SPE-217610-MS

Advancements in Applications of Machine Learning for Formation Damage

Predictions

Downloaded from http://onepetro.org/SPECTCE/proceedings-pdf/23CTC/3-23CTC/D031S022R011/3329096/spe-217610-ms.pdf/1 by Petrobras user on 27 June 2024

T. E. Abdulmutalibov, Y. Y. Shmoncheva, and G. V. Jabbarova, Azerbaijan State Oil and Industry University, Baku,
Azerbaijan

Copyright 2023, Society of Petroleum Engineers DOI 10.2118/217610-MS

This paper was prepared for presentation at the SPE Caspian Technical Conference and Exhibition scheduled to be held in Baku, Azerbaijan, 21 – 23 November 2023.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents
of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect
any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written
consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may
not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
Reservoir damage is a critical a major concern within the oil and gas sector that has the potential to have
a significant impact reduce reservoir productivity. Traditional methods of repairing formation damage are
frequently requiring a substantial amount of manual effort and consuming a considerable amount of time.
This study delves into the utilization of machine learning methods as a promising solution for predicting,
mitigating, and managing reservoir damage. The study begins with a discussion of the various elements
that lead to the occurrence of formation damage, including rock-fluid interactions, drilling operations, and
production processes. It then highlights the limitations of traditional methods and emphasizes the need for
data-driven approaches. Machine learning models such as support vector machines, regression analysis,
and neural networks are introduced as tools for analyzing large data sets derived from reservoir modeling,
wellbore data, and production history. These models identify key parameters and patterns associated with
formation damage, which helps predict potential damage. Additionally, this research paper investigates
the application of machine learning for optimizing drilling and completion strategies with the aim of
reducing the likelihood of formation damage. It addresses the incorporation of real-time data monitoring
and predictive analytics to enhance reservoir management methodologies.
The paper presents case studies and practical implementations of machine learning aimed at mitigating
formation damage. These examples illustrate the potential for enhancing reservoir performance, cutting
operational expenses, and boosting hydrocarbon production. It also outlines challenges and future directions
for research in this area, highlighting the importance of continued innovation in machine learning and data
mining methods to promote the sustainable growth of the oil and gas sector.
In conclusion, the application of machine learning for formation damage management represents a
transformative approach to address a critical challenge in the oil and gas sector. This research contributes
to the development of knowledge and practical implementation of machine learning methods to optimize
reservoir performance while minimizing the effects of reservoir damage.
2 SPE-217610-MS

Introduction
Formation damage, a phenomenon encountered in the oil and gas industry, presents a persistent challenge
that can have profound implications for reservoir productivity and profitability (Liu & Civan, 1993).
This occurs when various factors, such as drilling operations, formation fluid interactions, and production
activities, lead to a deterioration in the formation’s ability to efficiently deliver hydrocarbons. Traditional
methods of combating formation damage are often inadequate, requiring innovative approaches to
prediction, prevention and mitigation (Kalam et al., 1997).
In recent years, the intersection of artificial intelligence and the oil and gas sector has opened up promising

Downloaded from http://onepetro.org/SPECTCE/proceedings-pdf/23CTC/3-23CTC/D031S022R011/3329096/spe-217610-ms.pdf/1 by Petrobras user on 27 June 2024

opportunities to overcome the complexities associated with reservoir damage (Cundar et al., 2023). Machine
learning, a subset of artificial intelligence, has gained significant traction as a powerful tool for analyzing
large and intricate datasets, identifying hidden patterns, and making informed decisions (Effiong et al.,
2021). Its potential applications in reservoir management and formation damage mitigation have sparked
considerable interest among researchers, engineers, and industry professionals (Effiong et al., 2021; Jiang
et al., 2022; Yang et al., 2023).
This review article embarks on a comprehensive exploration of the burgeoning field of machine learning
as applied to formation damage in the oil and gas industry (Effiong et al., 2021). By examining the latest
research, case studies, and practical implementations, we aim to provide a deep understanding of the role that
machine learning can play in revolutionizing the solution to formation damage. From predictive analytics
to real-time monitoring (Zuluaga, 2000) and optimization of drilling strategies (Al Gharbi et al., 2018),
machine learning offers the promise of enhancing reservoir management practices, reducing operational
costs, and ultimately maximizing hydrocarbon recovery (Gholami et al., 2012).
Throughout that review, we delved into the key concepts, methodologies, and successes of machine
learning in formation damage management, shedding light on the transformative potential of this
technology. We also addressed the challenges and limitations that arose when integrating machine learning
into the industry, emphasizing the need for ongoing research and collaboration to fully exploit its capabilities
(Ikiensikimama & Azubuike, 2012).
To summarize, formation damage remains a significant challenge within the oil and gas sector, affecting
reservoir productivity and profitability (Yuan & Wood, 2018). Traditional methods of solving this problem
are failing, leading to a shift towards innovative solutions (Yuan & Wood, 2018). This review explores
the burgeoning application of machine learning as a transformative tool to predict, prevent, and mitigate
formation damage (Effiong et al., 2021). We delve into the latest research and practical implementations,
demonstrating how machine learning can improve reservoir management, reduce operating costs and
optimize hydrocarbon production. By acknowledging the challenges, our journey reveals the potential of
machine learning to redefine industry standards and enable more efficient and sustainable oil and gas
production (Nnamdi & Adelaja, 2019).

