Petroleum 8 (2022) 158e166

Contents lists available at ScienceDirect

Petroleum
journal homepage: www.keaipublishing.com/en/journals/petlm

Application of ML & AI to model petrophysical and geomechanical

properties of shale reservoirs e A systematic literature review
Fahad I. Syed, Abdulla AlShamsi, Amirmasoud K. Dahaghi*, Neghabhan S.
The University of Kansas, School of Engineering, Chemical and Petroleum Engineering, 1450 Jayhawk Blvd., Lawrence, 66045, USA

a r t i c l e i n f o a b s t r a c t

Article history: Extensive reviews and cross-comparison studies are essential to analyze the emerging developments in a
Received 10 June 2020 specific field of research. In the past decade, hydrocarbon exploration and exploitation from the shale
Received in revised form reservoirs have been the most discussed and researched area around the globe. A dramatic development
30 November 2020
in shale formations became the game-changer, especially in the US. On the other hand, Artificial Intel-
Accepted 1 December 2020
ligence (AI) and Machine Learning (ML) are playing an important role in the rapid development in all the
industries through automating most of the routine operations.
Keywords:
The oil industry is also getting equal benefits of AI and ML for the reservoir development planning and
AI
ML
its operational accuracy through a series of automated systems. For the field development, computerized
Petrophysics static and dynamic simulation models are generated based on several Petrophysical and Geomechanical
GeoMechanics properties gathered through different resources that are quite time-taking and expensive. AI and ML
have made this process much easier, faster, and economical by means of learning through uncounted
experiences from already explored and developed reservoirs, their rock properties, and the cross-
ponding fluid flow behavior under different circumstances and hence, predicts accordingly.
This article provides a comprehensive literature review in the area of AI and ML applications to model
Petrophysical and Geomechanical properties using different approaches and algorithms. Also, a sys-
tematic publication counts in each field of subject study per year in different literature databases are
presented that infect reflects the trending interest in this subject. Finally, multiple AI and ML techniques
are discussed in detail which have been tested in the last decade for the sake of achieving higher ac-
curacy in Petrophysical and Geo-Mechanical simulation models.
© 2020 Southwest Petroleum University. Publishing services by Elsevier B.V. on behalf of KeAi
Communications Co. Ltd. This is an open access article under the CC BY-NC-ND license (http://
creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction dependent on machines ever for minor jobs.

AI is a computer science that works with the capabilities of
We witness countless applications of AI and ML in our daily life machines on a similar pattern of human behavior. AI provides a
and it keeps increasing over time in almost all industries. Self- platform for computers to process a large amount of information
driving cars, instant machine translation, smartphones, and self- and data to come up with a computer-generated conclusion in no
talking virtual personal assistants like Siri and Alexa, etc. are few time. In parallel, ML enables computers to learn to respond to
examples of AI and ML that are making our daily life much easier desired needs or human behavior and also, learn how to learn using
than ever. But on the other hand, it’s making humans more self-defining algorithms. And no doubt, these advancements are
revolutionary for modern life and quick business decision. Simi-
larly, the oil industry is also getting an advantage to modernize
* Corresponding author. technology using AI and ML. The application of AI & ML is growing
E-mail address: masoud@ku.edu (A.K. Dahaghi). on daily basis in the industry from exploration to production and to
Peer review under responsibility of Southwest Petroleum University. gain insights from the vast amounts real-time data being collected
from air using drones, in the well during production through fiber
optics and as well as during drilling to remain on track as planned
using smart sensors, etc. [1,2].
Production and Hosting by Elsevier on behalf of KeAi

https://doi.org/10.1016/j.petlm.2020.12.001
2405-6561/© 2020 Southwest Petroleum University. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd. This is an open access article under the CC
BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
F.I. Syed, A. AlShamsi, A.K. Dahaghi et al. Petroleum 8 (2022) 158e166

