Machine Learning Approach for Petroleum Reservoir Engineering: A Review

Rakhi Agarwala, Devarati Mitraa, Rakesh Kumar Pandeya, Anil Kumarb, Ajay Mandalc

a
Department of Petroleum and Energy Studies, DIT University, Dehradun – 248009, India

b
Department of Information Technology, School of Computing, DIT University, Dehradun –
248009, India

c
Department of Petroleum Engineering, IIT (ISM), Dhanbad – 826004, India

*Corresponding Author: E-Mail: xxxxxxxxxxxxxxxxxxxxxxxx; Tel.: +91-xxxxxxxxxxxxxxxx

ABSTRACT

The oil and gas industry faces many challenges and difficulties in handling massive datasets and
to overcome these problems, machine learning has been applied. The models that are generated
can analyze large, complex data and produce quicker and more reliable results even in large
scales. It can give smart solutions that can enhance reservoir modeling. The difficulties and costs
of analyzing data-sets have led to the development of advanced methods for machine learning in
this field. Among all the models, we have found that ANN based models were most widely used.
In oil & gas industries ML finds its use in different fields such as in reservoir engineering, the
trainings of a machine learning model using noisy data provide a strong prediction despite some
history of flow rate. We have performed a detailed review using ML on reservoir engineering,
well testing, and well logging. We have also discussed about the statistical parameters used to
determine the accuracy of prediction of the ML based models. We have reviewed the papers
which implements ML techniques such as GA-ANN, SVM, combination of SVR and RVR, DP,
LSSVM, RSM, MLP, RBF used in reservoir engineering.

Keywords: Machine Learning, Data Mining, Prediction Model, Statistical Analysis, Performance
Measures, Oil and Gas, Reservoir Performance.

1. INTRODUCTION
Machine learning is an artificial intelligence technology (AI) that allows systems to learn from
their experience automatically and to evolve without being specifically programmed. Machine
learning focuses on designing applications that can access and use data by itself. In petroleum
industry, machine learning is used for analyzing the algorithms to learn from examples and data
such that output is increasingly enhanced on a given task without being programmed directly.
ML algorithms are used for enhancing oil recovery using different methods and algorithms such
as random forest (RF) and decision tree (DT) are used for prediction. Akmal Aulia et. al [1] has
used RF algorithm which is a collection of DT to determine the oil recovery factor. RF
algorithm has been found to be feasible and has good learning ability. This has been also shown
by Yile Ao et. al, [2] who used a linear random forest algorithm and analyzed its applications,
and concluded the supremacy of the random linear forest for the analysis of regression.

Machine learning defines probabilistic laws as well as statistical regularities based on the data
observed to assess physically the environment, and this is highly useful for handling large
datasets that are required to be assessed. M.R. Brule et. al [3] used the analytic methods to
overcome the data management problems, and made a strategy for collecting and storing the
data. In another study, Yingjie Wang et. al [4] constructed a database management system based
on SQL server database system. This database system provided a significant technical support
for the development of gas storage project. Fuzzy logic has been used extensively to deal with
the uncertainties that occurs in exploration and production operations. Sara Eghbali et. al [5]
introduced an EOR method recommended by a screening tool. Four popular methods were used
for screening of expert fuzzy logic, this expert systems were used to estimate the success of EOR
methods. Computer research is highly necessary for today's technology, and it grows really
quickly every day. Machine learning finds its application in image, face and speech recognition,
email spam and malware prediction, online fraud detection, medical diagnosis. Its application in
the petroleum domain is accurate modelling, improving drilling operation, locating the site to be
drilled, process optimization, providing well placement solutions, improving sub-surface
characterization. The resurgent interest in machine learning arises from the same reasons that
have boosted the success of data mining and Bayesian analysis. Issues like the increase of data
volumes and sizes, cheaper, more efficient and cost effective computer processing. All this
means that models can be generated quickly and automatically which can analyze larger, more
complex data and generate quicker, more reliable results, even on a large scale. Through detailed
models, a company has a greater ability to find potential opportunities – or to eliminate uncertain
risks.

2. EVOLUTION OF MACHINE LEARNING IN PETROLEUM INDUSTRY

Machine learning is one of the subfields of the AI which has applied to contemporary businesses
since the 1990's and helps hydrocarbon sector in the oil price decline and machines without
being specifically programmed. The upstream field is the obvious option for all areas of the oil
and gas industry that are capable of roll outing machine learning. Based on Bayesian statistics, a
mathematical branch that uses "degrees of belief" to interpret probability, machine learning
generates algorithms and utilizes them to predict data. The discovery process depends on how
the knowledge layers are perceived. Computers can analyze vast quantities of data by means of
machine learning and make choices to solve problems in a way that resembles the behavior of
human brain. It is done, however, more efficiently and effortlessly. Machine learning are built so
that when new data are presented they can learn and improve. Therefore, the predicted effects on
the basis of new data without new programming are continuously improved. Machine learning
hence provides a huge advantage in managing the data sets and performing the operations.

3. APPLICATIONS OF MACHINE LEARNING ALGORITHMS IN DIFFERENT

FIELDS OF PETROLEUM INDUSTRY

This section presents an overall survey of different research work carried out in different
areas of petroleum industry. The review is done to highlight the use of machine learning
algorithms and machine learning based models. This review paper focuses mainly on EOR,
well testing, simulation, well logging and reservoir engineering.

