ArticlePDF Available

Abstract and Figures

Artificial intelligence has been back on the stage of research works in all the walks in recently years, the sharply increase of AI-based work have shown its potential to be a future direction for almost all disciplines. In oil and gas industry, AI technology is also doubtlessly a new shining star that draws attention from researchers devoted themselves into it. In order to dig up more about the applications of artificial intelligence in oilfield development for a hint of the future trend of this exciting technology in oil and gas industry, literature investigation of a large amount of AI-based work reported has been conducted in this work. Based on the investigation, the application of AI in important issues in oilfield development including oilfield production dynamic prediction, developing plan optimization, residual oil identification, fracture identification, and enhanced oil recovery are specifically investigated and summarized, the backs and cons of existing AI algorithms has been compared. Based on the analysis and discussion, current situation of the application of AI in oilfield development is concluded, and suggestions and potential directions of future work AI application in oil and gas developing are provided.
Content may be subject to copyright.
1 3
Archives of Computational Methods in Engineering (2021) 28:937–949
Applications ofArtificial Intelligence inOil andGas Development
HongLi1· HaiyangYu1· NaiCao1· HeTian1· ShiqingCheng1
Received: 10 July 2019 / Accepted: 2 January 2020 / Published online: 16 January 2020
© CIMNE, Barcelona, Spain 2020
Artificial intelligence has been back on the stage of research works in all the walks in recently years, the sharply increase of
AI-based work have shown its potential to be a future direction for almost all disciplines. In oil and gas industry, AI technol-
ogy is also doubtlessly a new shining star that draws attention from researchers devoted themselves into it. In order to dig
up more about the applications of artificial intelligence in oilfield development for a hint of the future trend of this exciting
technology in oil and gas industry, literature investigation of a large amount of AI-based work reported has been conducted
in this work. Based on the investigation, the application of AI in important issues in oilfield development including oilfield
production dynamic prediction, developing plan optimization, residual oil identification, fracture identification, and enhanced
oil recovery are specifically investigated and summarized, the backs and cons of existing AI algorithms has been compared.
Based on the analysis and discussion, current situation of the application of AI in oilfield development is concluded, and
suggestions and potential directions of future work AI application in oil and gas developing are provided.
ANFIS Adaptive network-based fuzzy inference system
ANN Artificial neural network
APSO Adaptive particle swarm optimization
BDA Big data analytics
BP Back propagation
CNN Convolutional neural network
DM Data mining
FCM Fuzzy clustering method
GA Genetic algorithm
GNN Graph neural network
HIS Hybrid intelligent system
Iot Internet of things technology
IPSO Improved particle swami optimization
LSSVM Least squares support vector machine
MAPE Mean absolute percent error
ML Machine learning
MLPNN Multi-layer perceptron neural network
MSE Mean squared error
NARX Nonlinear auto regressive model with
PCA Principal component analysis
PNN Polynomial neural network
PSO Particle swarm optimization
QPSO Quantum particle swarm optimization
RMSE Root mean squared error
SD Standard deviation
SOM Self-organizing maps
SRM Surrogate regulation model
SVM Support vector machine
WOB Weight on bit
1 Introduction
Artificial intelligence has a long history began since the year
1950, when the British Mathematician Alan Turing asked
the famous question “can machines think?” [1]. Artificial
intelligence (AI) was formally proposed and defined as a
new research field at the 1956 Dartmouth academic confer-
ence. Then came the first spring of artificial intelligence,
when AI was quickly applied in various fields [2, 3]. Arti-
ficial intelligence laboratories began to be set up in many
countries, and experts back then believed that machines
would soon replace humans in various areas. However, when
it came to the 1970s, due to the limitations of artificial intel-
ligence algorithms at that time, the development of artificial
* Haiyang Yu
Hong Li
1 State Key Laboratory ofPetroleum Resources
andProspecting, China University ofPetroleum,
Beijing102249, China
938 H.Li et al.
1 3
intelligence was limited due to the inability to implement
large-scale or complex work. A few years later, with the
widespread use of the “expert system”, AI began to flour-
ish again [4]. However, since the “expert system” required
strong knowledge processing ability and high maintenance
cost, there was no significant breakthrough in the develop-
ment of AI. The rapid development of computers in the late
1990s seems to have ushered in a new spring for AI. After
more than 60years’ ups and downs, AI came back to the
focus with the victory of AlphaGo over Lee Se-dol [1]. Then
in the year 2017, AlphaGo Zero showed up with high-speed
self-training without any human input, which aroused high
attention from all walks of life, and brought new thoughts to
the development of AI in various fields [5]. With the devel-
opment of cloud computing, big data, artificial neural net-
work, deep learning and other new technologies, it could be
said that AI has achieved a new leap and changed our daily
life as well. Driverless cars, accurate face recognition and
other artificial intelligence applications are no longer just
figments in the sci-fi movies. AI has been applied in almost
all the aspects of daily life, as the core of the traditional
energy industry, the oil and gas industry also embraces AI
to bring new technological breakthroughs for oil and gas
exploration, development and production [68].
Application of AI in the field of petroleum engineering
was proposed on the forum by international association as
early as in the 1970s. In 2009, SPE established the branch
of “artificial intelligence in prediction analysis” to promote
the application of AI technology in the petroleum field, in
order to organize regular relevant discussions. Based on the
search result from the Onepetro platform, the number of
articles on AI has increased significantly since 2000, whose
main algorithms include the artificial neural network (ANN),
fuzzy logic, support vector machine (SVM), hybrid intelli-
gent system (HIS), genetic algorithm (GA), particle swarm
optimization (PSO), etc. This suggests an increasing interest
of the researchers in the application of artificial intelligence
in the oil industry, and among all the algorithms, the ANN
is the most studied one (Fig.1).
At present, the application of AI in the oil and gas indus-
try is rapidly developing, as the concept of AI gradually
penetrates various stages of the industry, intelligent drill-
ing, intelligent production, intelligent pipeline, intelligent
refinery, etc., and it will become the future research direc-
tion. With artificial intelligence algorithms, developers have
developed a series of practical application technologies in
exploration and development. In the field of exploration,
utilization of the ANN method has already achieved good
results in reducing exploration risks and improving the suc-
cess rate of exploration wells [8]. In the field of drilling,
new equipment such as automatic drilling rig and intelli-
gent drill pipe have improved the drilling level and reduced
the cost significantly [9]. In the oilfield development, based
on the historical data of oilfield production to optimize the
development plan is the main application mode of AI tech-
nology. In addition, AI has provided more accurate method
Fig. 1 AI the statistic of
increasing number of articles on
AI algorithms. Source: https ://
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018
939Applications ofArtificial Intelligence inOil andGas Development
1 3
for fracturing scheme design and the selection of operating
wells and target layers [10].
Though many scholars have their achievements shared,
due to the abundant methods and contents of AI, there has
been little specific summary and conclusion on the AI appli-
cation in the field of oilfield development. Production per-
formance prediction, development optimization, residual oil
identification, enhanced oil recovery, and correct identifica-
tion and prediction of artificial fractures in unconventional
reservoirs, which are particularly important, are the core
tasks in oilfield development. To embrace AI technology in
these works, it is of great significance to analyze and sum-
marize the existing achievements.
In order to dig up more about the applications of artificial
intelligence in oil and gas field development, in this work,
a great deal of related investigations have been conducted.
Of course, the content in one single paper is limited, so we
focused on certain important issues in oil and gas field devel-
opment to narrow the scope. The issues discussed in this
paper included the comparison of main existing algorithms,
the application of AI in oilfield production dynamic predic-
tion, in development optimization, in residual oil identifica-
tion, in fracture identification, and the application of AI in
enhanced oil recovery.
In each section, the AI algorithms mainly adopted dealing
with specific issue were summarized, the backs and cons of
which were compared, and the most suitable algorithm was
suggested. Finally, based on the analysis and investigation,
recommendations and potential directions of future work in
the application of artificial intelligence in oilfield develop-
ment was proposed.
2 From Digital Oil Field toAI Oil Field
The transition from digital oilfield to AI oilfield is inevitable.
The oil oilfield development involves huge data volume and
unpredictable emergencies, without resource integration and
automatic management, ideal results could not be achieved
and unexpected situations may even occur, though a good
many material and financial resource would be consumed.
Digital oilfield is advanced overall management supported
by information technology, which includes a series of pro-
cesses from exploration to production. This management can
obtain data timely, share completely and achieve delicacy
management through analysis of production optimization
[11, 12]. Digital oilfield enterprises are therefore more effi-
cient, creative and competitive. AI oilfield is an advanced
version of digital oilfield, it is an advanced automatic iden-
tification treatment system that covers all the aspects in
the oilfield based on the advanced computer technology,
automation technology, sensor technology and the profes-
sional technology. The treatment system could achieve more
efficient and sustainable development of oilfield by compre-
hensive perception of oilfield dynamic change, automatic
manipulation, prediction and optimization of oilfield [13,
14]. AI oilfield plays a leading and guiding role in the infor-
matization of oil fields at all levels and is the future trend of
oil enterprises [15], for it can greatly reduce the oil produc-
tion cost, improve the average oil field recovery, improve
the management efficiency of enterprises, and indirectly
promotes the economic and social development at the same
time. Trending to mature stage, the main future development
direction of the digital oil field system is to deeply dig the
oilfield data and integrate digital platform based on existing
digital oilfield; to establish AI oilfield with the ability of
prediction, warning, efficient analysis and optimization with
the help of emerging technologies, such as internet of things
(IoT) technology, the cloud computing technology, and the
big data technology, etc. Deep integration of informatiza-
tion and industrialization is also a necessity to reduce costs
and to improve quality and efficiency. A truly competitive
oilfield should be one that can fully perceive, automatically
control, predict trends and optimize decisions [16, 17].