Review
Diagnosing formation damage is an intricate task due to the wide array of potential issues that can arise,
demanding substantial practical experience and technical expertise. Becoming an expert in this domain
entails the ability to comprehend and retain findings from various technical publications and apply this
knowledge in a methodical manner. Recently, there have been developments in the form of programs
referred to as "Expert" or "Artificial Intelligence" (AI) systems, which can perform tasks akin to those
executed by qualified experts in specialized fields. To utilize these programs, a "knowledge base" must
initially be encoded and stored in the computer’s memory, after which the Expert System can roughly
simulate the analysis conducted by an expert (Garrouch et al., 2006; Moridis et al., 2011; Nashawi &
Malallah, 2009).
SPE-217610-MS 3

This paper (Alegre & Dougherty, 1988) presents an update on the progress of an expert system designed
for analyzing formation damage issues. The objective behind its development is to showcase the feasibility
of providing an automated formation damage advisor to the average engineer. In conclusion, it is emphasized
that employing AI for the diagnosis of formation damage represents a promising and productive application
of this technology.
Traditionally, water injection techniques in mature oil fields located in the Middle Magdalena Valley of
Colombia have had limited success. However, the study (Nunez Garcia et al., 2015) offers a comprehensive
approach to reservoir damage analysis. This approach includes damage modeling, characterization of

Downloaded from http://onepetro.org/SPECTCE/proceedings-pdf/23CTC/3-23CTC/D031S022R011/3329096/spe-217610-ms.pdf/1 by Petrobras user on 27 June 2024

reservoir and injected water properties, and specialized laboratory testing. The primary goal is to develop an
effective methodology for selecting candidate wells by identifying and quantifying mechanisms that cause
reservoir damage (such as fines mobilization, solids intrusion, and emulsion formation). The ultimate goal
is to make the optimal choice of stimulation procedures.
The formation damage affecting the field’s injection wells is believed to have been caused during drilling
and completion of the wells, as well as during the initial water injection initiated by the operator thirty years
ago (Nunez Garcia et al., 2015). To address this issue, the following steps were taken:
1. Comprehensive data collection and analysis, including compatibility tests between formation fluids
and drilling/completion fluids.
2. Recovery and characterization of solids plugging the injection ports.
3. Core sampling from representative areas to assess mineralogical composition and damage.
4. Evaluation of the possibility of mineral deposits caused by the mixture of injection and formation
water in downhole conditions using specialized software.
5. Development of a complex reservoir damage model using geochemical simulator software that
incorporates numerical modeling, reservoir studies, nodal analysis, mineral/organic studies, and
oxidation studies. This model allowed for the identification and quantification of each damage
parameter, including 3D modeling of some aspects.
After determining the parameters of man-made formation damage, a series of laboratory studies were
planned and carried out to diagnose, determine and evaluate the most effective methods for dissolving and
mitigating the damage. These studies included assessing the effects of formation/injection water salinity, as
well as testing the sensitivity of the formation to fine particle dissolution treatments. Key factors considered
in the analysis included critical velocity, residual permeability, core blowouts.
A paper (Nunez Garcia et al., 2015) presented the full field and methodology results of a perfect
application on a select group of injection wells. These wells had recovered desired water injection rates
beyond expectations and had contacted new zones to drain new oil reserves.
Investigating the preservation (Jiaojiao et al., 2010) of target formations with low porosity and low
permeability represents an advanced research initiative in the realm of petroleum engineering, carrying
significant implications for enhancing reserves and production levels. Traditionally, researchers have
employed core flooding tests to gain insights into the causes of formation damage. However, conducting
these tests becomes challenging due to the exceptionally low permeability of such formations, and obtaining
representative core samples can prove difficult in certain reservoirs.
In this research study (Jiaojiao et al., 2010), the primary focus was on conventional formations
characterized by low porosity and low permeability. The study successfully developed mathematical models
to quantitatively anticipate five distinct sensitivities of these formations and evaluate the damage incurred
from water blocking. These models were established by drawing upon a wealth of experimental data and
applying principles from petrophysics and interfacial chemistry specific to reservoirs, along with utilizing
intelligent techniques such as Artificial Neural Networks (ANN).
4 SPE-217610-MS