3.1 EOR

EOR is the extraction of crude oil from an oilfield that is not otherwise produced, and is often
known by the name of the tertiary recovery. EOR can extract between 30 and 60% of the
reservoir oil and between 20 to 40% by primary and secondary recovery. According to Tayfun
Babadagli [6] the three important parameters for successful EOR projects are: effective reservoir
characterization and geological descriptions and their effect on the EOR process, efficient design
and effective monitoring of the proposed EOR method and the efficiency of the human expertise.
EOR leads to optimizing the extracted oil reserves, rising the field life and recovery factor. It is
an important tool for companies that help to maintain production and increase returns on older
investments. There are three main EOR techniques: gas injection, thermal injection and chemical
injection. The results of the research work conducted by Zhao-xia Liu et. al [7] shows that
thermal recovery for heavy crude oils is the most efficient EOR technique.

3.1.1 Case Studies

1. C.H. Prada et. al [8] proposed a methodology by using ANN to build a neuro simulation
tool for the screening of IOR methods. For determining the most suitable process, the
authors evaluated the reservoir characteristics. The results indicate that a strong
correlation of networks was recognized between displacement mechanism and reservoir
characteristics.
2. Guangying Chen et. al [9] used CO2 oil MMP as a key parameter in CO2-EOR. A fast
and vigorous method was developed using ANN and as a result the developed model
provided a reliable basis for CO2 flooding.

CO₂-MMP

MSE

AARE

ARE

0 2 4 6 8 10 12

Impure Pure

[22, 2013]

3. B.J. Adeyemi et. al [10] introduced a reliable method for enhancing oil recovery. In this
research, feasibility of EOR projects was provided using high level data mining
 technology. Result showed that data mining is a robust tool for taking decisions on
investments, thus this concept makes EOR projects cost effective with less risks.
4. Steven B. Hawthrone et. al [11] developed a method for determining the MMP of crude
oils. The approach was used to test the reservoir’s impact on MMP, the suspended oil
was collected to analyze mass of mobilized oil and molecular weight distribution but at
lower pressures, the hydrocarbon would stay largely in the reservoir during EOR floods.
5. Munqith Aldhaheri et. al [12] used a machine learning technique to develop an advanced
methodology for selecting polymer gels. Different in-situ gels historical data was used for
training of models for logistic regression. Including different treating technologies three
probabilistic models have been developed and also a variant model was constructed
without water cut. Results showed that the prediction of gel technology was more than
85%.
6. Qian Sun et. al [13] considered an ANN based model for forecasting capabilities. The
main focus of this work was to develop a robust model. It also includes ANN model
training for better understanding of complex data structures. The paper showed that this
model was used as a classification tool as well as a nonlinear regression tool.
7. The dimensionless breakthrough time of water coning is predicted by the three models
developed by Arash Kamari et. al. [14] Among the models developed, namely LSSVM,
regression decision tree and ANN, the LSSVM model has a better accuracy.

AARD %

PSO-ANN

DT

LSSVM

0 2 4 6 8 10 12 14 16 18 20
 [18, 2016]

8. Jie Xiao et. al [15] with the help of big data has developed a model to predict the
hydrodynamics of EOR. Due to the huge data required, the authors have shown that big
data successfully integrated different data sources and was used to calculate production
index, PVT parameters etc. that served as the input parameters of existing numerical
simulation models. The results show that the prediction error was reduced by 46% when
compared with previous reservoir numerical simulation, thereby enabling reservoir
engineers to store and monitor the results in a chronological order.
9. Geraldo A.R. Ramos et. al [16] implemented neuro fuzzy (NF) simulation study to screen
suitable reservoirs for EOR projects. The required information was extracted by the
combined techniques of the searching potential of fuzzy logic and the learning capability
of neural networks. The extracted information was then validated against data from
successful EOR projects. The results that were obtained from the NF simulation
technique show that the suitable EOR techniques are polymer, hydrocarbon gas and
combustion.
10. Menad Nait Amar et. al [17] have been developed many correlations because of time
consuming and expensive methods. The objective of this study was to establish a global
model to forecast MMP in EOR process, and ABC was used to find hyper parameters of
SVR. The results showed that in the end a leverage approach was used to investigate the
forecasting capabilities of the new model.
11. Using least square support vector machine (LS-SVM), Mohammad Ali Ahmadi et. al [18]
has developed a model to calculate the CO 2 oil swelling factor and by using generic
algorithm the hyper parameters of the model has been optimized. The results indicate a
high coefficient of determination (0.9953) and low mean-squared error (0.0003), hence
making this model accurate.
12. Nastaran Khazali et. al [19] showed that the screening of EOR can be seen as a tool to
suggest the appropriate EOR procedures, they researched the usefulness of EOR
processes. The authors introduced a fuzzy decision tree method for screening of EOR.
This tree helps to build a network of expert system. This method was implemented on a
dataset. This study predicts the ranks of the most relevant EOR methods.
 13. Mohammad Madani et. al [20] has showed the importance of screening before any
decision making process for EOR methods due to the high cost and technical complexity
involved. The authors have used TOPSIS method which is one of MCDM techniques to
rank the 10 different EOR methods for 65 Iranian oil reservoirs. By calculating the
relative closeness to ideal solution, the results show that 74% of the reservoirs were
suitable for CO2 injection whereas 86% of the reservoirs were unsuitable for miscible N 2
method due to inconsistent reservoir pressure. The proposed method of the authors are
less expensive than a full field simulation study.
14. Lazreg Belazreg et. al [21] showed that the purpose of their research was to develop a
predictive model for water alternating gas (WAG) incremental recovery using reservoir
simulation and data mining techniques to help reservoir engineers preform quick
evaluation of WAG performance. The authors developed 1000 reservoir simulation
models and then fed the results into two data mining techniques, namely regression and
group method of data handling (GMDH). The results show that the correlation coefficient
for regression and GMDH were 0.766 and 0.853 respectively, making GMDH a more
capable method for developing accurate models.
15. Shabnam Sedghi et. al [22] has presented a data driven approach for developing and
implementing soft sensors for fluid flow rates of SAGD wells whose main objective is to
solve difficulties and monitor production in real time. The results show that the correction
strategy improves performance for well pairs with less acceptable prediction whereas it
maintains the performance of well pairs with good prediction.