Many oil companies have launched intelligent oilfield
projects to improve the quality of decision-making and man-
agement. Exploration and development multi-dimensional
environment platform software launched by Schlumberger
can realize automatic drilling design utilizing IoT and other
technologies [18]. Kuwait national petroleum corporation’s
digital oilfield (KwIDF) has been upgraded to an advanced
intelligent workflow integrated aboveground and under-
ground systems [19]. In addition, AI-based intelligent man-
agement assistants, such as oilfield robot, virtual oilfield
assistant and intelligent oilfield APP, can not only replace
human beings to deal with high-risk work, but also reduce a
great many repetitive work [20] (Fig.2).
3 AI inOil andGas Development
The mainly adopted application of AI in reservoir devel-
opment and exploitation is the optimization of develop-
ment based on historical oilfield production data. Entering
the big data era, it is necessary for oilfield development to
fully explore the huge potential value of big data and to
reveal hidden, previously unknown and potentially valuable
information, which is also one of the hot issues in AI oil-
field development research [21]. At present, AI has been
widely used in many industries (such as communication,
e-commerce, etc.), but it still has a long way to go in the
oil industry. This is mainly due to the particularity of data
and applications in oilfield development. For example,
large data amount, data with high dimension and coupling,
multi-source data with complex format, unstructured data,
strong heterogeneity of the research objects, etc. [22]. With
940 H.Li et al.
1 3
the development and application of big data and continu-
ous improvement of various related algorithms, AI plays
an increasingly important role in the field of oilfield devel-
opment. Combined with other relevant new technologies
(cloud computing, Internet of things, virtual reality, etc.),
new technologies and systems involved with AI will be con-
stantly proposed, which will surely become an important
way to reduce costs and improve efficiency. Thus, researches
on AI and its application in oilfield development have a no
doubt promising future and prospect. In this section, main
existing algorithms, the application of AI in oilfield produc-
tion dynamic prediction, in development optimization, in
residual oil identification, the in fracture identification, and
the application of AI in enhanced oil recovery are discussed
in detail respectively.
3.1 Backs andCons
The proposer and year of the mainly adopted artificial intel-
ligence algorithms are summarized by Agwu in Table1 [23].
Almost all algorithms were proposed in the 1990s, and the
earliest and most mature one is the artificial neural network
algorithm, which has been applied in all walks of life with
good results. The purposes of the following algorithms are
mostly algorithm optimization. Since the current research
problems tend to be more complicated, which could be
hardly solved by solely using one single algorithm, com-
bination of multiple algorithms and invention of new and
more intelligent algorithms have become the new targets
for researchers.
Based on the history of artificial intelligence, in order
to better select a suitable algorithm and have it applied
properly, understanding the advantages and disadvantages
of each algorithm is more important than just knowing the
Fig. 2 KwlDF workflow tools
architecture [19]
Table 1 Evolutionary trend of various artificial intelligence tech-
niques [23]
AI technique Year of
Artificial neural network 1943 McCulloch and Pitts
Fuzzy logic 1965 Lofti A. Zadeh
Genetic algorithm 1970 John Holland
Case based reasoning 1977 Schank and Abelson
Support vector machines 1995 Vapnik V
Particle swarm algorithm 1995 Eberhart and Kennedy
Hybrid intelligent systems
941Applications ofArtificial Intelligence inOil andGas Development
1 3
history itself. For this purpose, we summarize the advan-
tages and disadvantages of the commonly used algorithms
in the oil field as shown in Table2 to provide a clear view
of their advantages and disadvantages. ANN is the most
commonly used and simplest algorithm, but it has higher
requirements of input parameters. PSO is easy to be real-
ized and could be free from the problem information, but
the accuracy is relatively low. The utilization of fuzzy logics
doesn’t require mathematic model but its accuracy is low at
the same time. SVM is suitable for small sample learning
and is sensitive to real data. The GA method has a great
parallelism, it could quickly search and be easily combined
with other algorithms, but has a more complex program-
ming process and longer training time. Although the above
algorithms have certain limitations, suitable selection and
application of algorithm is the key to solve problems.
3.2 AI inHistory Matching ofOilfield
The most important task in oil gas development is to pre-
dict the future development based on the existing data and
make a reasonable development plan, history matching
is needed to facilitate the subsequent numerical simula-
tion before prediction. However, due to the complexity of
actual oilfield development that multiple factors interfere
simultaneously, which increases the difficulty of history
matching with conventional methods. With the develop-
ment of AI, it has been proved to be feasible to apply
neural network to history matching of oilfield development
[24]. With the history matching and training of the existing
data, it can effectively capture the nonlinearity of problems
with fast matching speed and good precision, and further
become a new history matching method.
Al-Thuwaini etal. [25] proposed a method for calcu-
lating root mean square (RMS), and performed history
matching based on artificial intelligence (ANN) combined
with self organizing maps (SOM) method, with examples
of history matching of model pressure and water contents
as illustration. The SOM method can use geological and
hydraulic (static and dynamic) information to gather grids
with similar behaviors, which can significantly improve
the matching quality and reduce the number of times
needed to achieve target matching.
Mohaghegh etal. [26]. adopted data mining technol-
ogy to conduct a top-down reverse modeling of shale
reservoirs, which is quite different from the traditional
bottom-up modeling from geological model to dynamic
prediction. AI and data mining (neural network, genetic
algorithm and fuzzy logic, etc.) were carried out on the
established data to establish the cohesive modeling of the
whole model, based on which the continuous fuzzy recog-
nition algorithm was used to transform discrete data into
reservoir viscosity model. The top-down reverse modeling
was validated with actual cases.
Costa etal. [27] used artificial neural network (ANN)
combined with GA to conduct production history match-
ing, which showed that the neural network could effectively
capture the nonlinearity of problems with a validation by
synthetic reservoirs with actual reservoir characteristics. The
work showed that neural network, as an agent model, has
a good application prospect in history matching of oilfield
with less simulation times and better fitting effect.
Table 2 The comparison between the advantages and disadvantages of algorithms
Method Advantages Disadvantages
ANN High classification accuracy, strong parallel distributed pro-
cessing ability, strong distributed storage and learning ability,
strong robustness and fault tolerance to noise nerves, full
approximation to complex nonlinear relations, associative
memory function, etc.
Many parameters are required, such as network topology, initial
values of weights and thresholds, output difficult to be inter-
preted, long training time, etc.
PSO Free from the problem information, solve problem with real
numbers, strong universality, few parameter adjustment,
simple theory, easy to achieve, collaborative search, fast
Low accuracy, prone to divergence, has certain dependence on
parameters, imperfection of the theory
Fuzzy logics Precise mathematical models are not required, strong robust-
ness, easy to achieve
Low accuracy, lack of systematic design
SVM Suitable for small sample machine learning problems; can
improve generalization performance, solve high dimensional
problems, solve nonlinear problems, and avoid Neural net-
work structure selection and local minimum point problem
Sensitive to missing data, no general solution for nonlinear
GA Fast random search capability unconfined in problem domain,
potential parallelism and good robustness, simple process,
use probability mechanism to conduct iteration with certain
randomness, extensible and easy to be combined with other
Complex programming, problem decoding is required after the
determination of optimal solution, accurate solution requires
more training time, etc.
942 H.Li et al.
1 3
Shahkarmi etal. [28] developed the surrogate regulation
model (SRM) as the driver of history matching process with
the pattern recognition capability of artificial intelligence
and data mining (AI and DM). With history matching of 24
production wells in heterogeneous field had been operat-
ing for 30years, it showed that the high speed and accu-
racy make SRM a fast and effective tool to assist history
The research on history matching has been relatively
mature but still needs to be further studied. It can be seen
from the above studies that artificial neural network (ANN)
is commonly utilized to conduct training and history match-
ing oilfield development, and combine genetic algorithm
(GA) is often combined to optimize and improve the match-
ing speed and accuracy. As we know, each step of oilfield
development involves large number of parameters, which
is typical “big data”. It is urgent to develop artificial intel-
ligence algorithms that can solve practical engineering prob-
lems to improve efficiency and reduce costs and risks. The
problems or parameters targeted by various researchers are
uneven, but since current research has shown that the speed
and accuracy of history matching can be improved by arti-
ficial intelligence algorithm, meaning that AI application is
feasible, what awaited to be done is in-depth and integration
of research. Based on the summarization and discussion, for
history matching, we concluded and came up with the fol-
lowing problems and potential suggestions:
The history matching process relies too much on artifi-
cial intelligence algorithm and lacks analysis of physical
significance of the results;
The production law of oilfield development should be
summarized after history matching to make prediction
and optimization of the developing plan;
The traditional history matching ignores the combina-
tion with geological data, such as seismic data, logging
data, etc., so the adjustable parameters and their variation
range should be determined on the basis of fully under-
standing the geological background.
