Environmental Science and Pollution Research

https://doi.org/10.1007/s11356-021-15379-z

GREEN ENERGY FOR ENVIRONMENTAL SUSTAINABILITY

A comprehensive study on artificial intelligence in oil and gas sector

Devansh Gupta 1 & Manan Shah 2

Received: 24 May 2021 / Accepted: 6 July 2021

# The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2021

Abstract
The authors investigate how artificial intelligence modifies a huge piece of the energy area, the oil and gas industry. This paper
attempts to evaluate technical and non-technical factors affecting the adoption of machine learning technologies. The study
includes machine learning development platforms, network architecture, and opportunities and challenges of adopting machine
learning technologies in the oil and gas industry. The authors elaborate on the three different sectors in this industry namely
upstream, midstream, and downstream. Herein, a review is presented to evaluate the applications and scope of machine learning
in the oil and gas industry to optimize the upstream operations (including exploration, drilling, reservoir, and production),
midstream operations (including transportation using pipelines, ships, and road vehicles), and downstream operations (including
production of refinery products like fuels, lubricants, and plastics). Enhanced processing of seismic data is illustrated which
provides the industry with a better understanding of machine learning applications. Basing on the investigation of AI imple-
mentation prospects and the survey of subsisting implementations, they diagram the latest patterns in creating AI-based instru-
ments and distinguish their impacts on speeding up and de-gambling measures in the business. They examine AI proposition and
calculations, just as the job and accessibility of information in the portion. Furthermore, they examine the principal non-
specialized difficulties that forestall the concentrated use of man-made brainpower in the oil and gas industry (OGI), identified
with information, individuals, and new types of joint effort. They additionally diagram potential situations of how man-made
reasoning will create in the OGI and how it might transform it later on (in 5, 10, and 20 years).

Keywords Artificial intelligence . Oil and gas . Upstream . Midstream

Introduction upstream, midstream, and downstream. This industry is highly

dynamic making its traditional centralized approach to work al-
The oil and gas sector has a staggering influence on the econo- most impossible to meet its ever-growing demand. The industry
mies of both oil-producing and oil-importing countries as well as necessitates advancement in technology for enhancing operation-
the everyday activities of humankind. The applications of oil and al potency and improving profits (Hanga and Kovalchuk 2019).
gas are directly influential to industrial growth worldwide and Several concrete assertions prove that this industry has been the
have been seen with the ease of transportation at both the global most influential source of energy for industries as well as human-
and local levels. Many of the household chores are solved using kind, one of which states that according to the 2011 fortune
oil and gas such as heating, generation of electricity, cooking, global 500 rankings, out of the top 10 world’s biggest companies,
and local mobility. The oil and gas industry (OGI) is one of the six are oil and gas companies, and out of these six, five have
largest industries involving three concrete sectors namely subsidiaries operating in different parts of the world. Showing its
evergreen merit and importance in human life and the worldwide
economy, no wonder it is called the “black gold.”
Responsible Editor: Nicholas Apergis
Artificial intelligence in OGI can be used to enhance the oil
and gas rate of production and in minimization of the cost of
* Manan Shah
[email protected]
lifting (Brelsford, 2018). Its applications include strengthen-
ing the modeling of reservoirs and maintenance prevention. It
1
Department of Chemical Engineering, Manipal Institute of is used in determining the most desired setup for defined
Technology, Manipal, India maintenance to circumvent failures by optimizing the sched-
2
Department of Chemical Engineering, School of Technology, Pandit ule of regular maintenance. Another important implementa-
Deendayal Energy University, Gujarat 382426 Gandhinagar, India tion of AI in the OGI is detecting anomalies; it helps in
 Environ Sci Pollut Res