To assist in the identification of damage mechanisms and guide the development of drilling fluids, a user-
friendly software tool has been designed for predicting formation sensitivity and water cut. This software
provides a practical alternative to and labor-intensive laboratory tests, as it only requires the input of relevant
formation parameters. Its widespread adoption has demonstrated a high degree of reliability, with more than
85% of the predicted outcomes aligning with measured results, affirming its effectiveness.
The research conducted (Jiaojiao et al., 2010) has enabled the identification of the primary mechanisms
responsible for causing damage in formations characterized by low porosity and permeability. These
mechanisms typically include water sensitivity and water blockage. The study (Jiaojiao et al., 2010)

Downloaded from http://onepetro.org/SPECTCE/proceedings-pdf/23CTC/3-23CTC/D031S022R011/3329096/spe-217610-ms.pdf/1 by Petrobras user on 27 June 2024

successfully developed an artificial neural network (ANN) model and software to predict reservoir
sensitivity. Extensive research has shown that more than 85% of predicted results match measured data.
The degree of hydraulic blockage damage is influenced by multiple factors, including permeability,
the interfacial tension between oil and water, porosity, and initial water saturation. Our gray ANN model
merges the strengths of gray static models with the nonlinear capabilities of a neural network to enhance
the precision of predictions related to water blockage.
Some surfactants or alcohols may be useful in reducing the surface tension of filtrates and the interfacial
tension between oil and water, as well as reducing capillary resistance and preventing damage due to water
blockage. In this study (Jiaojiao et al., 2010), a new drilling fluid was developed that effectively protects
typical low-permeability reservoirs and has excellent performance characteristics. The return permeability
of two cores treated with this drilling fluid was more than 88%.
A neural network was developed to predict skin factor, a critical factor in the context of formation damage
during drilling, well completion, and oil and gas production (Tan et al., 2020). The model’s performance
was assessed using four key metrics: goodness-of-fit (R2), mean square error (MSE), root mean square error
(RMSE), and average absolute percentage relative error (AAPRE) (Effiong et al., 2021). It is important to
note that all predictions made by the model closely matched the actual measurements.
Also, an assessment of the relative importance of the input parameters in the model showed that
permeability (k), flow rate (q), porosity (φ) and pressure drop (Δp) significantly influenced the calculated
values of the skin factor (S) obtained from well measurements. Therefore, if this developed model were
integrated into a downhole measurement tool capable of capturing fundamental reservoir parameters such
as pressure, flow rate, permeability, viscosity, and thickness, it could simplify the skin factor estimation
process in the petroleum industry, eliminating the need for a diagnostic approach.
Two mathematical models were developed (Erbas & Gumrah, 2001), incorporating a straightforward
genetic algorithm (GA), a widely-used artificial intelligence optimization technique. These models were
designed to simulate different aspects of petroleum reservoir behavior: one focused on damage caused by
geochemical processes and clay swelling, while the other aimed to enhance reservoir conditions through
acidization.
The first model’s purpose is to predict changes in permeability resulting from the acidization process. It
employs an analytical solution that relies on two critical parameters representing the acidization process:
Damkohler and acid capacity numbers. These parameters were fine-tuned using a basic genetic algorithm,
aligning calculated results with experimental data.
The second model is dedicated to mathematically simulating formation damage attributed to various
factors, including deposition, entrainment, dissolution, and clay swelling. This is accomplished by applying
fundamental balance equations. To optimize this model, a straightforward genetic algorithm was employed
to adjust three parameters: deposition, dissolution, and entrainment rate constants, which serve as indicators
of damage. This optimization process involved comparing calculated outcomes with experimental data.
Both of these models underwent validation (Erbas & Gumrah, 2001) through case studies, and they
effectively predicted changes in porosity and permeability.
SPE-217610-MS 5

ANN is a fundamental concept in the artificial intelligence and machine learning field`s. Using layers
of interconnected nodes or "neurons," ANNs can process complex patterns, recognize relationships in data,
and make predictions or decisions.
In this study (Denney, 2007), a neural network was employed to predict the initial pressure, permeability,
and skin factor of reservoir based on data obtained from pressure build-up tests.
1. Calculation of initial reservoir pressure
ANN was used to calculate the initial reservoir pressure. This is achieved using the Matlab toolbox.
2. Permeability Calculation:

Downloaded from http://onepetro.org/SPECTCE/proceedings-pdf/23CTC/3-23CTC/D031S022R011/3329096/spe-217610-ms.pdf/1 by Petrobras user on 27 June 2024

Permeability stands out as the most crucial rock property governing the flow of formation fluids.
In the context of field development, reservoir management, and the design of enhanced oil recovery
strategies, understanding rock permeability and its spatial distribution within the reservoir holds
paramount significance.
3. Skin Factor Calculation:
The pressure drop within a wellbore results from various contributing factors. In addition to
formation resistance and fluid viscosity, an additional form of resistance concentrated around the
wellbore plays a role. This phenomenon, known as the skin effect, diminishes the productivity of the
well. The extent of this damage or improvement is quantified using the skin factor, which can assume
either positive or negative values. We employed an artificial neural network (ANN) to compute the
skin factor of wells. Our training dataset comprises data from six wells, and the ANN’s learning
outcomes are outlined in Tables 1–3.