The combination of AI and reservoir engineers can effi-
ciently optimize the best parameter combination from
multiplicity of solutions.
The existing dynamic and static data should be deeply
mined, the artificial intelligence history matching system
should be formed by combining various algorithms, and
the relevant norms and standards should be established
by integrating resources of various platforms.
3.3 AI inProduction Dynamic Prediction ofOilfield
Production index prediction plays an important role in
research on reservoir engineering and oilfield develop-
ment. Oil field production dynamic analysis methods can
be divided into methods need production data (oilfield
numerical simulation method, characteristic curve method,
production decline method, material balance method, etc.)
and the ones need not (analogy method, empirical for-
mula method and chart method, etc.) [29]. Although the
above methods have been applied in oilfield production
for years, they still have obvious limitations due to vari-
ous and complex factors that affect the production index
dynamic prediction. Prediction method based on AI for
oilfield production performance has become an important
research orientation because of its rapid development in
recent years. At present, a common way is to combine
neural network with fuzzy theory or intelligent algorithm
and thereby reach a great fitting accuracy, with static and
dynamic production data involved. Production indexes are
mainly used to evaluate the current development status of
oil fields (such as injection volume, cumulative production
volume, etc.) and predict the dynamic change trend (such
as oil recovery rate, water cut rise rate, etc.) [30]. Eco-
nomic indexes are mainly used to evaluate the input cost
and economic benefits of oilfields. Artificial intelligence
algorithms such as neural network has already had certain
achievements in the applications in oilfield production
prediction, however, more practical and refined artificial
intelligence algorithm aiming at the particularity of data
gathered in the dynamic production of oil field, which is
research worthy, is awaited to be further explored to real-
ize efficient oilfield development.
To better understand the AI application in oilfield pro-
duction performance prediction at present, research results
in this aspect are summarized in Table3 [3139].
As shown in Table3, many artificial intelligence algo-
rithms have been applied in oilfield development dynam-
ics, the output parameters involved mainly focused on fluid
production (oil production or total production) and water
content, while a little attention was paid on oil production
speed. The input parameters varied widely, from one sin-
gle parameter to large number of static and dynamic data.
Based on this summarized situation, we came up with the
following potential suggestions:
BP artificial neural network is the most widely used and
the most mature AI algorithm in production dynamic
prediction at present, it could be optimized by combi-
nation with SVM and GA algorithm;
The input objects and output parameters varied widely
in reported works, the correlation among parameters,
such as parameters that have important influence on the
output results and the correlation among them, should
be clarified;
Prediction results should be summarized, reasonably
analyzed and interpreted;
943Applications ofArtificial Intelligence inOil andGas Development
1 3
Production prediction is a dynamic process, and a
dynamic AI algorithm should be established to realize
monitoring and prediction.
AI has the potential to develop a multi-well dynamic
prediction system considering inter-well and inter-layer
interference on the basis of improving the prediction
accuracy of a single well.
Existing big data should be fully exploited, the availabil-
ity of those data should be dug up more to support more
accurate prediction.
3.4 AI inDevelopment Plan Optimization
In the 1960s, linear programming method has already
applied optimization in aiming at improving production effi-
ciency of homogeneous reservoirs, some exploratory works
had been reported afterwards, but no attention was paid on
optimization of oilfield development back then. After the
year 1985, the developing situation of oilfields changed,
plan optimization was badly needed [40]. With long-term
water injection development, the oilfield gradually entered
the period of high water cut, and great dissimilitude showed
in the heterogeneity between layers. These could be solved
by recombination developing layers, the methods of which
included fuzzy cluster analysis, grey correlation analysis,
neural network algorithm, etc. Moreover, due to the pro-
gress of computer technology, evaluation and optimization
of developing plan began to draw more attention from major
companies in various countries since the oilfield developing
plan directly affected the life span and benefit of oilfields.
Due to the mutual influence of technical, economic and
social indicators in the adjustment and optimization of
developing plan, it is difficult to evaluate the performance
of the plan with one single index. Therefore, it is necessary
to establish an efficient and effective method for develop-
ing plan valuation and optimization. Traditional methods
are relatively one-sided, so it is urgent to apply new intel-
ligent optimization algorithm in solving the problems in
the oilfield development process.
To better understand the AI application in oilfield devel-
oping plan optimization, research results in this aspect are
summarized in Table4 [4146].
Developing plan is crucial for oilfield development, due
to the complexity of the situation in the developing pro-
cess, its optimization should be done based on actual situa-
tion. At present, AI application in developing plan optimi-
zation are mainly to improve production or its rate with the
consideration of economic factors. As summarized above,
the input parameters in different work varied greatly, and
the adopted algorithm are also flexible, based on which,
we put forward the following potential suggestions:
The most commonly used algorithm for oilfield devel-
oping plan optimization is the ANN + GA mode, which
can be further improved in algorithm;
The oilfield developing plan optimization should
include many aspects and multi-dimensions rather than
to evaluate solely from the economic aspect or produc-
tion result;
Table 3 The application of AI in the dynamic prediction in oilfield development
References Method Input parameters Output parameters Errors
[31] ANN Gas prices; GDP growth rate; annual depletion; wells
drilled; footage drilled and other properties.
Gas production 0.0034 (MSE)
[32] BP Temperature; pressure; superficial gas velocity; superficial
liquid velocity
Liquid holdup 8.544 (SD)
[33]GNN + IPSO liquid producing capacity Water content 1.37% (MAPE)
[34] BP Remaining geological reserves; total number of production
wells, monthly injection–production ratio; kernelfunc-
tion; number of open injection wells, newly opened
production wells and old wells with efficient treatment
Monthly oil production
and liquid producing
2% (MAPE)
[35]PCA + APSO + LSSVM Remaining geological reserves; injection–production ratio;
water content; number of open wells, open injection
wells, newly opened production wells and old wells with
efficient treatment
Oil production 0.1485 (RMSE)
[36] ANN Porosity; velocity; horizontal permeability etal. Oil production cumulative
[37] BP Deep; GR log; neutron log; density log; sonic log; deep
resistivity log; dizgenesis
Porosity; permeability
[38] MLPNN Distributed temperature sensing; distributed acoustic sens-
ing; daily flowing time
Gas production
[39]ANN + ANFIS Gamma ray; density; neutron; three different resistivity’s;
caliper; porosity
Water sateration 0.07 (MSE)
944 H.Li et al.
1 3
Combining geological background, historical matching,
dynamic monitoring and economic benefit to achieve
dynamic development plan optimization.
Instead of divergent research, a complete AI-based stand-
ard evaluation process and system should be established.
3.5 AI inIdentification inOilfield Development
3.5.1 The Application ofAI intheOilfield Development
AI identification technology has already been relatively
mature in face recognition, fingerprint recognition and
other aspects, it also can play a significant role in the oil
industry. In oilfield developing process, in addition to
development dynamic prediction and plan optimization,
residual oil identification could have the reservoirs to
be re-understood and provide necessary premise for the
adjustment of developing plan. The research methods of
residual oil distribution mainly include geological method,
reservoir engineering, well testing, numerical simulation
method and laboratory experiment technique, etc. The
research works on residual oil mainly focuses on the dis-
tribution description of residual oil; monitoring remaining
oil saturation; digging potential residual oil, etal [47]. The
factors influencing the distribution of residual oil could
be divided into two types, namely, the geological factors
(reservoir heterogeneity, structure, fault, etc.) and develop-
ment factors (injection–production relationship, well pat-
tern distribution, production dynamic, etc.).The synthetic
effects of these factors leads to the diversity of residual
oil distribution. The displacement mechanism for different
residual oil types are different, which set barriers to exca-
vate residual oil and improve oil displacement efficiency.
Traditional modeling methods include deterministic mod-
eling and stochastic modeling, but they cannot predict res-
ervoir parameters in time dimension [48]. Therefore, it is
of great significance to integrate AI into the identification
of residual oil for both its great potential in the identifica-
tion of residual oil morphology and model establishment.
Some related works are shown in Table5 [4951].
As shown in Table5, applications of Ai in research on
residual oil identification are not as so many as those in pro-
duction dynamic prediction and developing plan optimiza-
tions. However, we still believe that it is a research worthy
orientation, the correct identification of residual oil is of
great significance to oil and gas development. The potential
suggestions we concluded are as follow.