3.2 Well Testing

A well test is the execution of a set of scheduled data collection operations in the petroleum
industry. The data collected would be studied in order to extend the information and consider the
hydrocarbon characteristics and the characteristics of the underground hydrocarbon reservoir.
Using well test analysis, Rafael Osorio et. al [23] has developed type curves so as to determine
the absolute permeability of the heterolithic bed. His study also concludes that the generated type
curves can be effective in discerning the way the facies are changing. A fluid sample is the
primary source for evaluating a well. Additional grounds include calculating the initial strain,
determining the minimum storage capacity, evaluating the good permeability and impact of the
skin and defining heterogeneity and boundaries.

3.2.1 Case Studies

1. Z. Jeirani et. al [24] used ANN for estimation of various parameters of oil reservoirs
using pressure build up data. The results of this study shows that the application of ANN
is found to be very useful in well testing and it reduces the cost of the test.
2. Using ANN, reservoir models were determined from well test data by R. Kharrat et. al.
[25]. The authors have utilized 50 points on the normalized pressure derivate type curve
as input parameter of the ANN algorithm, and have determined the belonging and
probability of belonging of the data to a specific model.
3. Behzad Vaferi et al [26] has showed that multi-layer perception (MLP) which is a
structure of artificial neural network (ANN) can be used to identify the best suited
reservoir model. The results indicate that a two-layered MLP network was able to
successfully identify the best suited reservoir model out of 8 test reservoir models, with
an acceptable accuracy.
4. M. Hasanvand and S.M. Berneti [27] has developed a new method for prediction of oil
rates using artificial neural network. In their research, temperature and pressure has been
used as input variables to the network to obtain the flow rate as output.
5. Chuan Tian et. al [28] has used RNN and RNN-NARX to forecast reservoir performance
and identify reservoir model from down hole pressure gauge data.
 F-score

PNN

SVM

LR

RF

0 0.2 0.4 0.6 0.8 1 1.2

[1106, 2017]

6. By constructing nine ANN networks with each one able to differentiate the six reservoir
models, Ahmad M. AlMaraghi [29] has successfully developed a model for identification
of reservoir models. The results show that the proposed model can successfully identify
normal and complex reservoir models with acceptable accuracy.
7. By using CNN algorithm, Hongyang Chu et. al [30] has developed an automatic
classification method for well test curves. They have also shown, that when compared
with FCNN, the CNN model has shown better results.

3.3 Well Logging

Well logging is the method by which the detailed record of the sub-surface geological formations
is achieved. This can be done by obtaining sub-surface samples of the formation or by lowering
instruments down the borehole. Logging can be done during drilling, producing, completing or
abandoning a well. Wireline logging, coring and mudlogging are the various methods of logging.
The latest technologies can be implemented in new fields. It is a commodity. It is a liability for
old fields in which old logs have to be replaced with new logs in the field review.

3.3.1 Case Studies

1. Masoud Nikravesh et. al [31] established the next generation of smart software that
identifies the non-linear relation, this software was used to group data into data sets, and
dominant and important patterns were identified and graded.
2. Tan Fengqi et. al [32] has applied data mining techniques to identify eight types of
conglomerate reservoir lithology. By selecting oil saturation, water production rate and
oil productivity index as parameters, a multi-parameter model was devised to evaluate the
water-flooded layers of the oil bearing reservoir lithology.
3. Naser Golsanami et. al [33] focused on the NMR T2 distribution estimation and predicts
the logarithmic mean values using the models proposed in this study, the models were
developed using four training algorithms. As a result, GA optimized CMIS has better
performance than individual expert acting alone.
4. Fatai Anifowose et. al [34] used a selection based prediction of models and hybrid ML,
various hybrid models was proposed including SVM, FN etc for the estimation of
reservoir properties. Result showed that with the help of these models better prediction
accuracy was achieved. This paper presented the lessons learned from the challenges of
data limitations.
5. By applying multilayer neural network, Mehdi Mohammad Salehi et. al [35] has
developed a model for the prediction of resistivity and density logs. By comparing the
mean square error and the correlation coefficient, the authors have shown that their model
has better accuracy than other similar models.
6. The formation permeability is predicted by Shahab Aldin Taghipour et. al [36] by taking
five types of well-logging data as input data. Using LSSVM and Cuckoo optimization
algorithm, the authors have developed a model which achieves this purpose with 0.99602
correlation coefficient.
7. Using deep CNN, Seth Brazell et. al [37] has developed correlations to help interpreters
in constructing stratigraphic frameworks in the high and moderate well density areas. The
advantage that this model presents is the elimination of time and resource intensive
training.
8. David A. Wood et. al [38] has predicted the permeability, porosity and water saturation
by utilizing TOB algorithm.
 9. Kaibo Zhou et. al [39] introduced the basic principle of TDIP and the experimental
results showed the stability of TSVD-GCV, TSVD-L and tikhonov-I-L was useful for
inversion of the pore size distribution, a BG theory was introduced in result for
evaluation.