Table 1—Estimation of initial pressuresd

Well No. Initial pressure, MPa Initial pressure ANN, MPa Error, %

1 23 22.8988 0.44

2 15 14.934 0.26

3 25 24.89 0.12

4 30 29.868 0.09

5 29 28.8724 0.06

6 22.7 22.60012 0.03

Table 2—Permeability assessment

ANN permeability,
Well No. Permeability, millidarcy Error, %
millidarcy

1 100 99.56 6.5

2 269 267.8164 5.4

3 61 60.7316 5

4 112 111.5072 4.9

5 480 477.888 4.5

6 142 141.3752 4.1

6 SPE-217610-MS

Table 3—Skin factor assessment

Well No. Skin factor ANN skin factor Error, %

1 4.12 4.101872 7

2 3.8 3.78328 6.7

3 2.1 2.09076 5.9

4 3.3 3.28548 4.1

5 -1.2 -1.19472 3.9

Downloaded from http://onepetro.org/SPECTCE/proceedings-pdf/23CTC/3-23CTC/D031S022R011/3329096/spe-217610-ms.pdf/1 by Petrobras user on 27 June 2024

6 1.7 1.69252 3.2

It is evident from the results that there is excellent concordance between the pressure and permeability
data and the training data. Table 1 offers a comparison of the initial pressure readings, demonstrating their
substantial agreement. Table 2 presents the calculated permeability values, showcasing the neural network’s
ability to predict reservoir permeability accurately using pressure build-up data. Table 3 reports the outcomes
of skin factor calculations. Across all tables, the decreasing error values affirm that the program has been
meticulously trained and can be applied to simulate the initial pressure, permeability, and skin factor of any
reservoir effectively.
The primary source of production in the Shushufindi-Aguarico field (SSFD) located in Ecuador comes
from three stacked reservoir sands: T, U sands, and the basal Tena, as described (Rodriguez et al., 2014).
The SSFD is characterized as being undersaturated and exhibits two concurrent production mechanisms.
The initial mechanism is linked to the drive caused by gas dissolution, while second is connected to an
active bottom and lateral aquifer. This second mechanism results in relatively high recovery rates, typically
ranging between 24% and 29%, but it is also accompanied by a significant water cut in most wells, especially
those producing from the T sand.
The reservoir displays compartmentalization as a result of stratigraphic pinch out, resulting in varying
pressure regimes for each of the sands: The T sand ranges from 0.0145 to 0.0179 GPa, the U sand from
0.0096 to 0.0206 GPa, and the basal Tena sand from 0.00827 GPa. Natural reservoir pressure limits flow,
requiring artificial lift techniques (Rodriguez et al., 2014).
The main method of mechanized production at the field is an electric submersible pump (ESP), which
is used in 106 wells. There are also 5 wells utilizing hydraulic pumping, 1 well employing gas lift, and 1
well using beam pumping as alternative lift methods.
The saturation pressure of U and T sands varies from 0.0069 to 0.0073 GPa. In some cases, some wells
are operated with dynamic bottomhole pressure (Pwf) that drops below the saturation pressure, with Pwf
reaching approximately 0.0041 GPa.
Wells in the field were initially completed using 0.1397 m × 25.29 kg/m and 0.1778 m × 38.69 to 43.15
kg/m casing. However, in the past three years, wells have been completed with 0.2444 m × 69.94 to 78.87
kg/m casing and 0.1778 m × 43.15 kg/m or 38.69 kg/m liners.
The wells are configured with single bore completions, also targeting the T or U sand, or selectively
when both sands are to be produced sequentially, facilitated by a sliding door. In 2012, experiments
were conducted with concentric dual completions in four wells. However, this approach posed certain
complexities and limitations, leading to the potential for extended and risky future workover operations.
A critical aspect of production to take into account is the regulatory authority overseeing hydrocarbons
in Ecuador, ARCH, does not allow joint development of the U and T sands due to reservoir management
and oil accounting issues.
The operator, Consorcio Shushufindi (CSSFD) (Rodriguez et al., 2014), is considering intelligent
completion (IC) or compact IC to address technical and regulatory challenges. The objectives include
evaluating and testing ICs and compact ICs in the field.
SPE-217610-MS 7