Table 4 The application of AI in the oilfield developing plan optimization
References Method Input parameters Output parameters
[41] BP Net present value; profit investment ratio; investment payback period;
total investment; average cost; cumulative oil production; oil and
gas ratio
Expected value
[42] PNN +GA Realizable net present value; porosity; permeability; temperature;
initial pressure
Gas production
[43] BP Final recovery; maximum recovery rate; net present value; total profit
and internal profitability; annual decline rate; total investment;
dynamic payback period and average unit cost
Expected value
[44] NARX Energy production; energy use; final consumption expenditure; gross
domestic product; gdp growth; gold price; oil rent
Predicting oil price movements
[45] QPSO 12 measures and unit price Annual production; cumulative production ratio
[46] FCM Fuzzy clustering parameters mainly include porosity, permeability
variation coefficient, contained area, residual reserve abundance
and water flooding control extent
Oil production; recovery and water content
Table 5 The application of AI in the oilfield development identification
References Method Input parameters Output parameters
[49] BP Sand body type, injection–production relationship, connection status between
the estimated well and the injection–water well, distance from the injection–
water well, and water injection status
The water flooded degree
[50] BP Net present value, profit investment ratio, investment payback period, total
investment, average cost, cumulative oil production, oil and gas ratio
[51] BP Concavity, roundness, aspect ratio, rectangularity, eccentricity, and radius
Island shape, network shape, strip shape,
column shape, plug shape, membrane
945Applications ofArtificial Intelligence inOil andGas Development
1 3
More research on AI should be conducted, the classifica-
tion and displacement mechanism of residual oil should
be clarified to guide the development of residual oil.
Learning from more mature and advanced technical
methods applied in other fields (such as route optimiza-
tion in intelligent maps, rapid identification, etc.) could
expand the thinking of o oilfield developing identifica-
3.5.2 The Application ofAI inFracture Detection
Great influences on developing effect of reservoirs, espe-
cially the unconventional ones, could be brought by natural
fractures and artificial fractures formed by fracturing, so the
fracture identification work is great importance. Fractures
have always been a research topic in oilfield development.
Almost all the reservoirs have natural fractures, and in most
unconventional reservoirs, such as shale gas and tight oil,
many artificial fractures are formed by fracturing, making
fractures in the entire reservoir a major factor affecting pro-
ductivity [52]. The study on fractures could be generally
divided into qualitative and quantitative aspects–first quali-
tatively identify fractures and re-recognize the reservoirs
and then quantitatively calculate the distribution of fractures
(azimuth, length, openness, fracture porosity, etc.). Although
many researches have been conducted on fracture identifica-
tion, due to the strong heterogeneity of reservoirs and com-
plicated seepage characteristics caused by fractures, precise
identification of fractures remains to be a key problem await
to be solved. Some of the related works are summarized in
Table6 [5360].
Research on fractures is one of the hot spots in petroleum
field, which mainly focuses on the morphological character-
istics, property characteristics and classification character-
istics of fractures. The identification process is complicated
because the reservoir contains natural fractures, artificial
fractures and induced fractures, and the parameters involved
mainly rely on logging data. Application of AI in identifi-
cation of fractures is a new and reliable research method.
Based on above discussion, we came up with the following
The seismic data has the advantage of continuous
observation in space, so the inter well fractures may be
detected more accuracy by combining seismic data with
logging data.
Optimize and enrich the AI algorithm to improve the
identification accuracy to achieve dynamic identification;
The integrated study of formation, opening and closing
of fractures should be conducted with the application of
Reasonable analysis and interpretation of the identifica-
tion results should be made, and based on which reverse
guidance of optimization should be done.
3.5.3 The Application ofAI intheOilfield Diagnosis
In addition, safety ranks the priority in oilfield production.
Measures and other operations are carried out on con-
struction sites, which often rely on large number of data
monitoring systems. But due to the complexity and great
uncertainty of underground conditions, it is difficult for
Table 6 The application of AI in fracture detection
References Method Input parameters Output parameters
[53] BP Orientation, aperture, surface roughness, alteration
degree, mineral fillings, and other properties
The conductive state of individual fracture
[54] BP Neutron porosity, volume density, acoustic time dif-
ference, depth and shallow lateral resistivity, natural
gamma spectrum
Fracture density in well log interpretation and core
[55] BP Reflection intensity, Root Mean Squared amplitude
and arc length properties, coherence, curvature, ant
body and instantaneous frequency
Fracture density
[56] ML +DBA +SVM Surface treating pressure, clean and slurry pump-
rates, surface and downhole amounts of 100 mesh
and 30/50 mesh sand proppants
Automatically classify hydraulic fractures
[57]PSO + LSSVM Mud body integral number, resistivity differential
ratio, induced porosity, density ratio, acoustic time
difference ratio and compensated neutron ratio
No filling, argillaceous filling, siliceous filling and the
crystallization of filling
[58] CNN Borehole diameter, acoustic wave, neutron, density,
gamma ray, shallow laterolog, deep laterolog
Fracture reservoir types
[59] ANN Bottom pressure; the regularization parameter Closure pressure
[60] ANN weight on bit; rotation per minutes; rate of penetra-
tion; mud weight; pore pressure
Fracture pressure
946 H.Li et al.
1 3
traditional technology to spot the abnormal conditions in
time. However, the AI algorithm can carry out deep learn-
ing based on the background of big data and make timely
judgments according to the actual situation, so as to improve
the diagnostic accuracy of staff and save the cost of time and
economic. The identification involved are mainly the identi-
fication of abnormal well conditions, indicator diagram and
HSE warning in the production process. Some of the related
work are summarized in Table7 [6164].
Timely identification and diagnosis of the problems
encountered during the developing process is always the
goal. With AI technology, fast identification could be real-
ized, some achievements have already been applied in fields,
such as early warning with pressure detection and indicator
diagram. The follows are some suggestions based on above
summarization and discussion.
Improve the recognition accuracy and diagnosis accu-
racy of the algorithm, and make accurate interpretation
of results and right response to it;
A thorough intelligent early warning response system
should be formed, which can response and solve prob-
lems quickly.
3.6 AI inEOR
Since the 1950s, enhanced oil recovery (EOR) started its
important role in oilfield development. Up to the present
stage, the main EOR technologies applied in the field
include chemical flooding, thermal recovery, gas flooding,
etc. Exploring the application of AI in EOR may bring us
new ideas and technological breakthroughs.
Zhou etal. [65] used correlation analysis to select the
parameters could exert great impact on the EOR perfor-
mance of polymer flooding, and applied polynomial regres-
sion analysis and BP neural network to predict the nonlinear
and uncertain multivariable system.
Ni etal. [66] proposed a refined method to predict the
effect of steam flooding by a fusion of artificial bee swarm
algorithm and radial basis function (RBF) neural network
algorithm, the work showed that the proposed method has a
better nonlinear fitting ability and higher prediction accuracy
for the prediction of steam flooding development effect.
Shi etal. [67] predicted fracturing effect (cumulative oil
production) based on grey correlation analysis and BP neu-
ral network. Based on BP neural network, the selection and
optimization of water-controlled fracturing in bottom-water
reservoir was conducted by Yang etal. [68], which had a
more accurate optimization result and overcame the disad-
vantages of previous nonlinear factors.
The application of AI in EOR is mainly simulation pre-
diction, which can greatly improve the efficiency and accu-
racy. But it requires test data for verification. In addition, the
relationship between input and output parameters needs to
be further studied. Though AI algorithms in EOR research
is still on its way to be mature, its application potential is
tremendous, thus it is a research orientation worth researcher
to pay close attention to.
4 Conclusion
Based on the investigation of the application of AI in oil-
field development, it could be concluded that the intelligent
oilfield is on its way towards integration of business appli-
cation, coordination of decision and deployment, real-time
production management, visualization of comprehensive
research and sharing of information resources. AI oilfield
will eventually become an intelligent ecosystem integrated
the exploration, development, gathering, refining and man-
agement, etc. Based on the ecosystem, collaboration of all
levels, regions and disciplines could be realized to extend the
life cycle of oilfield, improve the decision-making efficiency
and quality, reduce cost and increase economic benefit, and
finally fulfill the transition from digital oilfield to AI oilfield.
The following conclusions are drawn based on our investiga-
tion and discussion.
Based SPE literature quantity survey, the number of AI-
related articles are increasing annually, research on AI
has become a research hotspot in recent years;
ANN is the most commonly used and simplest algorithm,
but it has higher requirements of input parameters. PSO
Table 7 The application of AI in the oilfield diagnosis
References Method Input parameters Output parameters
[61] ANN Active energy; kinematic viscosity; density; minimum pour point; molecular weight Wax appearance temperature
[62] BP Wellhead injection pressure Diagnostic output
[63] CNN Well fluid production, water content, oil pressure, casing pressure, stroke times, well-
head temperature, power consumption
Probability of occurrence of dif-
ferent fault types
[64] CNN Acquisition time, well ID, rod displacement, rod load, current, power, pressure, well-
head temperature, pump depth, pump diameter, stroke length, frequency of stroke,
Diagnosis the working conditions
947Applications ofArtificial Intelligence inOil andGas Development
1 3
is easy to be realized and could be free from the problem
information, but the accuracy is relatively low. Fuzzy
logics doesn’t require precise mathematic model but it
has a low accuracy. SVM is suitable for small sample
training and is sensitive to real data. GA could search
quickly and be combined with other algorithms easily,
but it requires long training time and complex program-
ming. Although all the algorithms have certain limita-
tions, suitable selection and application of algorithm is
the key to solve problems.
BP neural network is the most widely used and most
mature AI algorithm applied in oilfield. But to achieve
better performance in oil developing plan optimization,
GA or other algorithms should be combined; ANN is
suitable for identification and diagnosis, combination of
other algorithms could help to improve the result accu-
racy and algorithm speed.