In a reservoir, the exploration and the production strategies which is usually generated using synthetic sonic logs instead of
primarily depend on the reservoir rocks properties and the lithol- obtaining from the field directly. Here comes the role of ML and AI
ogy of overlying formations. With the concept of unconventional i.e. to generate those synthetic logs using regression principle and
reservoir particularly ultra-tight shale formations where the un- calculate the brittleness indicator by integrating information from
certainty level is quite high, AI is making the life much easier with the core obtained while drilling the well. Hence, the significance of
more accuracy [3]. The shale gas and its associated formation dif- ML and AI is increasing in the oil industry as its generating precise
ficulties are not new anymore to the world. The first shale gas information using integrating log and core data. This methodology
production was made back in early 19th century in Fredonia, while is of prime importance in predicting geo-mechanical properties of
Bakken shale has been producing shale gas since mid-1900s from shale which is considered the most heterogeneous rock with un-
the Williston Basin [4]. Nowadays the shale oil and gas are the most favorable wettability for hydrocarbons to flow [7,12,13]. Therefore,
producing hydrocarbon especially in the United States from un- ML and AI integrate and correlate data with more precision than an
conventional reservoirs. Based on recent activities in Shale forma- ordinary human can correlate. This method is robust, more effi-
tion development in the US, gas production from the tight to ultra- cient, and most importantly much more economical and faster than
tight and Shale formations is expected to contribute more than 80% manual integration and correlation. There are plenty of researchers
of the total gas production by 2040 with approx. 30 trillion cubic who have worked on the applications of ML and AI in integrating
feet as shown in Fig. 1 [5]. core and log data to calculate geo-mechanical properties and sweet
Shale gas reservoirs have been explored at different depths. The spots for production within Eagle Ford and Barnett Formations
depth can be as shallow as 1000 ft and deep up to 12000 ft with a [14e17].
variable range of Total Organic Content (TOC) from 1 to 12%. Also, Being an unconventional reservoir, the geo-mechanical prop-
the porosity usually ranges between 2 and 8% with ultra-tight erties of shale formations are being affected due to diagenetic
permeability [6]. The shale formation quality is being character- changes resulting from the depositional environment, temperature
ized on a number of factors including petrophysical properties and pressure. These changes result in the mineralogical alteration
(TOC, thermal maturity, saturation, etc.) and geo-mechanical that causes the change of rock composition and also directly affects
properties such as quartz considerable percentage or carbonate in the sediment compaction and lithification that makes it compara-
mineralogy, differential stress and friability [7]. Computer-aided tively difficult to predict the geo-mechanical properties of shale.
petrophysical and geo-mechanical studies such as log analysis, The rock mineralogy is being considered as one of the most vital
interpretation and integration are being used by petroleum engi- features because it directly controls the fracturing operation and
neers since the early 1980s [8,9]. Whereas, the AI and ML based the usage of proppant during hydraulic fracturing operation [18].
geomechanical and petrophysical analysis became more prominent Additionally, the pore pressure also gets affected due to diagenetic
within this decade that significantly resulted in much faster and changes in the rock that play an important role to introduce
successful development than ever in the history of the oil industry. porosity features in a geomechanical model [19]. It is also impor-
tant to consider that the pore pressure alteration shows different
1.1. Geomechanics properties log response. But it’s critical to distinguish the reason for different
log responses as the TOC also affects the log response [20]. How-
Understanding the geomechanical properties of rock is one of ever, the relationship of TOC and the log response is quite linear, but
the most important features to prepare a plan to perform a hy- the presence of gas in the pores would affect the compressional
draulic fracturing job. Especially in ultra-tight complex formations velocity data that is being used as an indicator for the sweet spots.
like shale where the operational risk is quite high due to formation ML and AI are applied to estimate the log response more robustly
uncertainty, thus, the most important thing is to minimize the for the generation of synthetic logs that are later applied to predict
operational risk and maximize the production with the least the seismic velocities using attributes like density, porosity and
possible expenses. mineralogy. This is a critical parameter to model changes in geo-
As a part of the strategy plan, a seismic analysis is being per- mechanical properties and fracture designs and to correctly pre-
formed with both the shear and compressional velocity waves to dict the brittleness of shale using hardness [16,17]. Techniques
determine the Young’s Modulus and other necessary geo- applied by machine learning and artificial intelligence are used to
mechanical properties [10,11]. Also, the Sonic logs are quite mark zones of shale brittleness depending on its hardness. The
important to understand the geo-mechanical behavior of the rock, hardness quality is being defined based on minimal contents i.e.
quartz dominant mineralogy provides more hardness while the
shale rock with higher clay content defines comparatively lower
hardness.

1.2. Shale gas petrophysics

Petrophysics is an integrated and another most important

discipline that connects both the engineering and the geosciences
for reservoir characterization and field development. Due to recent
technological advancements, a large volume of high-resolution
petrophysical data is accessible. The most common petrophysical
data have ‘7Vs’ characteristics including velocity, value, veracity,
variety, volume, variability and visualization [21]. ML and AI tech-
niques are deeply involved in the applications of petrophysical
data-driven analytics (PDDA) methods. Such methods are used for
image type data set interpreting and analyzing petrophysical in-
formation such as maps, image logs and cross-sections [22]. Fig. 2
shows a generic PDDA framework defined by Xu et al. [21] for the
Fig. 1. Projected shale gas production growth in the US [5]. type algorithms usually applied in ML and AI to configure
159
F.I. Syed, A. AlShamsi, A.K. Dahaghi et al. Petroleum 8 (2022) 158e166

Fig. 2. A generic PDDA framework e ML and AI algorithms to predict geological-petrophysical properties [21].

petrophysical problems in a more efficient manner. However, for neural networks, theorem proving, constraint satisfaction, knowl-
the sake of TOC calculation in shale formations, multiple ML and AI edge representation, natural language understanding, the theory of
techniques are used including functional neural network (FNN), computation, machine learning, etc. [23], [24].
Takagi-Sugeno-Kang fuzzy interference system (TSK-FIS), Mamdani AI and ML frameworks learn from the initially given information
Fuzzy Interference System (M-FIS) and support vector machine and logics in a procedure called as preparing or figuring out how to
(SVM), etc. accordingly predict the new information or the most expected
sequence of information [12,13,25]. Examinations utilizing AI and
ML have two fundamental stages. The first stage tends to the choice
1.3. Artificial Intelligence and Machine Learning
of significant factors and models for the most expected forecast
based on the recent historical sequence of events, isolating a part of
Artificial intelligence and machine learning are the most heard
the information for the preparation and approval of the models and
terminologies these days that are usually used interchangeably.
finally their optimization. In the second stage, the model is being
However, these terms are interlinked but they mean different
exposed to the information expected for testing, hence estimating
things. AI is commonly referred to a concept of machines
and analyzing the forecasting.
mimicking human intelligence, thinking and logic. On the other
hand, ML is an approach to implement AI. Hence, AI and ML can be
defined together as a set of instructions, logics and reactions are AI 2. Systematic literature review
algorithms while ML is a machine’s ability to execute those algo-
rithms and make it self-learned with defined logic and situations. Systematic literature review (SLR) is a noteworthy strategy that
AI is broadly classified into multiple categories that include is generally used to recognize and assess significant pieces of
reasoning, programming, artificial life, distributed AI, expert sys- research for a specific issue or wonders of intrigue. In this piece of
tems, belief revision, data mining, genetic algorithms, systems, work, we tried to keep our focus on but not limited to the following
list of questions:

