1,000
150 150 150 150
800
140 140 140 140 600
130 130 130 130 400
120 120 120 120 200
110 110 110 110 0
120 140 160 180
0 50 100 150 -0.1 0 0.1 0.2 0.3 0.4 2.8 2.6 2.4 2.2 2 40 60 80 100 120 140
GR (gA PI) TNPH (V /V ) RHON (G/C3) DTCO (US/F) DTSM (US/F)
0 1 2 3 4 5 50 6
51
Star Plots
51
52
Rock Types in TMC#8 Core
Argillaceous
Bioturbated SST Cross-laminated
and Bioturbated
SST
Bioclastic-
laminated SST
5
Clay-laminated SST
Bioturbated Clay-
4 laminated SST
52
53
Rock Types in TMC#8 Core
53
54
Rock Types in TMC#8
SVW-1X SV-1XST SV-1X SV-4P SV-3P SV-5P SV-1P SV-2P
54
55
Rock Types in TMC#7 – TMC#9
SVW-1X SV-1XST SV-1X SV-4P SV-3P SV-5P SV-1P SV-2P
TMC#7
TMC#8
TMC#9
55
56
SOM Model in TMC#10 - TMC#13
4.4
4.3
4 main facies
4.2 Possible 6 facies
4.1
3.9
3.8
3.7
3.6
3.5
3.4
3.3
3.2
3.1
2.9
2.8
2.7
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Number of Clusters
56
57
SOM Map TMC#10 – TMC#13
57
58
Cross Plots
Multi-Curve Crossplot
GR (gAPI)
GR (gAPI)
GR (gAPI)
GR (gAPI)
80 80 80 80
600
60 60 60 60
400
40 40 40 40
200
20 20 20 20
0
0 50 100 150 0 0 0 0
-0.1 0 0.1 0.2 0.3 0.4 2.8 2.6 2.4 2.2 2 40 60 80 100 120 140 120 140 160 180
GR (gA PI) TNPH (V /V ) RHON (G/C3) DTCO (US/F) DTSM (US/F)
2,500
0.4 0.4 0.4 0.4
0.35 0.35 0.35 0.35
0.3 2,000 0.3 0.3 0.3
0.25 0.25 0.25 0.25
Frequency
TNPH (V/V)
TNPH (V/V)
TNPH (V/V)
TNPH (V/V)
0.2 1,500 0.2 0.2 0.2
0.15 0.15 0.15 0.15
0.1 1,000 0.1 0.1 0.1
0.05 0.05 0.05 0.05
0 500 0 0 0
-0.05 -0.05 -0.05 -0.05
-0.1 -0.1 -0.1 -0.1
0
-0.15 0 0.2 0.4 -0.15 -0.15 -0.15
0 50 100 150 2.8 2.6 2.4 2.2 2 40 60 80 100 120 140 120 140 160 180
GR (gA PI) TNPH (V /V ) RHON (G/C3) DTCO (US/F) DTSM (US/F)
2 2 1,200 2 2
1,100
2.1 2.1 2.1 2.1
1,000
2.2 2.2 900 2.2 2.2
RHON (G/C3)
RHON (G/C3)
RHON (G/C3)
Frequency
700
2.4 2.4 2.4 2.4
600
2.5 2.5 500 2.5 2.5
DTCO (US/F)
DTCO (US/F)
DTCO (US/F)
Frequency
100 100 100 2,500 100
90 90 90 90
2,000
80 80 80 80
1,500
70 70 70 70
1,000
60 60 60 60
500
50 50 50 50
0
40 40 40 40 60 80 100 120 140 40
0 50 100 150 -0.1 0 0.1 0.2 0.3 0.4 2.8 2.6 2.4 2.2 2 120 140 160 180
GR (gA PI) TNPH (V /V ) RHON (G/C3) DTCO (US/F) DTSM (US/F)
1,800
190 190 190 190
1,600
180 180 180 180
1,400
170 170 170 170
1,200
DTSM (US/F)
DTSM (US/F)
DTSM (US/F)
DTSM (US/F)
Frequency
160 160 160 160
1,000
150 150 150 150
800
140 140 140 140 600
130 130 130 130 400
120 120 120 120 200
110 110 110 110 0
120 140 160 180
0 50 100 150 -0.1 0 0.1 0.2 0.3 0.4 2.8 2.6 2.4 2.2 2 40 60 80 100 120 140
GR (gA PI) TNPH (V /V ) RHON (G/C3) DTCO (US/F) DTSM (US/F)
0 1 2 3 4 5 58 6
59
Star Plots
59
60
Rock Types in TMC#11 Core
Clay-laminated
SST
Bioturbated SST
Bioclastic
Laminated SST
Ripple Coss-
laminated SST
1
Coss-laminated
SST
5
60
61
Rock Types in TMC#11 Core
61
62
Rock Types in TMC#8
SVW-1X SV-1XST SV-1X SV-4P SV-3P SV-5P SV-1P SV-2P
62
63
Rock Types in TMC#10 & TMC#11
SVW-1X SV-1XST SV-1X SV-4P SV-3P SV-5P SV-1P SV-2P
TMC#10U
TMC#10L
TMC#11
63
Machine Learning Techniques
1. Neural Networks
3. Genetic Algorithm
4. Fuzzy Logic
64
Genetic Algorithm
65
How does Genetic Algorithm works?