locating patterns and forestalling failures based on the rela- generated reality, and so on), new advances and frameworks
tionship between adequate readings (Hanga and Kovalchuk associated with AI will be continually proposed, which will
2019). The implementation of AI in the OGI is rapidly grow- doubtlessly turn into a significant method to decrease costs
ing. It is being currently used in intelligent drilling, intelligent and improve proficiency (Li et al. 2020).
production, intelligent refinery, etc. Various practical applica-
tion technologies are used in exploration and development. Internet of things (IoT)
Excellent results have been pulled off in the exploration field
by reducing risks in exploration and improving the success In the OGI, the guarantee of IoT applications lies not with
rate of the exploration wells. AI has provided more accurate overseeing existing resources, supply chains, or client connec-
methods for fracture scheme design and selection of operating tions in any case, rather, in making new worth in data about
wells and target layers (Li et al. 2020). these (Burtescu et al. 2014). The Internet of things (IoT),
The paper uses learnings from many AI projects performed which fundamentally incorporates detecting, correspon-
with the authors’ association over the last 10–15 years for dences, and investigation capacities, has been stewing for a
driving oil and gas organizations around the world. The un- while. There are three business targets pertinent to IoT orga-
dertakings covered AI answers for the entire range of the nizations in the oil and gas (O&G) business: improving de-
sectors’ exercises—geological appraisal of the reservoirs, dril- pendability, streamlining activities, and making new worth.
ling improvement, reservoir designing/field advancement, Upstream organizations (e.g., investigation and creation) ze-
production optimization, transportation development, and re- roed in on advancement that can acquire new operational ex-
finery product creation. periences by examining assorted arrangements of material sci-
As oil and gas organizations are a lot faster to receive new ence, non-physical science, and cross-disciplinary information
advancements than to explore different avenues regarding and (Dietterich 1997). Midstream organizations (e.g., transporta-
change their plans of action, their AI’s essential objective (and tion, like pipelines and capacity) peering toward higher orga-
other digitalization) endeavors are to improve effectiveness. nization honesty and new plug openings will in general dis-
Practically speaking, that normally intends to speed up mea- cover huge advantages by building an information-
sures and diminish hazards. This paper expects to discuss empowered foundation. Downstream players (e.g., oil-based
exhaustively and show how AI is changing the oil and gas commodities purifiers and retailers) should see the most en-
upstream, midstream, and downstream. The authors will prin- couraging open doors in the income age by growing their
cipally zero in on the accompanying three inquiries: perceivability into the hydrocarbon inventory network and
focusing on advanced shoppers through new types of associ-
& What de-gambling in the oil and gas industry means and ated advertising (Slaughter and Mittal 2021).
how AI is assisting with it.
& Which cycles can be sped up by applying AI and how Data analytics
much.
& What has been already done and what are the expected Huge information incorporates amorphous (not coordinated and
advancements in the following years (Li et al. 2020). text-hefty) and multi-organized information (counting diverse
information designs coming about because of individual/
machine interactions) (Mohammadpoor and Torabi 2020).
This attribute of big data can be seen in different areas of
Advance technology in oil and gas sector OGI, like investigation, well treatment, and creation. During
oil and gas investigation, seismic information securing pro-
For the most part, embraced use of AI in repository improve- duces a lot of information utilized to create 2D and 3D pic-
ment and abuse is the streamlining of advancement dependent tures of the subsurface layers. For the seaward seismic inves-
on chronicled oilfield creation information. Entering the big tigations, narrow azimuth towed streaming (NATS) utilizes
data time, it is vital for oilfield improvement to completely the accumulated information to create pictures of the funda-
investigate the colossal likely estimation of huge information mental geography. Wide azimuth (WAZ) is a later advance-
and to uncover covered up, beforehand obscure, and possibly ment to catch more information and create greater pictures.
significant data, which is likewise one of the hot issues in AI Every one of these apparatuses and advancements is creating
oilfield advancement research (W.W. Weiss et al. 2002; Li more information which requires preparing and analysis
et al. 2020). With the turn of events and utilization of large (Mohammadpoor and Torabi 2020). Optical fibers joined with
information and constant improvement of different related different sensors are currently being utilized in wells—
calculations, AI assumes an undeniably significant part in rounded to record various boundaries like liquid pressing fac-
the field of oilfield advancement. Joined with other applicable tor, temperature, and composition during oil and gas creation
innovations (cloud computing, Internet of things, computer- (Feblowitz 2013; Mohammadpoor and Torabi 2020). Dealing
Environ Sci Pollut Res

