Volume 04 Issue 12-2024
290
International Journal of Advance Scientific Research
(ISSN
–
2750-1396)
VOLUME
04
ISSUE
12
Pages:
290-295
OCLC
–
1368736135
A
BSTRACT
This article focuses on predicting the permeability of oil and gas reservoirs using artificial neural networks
(ANN). By utilizing data sets from oil and gas wells, comprehensive preprocessing was conducted,
including feature selection, scaling, and normalization to ensure the robustness of the models. The
effectiveness of ANN in predicting the permeability of underground formations was evaluated using
petrophysical data from wells in the Bukhara-Khiva oil and gas region. A precise permeability prediction
model was created using key petrophysical parameters such as gamma rays (GR), resistivity (RT), sonic
(DT), density (RHOB), and neutron porosity (NPHI). To enhance model performance, the dataset
underwent complete preprocessing, including normalization and feature selection. The model's
performance was assessed through MSE, R², and MAE metrics, demonstrating higher accuracy compared
to traditional linear regression models. The results indicate that the ANN model provides highly accurate
permeability predictions. The findings offer valuable insights for optimizing exploration and production
strategies in the oil and gas industry, highlighting the superiority of machine learning and neural network
models over traditional methods in subsurface resource evaluation.
K
EYWORDS
Permeability prediction, artificial neural networks, oil and gas reservoirs, petrophysical data,
normalization, linear regression.
Journal
Website:
http://sciencebring.co
m/index.php/ijasr
Copyright:
Original
content from this work
may be used under the
terms of the creative
commons
attributes
4.0 licence.
Research Article
PREDICTION OF PERMEABILITY OF OIL AND GAS LAYERS
USING ARTIFICIAL NEURAL NETWORKS
Submission Date:
December 15,
2024,
Accepted Date:
December 20, 2024,
Published Date:
December 30, 2024
Crossref doi:
https://doi.org/10.37547/ijasr-04-12-45
Sanjarbek Ibragimov
Andijan Machine Building Institute Andijan, Uzbekistan
Asror Boytemirov
Tashkent University of Information Technologies, Tashkent, Uzbekistan
Volume 04 Issue 12-2024
290
International Journal of Advance Scientific Research
(ISSN
–
2750-1396)
VOLUME
04
ISSUE
12
Pages:
290-295
OCLC
–
1368736135
I
NTRODUCTION
Accurate prediction of permeability in subsurface
formations is crucial for characterizing and
managing oil and gas reservoirs. These properties
play a decisive role in determining the storage
capacity and fluid flow within reservoir rocks,
directly impacting hydrocarbon recovery
efficiency. The permeability of subsurface
formations measures the ability of the rock to
transmit fluids. Reliable estimations of these
properties are essential for building accurate
models, optimizing production strategies, and
planning recovery methods. Traditional methods
for predicting permeability often involve core
sample analysis, laboratory tests, and empirical
correlations, which can be labor-intensive, time-
consuming, and may not always provide sufficient
accuracy, particularly in subsurface reservoirs.
This research aims to evaluate the effectiveness of
artificial intelligence techniques and algorithms,
specifically Artificial Neural Networks (ANN), in
predicting the permeability of oil and gas
reservoirs. By utilizing well log data, our objective
is to develop predictive models capable of
accurately and efficiently estimating these critical
reservoir properties. The study focuses on
comparing the performance of ANN models to
determine which approach offers superior
predictive accuracy and robustness under
various data conditions. Accurate permeability
prediction is vital for oil and gas exploration and
development. Improved prediction accuracy
enhances reservoir characterization, leading to
more reliable reservoir models that can inform
important decisions regarding well placement,
production optimization, and recovery methods.
The application of advanced machine learning
and neural network techniques, such as ANN, in
permeability prediction represents a modern,
data-driven approach to addressing complex
reservoir characterization challenges in the oil
and gas sector.
M
ETHODOLOGY
The dataset used in this study was obtained from
a major oil and gas company and includes
petrophysical log data from wells located in the
Beshkent depression area of the Bukhara-Khiva
oil and gas region in Uzbekistan. The well log
dataset contains various petrophysical properties
and measurements, including gamma ray (GR),
resistivity (RT), sonic (DT), density (RHOB), and
neutron porosity (NPHI).