(1) What are the most applied ML and AI techniques in Shale gas
petrophysics and geomechanical modeling?

Table 1
Year-wise publications with each searched terminology.

Year ML in Shale Gas ML in Shale Gas AI in Shale Gas AI in Shale Gas

Geomechanics Petrophysics Geomechanics Petrophysics

2010 11 52 39 37
2011 18 71 45 71
2012 21 79 51 85
2013 29 126 71 111
2014 23 112 61 89
2015 56 177 93 142
2016 67 197 98 144
2017 98 222 126 162
2018 213 351 204 196
2019 330 583 268 346
2020 105 187 80 133
(Until
April)
Total 971 2157 1136 1516
Fig. 3. Total number of publications from 2010 to 2020 along with trend line.

160
F.I. Syed, A. AlShamsi, A.K. Dahaghi et al. Petroleum 8 (2022) 158e166

Fig. 4. Trend line of searched term ‘Machine learning in Shale gas Geomechanics’. Fig. 6. Trend line of searched term ‘Artificial Intelligence in Shale gas Geomechanics’.

Fig. 5. Trend line of searched term ‘Machine learning in Shale gas Petrophysics’. Fig. 7. Trend line of searched term ‘Artificial Intelligence in Shale gas Petrophysics’.

(2) How the ML and AI techniques are employed in the petro- data origin, data type (real field, lab experimented or synthetic
physics and geo-mechanical of shale gas? data), etc. are also a few quality control criteria.
(3) What equipment is being subjected to petrophysics and As we know that geomechanics and petrophysics are two
geomechanics of shale gas formation? different aspects of shale gas so that the choice of keywords was
(4) What are the data used to apply the petrophysics and geo- selected to find the most relevant literature from the database. For
mechanics of shale gas? the sake of systematic literature review execution was performed
(5) What type of data set (real or synthetic) is needed and its using specific keywords for both the databases as given below:
origin?
(1) Artificial intelligence in shale gas geomechanics
The SLR on the subject topic is conducted using the literature (2) Artificial intelligence in shale gas petrophysics
databases that provide a significant scope of community and peo- (3) AI learning in shale gas geomechanics
ple in the global oil industry. That includes mainly the google (4) ML in shale gas geomechanics
scholar and the Society of Petroleum Engineers based database, (5) Deep ML in shale gas geomechanics
named as One-Petro. The exclusion criteria adopted for this work (6) AI learning in shale gas petrophysics
including the literature to review from 2010 and onwards. Also, the (7) ML in shale gas petrophysics
literature with no conclusions with petrophysics and geomechanics (8) Deep ML in shale gas petrophysics, etc.
of shale gas formation is excluded. Quality control is the key to this
work being worthy and that is defined by only selecting the liter-
ature from journals and the conferences that compare results with 2.1. Publication distribution on yearly basis
petrophysics and geomechanics from shale gas formation using
different AI and ML techniques. Finally, the quality data extraction Fig. 3 shows the cumulative papers found from both the data-
is also being considered as an important quality control factor for bases, that are 5780, and by viewing the trend line it showed that
this paper. Application of classical or new ML and AI techniques, publications had been increased over the years and we have seen a
161
F.I. Syed, A. AlShamsi, A.K. Dahaghi et al. Petroleum 8 (2022) 158e166