The initiative has made significant progress (Rodriguez et al., 2014) in gathering information on candidate
wells, including architectural requirements, conceptual modeling for production assessment, operating
philosophy, application advantages and disadvantages, and international precedents. Furthermore, technical
presentations were delivered to management at different organizational levels. These presentations aimed
to elucidate the technical justification and advantages of adopting flash completions as a means to enhance
production, exploit reserves, and minimize reservoir damage stemming from factors like well interventions
such as recompletions.
Early discussions (Rodriguez et al., 2014) with regulators led to approval of a pilot trial involving the

Downloaded from http://onepetro.org/SPECTCE/proceedings-pdf/23CTC/3-23CTC/D031S022R011/3329096/spe-217610-ms.pdf/1 by Petrobras user on 27 June 2024

implementation and evaluation of IC on five wells in the SSFD field.
Artificial neural network (ANN) models (Kalam et al., 1996) are designed to simulate human processing
capabilities and include functions such as knowledge processing, speech, situation prediction, and control.
The ability of ANN systems to autonomously learn from input data, draw conclusions from imprecise and
uncertain data, and provide appropriate responses to new information has led to the growing acceptance of
this technology in all areas of the field, leading to numerous applications. This paper (Kalam et al., 1996)
presents an initial exploration of the utilization of this innovative technology, to forecast formation damage
by measuring the wettability and two-phase relative permeability of reservoirs.
The backpropagation-based ANN model is trained using various parameters derived from empirically
determined relative permeability curves (relperm). The input dataset comprises reservoir core characteristics
encompassing a wide spectrum of porosities and permeabilities across diverse lithologies that exhibit
varying degrees of wettability. (Kalam et al., 1996) Once trained, the model is tested using only a
small set of readily available input variables such as Swc (bound water saturation) and Sor (residual oil
saturation) to produce predictions regarding wettability and permeability curves. Alterations or deviations
in relative permeability curves are linked to shifts in wettability and may suggest the possibility of formation
damage occurring during drilling activities. The predicted wettability of rock-fluid system agrees with
the experimentally obtained values are accurate to an extent of 89%, while the predictions of the relative
permeability curves, particularly at their endpoints, closely match 84.5% of the measured outcomes. (Kalam
et al., 1996) Notably, the precision of these predictions can be significantly improved through further
model training using more precise reservoir data and well-defined formation lithologies. This demonstrates
the enormous potential of neural networks in predicting "relperm" curves and hence in estimating in-situ
reservoir damage.
ANN models are increasingly becoming important tools for predicting various parameters in field
development (Kalam et al., 1997). In the broader context of artificial intelligence, he fields of petroleum
engineering is witnessing a significant uptick in the adoption of expert systems and associated technologies,
particularly with a concentration on artificial neural networks (ANN). These relatively recent computer-
based tools offer solutions to numerous engineering challenges that are typically costly to experimentally
quantify. Simultaneously, they help mitigate uncertainties ingrained in numerous measurement methods.
Kalam and Al-Alawi (Kalam et al., 1997) effectively applied ANN-based models to assess damage to
limestone reservoirs and predict the wettability characteristics of cores obtained derived from multiple
carbonate reservoirs located in the Sultanate of Oman. Following this, the ANN predictions were compared
to real measurements and various empirically established correlations frequently employed in traditional
reservoir simulation investigations. The developed artificial neural network models by (Kalam et al. 1997)
provide quick and reasonably precise predictions for the endpoint relative permeability’s of reservoir
cores originating from heterogeneous reservoirs naturally fractured, naturally fractured. They achieve
this using foolproof input parameters obtained from well log data. In contrast, errors in estimating these
critical reservoir characterization properties using several widely accepted correlations are notably higher in
contrast to experimental observations. Relative permeability curves predicted by artificial neural networks
models are in good agreement with curves obtained from labor-intensive and relatively expensive SCAL
(Special Core Analysis Laboratory) programs. Therefore, these models help reduce uncertainties in reservoir
8 SPE-217610-MS

modeling (Kalam et al., 1997). Below, Table 4 shows a list of methods used in the articles. Table 5 shows
the advantages and disadvantages of the methods used.