It is common for AI algorithms to rely too much on
data and ignore the physical relations between various
parameters. In addition, dynamic AI research are far from
enough. Thus, instead of blindly trust the results provided
by AI, analysis and interpretation should be made to a
closed loop and solve practical problems more accurately.
The collection and processing of data is the key to realize
the application of AI in oil and gas fields.
More efforts are still needed to realize AI oilfield, the
most important is to have the capacity to utilize and share
big data of oilfield and integrate intelligent systems at
various stages.
Acknowledgements The authors are grateful for financial support from
the National Natural Science Foundation of China (51874317)andthe
National Science and Technology Major Projects of China (Grant Nos.
Compliance with Ethical Standards
Conflict of interest All authors declare no conflict of interest.
1. Wang XL (2017) Application of artificial intelligence in oil and
gas industry. Mod Inf Technol 3(1):117–119
2. Gilman H, Nordtvedt J E (2014) Intelligent energy: the past, the
present, and the future. In: SPE intelligent energy conference &
exhibition, 1–3 April. Society of Petroleum Engineers, Utrecht,
pp 185–190
3. Yu WJ (2016) On the development trend and necessity of intel-
ligent oilfield. Sci Wealth 10:500
4. Shi YJ (2016) Analysis of the research status of intelligent oilfield
in China. Straits Technol Ind 12:81–83
5. Ma SX (2018) Technology focus: formation evaluation. SPE
70(08):50. https ://
6. Liu W, Yan N (2018) Application and influence of artificial intelli-
gence in petroleum engineering area. Pet Technol Forum 4:32–40
7. Li DW, Shi GR (2018) Optimization of common data mining algo-
rithms for petroleum exploration and development. Acta Pet Sin
8. Giuliani M, Cadei L, Montini M etal (2018) Hybrid artificial
intelligence techniques for automatic simulation models matching
with field data. In: Abu Dhabi international petroleum exhibition
& conference, 12–15 November. Society of Petroleum Engineers,
Abu Dhabi, pp 1–11
9. Hojageldiyev D (2018) Artificial intelligence in HSE. In: Abu
Dhabi international petroleum exhibition & conference, 12–15
November. Society of Petroleum Engineers, Abu Dhabi, pp 1–9
10. Liu BJ (2015) Construction conception of intelligent oilfield. J
Shengli Oilfield Party Sch 06:99–101
11. Liu XD, Yang TT, Yan HY (2013) Forecast of oilfield indexes
based on elman neural network and genetic algorithm. Comput
Mod 2:150–152
12. Lien MO, Jansen JD, Brouwer DR (2006) Multiscale smart well
management. In: Intelligent energy conference and exhibition,
11–13 April. Society of Petroleum Engineers, Amsterdam, pp
13. Liu N (2014) Digital oilfield construction, smooth evolution to
intelligent oilfield. Silicon Val 4:191
14. Yousef AA, Liu JS, Blanchard GW etal (2012) Smart waterflood-
ing: industry. In: SPE annual technical conference and exhibition,
8–10 October. Society of Petroleum Engineers, San Antonio, pp
15. Wang HF, Wang SJ, Zhu SB (2018) Conception and exploration
of the smart oil and gas field construction in “internet +” era. Oil
Gas Field Surf Eng 8:1–8
16. Anderson S, Barvik S, Rabitoy C (2019) Innovative digital inspec-
tion methods. In: Offshore technology conference, 6–9 May, Hou-
ston, TX, pp 1–9
17. Denney D (2006) To support digital oil fields. SPE 58(10):71–72
18. Nadhan D, Mayani MG, Rommetveit R (2018) Drilling with digi-
tal twins. In: IADC/SPE Asia pacific drilling technology confer-
ence and exhibition, 27–29 August. Society of Petroleum Engi-
neers, Bangkok, pp 1–18
19. Rodriguez JA, Nasr H, Scott M etal (2013) New generation of
petroleum workflow automation: philosophy and practice. In: SPE
digital energy conference, 5–7 March. Society of Petroleum Engi-
neers, The Woodlands, pp 1–13
20. Tan AA, Potts JK (1995) Digital log management system. SPE
21. William WW, Balch RS, Stubbs BA (2002) How artificial intel-
ligence methods can forecast oil production. In: SPE/DOE
improved oil recovery symposium, 13–17 April. Society of Petro-
leum Engineers, Tulsa, pp 1–16
22. Gatta SR (1999) Decision tree analysis and risk modeling to
appraise investments on major oil field projects.In: SPE middle
east oil show and conference. Information Engineering Research
Institute, USA, pp 187–202
23. Agwu OE, Akpabio JU, Alabi SB etal (2018) Artificial intelli-
gence techniques and their applications in drilling fluid engineer-
ing: a review. J Pet Sci Eng 167:300–315
24. Tang HK, Li XY (2015) BP neural network is applied to the pro-
duction dynamic analysis of an oil field. Oil Gas Field Surf Eng
25. Al-Thuwaini J, Zangl G, Phelps RE (2006) Innovative approach
to assist history matching using artificial intelligence. In: Intel-
ligent energy conference and exhibition, 11–13 April. Society of
Petroleum Engineers, Amsterdam, pp 1–7
26. Mohaghegh SD, Grujic OS, Zargari S, Dahaghi AK (2011) Mode-
ling, history matching, forecasting and analysis of shale reservoirs
performance using artificial intelligence. In: SPE digital energy
948 H.Li et al.
1 3
conference and exhibition, 19–21 April. Society of Petroleum
Engineers, The Woodlands, pp 1–14
27. Costa LAN, Maschio C, Schiozer DJ (2014) Application of
artificial neural networks in a history matching process. SPE J
28. Shahkarami A, Mohaghegh SD, Gholami V, Haghighat SA (2014)
Artificial intelligence (AI) assisted history matching. In: SPE
western North American and Rocky Mountain joint meeting,
17–18 April. Society of Petroleum Engineers, Denver, pp 1–26
29. Denney D (2011) Modeling history matching, forecasting, and
analysis of shale-reservoir performance with artificial intelligence.
SPE 63(09):60–63
30. Zhu QJ, Gu YB, Lu SL etal (2002) Application of artificial neu-
ral network model in oil resource prediction. East China Geol
31. Al-Fattah SM, Startzman RA (2001) Predicting natural gas pro-
duction using artificial neural network. In: SPE hydrocarbon eco-
nomics and evaluation symposium, 2–3 April. Society of Petro-
leum Engineers, Dallas, pp 1–10
32. Osman ESA (2001) Artificial neural networks models for identify-
ing flow regimes and predicting liquid holdup in horizontal mul-
tiphase flow. In: SPE middle east oil show, 17–20 March. Society
of Petroleum Engineers, Manama, pp 1–8
33. Yan HY, Fu JY, Dong JH etal (2014) Forecast model of oilfield
indexes based on IPSO GNN. Sci Technol Eng 14(15):197–202
34. Xu SH, Bi CC, Zhang Y etal (2015) Oilfield development indi-
cators prediction based on radial basis process neural network.
Comput Technol Autom 3:52–54
35. Feng GY, Han JX (2015) The oilfield production prediction model
based on principal component analysis and least squares support
vector machine. Comput Knowl Technol 11(31):144–147
36. Gaurav A (2017) Horizontal shale well EUR determination
integrating geology, machine learning, pattern recognition and
multivariate statistics focused on the Permian basin. In: SPE liq-
uids-rich basins conference—North America, 13–14 September.
Society of Petroleum Engineers, Midland, pp 1–19
37. Salem KG, Abdulaziz AAM, Abdel Sattar A, Dahab ASD (2018)
Prediction of hydraulic properties in carbonate reservoirs using
artificial neural network. In: Abu Dhabi international petroleum
exhibition & conference, 12–15 November. Society of Petroleum,
Abu Dhabi, pp 1–18
38. Ghahfarokhi PK, Carr T, Bhattacharya S, Elliott J, Shahkarami
A, Martin K (2018) A fiber-optic assisted multilayer perceptron
reservoir production modeling: a machine learning approach in
prediction of gas production from the Marcellus shale. In: SPE/
AAPG/SEG unconventional resources technology conference,
23–25 July, Houston, Texas, USA pp 1–10
39. Khan MR, Tariq Z, Abdulraheem A (2018) Machine learning
derived correlation to determine water saturation in complex
lithologies. In: SPE Kingdom of Saudi Arabia annual technical
symposium and exhibition, 23–26 April. Society of Petroleum
Engineers, Dammam, pp 1–10
40. Zhang ZX, Kang XF, Zhang FF (2012) Application of grey theory
for oil and gas reservoir evaluation program optimization. Adv
Mater Res 616–618:1008–1012
41. Kang XJ, Yuan AW, Gao J etal (2006) Neural network method
for comprehensive evaluation of oilfield development planning
scheme. Spec Oil Gas Res 13(2):48–50
42. Park H, Lim JS, Kang JM, Roh J, Min B (2006) A hybrid arti-
ficial intelligence method for the optimization of integrated gas
production system. In: SPE Asia pacific oil & gas conference and
exhibition, 11–13 September. Society of Petroleum Engineers,
Adelaide, pp 1–9
43. Xiao DR, Pan H (2010) Optimization of designing of the oil field
exploiting based on fuzzy mathematics and BP neural network.