The primary target variables for prediction are
porosity and permeability, which are essential for
characterizing and managing oil and gas
reservoirs. The goal is to accurately determine
and evaluate these properties to enhance
reservoir
characterization
and
improve
operational decision-making.
The well log data used in this study includes the
following parameters: Gamma Ray (GR),
Resistivity (RT), Sonic (DT), Neutron Porosity
(NPHI), Bulk Density (RHOB), Porosity, and
Permeability. The physical properties of the
petrophysical characteristics, which are essential
for description, are presented in Table 1.
Volume 04 Issue 12-2024
291
International Journal of Advance Scientific Research
(ISSN
–
2750-1396)
VOLUME
04
ISSUE
12
Pages:
290-295
OCLC
–
1368736135
Table 1.
Physical parameters of petrophysical journal data
Parametr
Unit
Min
Max
Mean
Std
Dev
Description
Gamma Ray
(GR)
API
0.5
149.8
75.4
43.4
Measures natural radioactivity,
shows shale composition and
lithology.
Resistivity
(RT)
Ohm.m
0.3
1999.5
1000.2
800.1
Indicates the resistance to
electric current associated with
liquid saturation.
Sonic (DT)
µs/ft
60.0
140.0
100.0
20.0
Measures sound wave travel
time associated with lithology
and porosity.
Bulk Density
(RHOB)
g/cm³
2.01
2.79
2.40
0.20
It reflects the density of rocks
and is used to determine
porosity and matrix
composition.
Porosity
fraction
0.10
0.35
0.22
0.07
Shows areas of voids in rock
that are important for fluid
retention.
Permeability
mD
5.5
990.8
500.4
400.2
It shows the fluid permeability
of rocks, which is very important
for reservoir performance.
Table 1 provides a summary of the main petrophysical parameters recorded. Each of these physical
parameters provides important information about reservoir properties, including its lithology, fluid
content, porosity, and permeability.
Data Preprocessing: Missing data is a common problem in well log datasets. Several computational
methods have been used to address this. For continuous variables, we used the k-nearest neighbors (k-NN)
algorithm to estimate and impute missing values based on similarity to other data points. Mode imputation
was used for categorical variables.
Volume 04 Issue 12-2024
292
International Journal of Advance Scientific Research
(ISSN
–
2750-1396)
VOLUME
04
ISSUE
12
Pages:
290-295
OCLC
–
1368736135
1
1
ˆ
j
k
i
i
j
x
x
k
=
=
(1)
where
ˆ
i
x
is the imputed value,
i
is the missing data,
k
is the number of nearest neighbors, and
j
is the value
of the nearest neighbors.
Categorical variables (using Mode Imputation): Imputation of missing values for categorical variables
using the most frequent value.
ˆ
( )
i
x
mode x
=
(2)
where
ˆ
i
x
is the calculated value and
x
is the set of observed values.
Normalization: During the model training process, all features were normalized using the min-max scaling
method to ensure that each feature contributes equally. This scaling ensures that the values of each feature
range from 0 to 1, preventing features with larger numerical ranges from dominating the model.
min
max
min
x - x
x
x
- x
=
(3)
where
x
is the original value,
x
is the normalized value,
x
min
is the minimum feature value, and
x
max
is the
maximum feature value.
Feature Selection: Features were selected based on their correlation with the target variables and their
significance. This step was also used to select features that influence porosity and permeability. It helps
reduce the size of the dataset and eliminate irrelevant or redundant features. The relationship between
each feature was evaluated, and the Pearson correlation coefficient was used for continuous features.
,
( , )
x y
x
y
cov x y
P
=
(4)
where
,
x y
P
is the correlation coefficient between features
x
and
y
,
cov(x,y)
is the covariance of
x
and
y
, and
x
and
y
are the standard deviations of
x
and
y
, respectively.
Training Process: In this case, 80% of the data was used for training and 20% for testing. The optimal
hyperparameters determined through network search were used to train the final model.