Table 4—Artificial İntelligence Methods used in articles

Source Method

(Nnamdi & Adelaja, 2019) ANN

(Kalam et al., 1996) ANN

(Erbas & Gumrah, 2001) ANN and ML

Downloaded from http://onepetro.org/SPECTCE/proceedings-pdf/23CTC/3-23CTC/D031S022R011/3329096/spe-217610-ms.pdf/1 by Petrobras user on 27 June 2024

(Jiaojiao et al., 2010) ANN

(Effiong et al., 2021) ANN

Table 5—Advantages and disadvantages of artificial intelligence methods that have been and can be used for forecasting

Method Advantages Disadvantages

1. Simple and clear to interpret. 1. A linear relationship between predictors

2. Can be used for both prediction and and outcome is assumed.
inference. 2. May be sensitive to outbursts.
Linear regression
3. Can work with both numeric and 3. Unable to cope with nonlinear
categorical predictors. relationships between predictors and
outcome.

1. Simple and clear to interpret. 1. A linear relationship is assumed between

2. Can work with both numeric and predictors and logarithmic outcome
categorical predictors. coefficients.
3. Can make probabilistic predictions. 2. Cannot cope with non-linear
Logistic regression
relationships between predictors and
logarithmic outcome coefficients.
3. May be susceptible to overfitting if
model is too complex.

1. Intuitive and easy to interpret. 1. May be subject to overfitting if the tree

2. Can work with both numeric and is too deep or the data set is too small.
Decision Tree (DT) categorical predictors. 2. May not generalize well to new data.
3. Can capture interactions between 3. May be unstable and sensitive to small
predictors. data changes.

1. Can capture nonlinear relationships 1. Training and assessment can be

between outcome and predictor computationally expensive.
variables. 2. It can be hard to interpret and
Random Forest (RF) 2. Can handle multi-dimensional data. understand how the model makes its
3. Copes well with a wide range of tasks. predictions.
3. May work poorly with highly correlated
predictors.

1. SVM usually performs well on a wide 1. SVM can be computationally expensive

range of tasks, especially when the data to train and evaluate, particularly with
can be partitioned. large datasets.
2. Can handle large size data. 2. The choice of kernel function and
Support Vector Machine (SVM)
3. The use of kernels allows SVM to hyperparameters can significantly affect
capture complex nonlinear relationships the model’s performance.
between predictors and outcome 3. SVM may perform poorly on highly
imbalanced datasets.

1. Simple and accessible to implement. 1. The input objects are assumed to be

2. Can handle large size data. conditionally independent given the
3. Can work well with small data sets. class variable, which may not be true in
practice.
Naive Bayes classifier (NB)
2. May not work properly if independence
assumption is violated.
3. May be sensitive to the selection of prior
probabilities.

1. Simple and accessible to implement. 1. Training and assessment can be

K-Nearest Neighbors (KNN)
computationally expensive.
SPE-217610-MS 9

Method Advantages Disadvantages

2. Can capture nonlinear relationships 2. Can work poorly with high dimensional
between predictor variables and data.
outcome. 3. Can be sensitive to the choice of K and
3. Works well with small data sets. distance metric.

1. Can handle large size data. 1. Training and assessment can be

2. Copes well with a wide range of tasks. computationally expensive.
3. Less prone to overfitting than individual 2. Can be susceptible to noisy data and
AdaBoost
weak classifiers. outliers.
3. May work poorly with highly correlated
predictors.

Downloaded from http://onepetro.org/SPECTCE/proceedings-pdf/23CTC/3-23CTC/D031S022R011/3329096/spe-217610-ms.pdf/1 by Petrobras user on 27 June 2024

1. Copes well with a wide range of tasks. 1. Training and assessment can be
2. Can handle large size data. computationally expensive.
3. Less susceptible to overfitting than 2. Can be susceptible to noisy data and
XGBoost
standalone decision trees. outliers.
3. May work poorly with highly correlated
predictors.

1. Can capture complex and nonlinear 1. Training and assessment can be

relationships between predictor variables computationally expensive.
and outcome. 2. Can be sensitive to choice of network
Artificial Neural Networks (ANN)
2. Copes well with a wide range of architecture and hyperparameters.
problems. 3. May work poorly with small data sets.
3. Can handle large size data.

1. Easily recognizes trends and patterns. 1. Data acquisition

2. Works without human intervention 2. Time and resource allocation
(automated) 3. Interpretation of findings
Machine Learning (ML) 3. Continuous improvement. 4. High vulnerability to errors
4. Processes complex multidimensional
data.
5. Wide range of applications.