Table 2 sharp rise in publications in 2018 and followed the exponential

Publications from Conference proceedings. trend in 2019. By looking at this dramatic increasing trend one can
Conferences Publications easily understand the growing interest in this field globally. In both
SPE 2984
the literature databases, four different search terms are used that
SPWLA 154 are listed in Table 1 on a yearly basis.
SEG 212 The histogram are plotted for each searched term as discussed
Unconventional Resources Technology Conference 432 earlier to visualize the trend lines that observed is increasing each
year with the number of publications. For example, total of 971
publications are found for the searched term ‘Machine learning in
Table 3 shale gas geomechanics’ including conference papers as shown in
Three major journals with most publications.
Fig. 4. Similarly, Fig. 5 represents the trend of total 2157 publica-
Journal Publications tions for the term ‘Machine learning in shale gas petrophysics’,
SPE journal 652 Fig. 6 for the term ‘Artificial intelligence in shale gas geomechanics’
Petrophysics of SPWLA 115 with total 1136 publications, and Fig. 7 shows the total 1516 pub-
Journal of Petroleum Technology 165 lications distribution per year for the term ‘Artificial intelligence in
shale gas petrophysics’. Comparatively, a more rapid increase in ML
and AI in shale gas geomechanics is observed than the petrophysics
Table 4 in the last couple of years.
Summary of ML and AI algorithms used in Shale gas petrophysical and geo-
mechanical modeling projects (Wikipedia: Outline of machine learning, 2018,
SPWLA Petrophysics Journal, Dec. 2018). 2.2. Publication distribution among journals & conferences
Commonly used ML and AI algorithms for shale gas References
petrophysics and geomechanics The data was further refined by considering the fact publications
in journals and in conferences. These papers have been published
Artificial neural networks (ANN) [27]
The genetic algorithm [28,29] based on ML and AI being applied in shale gas petrophysics and
Support vector machine (SVM) [30,31] geomechanics all around the world in different reservoirs. Around
Least squares-support vector regression (LS-SVR) [30,31] 70% of publications are found in proceedings published for different
Extreme machine learning [30,31]
conferences around the globe. With four major conferences con-
Convolutional neural network (CNN) Alqahtani et al.,
2018; [32]
ducted under the flag of the Society of Petroleum Engineering
CNN e deep learning [22] (SPE), Society of Petrophysics and Well Logging Analyst (SPWLA),
Wavelet decomposition and CNN e deep learning [33] Society of Exploration Geophysicist (SEG) and Unconventional Re-
K-means clustering [16,17,34] sources Technology Conference are shown in Table 2. Table 3
Random forest (RF) [35,36]
summarizes the publications listed in 2010e2020 in well-known
Bayesian network (BN) [35,37,38]
Integrated neural networks (NNs) and fuzzy logic (FL) [39] journals associated mainly with SPE, SPWLA, and the Journal of
Fully connected neural network [40] Petroleum Technology (JPT).
Forward & inverse neural network [41]
Fuzzy interference system [42]
OLS, PLS, LASSO, MARS, ANN. [43] 3. Most applied AI & ML algorithms
Functional network [44]
Neural network, gradient-boost regression tree, gaussian [45] During the last decade, AI and ML remained the most researched
process regression
topic in academics and commercial R&D institutions. Also, most of
Transdimensional Bayesian method [38]
PCA & SVM [30] the oil companies invested in upgrading their operating systems to
Automation workflows [46] be based on AI and ML for quick and efficient operational and
business decisions. It is very important to highlight here that in
most of the publications, the actual field data is found to be used

Fig. 8. ANN Model with hidden layers [49].

162
F.I. Syed, A. AlShamsi, A.K. Dahaghi et al. Petroleum 8 (2022) 158e166

(1) Rock typing based on petrophysical rock characteristics and

geological facies classification
(2) Petrophysical modeling using seismic rock properties
(3) Petrophysical, geochemical and geo-mechanical properties
modeling and forecasting
(4) Logging tools’ physical modeling
(5) Reservoir surveillance and well monitoring
(6) Smart operational data quality control
(7) Proxy models generation and
(8) Logging or coring operation guidance systems

3.1. Artificial neural network

Artificial Neural Network (ANN) is a scientific model of ML that

is designed with the inspiration of the human neural system [47]. It
is made up of an enormous number of neurons which are conveyed
in various layers based on the unmistakable capacities [48]. ANN is
based on three main layers including input, hidden and the output
layer. The hidden layer could be designed with several sub-layers.
All these layers consist of numerous neurons. A schematic of a
typical ANN model structure is presented in Fig. 8. The most
important application of this method is to approximate any func-
tion with arbitrary precision. An activated function is applied as a
result of a response from inputs to generate outputs. In the output,
the generated errors are rectified by comparing values to target
values and the algorithm repeats itself until the generated errors
reduce to a minimum acceptable value.
The ANN could be trained via many algorithms, but most
commonly the back-propagation algorithm is used that is also
known as the BP algorithm. ANN is used in the enhancement of
shale gas reservoir characterization by predicting the TOC in
boreholes where the core or the well log data are not available. ANN
method that is applied in Multilayer Perceptron (MLP) for the
estimation of TOC in shale gas [50]. The input layer is usually made
up of 5 neurons consisting of Sonic S-wave, Gamma-ray, neutron
porosity, etc. The hidden layer consist of nine neurons and the
output layer comprises a single neuron corresponding to the TOC
log. The subject algorithm is designed to generate synthetic geo-
mechanical logs from the conventional logs. This model is quite
critical for the hydraulic fracturing operation as it directly deals
with the understanding of the geomechanical behavior of shale gas
reservoir rock properties. Parapuram et al. [27] presented their
prediction for the geo-mechanical properties of Upper Bakken
shale by using the same methodology. They also proposed a
framework to predict shear wave velocity and the geomechanical
Fig. 9. Genetic Algorithm Workflow [53].
properties from the conventional well logs. Finally, the ANN
models showed a better line of best fit than linear regression and
root mean square error, as well as the average absolute error, which
which means all the algorithms that match the field operations and was observed to be less than the linear models.
their models’ characteristics are already verified and tested to be
applied in reservoirs with similar petrophysical and geo- 3.2. Fuzzy rule-based systems or fuzzy inference system
mechanical properties. Prytz et al. [26] concluded in his paper
that synthetic data does not represent the real data and the envi- The Fuzzy Rule-Based System was introduced by Ref. [51]. The
ronment so that it is necessary to test all the algorithms on real field objective of this system is to figure out the uncertainty that resulted
data to be verified. Table 4 summarizes the most recent and in the inaccuracy of data and to determine their reasons for un-
commonly used ML and Al algorithms in shale gas petrophysics and certainty. The yield or result of an occasion in an arbitrary pro-
geomechanical projects. cedure and profoundly subject to the possibility or probability of
In the most recent literature on petrophysics, AI and ML have the event. Consequently, likelihood hypothesis usage is appropriate
been highlighted with some major benefits over the conventional in taking care of an issue when the vulnerability is a resultant of
methodology been used previously in the following list of technical occasion haphazardness. The subject system makes several at-
areas [21]: tempts to recognize an association between the given input and