Microcomput Inf 26(6):209–211
44. Godarzi AA, Amiri RM, Talaei A etal (2014) Predicting oil price
movements: a dynamic artificial neural network approach. Energy
Policy 68(5):371–382
45. Sun H, Li PC (2016) Measures optimization for oil and water well
based on quantum particle swarm optimization. Comput Technol
Dev 26(9):78–82
46. Feng GQ, Pan LY, Kong B etal (2018) Hierarchical optimization
research based on fuzzy clustering analysis. Eval Dev Oil Gas
Reserv 3:30–39
47. Roueché JN, Karacan CO (2018) Zone identification and oil satu-
ration prediction in a waterflooded field: residual oil zone, East
Seminole field, Texas, USA, Permian Basin. In: SPE improved
oil recovery conference, 14–18 April. Society of Petroleum Engi-
neers, Tulsa, pp 1–14
48. Denney D (2000) Artificial neural networks identify restimulation
candidates. SPE 52(02):44–45
49. Du QL, Zhu LH (2009) Remaining oil distribution and its descrip-
tion technology during ultra-high water cut stage in La-Sa-Xing
oilfi1eld. Daqing Pet Geol Dev 5:99–105
50. Liu J, Gao HY, Wang TT etal (2013) Parameters prediction of
remaining oil of aII oil region in Daqing based on BP neural net-
works. Inner Mong Petrochem Ind 23:47–50
51. Mao GQ, Teng QZ, Wu Y etal (2014) Shape recognition of
remained oil based on Bp neural network. J Terahertz Sci Electron
Inf Technol 12(6):858–864
52. Denney D (2006) Treating uncertainties in reservoir-performance
prediction with neural networks. Soc Pet Eng 58(06):69–71
53. Thomas AL, La Pointe PR (1995) Conductive fracture identifi-
cation using neural networks. In: The 35th U.S. symposium on
rock mechanics (USRMS), 5–7 June. American Rock Mechanics
Association, Reno pp 1–6
54. Zhang JC, Xing YZ, Zheng LH (2005) Using artificial intelligent
technique to identify fractures. Well Log Technol 29(1):52–54
55. Huang FX, Xia ZY, Gui HB etal (2016) The application of BP
neural network to DMT hill metamorphic fracture prediction.
Chin J Eng Geophys 13(4):483–490
56. Anderson RN, Xie B, Wu L, Kressner AA, Frantz JH, Ockree
MA, McLane MA (2016) Using machine learning to identify the
highest wet gas producing mix of hydraulic fracture classes and
technology improvements in the Marcellus shale. In: Unconven-
tional resources technology conference, pp 1–13
57. Xie W (2017) Pso-LSSVM based identifying method for the
fracture-vug fillings. Daqing Pet Geol Dev 2:135–142
58. Wu ZY, Mo XW, Liu JH etal (2018) Convolutional neural net-
work algorithm for classification evaluation of fractured reser-
voirs. Pet Explor 57(04):135–143
59. Nande S (2018) Application of machine learning for closure pres-
sure determination. In: SPE annual technical conference and exhi-
bition, 24–26 September. Society of Petroleum Engineers, Dallas,
pp 1–10
60. Ahmed SA, Elkatatny S, Ali AZ, Abdulraheem A, Mahmoud M
(2019) Artificial neural network ANN approach to predict fracture
pressure. In: SPE middle east oil and gas show and conference,
18–21 March. Society of Petroleum Engineers, Manama, pp 1–9
61. Adeyemi BJ, Sulaimon AA (2012) Predicting wax formation using
artificial neural network. In: Nigeria annual international confer-
ence and exhibition, 6–8 August. Society of Petroleum Engineers,
Lagos pp 1–8
62. Li TN (2015) Study on diagnosis model of oilfield abnormal well
based on feedback dynamic neural network. Comput Technol
Autom 2:114–116
63. Liu BJ (2018) Research on diagnostic technique of indicator dia-
gram based on CNN convolution neural network. J Xi’an Pet Univ
64. Wang X, He Y, Li F, Dou X, Wang Z, Xu H, Fu L (2019) A work-
ing condition diagnosis model of sucker rod pumping wells based
949Applications ofArtificial Intelligence inOil andGas Development
1 3
on big data deep learning. In: International petroleum technology
conference 26–28 March, Beijing, China, pp 1–10
65. Zhou CC, Li J, Zhang XG etal (2008) Predication for EOR by
polymer flooding based on artificial neural network-comparison
between ANN and quadratic. Polynomial Stepwise Regres Method
66. Ni HM, Liu YJ, Fan YC etal (2014) Optimization of steam flood-
ing injection and production based on improved particle swarm
optimization. Acta Pet Sin 35(1):114–117
67. Shi SZ, Yu HY, Sun ZL etal (2014) Forecast of fracturing effect
based on gray correlation analysis and BP neural network. J
Changjiang Univ (Self Publ Ed) 31:154–156
68. Yang ZH, Li ZP (2017) A new method for choice of water-control
fracturing segments in horizontal wells based on the BP neural
network system. Geol Explor 53(4):0818–0824
Publisher’s Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
... They are particularly useful in solving problems where the relationships between variables are unknown or poorly understood [51,52]. Examples of the application of evolutionary algorithms in reservoir engineering include modelling and production optimization [53][54][55], estimation of effectiveness and optimization of EOR methods (including WAG) [56][57][58][59], estimating values of parameters such as MMP [60], the formation volume factor [61] or the emulsion viscosity [62], and issues related to the reservoir development [63][64][65][66][67]. ...
Full-text available
Production from mature oil fields is gradually declining, and new discoveries are not sufficient to meet the growing demand for oil products. Hence, enhanced oil recovery is emerging as an essential link in the global oil industry. This paper aims to recognize the possibility of increasing oil recovery from Polish carbonate reservoirs by the water alternating gas injection process (WAG) using various types of gases, including CO2, acid gas (a mixture of CO2 and H2S of 70/30% vol/vol) and high-nitrogen natural gases occurring in the Polish Lowlands. A series of 17 core flooding experiments were performed under the temperature of 126 °C, and at pressures of 270 and 170 bar on composite carbonate cores consisting of four dolomite core plugs. Original reservoir rock and fluids were used. A set of slim tube tests was conducted to determine the miscibility conditions of the injected fluids with reservoir oil. The WAG process was compared to continuous gas injection (CGI) and continuous water injection (CWI) and was proven to be more effective. CO2 WAG injection resulted in a recovery factor (RF) of up to 82%, where the high nitrogen natural gas WAG injection was less effective with the highest recovery of 70%. Based on the core flooding results and through implementing a genetic programming algorithm, a mathematical model was developed to estimate recovery factors using variables specific to a given WAG scheme.
... Such a tool can further enhance the gas recovery, decrease production costs, and effectively protect the environment and reduce carbon emissions. With recent advances in artificial intelligence (AI) techniques (e.g. machine learning and data analytics), opportunities have opened up for meta-model development for problems involving multiphysics processes associated a large number of properties [32][33][34][35][36][37][38][39]. The main goal of the meta model is to provide an alternative tool to replace a conventional THM simulator in a more timely and computationally effective way. ...
Full-text available
Natural gas hydrate, a crystalline solid existing under high-pressure and low-temperature conditions, has been regarded as a potential alternative energy resource. It is globally widespread and occurs mainly inside the pores of deepwater sediments and sediments under permafrost area. Hydrate production via well depressurization is deemed well-suited to existing technology, in which the pore pressure is lowered, the natural gas hydrate is dissociated into water and gas, and the water and gas are produced from well. This method triggers multiphysics processes such as fluid flow, heat transfer, energy adsorption, chemical reaction and sediment deformation, all of which are dependent on the amount of gas hydrates remaining in the pores. Therefore, modeling of hydrate production is computationally intensive and expensive. While back-analysis through observed production history is essential for better understanding of the reservoir characteristics and reliable prediction for future gas hydrate production, a large number of required simulations makes it impractical. This study employs Artificial Intelligence (AI) to achieve an efficient back-analysis of the gas hydrate production conducted at the offshore Nankai site, Japan, in 2013. The results show that the AI-based metamodel is capable of reproducing outputs of heavy computation of the multiphysics processes and thus performs back-analysis greatly efficiently. The efficient AI-based metamodel also makes it possible to carry out sensitivity analysis and it is found that the permeability and the preyield plasticity parameter are most influential to reservoir responses. The approach of this study can be applicable to other reservoirs and will reveal the ground truth in-situ properties and the most influential properties, contributing to better understanding of reservoir behavior for future gas hydrate production.
Artificial Intelligence (AI) has become one of the most popular technologies and technological tools. The capabilities and use cases of AI have only increased, both in scope and complexity over the past few years. What initially started as a simple idea in the 1950s has now grown into perhaps the most influential twenty-first century technologies. In the industrial sector, the use of AI has become more and more common and even popular, given the unique problems the sector faces. In this paper, the potential of applying AI in the oil and gas sector is explicated. This is as a consequence of the significant impact that AI is bound to have on advancing the oil and gas sector moving into the future. It was established that the adoption of AI will provide credible solutions to the most pressing challenges faced by the sector.