Conclusion
In conclusion, the review article offers an extensive survey of the utilization of machine learning methods
in tackling formation damage issues within the oil and gas sector. Reservoir damage remains a serious
problem, potentially hampering hydrocarbon production, increasing operating costs and reducing reservoir
life. Traditional methods have limitations in their ability to accurately predict and mitigate formation
damage, making the integration of machine learning an interesting and promising area of progress. Through
a thorough examination of existing literature and case studies, it is clear that machine learning has the
potential to revolutionize the approach to reservoir damage. Machine learning models such as support vector
machines, regression analysis and neural networks demonstrated their effectiveness in analyzing large and
complex data sets, identifying hidden patterns, and accurately predicting formation damage risks. Moreover,
machine learning models can be employed to optimize drilling and completion strategies, leading to more
efficient operations and reduced formation damage incidents.
The successful applications of machine learning discussed in this review highlight the substantial
benefits it can bring to the oil and gas industry. These benefits include improved reservoir management
practices, enhanced hydrocarbon recovery rates, and significant cost savings. Furthermore, the integration
of real-time data monitoring and predictive analytics can provide operators with valuable insights into
reservoir behavior, enabling them to proactively address potential formation damage issues. However, it
is important to acknowledge that challenges remain. Data quality, accessibility, and privacy concerns must
be carefully considered when implementing machine learning solutions. Additionally, ongoing research
and development efforts are essential to refine existing algorithms, develop new ones, and adapt machine
learning methods to evolving reservoir conditions.
In summary, a literature review demonstrates that machine learning has emerged as a powerful tool
for mitigating and managing formation damage within the oil and gas sector. Its potential to transform
10 SPE-217610-MS

reservoir management, optimize drilling practices, and improve hydrocarbon recovery rates cannot be
understated. As we look to the future, continued collaboration between data scientists, engineers, and
industry professionals will be crucial in harnessing the full potential of machine learning to ensure the
sustainability and profitability of oil and gas reservoirs. By embracing this technological advancement, the
industry is poised to overcome one of its most persistent challenges and usher in a new era of efficient and
data-driven reservoir management.

References

Downloaded from http://onepetro.org/SPECTCE/proceedings-pdf/23CTC/3-23CTC/D031S022R011/3329096/spe-217610-ms.pdf/1 by Petrobras user on 27 June 2024

Al Gharbi, S., Liao, Q., Elkatatny, S., & Abdulraheem, A. (2018). IIncreasing ANN Accuracy, by Improving the Training
Dataset
Criteria. Case Study: Identify the Formation Density from The Drilling Surface Parameters in Real-Time. In SPE Kingdom
of Saudi Arabia Annual Technical Symposium and Exhibition (p. SPE-192363-MS). https://doi.org/10.2118/192363-
MS
Alegre, L., & Dougherty, E. L. (1988). Applicability of Expert Systems To Diagnose Formation Damage Problems: A
Progress Report. In SPE California Regional Meeting (p. SPE-17460-MS). https://doi.org/10.2118/17460-MS
Cundar, C., Guerrero-Benavides, C., Aristizabal, J. D., Moncayo-Riascos, I., Rojas-Ruiz, F. A., Orrego-Ruiz, J. A.,
Cañas-Marín, W., & Osorio, R. (2023). A Comprehensive Approach to Organic Precipitation Damage by CPA EoS
from Monte Carlo, and Machine Learning Methods. In SPE Latin American and Caribbean Petroleum Engineering
Conference (p. D011S006R005). https://doi.org/10.2118/213163-MS
Denney, D. (2007). Technology Integration in the Caspian. Journal of Petroleum Technology, 59(12), 69–71. https://
doi.org/10.2118/1207-0069-JPT
Effiong, A. J., Etim, J. O., & Okon, A. N. (2021). Artificial Intelligence Model for Predicting Formation Damage in
Oil and Gas Wells. In SPE Nigeria Annual International Conference and Exhibition (p. D021S009R005). https://
doi.org/10.2118/207129-MS
Erbas, D., & Gumrah, F. (2001). The Use of Genetic Algorithms as an Optimization Tool for Predicting Permeability
Alteration in Formation Damage and Improvement Modelling. In Canadian International Petroleum Conference (p.
PETSOC-2001-052). https://doi.org/10.2118/2001-052
Garrouch, A. A., Malallah, A. H., & AlEnizy, M. M. (2006). A Comprehensive Expert System for Diagnosing and
Assessing Formation Damage. In SPE Europec/EAGE Annual Conference and Exhibition (p. SPE-99277-MS). https://
doi.org/10.2118/99277-MS
Gholami, R., Shahraki, A. R., & Jamali Paghaleh, M. (2012). Prediction of Hydrocarbon Reservoirs Permeability Using
Support Vector Machine. Mathematical Problems in Engineering, 2012, 670723. https://doi.org/10.1155/2012/670723
Ikiensikimama, S. S., & Azubuike, I. I. (2012). Modeling Approach for Niger-Delta Oil Formation Volume Factor
Prediction Using Artificial Neural Network. In Nigeria Annual International Conference and Exhibition (p.
SPE-162987-MS). https://doi.org/10.2118/162987-MS
Jiang, Y., Li, X., Luo, H., Yin, S., & Kaynak, O. (2022). Quo vadis artificial intelligence? Discover Artificial Intelligence,
2(1), 4. https://doi.org/10.1007/s44163-022-00022-8
Jiaojiao, G., Jienian, Y., Zhiyoong, L., & Zhong, H. (2010). Mechanisms and Prevention of Damage for Formations
with Low-porosity and Low-permeability. In International Oil and Gas Conference and Exhibition in China (p.
SPE-130961-MS). https://doi.org/10.2118/130961-MS
Kalam, M. Z., Al-Alawi, S. M., & Al-Mukheini, M. (1996). Assessment of Formation Damage using Artificial Neural
Networks. In SPE Formation Damage Control Symposium (p. SPE-31100-MS). https://doi.org/10.2118/31100-MS
Kalam, M. Z., Al-Alawi, S. M., & Al-Shekaili, S. (1997). A Novel Technique for Predicting End-Point Relative
Permeabilities of Heterogeneous Limestones from Log Derived Input Data. In Middle East Oil Show and Conference
(p. SPE-37694-MS). https://doi.org/10.2118/37694-MS
Liu, X., & Civan, F. (1993). Characterization and Prediction of Formation Damage in Two-Phase Flow Systems (Formation
Damage and Filter Cake in Laboratory Core Tests - Part II: Model Assisted Analaysis). In SPE Production Operations
Symposium (p. SPE-25429-MS). https://doi.org/10.2118/25429-MS
Moridis, G. J., Kuzma, H. A., Reagan, M. T., Blasingame, T. A., Huang, Y. W., Santos, R., Boyle, K., Freeman,
C. M., Ilk, D., Cossio, M., Bhattacharya, S., & Nikolaou, M. (2011). SeTES: A Self-Teaching Expert System
for the Analysis, Design, and Prediction of Gas Production From Unconventional Gas Resources. In Canadian
Unconventional Resources Conference (p. SPE-149485-MS). https://doi.org/10.2118/149485-MS
Nashawi, I. S., & Malallah, A. (2009). Improved Electrofacies Characterization And Permeability Predictions In Sandstone
Reservoirs Using a Data Mining And Expert System Approach. Petrophysics - The SPWLA Journal of Formation
Evaluation and Reservoir Description, 50(03).
SPE-217610-MS 11