with enormous informational indexes which are coming from business, including the subsurface (mining) operations,
a mind-boggling processing issue is complex, and tracking followed by advancement in fields and production of the un-
down the basic pattern can be testing. For these issues, big refined petroleum/gas. Investigation of oil and gas reserves is
data apparatuses can be extremely useful. a bunch of activities bringing about a 3D topographical model
of an oil/gas field or supply. The activities incorporate
Blockchain petrophysical and geophysical experiments and preparing
the information procured during the examinations. The up-
The blockchain has an incredible perspective in the OGI prin- stream includes natural gas and crude oil creation. It incorpo-
cipally in the accompanying four perspectives: cybersecurity, rates probing for likely underground or submerged natural gas
decision-making, trading, and management. The utilization of and crude oil fields, drilling exploratory wells, and hence dril-
blockchain innovation in oil and gas exchanging predomi- ling and working the wells used to heave the unrefined petro-
nantly incorporates transactions and smart contracts. A smart leum or crude natural gas to the surface. The upstream is
contract is a sort of agreement that documents terms with code specifically compelling as it is the most capital-concerted
rather than lawful language, and it quite possibly elicits the and significant of the three sections in the OGI business
main ideas in Ethereum (Buterin 2014; Lu et al. 2019). (Shafiee et al. 2019). Companies from the sector that manage
Ethereum upholds the improvement of smart contracts colossal vulnerabilities took care of manually and depended
through Turing complete dialects (Serpent, solidity, Viper). on expert information, not the real information. The vulnera-
As an implementation gushing in the Ethereum Virtual bilities need taking care of when settling on multibillion
Machine, the smart contract can get exchange solicitations choices on how and where to put resources into the coming
and occasions from outside and create new exchanges and 10–15 years. Nonetheless, regardless of the intricate and un-
occasions by setting off the running code rationale ahead of sure character of the executives’ issues in the area, the single-
time. The consequences of the smart agreement can be measure proceedings have truly overwhelmed decision-
refreshed for the situation with the record on the Ethereum making (Strantzali and Aravossis 2016; Koroteev and Tekic
organization, and these adjustments cannot be fashioned and 2021). To utilize existing field information to represent vul-
altered once affirmed. Due to the gigantic and compound na- nerabilities related to professionals’ emotional discount and
ture of the OGI, long and confounded agreements may emerge decision-making dependent on experience, the initial phases
in the exchange, everything being equal, and the number of in utilizing AI and ML in the upstream are made, turning out
agreements will be extensive. Smart contracts can incredibly to be progressively popular (Table 1).
diminish administrative work, work on the interaction, im- Mao et al. (2019) carried out surveying of seismic proper-
prove effectiveness, and save costs (Lu et al. 2019). In the ties that manufactures a set of sensor recordings called seismic
climate of oil value change, many oil and gas undertakings traces. These traces were time series addressing the robustness
are confronting enormous strain to decrease costs and enhance of elastic waves reflected from boundaries which were started
efficiency, to keep satisfactory overall revenue. In the oil and by a vibrator at the surface separating different subsurface
gas exchange, the conventional way provides the exchange development layers. These documented time series along with
unavoidably bringing about mistakes, and the exchange is spatial directions of the relating sensors and the vibrator were
inclined to extortion and bargain. Blockchain innovation can put to a distinctive re-modeling program bringing about noisy
take care of this difficulty well (overview of blockchain for 3D pictures representing a portion of the boundaries that were
energy and commodity trading, London, UK, 2017). It can reflected. The reconstruction algorithm is emphatically offline
likewise make the exchange more straightforward. The two because of critical necessities for high-performance process-
factions of the exchange can see all the exchange records and ing. Machine learning–based intelligence and relating-
assessments of the opposite side, which in this way can en- centered investigations were used to accelerate this process.
hance the achievement pace of the exchange. Moreover, the The 3D pictures are called seismic cubes. These were studied
two sides of the exchange can likewise perceive the particular by them, which can likewise be engaged with setting the past
circumstance of every stage in the exchange interaction, to be reconstruction phase variables. They fragmented the 3D pic-
more ready to command the general circumstance (Jordan and tures by choosing the surfaces, lines, and points inside the 3D
Mitchell 2015). cube, which were unquestionably identified with the limits
between the different layers in the subsurface development.
Wood 2020) evaluated well logging to collect more accu-
Application of artificial intelligence rate details about numerous subsurface physical properties
in upstream along a well bore. The resoluteness of well logging is down
to centimeters. The well logging sensors quantified natural
The OGI partitions into upstream, midstream, and down- gamma-beam force, electrical resistivity reaction to magnetic
stream. The upstream sums up the exploration piece of the excitation (nuclear magnetic reverberation study), and neutron
 Environ Sci Pollut Res

Table 1 A comprehensive study on artificial intelligence in petroleum upstream

Optical technique Application Method Reference

1. ANN, ML Selection of drill bits Decoding data on user-defined info database and learning National Oilwell
the codes to choose the acceptable bit for a specific Varco, 2013
drilling environment
2. GRNN, ML Fracture and mud prediction of Use of depth, Poisson’s ratio, and overburden gradient Sadiq and Nashawi,
gradient taken to plot the predicted mud/fracture gradient for data set 2000
3. BPNN, ML Casing collapse occurrence and Back simulation of user network defined number of inner Bello et al., 2016
depth determination hidden layers joined to input and output layers
4. DRIFT spectrum, Performance and quality of oilfield Producing infrared spectrum of cement powder to provide Fletcher et al., 1995
ANN, ML cement slurries refined info about nature of cement
5. BP ANN, ML Selection of offshore platform Nonlinear BPNNs with multiple input nodes, a preparing Wang et al., 2011
hidden layer, and model functions used for deepwater
floating platform selection
6. ANN, BPNN, ML Accurate prediction of frictional Two-layer BPNN with adequate training used to determine Sadiq and Gharbi,
drag bottom contact force and frictional drag as a function 1998
of various elasticity parameters
7. BPNNs, ML Prediction of hole cleaning Aimed to keep down damage of formation, stabilizer and Rooki et al., 2014
bit problems in drilling
8. ANN, ML ROP improvement Managing command parameters on real-time data to get Gidh et al., 2012
max bit possible administration length at highest possible
rate of drilling
9. ANN, ML Operational trouble shooting Forecasting the risk of stuck pipe during well planning and Shadizadeh et al.,
in real time 2010
10. Case-based Procedural decision taking Aiding in the choosing of the most favorable hole cleaning process in Popa et al., 2008
reasoning, ML unconsolidated shallow sands