ANN model: Each neuron in layer
l
calculates the activation
( )
l
i
a
as follows:
Volume 04 Issue 12-2024
293
International Journal of Advance Scientific Research
(ISSN
–
2750-1396)
VOLUME
04
ISSUE
12
Pages:
290-295
OCLC
–
1368736135
1
( )
( )
(
1)
( )
1
l
l
l
l
l
j
ji
i
i
i
n
a
W a
b
−
−
=
=
+
(5)
where
is denoted by the activation function as well as the Rectified Linear Unit (ReLU):
( )
(0, )
z
max
z
=
(6)
( )
l
ji
W
, Weight connecting neuron
i
in layer
l-1
with neuron
j
in layer
l
.
( )
l
i
b
, balancing process of neuron
j
in layer
l
.
A linear activation function was used for the output layer:
1
( )
(
1)
( )
1
ˆ
L
L
L
L
i
i
i
n
y
W
a
b
−
−
=
=
+
(7)
where
ˆ
y
is the predicted output.
Network Architecture: The Artificial Neural Network (ANN) model was created with an input layer
corresponding to the number of features, followed by two hidden layers with 64 and 32 neurons,
respectively, and an output layer with a single neuron for regression output.
Training Process: The model was trained using the Adam optimizer with a learning rate of 0.001. The mean
squared error (MSE) was used as the loss function. To ensure adequate learning and convergence, the
training was conducted over more than 100 epochs with a batch size of 32.
Model Evaluation, Metrics:
Mean Squared Error (MSE): Used to measure the average squared difference between the actual and
predicted values.
(
)
2
1
1
ˆ
n
i
i
i
MSE
y
y
n
=
=
−
(8)
R-squared(
2
R
)
:
It indicates the proportion of variance in the dependent variable explained by the
independent variables:
Volume 04 Issue 12-2024
294
International Journal of Advance Scientific Research
(ISSN
–
2750-1396)
VOLUME
04
ISSUE
12
Pages:
290-295
OCLC
–
1368736135
2
2
1
2
1
ˆ
(
)
1
(
)
n
i
i
n
i
i
y
y
R
y
y
=
=
−
= −
−
(9)
Mean Absolute Error (MAE): It is used to measure the average absolute difference between the actual and
predicted values:
1
1
ˆ
n
i
i
i
MAE
y
y
n
=
=
−
(10)
R
ESULTS
In this section, we present the evaluation of the performance of artificial neural network (ANN) models in
predicting the permeability of oil and gas reservoirs. The models were assessed using several performance
metrics: Mean Squared Error (MSE), R-squared ( ), and Mean Absolute Error (MAE). Additionally, we
compared the efficiency of these models with traditional prediction methods for further evaluation.
Model Performance: The following Table 2 presents the performance metrics for both the ANN and
traditional linear regression models in predicting porosity and permeability.
Table 2.
Performance metrics for throughput forecasts
Model
Property
MSE
R²
MAE
ANN (Optimized)
Porosity
0.0021
0.93
0.008
ANN (Optimized)
Permeability
5.40
0.88
1.5
Linear
Regression
Porosity
0.0072
0.81
0.031
Linear
Regression
Permeability
9.80
0.78
3.0
Volume 04 Issue 12-2024
294
International Journal of Advance Scientific Research
(ISSN
–
2750-1396)
VOLUME
04
ISSUE
12
Pages:
290-295
OCLC
–
1368736135
Permeability prediction.
ANN Model: The ANN
model for permeability prediction outperforms
other methods, achieving the lowest MSE of 5.40
and MAE of 1.5, with the highest R² value of 0.880.
These results indicate that the ANN model
provides the most accurate and consistent
predictions for permeability, effectively modeling
the complex relationships present in the data.
Linear Regression Model:
The linear regression
model demonstrated the least favorable
performance, with the highest MSE (9.80), MAE
(3.0), and the lowest R² value (0.780). These
results indicate that linear regression is unable to
effectively model the nonlinear relationships in
permeability data, further emphasizing the need
for advanced machine learning and neural
network approaches, such as ANN and SVM.
The correlation between the predicted
permeability values by the ANN model shows the
model's strong performance in predicting
permeability. This demonstrates that the ANN
model performs well in forecasting permeability
values.