3.4 Reservoir performance

Reservoir engineering is an oil and gas engineering branch which applies scientific principles to
porous fluid flow during oil and gas reservoir development and production, in order to achieve a
high economic recovery. In order to determine location and quantity of fuel in underground
reservoirs, knowledge in geology and fluid mechanics is used. Often researchers are focused on
the position of oil and natural gas reserves with specialized tools, such as computer modeling and
imaging.

Reservoir simulation is a field in petroleum engineering that utilizes computer modelling to

represent the operation that has to be performed in various production phases. Simulation is
based on production and reservoir engineering, geosciences, petro physics, economics and
computer science. Geological model, flow model, gridding, integrated reservoir models etc. are
used to study the behavior of the reservoir. The purpose of the simulation is to estimate field
output in one or more production schemes, whereas it can only be generated once, a model can
be made or run several times at low cost over a short period at considerable expense. V.S. Rios
et. al [40] has proposed, tested and evaluated a workflow that can be applied to any reservoir for
reducing the time spent in numerical optimization and enhancing the performance of the
simulation models.

3.4.1 Case Studies

1. Using ARULES to analyze the datasets formed by several simulations and ANN
algorithm to rank the importance of the generated parameters, Akmal Aulia et. al [41] has
performed several simulations using Eclipse software to obtain oil recovery factor.
2. Ruyi Wang et. al [42] overcame the challenge of reservoir rock typing. They presented an
overlapping index to define the overlapping relationships and a GDOH algorithm was
used along with overlapping index from which they got four reservoir rock types. The
results showed that the GDOH algorithm verified the flow behavior and the reservoir
quality.
3. An interpretation technique developed by the combination of soft-computing techniques
and statistical technique is utilized by Masoud Nikravesh et. al [43] to predict mapping
between production log data and 3-D seismic data, recognition of pay-zone and reservoir
connectivity. The prediction of optimal well placement using clustering techniques is the
main goal of this research.
4. Mohsen Saemi et. al [44] has developed a method for the auto-design of ANN using
generic algorithm (GA). The comparison of trial and error approach calibrated model
with the GA-ANN model shows that the latter is a better alternative for estimating the
permeability of the reservoir with a high correlation coefficient.
5. Dong Xiucheng et. al [45] has developed a model using ANN and generic algorithm to
accurately predict the saturates of vacuum gas oil and the results indicate that this model
has better accuracy when compared with conventional ANN models to predict the
saturates.

6. The prediction of oilfield development indices and the development index of the high
water-cut oil fields have been successfully predicted by Zhong Yihua et. al [46] with the
help of their developed model using support vector machine (SVM).

7. P.A Fokker et. al [47] reports a field study in which subsidence data have been used to
reduce reservoir design uncertainty. The compaction in the reservoir was estimated with
subsidence data by the authors, they used previously developed method in which prior
information and experiences are mixed. Results showed that a thorough inversion can
significantly reduce uncertainties.
8. Using generic algorithm in combination with Relevance vector regression (RVR) and
Support vector regression (SVR), Raoof Gholami et. al [48] has concluded that density,
porosity, gamma ray, deep latero-resistvity and sonic logs are the most suitable logs
 related to permeability. Their study also indicates that RVR method outperforms the SVR
method in prediction of permeability.

Permeability

GA-SVR

GA-RVR

0 0.0001 0.0002 0.0003 0.0004 0.0005 0.0006

Best fitness error Mean fitness error

[910, 2014]

9. Using radial basis function with GA, Afshin Tatar et. al [49] developed a model for the
prediction of reservoir brine properties. The authors concluded that the most important
properties of brine are: enthalpy, density and vapor pressure.

RMSE

Vapor pressure

Enthalpy

Density

0 0.2 0.4 0.6 0.8 1 1.2 1.4

[995, 2015]
10. Using porosity, permeability, well spacing density, ratio of oil and water wells, effective
thickness and crude oil viscosity as input parameters, Bing Han and Xiaoqiang Bian [50]
has developed a PSO-SVM model to predict the oil recovery factor. When compared with
empirical model from Oil & Gas Company and PSO-BP neural network model, the PSO-
SVM model gave the best result.