Genetic algorithm based on a Darwin’s theory of evolution which is survival-of-the-fittest
strategy with sexual reproduction.
1. Initial population (a set of individuals or candidates)
Crossover Mutation
66
Why is Genetic Algorithm?
GA is good for global
Optimization Problems
Search Space
67
Genetic Algorithm for Peameability
There is a relationship between sonic log (or any porosity log) and gamma ray log vs. permeability f(s,g)
following this equation (Brown, 2000):
as b cg d e
f ( s, g ) 10
1 chromosome
a, b, c, d and e are numerical constants to be determined (-0.5,0.5) Allele 1 Allele 6
is an operator representing either + or * 1 -0.2 0.3 -0.4 0.1 0.3
s, g are sonic log and gamma ray log, respectively
+/* a b c d e
Population size is 100.
Chromosome 1 xi Offspring 1 zi
1 -0.2 0.3 -0.4 0.1 0.3 0 -0.2 0.3 -0.4 0.15 0.3
Chromosome 2 Crossover Offspring 2
1 -0.1 -0.2 0.3 0.2 -0.4 1 -0.1 -0.2 0.3 0.15 -0.4
68
Genetic Algorithm for Peameability
- Selection is linear ranking selection.
- Fitness function is
Minimise : pi f ( si , g i )
f
i
pi, si and gi denote respectively the measured permeability (from available side wall cores),
sonic transit time and gamma-ray intensity at depth i.
s 8 g 8 4
f ( s, g ) 10
69
Genetic Algorithm for Peameability
11
1
1000
463
0.01
100
0.1
10
1
1000 1000
100 100
Predicted Permeability
10 10
1 1
0.1 0.1
0.01 0.01
0.01
1000
0.1
100
10
1
Core Permeability
1. Neural Networks
3. Genetic Algorithm
4. Fuzzy Logic
71
Crisp logic and Fuzzy logic
72
Fuzzy Set Theory
A fuzzy set is a class of objects with a continuum of grades of membership. Such a set is characterized by a
membership function which assigns to each object a grade of membership ranging between zero and one.
The boundary of a fuzzy set is not precise Several shapes of Membership Functions
73
Fuzzy Logic for litho-facies prediction
Litho-facies type f
P()
x
P(x) =
- +
: mean; : standard deviation Relative fuzzy possibility R(xf) of a value x (GR
or…) belonging to litho-facies type f
74
Fuzzy Logic for litho-facies prediction
Litho-facies type f
- +
: mean; : standard deviation
If Cf is MAXIMUM then Litho-facies type of a precise depth is f
75
Fuzzy Logic for litho-facies prediction
Mud flat
Channel / Crevasse splay
Temporal lake
76
Fuzzy Logic for litho-facies prediction
A
8.5m Core
B
17.4m Core
D
Litho-facies analysis in Lab from conventional 25.6m Core
cores of 4 wells A, B, C, D C
27.4m Core
77
Fuzzy Logic for litho-facies prediction
78
Fuzzy Logic for litho-facies prediction
A B
79
Fuzzy Logic for litho-facies prediction
C D
80
Fuzzy Logic in IP
81
Fuzzy Logic in IP
82
Fuzzy Logic – Facies Model Build
83
Fuzzy Logic - Equations
84
Fuzzy Logic - Prediction
85
Fuzzy Logic - Permeability
86
Fuzzy Logic in IP
– Curve to be predicted.
87
Fuzzy Logic - Permeability
88
Fuzzy Logic - Permeability
89