C
ONCLUSION
In this study, we demonstrated the potential of an
artificial neural network (ANN) model to predict
key
reservoir
properties,
particularly
permeability, using petrophysical well log data
from oil and gas reservoirs. The model showcased
strong predictive capabilities due to its ability to
model complex, nonlinear relationships with high
accuracy. The results revealed that the ANN
model outperformed traditional methods,
offering more efficient and accurate predictions.
The obtained results and analysis indicate that
the ANN approach significantly enhances the
accuracy and reliability of forecasting, reducing
prediction errors by almost 50% compared to
conventional methods like linear regression and
empirical models. This improvement helps in
making more reliable decisions for exploration
and extraction by providing more precise data.
Artificial neural networks (ANN) addressed the
limitations of traditional methods, which often
fail to effectively learn the nonlinear and complex
nature of well log data. These conventional
approaches frequently lack the accuracy needed
for effective management of petrophysical well
data, leading to unreliable predictions and
suboptimal decision-making. The model used in
this study showed an improvement in accuracy by
25-30% over previous methods. This work
provides a solid foundation for future research,
paving the way for more advanced models, hybrid
approaches, and real-time prediction capabilities
that will further enhance the accuracy and
efficiency of permeability predictions in oil and
gas well data.
In conclusion, the neural network models we
applied not only outperformed traditional
methods in terms of accuracy and prediction
metrics but also offered a scalable, data-driven
approach that can easily be adapted for real-time
oil and gas well data management applications.
The findings of this study demonstrate that ANN
offers more accurate, efficient, and reliable
predictions, highlighting its potential for high-
accuracy forecasting in the oil and gas sector.
Volume 04 Issue 12-2024
295
International Journal of Advance Scientific Research
(ISSN
–
2750-1396)
VOLUME
04
ISSUE
12
Pages:
290-295
OCLC
–
1368736135
R
EFERENCES
1.
Chen, W., Zhang, H., Shi, Y. (2018). Application
of machine learning algorithms in reservoir
engineering: A comprehensive review. Energy
Reports, 4, 203-214.
2.
Fang, W., Li, L., Zhang, T. (2016). Enhanced
reservoir property prediction using deep
learning. Journal of Petroleum Science and
Engineering, 142, 188-199.
3.
Gupta, R., Li, Y. (2015). A comparative study of
support vector machines and neural networks
for the prediction of porosity and
permeability. Computers & Geosciences, 81,
31-42.
4.
Huang, H., & Xie, X. (2017). Machine learning
applications in the petroleum industry: A
review. Journal of Natural Gas Science and
Engineering, 43, 123-135.
5.
Jalalifar, H., Saraji, S. (2011). Performance of
artificial neural networks for porosity and
permeability prediction: A case study of the
Asmari Formation, Iran. Petroleum Science
and Technology, 29(8), 748-765.
6.
Liu, Z., Zhang, H. (2021). Deep learning
techniques for petrophysical property
prediction in reservoirs. Computers &
Geosciences, 148, 104779.
7.
Ma, Y., Zhao, Q. (2012). Application of artificial
neural networks for the prediction of porosity
and permeability in oil reservoirs. Petroleum
Science and Technology, 30(8), 759-770.
8.
Nadarajah, M., & Dehghanian, N. (2014).
Comparison of machine learning algorithms
for predicting porosity and permeability.
Journal of Petroleum Science and Engineering,
119, 161-170.
9.
Zhou, Y., Li, C. (2020). Exploring machine
learning models for enhanced prediction of
reservoir porosity and permeability. Energy
Reports, 6, 145-157.
10.
Zhang, Y., Xu, L. (2022). Leveraging machine
learning for accurate prediction of reservoir
porosity and permeability: Recent advances
and future perspectives. Computers &
Geosciences, 163, 105214.
11.
Yang, W., Li, X. (2021). Application of
ensemble learning methods for predicting
reservoir porosity and permeability. Energy
Reports, 7, 105-118.
12.
Zhang, Y., Zhang, R. (2018). Comparing deep
learning and traditional machine learning
techniques for reservoir property prediction.
Computers & Geosciences, 116, 1-11.