163
F.I. Syed, A. AlShamsi, A.K. Dahaghi et al. Petroleum 8 (2022) 158e166

Table 5
The summary of all methods that are being used in shale gas petrophysical and geo-mechanical modeling are summarized.

Algorithms Description Pros Cons or limitations

Fuzzy Fuzzy systems are used to predict permeability in shale Fuzzy logic (FL) even produces a valid response if input Not feasible in the selection of
interference gas reservoirs using well-log responses. information is incomplete, noisy, vague, and imprecise. inputs parameter.
system
Integrated neural NNS and FL was applied to predict the friability of shale Used for complex modeling and high nonlinear Not work properly in the selection
networks (NNs) in shale gas reservoir problems whose can’t be resolved by a common of optical parameters.
and fuzzy logic statistical and analytical approach
(FL)
Genetic GA was used to predict shear and compressional GA functions are discontinuous, highly nonlinear, non- Limited by number of data sets, the
algorithms (GA) velocities in a shale gas reservoir for finding differentiable and evenly stochastic more data sets the more accurate
geomechanical properties results will be
Artificial neural Predicts the geomechanical properties of Upper Bakken ANN had a better line of best fit than linear regression The ANN had to be trained using
networks (ANN) shale by adopting the ANN method and root mean square error and average absolute error backpropagation technique with
was less than linear models for ANN multilayer perceptron for better
results.
Extreme machine ELM method was used for prediction of the brittleness of ELM models show its great advantage in running speed When compared with BPANN, the
learning (ELM) shale in shale gas reservoir and required less time when processing brittleness accuracy was found less and is
limited to 3 logs (SP, DEN, and GR)
for better performance.
Least squares- extension of support vector machine to predict the The LS-SVR model yielded smaller errors than the LS-SVR model needs to determine
support vector geomechanical properties (Brittleness Index of shale) BPANN model when comparing and give more accurate the parameters by cross-validation,
regression (LS- results which is time-consuming
SVR)
Gaussian mixture clustering approach was adopted in shale gas GMM found to have the highest consistency with the It is sensitive to fluid distribution in
model (GMM) petrophysics for fluid characterization on T1-T2 maps theoretical fluid boundary than other clustering pores of shale
(clustering) algorithms k-means and spectral clustering and is
suitable for multiple fluid components present in
different pore systems of shale
Random forest Geomechanical studies by random forest included RF is successful in predicting facies changes and the Unlike Bayesian Network, RF is not
(RF) variations in properties as a result of facies changes and presence of fractures in the shale gas reservoir with high the best when it comes to the data
the presence of fractures in shale gas accuracy. It could be applied to large projects that have insight
limited core data Computational time is more for RF
compared to BN
Bayesian To estimate, sort, and predict the existence of facies in Less computational time than RF for the same work BN is less accurate than RF when
network (BN) the Marcellus and incidence of fractures. Also, it “was BN was more relevant in terms of petrophysical data and compared with each other
applied to learn the petrophysical data pattern associated output.
associated with different facies in shale gas reservoirs”
(Shuvajit, 2017)
Convolutional CNN was used to generate NMR T2 by processing CNN showed detection robustness regarding real images Require the use of larger and more
neural network conventional well logs for improving pore size Robust to noise in input logs realistic datasets in the
(CNN) distribution in shale gas reservoir and detection of Convolutional neural network
fractures. training process
Support vector SVM is applied in estimating total organic carbon (TOC) SVM provided a convenient method to extend TOC Choice of attributes is crucial and
machine (SVM) in a Barnett Shale using triple combo logs coverage to a greater area. requires great insight in applying
this method to generate TOC
volume.