Nnamdi, D. N., & Adelaja, V. O. (2019). Dynamic Production Forecasting using Artificial Neural Networks customized
to historical well Key Flow Indicators. In SPE Nigeria Annual International Conference and Exhibition (p.
D033S027R004). https://doi.org/10.2118/198756-MS
Nunez Garcia, W., Kleber, M., Polo, R., Franco, C. A., Escobar, M. A., Sierra, A., & Arango, M. (2015). Comprehensive
Methodology to Identify, Quantify and Eliminate the Formation Damage Mechanisms, Succesfully Applied for the first
Time by the Operator in a Colombian Mature Field; Including formation Damage Modeling, Well Candidate Selection,
Stimulation T. In SPE Latin American and Caribbean Petroleum Engineering Conference (p. D021S015R003). https://
doi.org/10.2118/177090-MS
Rodriguez, J. C., Dutan, J., Serrano, G., Sandoval, L. M., Arevalo, J. C., & Suter, A. (2014). Compact Intelligent
Completion: A Game Change for Shushufindi Field. In SPE Latin America and Caribbean Petroleum Engineering

Downloaded from http://onepetro.org/SPECTCE/proceedings-pdf/23CTC/3-23CTC/D031S022R011/3329096/spe-217610-ms.pdf/1 by Petrobras user on 27 June 2024

Conference (p. D021S013R003). https://doi.org/10.2118/169483-MS
Tan, C., Yan, W., Tang, Q., Wu, H., Bu, H., Kambi, S. J., & Liu, J. (2020). Casing Damage Prediction Model Based on the
Data-Driven Method. Mathematical Problems in Engineering, 2020, 8315908. https://doi.org/10.1155/2020/8315908
Yang, Y., Hu, Y., Jiang, M., Yang, Y., & Xiao, C. (2023). Development scheme screening considering reservoir damage
based on the AHP model. SN Applied Sciences, 5(4), 123. https://doi.org/10.1007/s42452-023-05337-4
Yuan, B., & Wood, D. A. (2018). A comprehensive review of formation damage during enhanced oil recovery. Journal
of Petroleum Science and Engineering, 167, 287–299. https://doi.org/https://doi.org/10.1016/j.petrol.2018.04.018
Zuluaga, E. (2000). Prediction of Permeability Reduction by External Particle Invasion Using Artificial Neural
Networks and Fuzzy Models. In Canadian International Petroleum Conference (p. PETSOC-2000-002). https://
doi.org/10.2118/2000-002