Oil recovery factor

RMS

Emax

R2

AARD

0 1 2 3 4 5 6 7 8 9 10

SVM BP

[998, 2017]

11. Palash Panja et. al [51] have developed three models based on ANN, LSSVM and RSM
to predict the oil recovery from hydrocarbon shales. The results show that RSM based
model has a higher accuracy in prediction of oil recovery, followed by LSSVM based
model whereas LSSVM model shows a better accuracy to predict gas-oil ratio.
12. Egbadon Udegbe et. al [52] presents a new paradigm for fast and robust classification of
production data, adapted from an algorithm for real time face sensing. The findings
indicate improvements in comparison to current type curve approaches.
13. Aref Hashemi Fath et. al [53] developed two models by using MLP and RBF for the
prediction of solution gas oil ratio. The results show that MLP and RBF models
outperformed other standard empirical relations and when compared among themselves,
RBF had an error of 11.95% making it a better model over MLP which had an error of
14.897%.
14. Using a hybrid clustering-fuzzy arithmetic algorithm, Pedram Masoudi et. al [54] has
used clustering techniques to analysis porosity uncertainty then with the help of fuzzy
arithmetic they have projected the uncertainty to the permeability and irreducible water
saturation.
15. By using artificial neural network, Aref Hashemi Fath et. al [55] has developed a model
which can effectively predict the bubble point pressure for crude oils with solution gas-
oil ratio ranging from 8.61 to 3298.66 SCF/STB. Reservoir temperature, specific gravity
of gas, oil gravity and solution gas-oil ratio are used as an input data to the ANN model.

Bubble point pressure

SD

RMSE

AAPRE

0 50 100 150 200 250 300 350

[913, 2018]
16. Xidong Wang et. al [56] has applied the data mining techniques to organize and extract
information from the disorganized distribution of different pore structure parameters. By
the analysis of MICP data, the authors have extracted seven parameters and has classified
the reservoir pore structure into four types. The results indicate that the predicted pore
structure model has a good effect and has an accuracy rate of 90.9%.
17. Salaheldin Elkatatny et. al [57] has shown that neutron porosity, bulk density and
compressional time are utilized as inputs in predicting the reservoir porosity using
artificial intelligence techniques. Their research indicates that artificial neural network
can effectively predict porosity when compared with adaptive neuro fuzzy inference
system and support vector machine.
 4. CONCLUSION

This paper has reviewed the machine learning algorithms used in different field of petroleum
industry. Starting from the evolution of machine learning in oil and gas industry, this paper has
reviewed the various applications of machine learning. After conducting the review, the
following conclusions are drawn:
1. In the areas where complex data-sets, optimization and decision-making was required,
machine learning provided a robust solution to overcome these complexities.
2. From the above review, it has been observed that ANN based models has been most
widely used in the field of EOR and reservoir.
3. A single machine learning technique cannot achieve the purpose as presented in this
review. With the combination of other ML techniques, the task is achieved efficiently.
4. Though machine learning algorithms have been a great tool for the petroleum industry,
they are time consuming, requires careful selection of algorithms and requires a lot of
resources.

NOMENCLATURE

AAPE Average Absolute Percentage Error

AAPRE Average Absolute Percentage Error

AARD Absolute Average Relative Deviation

AARE Average Absolute Relative Error

ABC Artificial Bee Colony

AI Artificial Intelligence

ANFIS Adaptive Neuro Fuzzy Inference System

ANFIS Adaptive Neuro Fuzzy Interference System

ANN Artificial Neural Network

API American Petroleum Institute

ARE Average Relative Error

ARULES Association Rules

AUC Area Under ROC Curve

BHP Bottom Hole Pressure

BP Back Propagation

CMIS Coefficient of committee Machine Intelligent System

CNN Convolutional Neural Network

CSS Cascading Style Sheets

DL Deep Learning

DP Dynamic Programming

DT Decision Tree

EOR Enhanced Oil Recovery

Ep Porosity

F- score Measure of model’s accuracy

FCNN Fully Connected Neural Networks

FL Fuzzy Logic

FN False Negative

GA Generic Algorithm

GCV Generalized Cross-Validation

GDOH Grid Density Overlapping Hierarchical algorithm

GMDH Group Method of Data Handling

GOR Gas Oil Ratio

IFT Interfacial tension

IOR Improved oil recovery

Ke Permeability

KNN K-Nearest Neighbors

LAS Log ASCII Standard

LLD Limited Labeled Data

LR Logistic Regression

LSSVM Least Square Support Vector Machine

MAE Mean Absolute Error

MAPE Mean Absolute Percentage Error

MCDM Multiple Criteria Decision Analysis

MICP Mercury Injection Capillary Pressure

ML Machine Learning

MLP Multilayer Perception

MLR Multivariate Linear Regression

MMP Minimum Miscibility Pressure

MRE Mean Relative Error

MSE Mean Square Error

MWC5+ Molecular weight of pentane plus fraction of crude oil

NARX Non-linear Autoregressive exogenous model

NDEI Non Dimensional Error Index

NMR Nuclear Magnetic Resonance

NPHI Neutron Porosity

NRMSE Normalized Root Mean Square Error

NS Neuro fuzzy

PHIT Total Porosity

PI Production index

PLS Partial Least Square

PNN Probabilistic Neural Network

POD Proper Orthogonal Decomposition

PSO Particle Swarm Optimization

PVT Pressure Volume Temperature

R Correlation coefficient

R2 Coefficient of determination

RBF Radial basis function

RF Random Forest

RHOB Bulk Density

RMSE Root Mean Square Error

RNN Recurrent Neural Networks

ROC Receiver Operating characteristics Curve

RSD Relative Standard Deviation

RSM Response Surface Model

RVR Relevance Vector Regression

SAGD Steam Assisted Gravity Drainage

SD Standard Deviation

SQL Structured Query Language

SSE Sum Squared Error

SVM Support Vector Machine

SVR Support Vector Regression

Sw Water Saturation

Tc Critical temperature of injection gas

TOB Transparent Open Box

TOPSIS Technique of Order Preference Similarity to the Ideal Solution

TSVD Truncated Singular Value Decomposition

VIT Vanishing Interfacial Tension

WAG Water Alternating Gas

REFERENCES

[1] A. Aulia, A. Rahman, and J. J. Quijano Velasco, “Strategic well test planning using
random forest,” Soc. Pet. Eng. - SPE Intell. Energy Int. 2014, no. April, pp. 140–162,
2014.