output values through characterizing a few if rules and is made of variables, which include hydraulic-fracturing stages, hydraulic frac-
certain components that are depicted in the system. Ordinarily, the turing spacing, hydraulic fracturing half-length, the spacing of
input and output information is represented by a curve. The curves wellbore, and stochastic gradient-free optimization methods that
describe a few loads to the input and characterize how it tends to be include genetic algorithms. This method had been applied to a
mapped on the resulted output. The rules are the directions that synthetic shale gas reservoir that has properties similar to Barnett
characterize the contingent articulations. They, alongside the shale. A detailed workflow of the Genetic Algorithm is given in Fig. 9
membership functions, map input onto the output. In 2016, Gomide [53].
gave the complete review on a fuzzy system that explains, by using
well-logs, fuzzy systems can be used to predict the permeability in
sedimentary rocks. This approach shows a global nonlinear rela-
tionship between permeability and a log response that represents a 3.4. Integrated neural networks (NNs) and fuzzy logic (FL)
smooth finite flexible local sub-model. Fuzzing Clustering algo-
rithms are also used to obtained local input-output relations Neural Networks (NNs) are significantly less complex that has
automatically from the observed measurements. It’s very simple to been inspired by the human brain. It is normally used for complex
contract fuzzy systems without computational efforts. This modeling and higher nonlinear problems that can’t be resolved by a
approach is applied in the Lake Maracaibo basin, Venezuela [52]. common statistical and analytical approach [54]. In Neural net-
works, the input data is commonly referred to as hierarchical nodes
3.3. The genetic algorithm called neurons that are trained. Some functions are operating to
map this input for the desired outputs that are initially generated
The primary objective of the genetic algorithm in shale gas geo- randomly. While the Fuzzy logic is an algorithm-based approach
mechanics is to calculate Net Present Value (NPV) in which the used for computing grounded on degrees of relevance to the actual
pseudo-pressure method is integrated with the Wattenbarger Slab input data rather than the usual true or false approach on which the
model [28]. Depending on the geomechanics, there are decision modern computer works.
164
F.I. Syed, A. AlShamsi, A.K. Dahaghi et al. Petroleum 8 (2022) 158e166