Conference Paper
Natural gas is one of the alternative energy sources for oil. It is a highly efficient, low-cost, low emission fuel. Natural gas is an essential energy source for the chemical industry and is a critical component of the world's energy supply. It is also considered one of the cleanest, safest, and most beneficial sources of energy available. Therefore, this paper attempts to build a natural gas forecast optimization model as well as find out what type of gas is connected to the networks using intelligent data analysis. In this study; We will build the HPM-STG system, which consists of integrating two technologies together: XGBoost prediction technology and GSK optimization technology. This study will solve some problems, including the problem of data coming from natural gas that is collected manually, so the error rate will be because the prediction principle must be correct data, so we will solve it by collecting data through sensors, and the second problem that we will solve is the algorithm problem. Although the XGBoost algorithm is one of the best prediction algorithms, it faces many problems, so its core will be replaced by one of the optimization algorithms, and these two technologies together will give more accurate results.
Full-text available
The typical carbonate reservoir in the Middle East has huge potential, but due to low permeability and high gas-oil ratio, the characteristics of reservoir saturation and pressure distribution are not clear, which makes reservoir development very difficult. As an intelligent prediction and analysis technique, deep learning provides a novel solution to solve this problem. In this study, firstly, the actual static geological data and dynamic production data from the target reservoir were gathered to provide a data foundation for model training and analysis. And then, Random Forest algorithms, Reservoir Engineering methods, and Kriging interpolation algorithms were selected to predict and supplement the missing data of pressure and saturation field. After that, the Convolutional Long Short-term Memory (ConvLSTM) neural network algorithm was employed to build an optimal model to forecast the oil/water/gas saturation distribution and pressure distribution fields. The results showed that the optimal ConvLSTM model achieved good prediction performance. Not only the calculation efficiency, but also the prediction accuracy was higher, in which the average R2 was 0.8825 and the average accuracy was 0.865. Furthermore, we proposed five development optimizations for the remaining oil potential tapping strategies according to the prediction results. The workflow proposed in this study provides a new direction for the characterization of saturation and pressure distribution in a typical carbonate reservoir.
Full-text available
Learning Outcomes of Classroom Research. Multiple aspects of Artificial Intelligence and their applications were studied.
Conference Paper
Monitoring the working conditions of the sucker rod pumping wells in a timely and accurate manner is important for oil production. With the development of smart oil fields, more and more sensors are installed on the well. The variety and volume of the monitoring data are big. The oilfield big data can be utilized to improve the diagnostic performance. In this work, we aim to utilize the big data collected during oil well production and a deep learning technique to build a new generation of intelligent diagnosis model to monitor working condition of sucker rod pumping wells. Over 5 million of well monitoring records, which covers information about one year of an oil field block, are collected and preprocessed. To show the dynamic changes of the working conditions for the wells, the overlay dynamometer card is proposed and plotted for each data record. The overlay dynamometer card stacks two dynamometer curve at different times. Based on the overlay dynamometer cards, the working conditions are divided into 30 types, and the corresponding dataset are created. An intelligent diagnosis model using the convolutional neural network (CNN), one of the deep learning framework, is proposed. By the convolution and pooling operation, CNN can extract features of an image implicitly without human effort and prior knowledge. That makes CNN very suitable for the recognition of the overlay dynamometer cards. The architecture for working condition diagnosis CNN model is designed. The CNN model consists of 14 layers with six convolutional layers, three pooling layers, and three fully connected layers. The total number of neurons is over 1.7 million. The overlay dynamometer card dataset is used to train and validate the CNN model. The accuracy and efficiency of the model are evaluated. Both the training and validation accuracies of the CNN model are over 99% after ten training epochs. The average training elapsed time for an epoch is 8909.5 seconds and the average time to diagnosis a sample is 1.3 milliseconds. Based on the trained CNN model, a working condition monitoring software for suck rod pumping well is developed. The software runs 7×24 hours to diagnosis the working conditions of wells and post warning to users. It also has a feedback learning workflow to update the CNN model regularly to improve its performance. Through 3 months of on-site test run show that the actual accuracy of the CNN model is over 90%.
Conference Paper
While there is a lot of talk about Big Data Analytics, Internet of Things (IoT), Artificial Intelligence (AI), Real Time Monitoring (RTM), Digital Twin and other methodologies, all of them require, not only data, but accurate, reliable data. This paper describes a new and innovative inspection methodology that combines 3D laser scanning and precise 3D Geometric Dimensioning & Tolerancing (GD&T) metrology data with advanced Non-Destructive Testing (NDT) results. This data is then combined in digital 3D space to give an accurate representation of current equipment condition and mechanical integrity of critical offshore assets. Such inspection and testing can be conducted during manufacturing as a quality check, creating digital baseline records, or on deck during operations, saving significant downtime and costs. By including metrology and phased array, the described inspection methodology can provide precise digital data and specialized 3D reports that will satisfy not only compliance and regulatory efforts in a more objective manner, but also assist original equipment manufacturers (OEMs), drilling contractors and operators by supplying conclusive and accurate data of equipment condition. This data, based on in-situ NDT and Geometric Dimensioning & Tolerancing measurement information, will support Digital Twin data and operational and maintenance decisions that will preserve the integrity, safety, and availability of the assets. This innovative inspection solution can form part of a Condition-Based Maintenance (CBM) program where operators can move from time-based programs and use the digital data to determine future equipment performance, work scope, and schedules while maintaining a complete and updated digital condition record throughout the lifecycle of the equipment. The program may predict and prevent problems at early stages, provide strategies that will simplify maintenance activities, and potentially identify manufacturing flaws. More importantly, it can create historical digital data that will change the way the drilling industry operates and satisfy what regulatory agencies have been seeking since the implementation of the new well control rule.
Conference Paper
Fracture pressure is a critical formation condition that affects efficiency and economy of drilling operations. The knowledge of the fracture pressure is significant to control the well. It will assist in avoiding problems associated with drilling operation and decreasing the cost of drilling operation. It is essential to predict fracture pressure accurately prior to drilling process to prevent various issues for example fluid loss, kicks, fracture the formation, differential pipe sticking, heaving shale and blowouts. Many models are used to estimate the fracture pressure either from log information or formation strengths. However, these models have some limitations such as some of the models can only be used in clean shales, applicable only for the pressure generated by under-compaction mechanism and some of them are not applicable in unloading formations. Few papers used artificial intelligence (AI) to estimate the fracture pressure. In this work, a real filed data that contain only the real time surface drilling parameters were utilized by artificial neural network (ANN) to predict the fracture pressure. The results indicated that artificial neural network (ANN) predicted the fracture pressures with an excellent precision where the coefficient of determination (R2) is greater than 0.99. In addition, the artificial neural network (ANN) was compared with other fracture pressure models such as Matthews and Kelly model, which is one of the most used models in the prediction of the fracture pressure in the field. Artificial neural network (ANN) model outperformed the fracture models by a high margin and by its simple prediction of fracture pressure where it can predict the fracture pressure from only the real time surface drilling parameters, which are easily available.
The indicator diagram of oil pumping well can directly reflect the operation situation of the well, and the analysis and research of it is the most direct and effective method for the analysis of the working condition and parameter optimization of the oil well. Through the digital description of the oil well parameters and indicator diagram, combined with the convolution neural network technology, the indicator diagram diagnosis model was established, and the corresponding software program was developed to realize the intelligent diagnosis of the working condition of oil pumping well. The application result shows that the diagnostic accuracy of the model is 89.3% for the common working conditions, such as insufficient liquid supply, gas influence, eccentric wear, packing tight, etc. The model has become an effective technical means for oil well condition analysis and production optimization. © 2018, the Editorial Department of Journal of Xi'an Shiyou University. All right reserved.
Conference Paper
Asset optimization has recently become a crucial issue in Oil&Gas industry, considering oil price conjuncture and an increased awareness on environmental aspects. In this paper, an Artificial Intelligence (AI) technique is presented, which is able to manage big dataset to automatically match the entire production model against measured field data. The tool is based on a hybrid in-house developed AI technique, integrating deep neural networks, biogenetical algorithms, commercial simulators and real-time data. The workflow starts with the modeling of the production system through physics-based commercial simulators. A sensitivity analysis identifies the critical variables, which are then randomly varied with a Sobol distribution, exploring the entire solution domain. With these data, a proxy model to the commercial software is generated using an artificial neural network. Finally, the AI tool fed by real-time data is used to match the field behavior: uncertain parameters are modified through a differential evolution algorithm that minimizes the error between calculated and measured variables. The matching parameters are, then, passed to the simulators achieving a field representative model. The tool has been developed considering an operating field in offshore western Africa. The typical uncertain parameters in this kind of field are related to the fluid characteristics, in particular densities and compositions, but also to the physical characterization of the pipelines such as roughness and heat transfer characteristics. The matching process has been performed coupling the proxy model, which is a neural network able to replicate the field behavior, and a differential evolution algorithm as the optimization algorithm. The fitness function to be minimized is a Mean Absolute Percentage Error (MAPE) that represents the distance between the actual field production parameters and the modelled ones. The best configuration of both the neural network and the differential evolution algorithm required a computational time of 6 seconds with a MAPE equal to 2.6%. These results are compared to the one obtained coupling the same differential evolution algorithm with the commercial simulator to perform the matching. The required computational time is equal to about 20 hours (70400s) and a MAPE equal to 2.2%. The big gain with the novel approach is clearly the knocking down of computational time with a comparable error. In this paper, it has been shown how substituting the physical model with a proxy one can give substantial advantages in terms of computational time. In principle, with the velocity of the tool implemented, the matching procedure could be done on a daily basis. This is a breakthrough because it allows having the simulator model always tuned and ready to be utilized.