[2] Y. Ao, H. Li, L. Zhu, S. Ali, and Z. Yang, “The linear random forest algorithm and its
advantages in machine learning assisted logging regression modeling,” J. Pet. Sci. Eng.,
vol. 174, pp. 776–789, 2019, doi: 10.1016/j.petrol.2018.11.067.

[3] M. R. Brulé, “SPE-173445-MS The Data Reservoir: How Big Data Technologies
Advance Data Management and Analytics in E&P Introduction-General Data Reservoir
Concepts,” pp. 3–5, 2015.

[4] Y. Wang, J. Liu, X. He, and B. Wang, “Design and realization of rock salt gas storage
database management system based on SQL Server,” Petroleum, vol. 4, no. 4, pp. 466–
472, 2018, doi: 10.1016/j.petlm.2017.10.001.

[5] S. Eghbali, S. Ayatollahi, and R. Bozorgmehry Boozarjomehry, “New expert system for
 enhanced oil recovery screening in non-fractured oil reservoirs,” Fuzzy Sets Syst., vol.
293, pp. 80–94, 2016, doi: 10.1016/j.fss.2015.05.003.

[6] T. Babadagli, “Journal of Petroleum Science and Engineering Philosophy of EOR,” vol.
188, no. November 2019, 2020, doi: 10.1016/j.petrol.2020.106930.

[7] Z. Liu, Y. Liang, Q. Wang, Y. Guo, and M. Gao, Journal of Petroleum Science and
Engineering Status and progress of worldwide EOR field applications, vol. 193, no.
December 2019. Elsevier B.V., 2020.

[8] C. H. Parada and T. Ertekin, “A new screening tool for improved oil recovery methods
using artificial neural networks,” Soc. Pet. Eng. West. Reg. Meet. 2012, pp. 225–241,
2012, doi: 10.2118/153321-ms.

[9] G. Chen et al., “Simulation of CO2-oil minimum miscibility pressure (MMP) for CO2
enhanced oil recovery (EOR) using neural networks,” Energy Procedia, vol. 37, pp.
6877–6884, 2013, doi: 10.1016/j.egypro.2013.06.620.

[10] B. J. Adeyemi, V. I. Ajilo, O. A. Falode, and E. O. Falobi, “Enhancing oil recovery

through high level data mining technology,” Soc. Pet. Eng. - North Africa Tech. Conf.
Exhib. 2013, NATC 2013, vol. 2, pp. 1327–1338, 2013, doi: 10.2118/164761-ms.

[11] S. B. Hawthorne et al., “A rapid method for determining CO2/oil MMP and visual
observations of CO2/oil interactions at reservoir conditions,” Energy Procedia, vol. 63,
pp. 7724–7731, 2014, doi: 10.1016/j.egypro.2014.11.806.

[12] M. Aldhaheri, M. Wei, B. Bai, and M. Alsaba, “Development of machine learning

methodology for polymer gels screening for injection wells,” J. Pet. Sci. Eng., vol. 151,
pp. 77–93, 2017, doi: 10.1016/j.petrol.2016.12.038.

[13] Q. Sun and T. Ertekin, “Structuring an artificial intelligence based decision making tool
for cyclic steam stimulation processes,” J. Pet. Sci. Eng., vol. 154, no. August, pp. 564–
575, 2017, doi: 10.1016/j.petrol.2016.10.042.

[14] A. Kamari, A. H. Mohammadi, and A. Bahadori, “On the determination of breakthrough

time for water coning phenomenon in oil reservoirs,” Pet. Sci. Technol., vol. 34, no. 1, pp.
44–49, 2016, doi: 10.1080/10916466.2015.1115874.
[15] J. Xiao and X. Sun, “Big data analytics drive EOR projects,” Soc. Pet. Eng. - SPE
Offshore Eur. Conf. Exhib. 2017, no. September, pp. 5–8, 2017, doi: 10.2118/186159-ms.

[16] G. A. R. Ramos and L. Akanji, “Data analysis and neuro-fuzzy technique for EOR
screening: Application in angolan oilfields,” Energies, vol. 10, no. 7, pp. 9–13, 2017, doi:
10.3390/en10070837.

[17] M. Nait Amar and N. Zeraibi, “Application of hybrid support vector regression artificial
bee colony for prediction of MMP in CO2-EOR process,” Petroleum, no. June 2018, pp.
1–8, 2018, doi: 10.1016/j.petlm.2018.08.001.

[18] M. A. Ahmadi, S. Zendehboudi, and L. A. James, “Hybrid connectionist model

determines CO2–oil swelling factor,” Pet. Sci., vol. 15, no. 3, pp. 591–604, 2018, doi:
10.1007/s12182-018-0230-5.

[19] N. Khazali, M. Sharifi, and M. A. Ahmadi, “Application of fuzzy decision tree in EOR
screening assessment,” J. Pet. Sci. Eng., vol. 177, no. October 2018, pp. 167–180, 2019,
doi: 10.1016/j.petrol.2019.02.001.