3.5. Least Squares-Support Vector Regression (LS-SVR) Geology, vol. 2, American Association of Petroleum Geologists, Tulsa, OK, USA,
1994.
[10] F.I. Syed, M. Boukhatem, A.A. Al Kiyoumi, Lean HC gas injection pilots analysis
Least Squares-Support Vector Regression (LS-SVR) is an exten- and IPR back calculation to examine the impact of asphaltene deposition on
sion of a support vector machine that predicts the geo-mechanical flow performance, Petrol. Res. 4 (1) (2019) 84e95.
properties i.e. Brittleness Index of shale. In this method, the well [11] F.I. Syed, S. Al Saadi, O. Khedr, M.Y. Khan, R. Yusuf, K.K. Thakur, November.
Successful application of a mechanistic coupled wellbore-reservoir dynamic
logs are considered as inputs, output as brittleness index, and they simulation model to history match and plan cleanup operation of long hori-
compare their model with another artificial intelligence-based zontal wells, in: Abu Dhabi International Petroleum Exhibition & Conference,
model through backpropagation artificial neural network Society of Petroleum Engineers, 2016.
[12] F.I. Syed, et al., Numerical Validation of Asphaltene Precipitation and Depo-
(BPANN). The accuracies of both models are then tested with sition during CO2 Miscible Flooding, Petroleum Research, 2020, https://
actually measured findings. After several attempts, LS-SVR is found doi.org/10.1016/j.ptlrs.2020.04.002.
to be more accurate than BPANN [55]. [13] F.I. Syed, M. Alshamsi, A.K. Dahaghi, S. Neghabhan, Artificial Lift System
Optimization Using Machine Learning Applications, Petroleum, 2020.
[14] F.I. Syed, A.H. Tunio, N.A. Ghirano, Compositional analysis and screening for
3.6. Extreme learning machine (ELM) algorithm enhanced oil recovery processes in different reservoir and operating condi-
tions, Int. J. Appl. 1 (4) (2011) 143e160.
[15] Sagar Kale, Chandra Rai, Carl Sondergeld, Rock typing in gas shales, in: SPE
Extreme machine learning is regarded as the Generalized Annual Technical Conference and Exhibition. Society of Petroleum Engineers,
Single-Layer Feedforward Neural Network and ELM works with the 2010.
random initialization between output and input layer weights [16] I. Gupta, C. Rai, D. Devegowda, C. Sondergeld, Use of data analytics to optimize
hydraulic fracture locations along borehole, Soc. Petrophysicists Well-Log
[53,56]. Anal. 59 (06) (2018) 811e825.
[17] Ishank Gupta, Carl Sondergeld, Chandra Rai, Applications of nano indentation
for reservoir characterization in shales, in: 52nd US Rock Mechanics/Geo-
4. Summaries and conclusions mechanics Symposium, American Rock Mechanics Association, 2018.
[18] Xiaodong Ma, Mark D. Zoback, Lithology-controlled stress variations: a case
All methods that are being used in shale gas petrophysical and study of the woodford shale, Oklahoma, in: Unconventional Resources
Technology Conference, Austin, Texas, 24-26 July 2017, Society of Exploration
geomechanical modeling are summarized in Table 5. It is a quan-
Geophysicists, American Association of Petroleum Geologists, Society of Pe-
titative piece of work covering shale gas petrophysics and geo- troleum Engineers, 2017.
mechanics and tries to answer the defined set of research questions [19] Naeini, Ehsan Zabihi, et al., An integrated deep learning solution for petro-
physics, pore pressure, and geomechanics property prediction, Lead. Edge 38
by using machine learning and artifificial intelligence. It’s been
(1) (2019) 53e59.
studied that the number of publications has been increasing [20] Q.R. Passey, et al., A practical model for organic richness from porosity and
through the years and most of the work is presented in conferences. resistivity logs, AAPG Bull. (1990) 1777e1794, 74.12.
SPE, SPWLA, and Unconventional Resources Technology Confer- [21] Xu, Misra, When petrophysics meets big data: what can machine do?, in: SPE
Middle East Oil and Gas Show and Conference Society of Petroleum Engineers,
ence shared most publications on ML and AI shale gas petrophysics 2019.
and geomechanics. It is also noted that ML and AI techniques have [22] T.D. Jobe, E. Vital-Brazil, M. Khaif, Geological Feature Prediction Using Image-
been increasing in shale gas petrophysics and geomechanics more Based Machine Learning, Society of Petrophysicists and Well-Log Analysts,
2018.
significantly in the last couple of years. This significant increase is [23] X. Zhou, B. Liu, Z. Wu, Y. Feng, Integrative mining of traditional Chines
due to global growing interest in shale gas reservoirs, as conven- medicine literature and MEDLINE for functional gene networks, Artif. Intell.
tional reservoirs are being depleted. Med. 41 (2) (2007) 87e104.
[24] S. Wang, Y. Wang, W. Du, F. Sun, X. Wang, C. Zhou, Y. Liang, A multi-ap-
In this review, it is also noted that the ML and AI are being proaches-guided genetic algorithm with application to operon prediction,
applied in petrophysics and geomechanical modeling to identify Artif. Intell. Med. 41 (2) (2007) 151e159.
reservoir zones and predict the geo-mechanical properties such as [25] Y. Xiao, J. Xiao, F. Lu, S. Wang, Ensemble ANNs-PSO-GA approach for day-
ahead stock e-exchange prices forecasting, Int. J. Comput. Intell. Syst. 6 (1)
fracability, which is a critical parameter during hydraulic fracturing.
(2013) 96e114.
Additionally, it is remarked that there are multiple techniques like [26] R. Prytz, S. Nowaczyk, T. Ro €gnvaldsson, S. Byttner, Predicting the need for
ANN, Fuzzy system, Genetic Algorithms, etc. that have been suc- vehicle compressor repairs using maintenance records and logged vehicle
data, Eng. Appl. Artif. Intell. 41 (2015) 139e150.
cessfully applied on different shale gas operational wells data set,
[27] G.K. Parapuram, M. Mokhtari, J.B. Hmida, Prediction and analysis of geo-
and these techniques have been used to discover new prospects of mechanical properties of the upper bakken shale utilizing artificial intelli-
shale gas reservoirs. gence and data mining, Prediction and Analysis of Geomechanical Properties
of the Upper Bakken Shale Utilizing Artificial Intelligence and Data Minin 10
(2017).
References [28] Hamid Rahmanifard, Tatyana Plaksina, Application of fast analytical approach
and AI optimization techniques to hydraulic fracture stage placement in shale
[1] Hiren Maniar, et al., Machine-learning methods in geoscience, in: SEG Tech- gas reservoirs, J. Nat. Gas Sci. Eng. 52 (2018) 367e378.
nical Program Expanded Abstracts 2018, Society of Exploration Geophysicists, [29] M. Rajabi, M. Tingay, November. Applications of intelligent systems in pe-
2018, pp. 4638e4642. troleum geomechanics-Prediction of geomechanical properties in different
[2] Mohammad Anemangely, et al., Machine learning technique for the prediction types of sedimentary rocks, in: International EAGE Workshop on Geo-
of shear wave velocity using petrophysical logs, J. Petrol. Sci. Eng. 174 (2019) mechanics and Energy, 2013.
306e327. [30] A.R. Venna, Y.Y. Ang, N. Nguyen, Y. Lu, D. Walters, Support-vector-machine
[3] Hao Li, Jiabo He, Siddharth Misra, Data-driven in-situ geomechanical charac- phase classification of downhole leak flows based on acoustic signals, Soc.
terization in shale reservoirs, in: SPE Annual Technical Conference and Exhi- Petrophysicists Well-Log Anal. 59 (06) (2018) 841e848.
bition, Society of Petroleum Engineers, 2018. [31] T. Zhao, S. Verma, D. Devegowda, V. Jayaram, December. TOC estimation in the
[4] Benjamin K. Sovacool, Cornucopia or curse? Reviewing the costs and benefits Barnett Shale from triple combo logs using support vector machine, in: 2015
of shale gas hydraulic fracturing (fracking), Renew. Sustain. Energy Rev. 37 SEG Annual Meeting, Society of Exploration Geophysicists, 2015.
(2014) 249e264. [32] H. Li, S. Misra, Long short-term memory and variational autoencoder with
[5] Eia report, 2014, http://www.eia.gov. convolutional neural networks for generating NMR T2 distributions, Geosci.
[6] Craig L. Cipolla, Elyezer Lolon, Michael J. Mayerhofer, Reservoir modeling and Rem. Sens. Lett. IEEE 16 (2) (2018) 192e195.
production evaluation in shale-gas reservoirs, in: International Petroleum [33] L. Zhu, H. Li, Z. Yang, C. Li, Y. Ao, Intelligent logging lithological interpretation
Technology Conference, International Petroleum Technology Conference, with convolutional neural networks, Soc. Petrophysicists Well-Log Anal. 59
2009. (06) (2018) 799e810.
[7] Carl H. Sondergeld, et al., Petrophysical considerations in evaluating and [34] J. Han, H. Daigle, X. Tian, M. Pyrcz, C. Griffith, B. Zhang, A comparison of
producing shale gas resources, in: SPE Unconventional Gas Conference, Soci- clustering algorithms applied to fluid characterzation using NMR T1-T2 maps
ety of Petroleum Engineers, 2010. of shale, Comput. Geosci. (2019).
[8] J.H. Doveton, Log Analysis of Subsurface Geology: Concepts and Computer [35] S. Bhattacharya, S. Mishra, Applications of machine learning for facies and
Methods, John Wiley & Sons, Inc., New York, NY, USA, 1986. fracture prediction using Bayesian Network Theory and Random Forest: case
[9] J.H. Doveton, Geologic Log Analysis Using Computer Method: Application in studies from the Appalachian basin, USA, J. Petrol. Sci. Eng. 170 (2018)