Conference Paper
Artificial Intelligence (AI) is a relatively new concept for oil and gas industry. Yet, its agile nature and massive potential give a promise for constantly evolving oil and gas business in tackling such issues as environmental protection. AI systems utilize a variety of technologies making it possible to produce comprehensive, nonetheless easy-to-use tools, which can be used by Health, Safety and Environment (HSE) professionals in their daily activities. This paper examines a so-called HSE AI Assistant - another tool aimed to provide support to HSE professionals in getting instant answers to questions on a variety of regulatory topics in occupational health, safety, and environmental protection. This AI tool will utilize latest NLP technologies to retrieve data from a bank of up-to-date HSE regulations and standards, and provide answers to required questions. The tool is a much more efficient alternative to traditional search engines and closed intranet data retrievals, as the AI system has an ability to understand human questions, hunt for applicable data in attribute files, and produce the most relevant answer to the questions. The system utilizes latest machine learning technologies, and has a capability of constant learning and evolvement. This will be especially useful in terms of changing local legislation and updates to the HSE regulations. The HSE AI Assistant can also be supplied with a capability to assist HSE specialists in classification of hazardous materials and help to perform regular HSE audits in the areas where hazardous materials are used and disposed. Developed locally in Turkmenistan by a team of HSE and IT professionals, the HSE AI Assistant can become a very accessible and easy to maintain reference and solution tool for HSE specialists working on COSHH audits and environmental impact assessment. The HSE AI can also assist those who control the chemical stock in the warehouse, and who manage the hazardous waste at the disposal sites.
Conference Paper
The industry is undergoing a transition into efficient technologies and it has digitalization written all over it. Digitalization not only should be about data, a fancy software, touchscreens and the internet, it is important that solutions are able to connect within existing work processes and with people for companies to truly lead to more efficient and safer drilling operations. Oil and gas industries are now moving towards using Digital Twin's during the life-cycle of well construction. The concept of Digital Twins was first introduced by Dr. Michael Grieves at the University of Michigan in 2002 through Grieves’ Executive Course on Product Lifecycle Management. Digital Twin is a digital copy of the physical systems and act as a connection between physics and digital world. The digital system gets the real-time data from the mechanical systems which include all functionality and operational status of the physical system. An example from another industry; A Formula 1 team uses data from many sensors used in the car, harnessing data and using algorithms to make projections about what's ahead, and apply complex computer models to relay optimal race strategies back to the driver. Ultimately, to drive faster and safer. By means of the digital twin of the drilling wells during the life cycle of the drilling by combining digital and real-time data together with predictive diagnostic messages there is seen a lot of advantageous in the improvement of accuracy in decision making and results. This again will help the industry to increase safety, improve efficiency and gain the best economic-value-based decision. A Digital Twin driven by real-time data helps to give operations the optimal plan with focus on safety, risk reduction and improved performance. In this paper, the concept will first be explained in creating and utilizing a Digital Twin of your well for drilling and how it will directly influence how Drilling/well engineers, managers and supervisors plan, prepare and monitor their drilling operations and then implement learnings on future wells; for faster and improved decision making with direct relation to predicting and avoiding/mitigating NPT while also optimizing operations along with it. Case examples will be shared, showing value from use of the Digital Twin from first introduced in 2008 up until now where operators around the globe have implemented it for multiple uses in the drilling lifecycle.
Conference Paper
Hydraulic fracturing is a method in which fluid is pumped at elevated pressures to break down the formation, and create a conductive pathway for production of hydrocarbon fluids. Understanding the stress environment of the rock is critical for a better design and successful execution of a fracturing treatment. More often than not, a formation breakdown pressure is equal to or in close approximation of minimum horizontal or in-situ stress, also termed as closure pressure. Various analytical methods such as G-Function plot, G-dP/dG plot, square-root of time plot etc. are used for the determination of closure pressure, and have been implemented since the inception of hydraulic fracturing as a way to better design fracture treatments. These methods are prone to have subjectivity due to the experience and knowledge of the person analyzing the data, which calls for a need to more objectively analyze such data, in order to better predict the closure pressure. Machine learning is a method to teach computers to implement a predesigned algorithm and execute tasks without having to explicitly program them. It helps create significantly complex mathematical models which automate processes based on critical learning parameters, and predict within a certain acceptable degree of accuracy. In this paper, Artificial Neural Networks (ANN), a machine learning methodology, has been applied in order to minimize the subjectivity in predicting the value of closure pressure. Artificial neural networks, similar to neural networks of the brain, are a system comprising of various neurons. These neurons are organized in layers namely input layer (consisting of input neurons), output layer (consisting of output neurons or results) and multiple hidden layers. The number of input, hidden and output neurons depend on the parameters affecting the end-result. An ANN has been designed, taking into consideration critical parameters on which closure pressure depends. The model identifies and learns from the patterns in the data and predicts the required output. This output is then compared with the actual results in order minimize the error. The objective is to minimize the error so as to get a close match for the given data. In this paper we have kept the ratio of learning to testing at 80:20, which means that of all the available data, 80% is used for training the model and the rest 20% is used for testing the model. Results from this work point to the fact that the ANN was able to predict the closure pressure with reasonable accuracy.
Conference Paper
Possibly the most underrated petrophysical parameter, the importance of water saturation cannot be emphasized enough with a whole range of petrophysical as well as reservoir engineering computations being dependent on its accurate determination leading to vital field development decisions; reserves estimation, waterflooding efficiency calculation and capillary pressure deduction. In 1942 Archie was first to present the equation to determine water saturation in a clean, non-clay reservoir. Ironically, ever since decades have passed with the intricacy of water saturation determination yet to be untangled in complex lithologies, especially in carbonates. Several researchers have tried to deconvolute the water distribution in composite formations by formulating empirical correlations that depend on log derived data which is not a very precise representation and hence no consensus exists among log analysts about which model can be universally used. The use of computer generated algorithms, fuzzy logic and neural networking is picking up pace in the petroleum industry. Consequently, in this paper we show how Machine Learning can be used to generate a correlation, to determine water saturation in carbonate reservoirs, which is simple and practical to use in the sense that has less uncertainty in the parameters that it employs compared to existing models. In this work, multiple machine learning techniques namely, Artificial Neural Network (ANN), and Adaptive Neuro-Fuzzy Inference System (ANFIS) are used to estimate water saturation using conventional wireline log data as input parameters and the output is core dean-stark data. The data comprised of more than 2000 well log points which were reduced to around 150 corresponding to available core data. All the developed models are compared after a rigorous sensitivity analysis based on various artificial intelligence algorithms. This work clearly shows that computer-based machine learning techniques can determine water saturation with a precision of approximately 94% when related to experimental core values. The developed correlation works extremely well in prediction mode with the shale affected log data as inputs. A comprehensive numerical and illustrative evaluation of the claimed accuracy is shown along with the error analysis between both the machine learning techniques used.
Technology Focus The future is here; a machine can learn games and beat the world’s best players. What a fascinating time we are living in—Industrial Revolution 4.0. In December 2017, news broke that AlphaZero decisively beat the world’s best players in chess after its older sibling, AlphaGo, defeated the world champions in the ancient game of go a couple of years earlier. These games have been won by machine learning and artificial intelligence. What effects will Industrial Revolution 4.0 have on the upstream oil and gas business in general and petrophysics in particular? We should see more intelligence and automation in measurements, processes, work flows, and operations, which should result in more-consistent results, better-quality answer products, and less nonproductive time (i.e., improved quality, efficiency, and productivity with less cost). Historically, petrophysics is based on physical principles or empirical relationships, as illustrated in paper SPE 191296 on predicting crude-oil viscosity. With Industrial Revolution 4.0, a new era in petrophysics has begun. The logging tools are smarter (as demonstrated in paper SPE 190062 dealing with environmental corrections to some of the deliverables of pulsed-neutron logs that can be performed automatically). And the operational and data-interpretation work flows are more automated (as shown in paper SPE 187040, which details formation testing and sampling jobs that can be done semiautomatically through standardizing terminologies, measurement uncertainties, and data quality-control criteria). In playing games such as go and chess, machines learn on the basis of man-made game rules. In petrophysics, rules are often data-driven, so data quality becomes critically important. It is always true that having bad data is worse than having no data. Density, representativeness, and coverage are other parameters of the data besides data quality that are required for data-driven petrophysics. Recommended additional reading at OnePetro: SPE 189807 Characterization of Reservoir Quality in Tight Rocks Using Drill Cuttings: Examples From the Montney Formation, Alberta, Canada by A. Ghanizadeh, University of Calgary, et al. SPE 188804 Low-Resistivity Pay Identification in Lower Cretaceous Carbonates, Onshore UAE by J.L. Ruiz, ADCO, et al. SPE 187371 Saturation Mapping in the Interwell Reservoir Volume: A New Technology Breakthrough by Alberto F. Marsala, Saudi Aramco, et al.