[20] M. Khojastehmehr, M. Madani, and A. Daryasafar, “Screening of enhanced oil recovery

techniques for Iranian oil reservoirs using TOPSIS algorithm,” Energy Reports, vol. 5, pp.
529–544, 2019, doi: 10.1016/j.egyr.2019.04.011.

[21] L. Belazreg, S. M. Mahmood, and A. Aulia, “Novel approach for predicting water
alternating gas injection recovery factor,” J. Pet. Explor. Prod. Technol., vol. 9, no. 4, pp.
2893–2910, 2019, doi: 10.1007/s13202-019-0673-2.

[22] S. Sedghi, R. Tan, and B. Huang, “Data analytics approach for online produced fluid flow
rate estimation in SAGD process,” Comput. Chem. Eng., vol. 136, p. 106766, 2020, doi:
10.1016/j.compchemeng.2020.106766.

[23] R. Osorio, A. Shari, E. Stephens, and H. Hamidi, “Journal of Petroleum Science and
Engineering Geological interpretation of channelized heterolithic beds through well test
analysis,” vol. 158, no. June, pp. 516–528, 2017, doi: 10.1016/j.petrol.2017.08.078.

[24] Z. Jeirani and A. Mohebbi, “Estimating the initial pressure , permeability and skin factor
of oil reservoirs using artificial neural networks,” vol. 50, pp. 11–20, 2006, doi:
 10.1016/j.petrol.2005.09.002.

[25] “Archive of SID www.SID.ir.”

[26] B. Vaferi, R. Eslamloueyan, and S. Ayatollahi, “Journal of Petroleum Science and

Engineering Automatic recognition of oil reservoir models from well testing data by using
multi-layer perceptron networks,” J. Pet. Sci. Eng., vol. 77, no. 3–4, pp. 254–262, 2011,
doi: 10.1016/j.petrol.2011.03.002.

[27] M. Hasanvand, S. M. Berneti, and P. Engineering, “Energy Sources , Part A : Recovery ,

Utilization , and Environmental Effects Predicting Oil Flow Rate due to Multiphase Flow
Meter by Using an Artificial Neural Network,” no. May 2015, pp. 37–41, 2011, doi:
10.1080/15567036.2011.590865.

[28] C. Tian and R. N. Horne, “SPE-187181-MS Recurrent Neural Networks for Permanent
Downhole Gauge Data Analysis,” no. October, pp. 9–11, 2017.

[29] A. M. Almaraghi, Q. P. Co, and A. H. El-banbi, “Automatic Reservoir Model

Identification using Artificial Neural Networks,” 2015.

[30] P. Based et al., “An Automatic Classification Method of Well Testing,” 2019.

[31] M. Nikravesh and F. Aminzadeh, “Mining and fusion of petroleum data with fuzzy logic
and neural network agents,” 2001.

[32] T. Fengqi, L. Hongqi, X. Changfu, and L. Qingyuan, “Quantitative evaluation methods for
water- fl ooded layers of conglomerate reservoir based on well logging data,” pp. 485–
493, 2010, doi: 10.1007/s12182-010-0092-y.

[33] N. Golsanami, A. Kadkhodaie-ilkhchi, Y. Sharghi, and M. Zeinali, “Journal of Petroleum

Science and Engineering Estimating NMR T 2 distribution data from well log data with
the use of a committee machine approach : A case study from the Asmari formation in the
Zagros Basin , Iran,” J. Pet. Sci. Eng., vol. 114, pp. 38–51, 2014, doi:
10.1016/j.petrol.2013.12.003.

[34] F. Anifowose, S. Adeniye, A. Abdulraheem, and A. Al-shuhail, “Journal of Petroleum

Science and Engineering Integrating seismic and log data for improved petroleum
 reservoir properties estimation using non-linear feature-selection based hybrid
computational intelligence models,” J. Pet. Sci. Eng., vol. 145, pp. 230–237, 2016, doi:
10.1016/j.petrol.2016.05.019.

[35] M. Mohammad, M. Rahmati, M. Karimnezhad, and P. Omidvar, “Estimation of the non

records logs from existing logs using artificial neural networks,” Egypt. J. Pet., 2016, doi:
10.1016/j.ejpe.2016.11.002.

[36] S. A. Taghipour, S. A. Hoseinpour, and B. Soltani, “Accurate estimation of formation

permeability by means of petrophysical logs,” Pet. Sci. Technol., vol. 35, no. 7, pp. 718–
725, 2017, doi: 10.1080/10916466.2016.1269125.

[37] S. Brazell, A. Bayeh, M. Ashby, and D. Burton, “A Machine-Learning-Based Approach to

Assistive Well-Log Correlation 1,” vol. 60, no. 4, pp. 469–479, 2019, doi:
10.30632/PJV60N4-2019a1.

[38] D. A. Wood, “Predicting porosity, permeability and water saturation applying an

optimized nearest-neighbour, machine-learning and data-mining network of well-log
data,” J. Pet. Sci. Eng., vol. 184, no. September 2019, p. 106587, 2020, doi:
10.1016/j.petrol.2019.106587.

[39] K. Zhou, S. Zhang, Z. Huang, and J. Zhang, “Journal of Petroleum Science and
Engineering An improved TSVD-GCV inversion algorithm of pore size distribution in
time-domain induced polarization using migration Hankel matrix,” J. Pet. Sci. Eng., vol.
183, no. July, p. 106368, 2019, doi: 10.1016/j.petrol.2019.106368.