165
F.I. Syed, A. AlShamsi, A.K. Dahaghi et al. Petroleum 8 (2022) 158e166

1005e1017. SPWLA 2018 STC Panel Discussions Summary, SPWLA Newsletter September
[36] O. Akinnikawe, S. Lyne, J. Roberts, Synthetic well log generation using ma- 2018, Houston TX, 2018. April 16-17.
chine learning techniques, Unconv. Resour. Technol. Conf. (2018), https:// [46] E. Frost, T. Quinn, Data preconditioning for predictive and interpretative al-
doi.org/10.15530/URTEC-2018-2877021. gorithms: importance in datadriven analytics and methods for application,
[37] S. Agar, R. Goteti, D. Jobe, S. Zhang, Y. Zaretskiy, Geologic Risking with Soc. Petrophysicists Well-Log Anal. 59 (06) (2018) 873e890.
Bayesian Methods: Fracture Case Study in an Unconventional Setting. [47] S. Biswal, G.R. Sabareesh, in: Design and Development of a Wind Turbine Test
GEO2018 Conference Held in Manama, 2018. Bahrain. Rig for Condition Monitoring Studies. 2015 International Conference on In-
[38] Q. Shen, J. Chen, H. Wang, Data-driven interpretation of ultradeep azimuthal dustrial Instrumentation and Control, IEEE, 2015, pp. 891e896. ICIC 2015.
propagation resistivity measurements: transdimensional stochastic inversion [48] X. Shi, G. Liu, S. Jiang, L. Chen, L. Yang, November. Brittleness index prediction
and uncertainty quantification, Soc. Petrophysicists Well-Log Anal. 59 (6) from conventional well logs in unconventional reservoirs using artificial in-
(2018). telligence, in: International Petroleum Technology Conference, International
[39] S. Karimpouli, A. Malehmir, Neuro-Bayesian facies inversion of prestack Petroleum Technology Conference, 2016.
seismic data from a carbonate reservoir in Iran, J. Petrol. Sci. Eng. 131 (2015) [49] C.I. Noshi, J.J. Schubert, The Role of Machine Learning in Drilling Operations; A
11e17. Review, Society of Petroleum Engineers, 2018, October 5, https://doi.org/
[40] T. Zimmermann, L. Liang, S. Zeroug, Machine-learning-based automatic well- 10.2118/191823-18ERM-MS.
log depth matching, Soc. Petrophysicists Well-Log Anal. 59 (06) (2018) [50] Sid-Ali Ouadfeul, Leila Aliouane, Shale gas reservoirs characterization using
863e872. neural network, Energy Procedia 59 (2014) 16e21.
[41] D. Chen, C. Jones, B. Dai, A. van Zuilekom, Fluid Optical Database Recon- [51] L.A. Zadeh, Information and control, Fuzzy Sets 8 (3) (1965) 338e353.
struction with Validated Mapping from External Oil and Gas Information [52] Jose Finol, Xu-Dong D. Jing, Permeability prediction in shaly formations: the
Source, Society of Petrophysicists and Well-Log Analysts, 2018. fuzzy modeling approach, Geophysics 67 (3) (2002) 817e829.
[42] M.M. Korjani, A.S. Popa, E. Grijalva, S. Cassidy, I. Ershaghi, Reservoir Charac- [53] M. Sayyafzadeh, A. Keshavarz, Optimisation of gas mixture injection for
terization Using Fuzzy Kriging and Deep Learning Neural Networks, Society of enhanced coalbed methane recovery using a parallel genetic algorithm, J. Nat.
Petroleum Engineers, 2016, https://doi.org/10.2118/181578-MS. Gas Sci. Eng. 33 (2016) 942e953.
[43] J. He, S. Misra, H. Li, Comparative Study of Shallow Learning Models for [54] C.M. Bishop, Neural Networks for Pattern Recognition, Oxford university
Generating Compressional and Shear Traveltime Logs, Society of Petrophysi- press, 1995.
cists and Well-Log Analysts, 2018. [55] M.A. Ahmadi, M. Pournik, A predictive model of chemical flooding for
[44] Z. Tariq, A. Abdulraheem, M. Mahmoud, A. Ahmed, A Rigorous Data-Driven enhanced oil recovery purposes: application of least square support vector
Approach to Predict Poisson’s Ratio of Carbonate Rocks Using a Functional machine, Petroleum 2 (2) (2016) 177e182.
Network. Society of Petrophysicists and Well-Log Analysts. Theys, Editions [56] G.B. Huang, An insight into extreme learning machines: random neurons,
Technip, Paris, France, 2018, p. 2011. Quest for Quality Data. random features and kernels, Cogn. Comput. 6 (2014) 376e390.
[45] C. Xu, S. Misra, Petrophysics Data-Driven Analytics: Theories and Applications.

166