[40] V. S. Rios, G. D. Avansi, and D. J. Schiozer, “Journal of Petroleum Science and

Engineering Practical workflow to improve numerical performance in time-consuming
reservoir simulation models using submodels and shorter period of time,” J. Pet. Sci. Eng.,
vol. 195, no. June, p. 107547, 2020, doi: 10.1016/j.petrol.2020.107547.

[41] A. Aulia, T. B. Keat, M. Sanif, B. Maulut, N. El-khatib, and M. Jasamai, “Smart Oilfield
Data Mining for Reservoir Analysis,” no. December, pp. 78–88, 2010.

[42] R. Wang, Z. Wang, A. Osumanu, G. Zhang, B. Li, and Y. Lu, “Grid density overlapping
hierarchical algorithm for clustering of carbonate reservoir rock types: A case from
 Mishrif Formation of West Qurna-1 oilfield, Iraq,” J. Pet. Sci. Eng., vol. 182, no. June, p.
106209, 2019, doi: 10.1016/j.petrol.2019.106209.

[43] M. Nikravesh, R. D. Adams, and R. A. Levey, “Soft computing : tools for intelligent
reservoir characterization ž IRESC / and optimum well placement ž OWP /,” pp. 239–262,
2001.

[44] M. Saemi, M. Ahmadi, and A. Yazdian, “Design of neural networks using genetic
algorithm for the permeability estimation of the reservoir,” vol. 59, pp. 97–105, 2007, doi:
10.1016/j.petrol.2007.03.007.

[45] D. Xiucheng, W. Shouchun, S. Renjin, and Z. Suoqi, “Design of arti fi cial neural
networks using a genetic algorithm to predict saturates of vacuum gas oil,” pp. 118–122,
2010, doi: 10.1007/s12182-010-0015-y.

[46] Z. Yihua, Z. Lei, L. Zhibin, X. Yao, and L. Rong, “Using a support vector machine
method to predict the development indices of very high water cut oil fi elds,” pp. 379–384,
2010, doi: 10.1007/s12182-010-0081-1.

[47] P. A. Fokker, K. Visser, E. Peters, G. Kunakbayeva, and A. G. Muntendam-Bos,

“Inversion of surface subsidence data to quantify reservoir compartmentalization: A field
study,” J. Pet. Sci. Eng., vol. 96–97, pp. 10–21, 2012, doi: 10.1016/j.petrol.2012.06.032.

[48] R. Gholami, A. Moradzadeh, S. Maleki, S. Amiri, and J. Hanachi, “Applications of

artificial intelligence methods in prediction of permeability in hydrocarbon reservoirs,” J.
Pet. Sci. Eng., vol. 122, pp. 643–656, 2014, doi: 10.1016/j.petrol.2014.09.007.

[49] A. Tatar, S. Naseri, N. Sirach, M. Lee, and A. Bahadori, “Prediction of reservoir brine
properties using radial basis function ( RBF ) neural network,” Petroleum, vol. 1, no. 4,
pp. 349–357, 2015, doi: 10.1016/j.petlm.2015.10.011.

[50] B. Han and X. Bian, “A hybrid PSO-SVM-based model for determination of oil recovery
factor in the low-permeability reservoir,” Petroleum, pp. 4–10, 2017, doi:
10.1016/j.petlm.2017.06.001.

[51] P. Panja, R. Velasco, M. Pathak, and M. Deo, “Application of artificial intelligence to

forecast hydrocarbon production from shales,” Petroleum, 2017, doi:
 10.1016/j.petlm.2017.11.003.

[52] E. Udegbe, E. Morgan, and S. Srinivasan, “From face detection to fractured reservoir
characterization: Big data analytics for restimulation candidate selection,” Proc. - SPE
Annu. Tech. Conf. Exhib., vol. 0, 2017, doi: 10.2118/187328-ms.

[53] A. Hashemi, F. Madanifar, and M. Abbasi, “Implementation of multilayer perceptron

( MLP ) and radial basis function ( RBF ) neural networks to predict solution gas-oil ratio
of crude oil systems,” Petroleum, no. November, pp. 0–1, 2018, doi:
10.1016/j.petlm.2018.12.002.

[54] P. Masoudi, T. Aïfa, H. Memarian, and B. Tokhmechi, “Journal of Petroleum Science and
Engineering Uncertainty assessment of porosity and permeability by clustering algorithm
and fuzzy arithmetic,” J. Pet. Sci. Eng., vol. 161, no. November 2017, pp. 275–290, 2018,
doi: 10.1016/j.petrol.2017.11.018.

[55] A. Hashemi, A. Pouranfard, and P. Foroughizadeh, “Development of an arti fi cial neural

network model for prediction of bubble point pressure of crude oils,” Petroleum, 2018,
doi: 10.1016/j.petlm.2018.03.009.

[56] X. Wang, S. Yang, Y. Zhao, and Y. Wang, Journal of Petroleum Science and Engineering
Improved pore structure prediction based on MICP with a data mining and machine
learning system approach in Mesozoic strata of Gaoqing fi eld , Jiyang depression, vol.
171, no. June. Elsevier B.V., 2018.

[57] S. Elkatatny, Z. Tariq, M. Mahmoud, and A. Abdulraheem, “AC SC,” Petroleum, 2018,
doi: 10.1016/j.petlm.2018.04.002.