thickness. The outcome of the well logging was vectors of advancement plans. The field improvement plan contained the
properties along the wellbore. He utilized well logging infor- arrangement for the drilling operations of the well. The out-
mation for his elucidation schedule, assessment of porosity come of each one of the runs of the reservoir modeling was a
and penetrability along the bore, rock composing along the conjecture of oil and gas production for approaching years
wellbore, and assessment of relative fluid saturation (a mea- (commonly 15–20 years) for a specific field advancement
sure of oil versus the measure of gas and measure of water) plan. Executing numerous runs, they selected the ideal field
along the bore. improvement plan and field advancement plan.
Baraboshkin et al. (2019) evaluated to study the core anal- Temirchev et al. (2020) applied machine learning in reser-
ysis by refining the seismic interpretation. The results of the voir engineering. They identified calculations with the help of
petrophysical understanding were utilized to refine seismic conventional reservoir modeling apparatuses. The instruments
translation. They extrapolated the obtained resources into the performed mathematical solutions of partial differential equa-
seismic cube from near-wellbore zones, saturating the 3D di- tions portraying the physics of reservoir streams. The calcula-
vided cube with permeability circulation, porosity, and esti- tions were implemented on the 3D matrix holding, normally,
mations of liquid saturation. This incorporation was likely the from 1 million to several billion cells. The calculations were
most tedious and instinctive piece of the entire geological somewhat long, even with the cutting-edge HPC servers and
displaying work process. With all the tuning working out, workstations restricting the number of potential runs. The in-
the cycle would require 2 or 3 months or several years. They crease in speed of reservoir modeling is one of the conspicu-
anticipated that, if appropriate preparation on numerous phys- ous bearings for AI innovations. Present-day proxy reservoir
ically directed extrapolation worked out, generative designs of models with another calculation motor dependent on profound
profound machine learning organizations could speed up the ML networks pack the numerical issue measurement and sur-
cycle by a factor of 1000+. mised the time subordinates guaranteed 100–1000 times the
Gasda and Celia (2005) performed upscaling to construct a regular models’ speedup while preserving comparative
reservoir model from a geological model. This lessened the usefulness.
measure of the 3D cells depicting the properties of the reser- Barker and Thibeau (1997) used the process of upscaling
voir by expanding the size of the cells from the geological which has a notable portion of art within. The subsequent
model. After the upscaling, they used reservoir modeling pro- chance in upscaling (i.e., putting forward the data acquired
gramming to prototype the reservoir streams at different field to a solitary geological and afterward hydrodynamical
Environ Sci Pollut Res

reservoir models from different calibrations of geophysical representative is not satisfactorily dressed for the specific ar-
examinations) was very effective. It acquainted a solid incli- rangement of tasks. Besides, prescient investigation cautioned
nation with the reservoir model. As there is no solitary right the administrators on the equipment’s robustness state,
and target technique for the process of upscaling, they consid- empowering proactive activities to forestall a fiasco with the
ered expanding objectiveness by summarizing the different results to wellbeing, security, and climate.
incidences with an AI smart tool. This was well possible with Qiao et al. (2017) presented a model projecting oil and gas
a profound learning algorithm of machine learning prepared production, which merged the least squares support vector
on numerous instances of manual upscaling. The result here machine (LS-SVM) and the particle swarm optimization
was not just the expanded goal yet additionally it sped up the (PSO) methods. Utilizing past information of oil and gas
upscaling process. organizations and MATLAB run simulations, the authors
Hegde and Gray 2017) tried to involve machine learning were able to investigate the factors affecting the production
to make history matching less biased and a lot quicker. The and construct an exact forecast model that was prepared at a
aim was to guarantee the greatest contact between the pro- rapid pace.
ductive piece of the development and the wellbore, and the Panja et al. (2018) created and compared an LS-SVM mod-
topmost pace of the entire process of construction of the el, a response surface model (RSM), and an artificial neural
well at negligible danger of failure. Modern drilling is an network (ANN) model to anticipate the oil creation and gas-
information-occupied operation. There were three kinds of to-oil proportion taking into account reservoirs that are ex-
sensors used. The first was the sensors on an exterior level posed to rate and time associated requirements. Some central
that recorded the mechanical variables of the process of factors were reviewed as input variables. The equivalent sets
drilling continuously. The second was the logging-while- of data were utilized to prepare the three models for anticipat-
drilling (LWD) sensors, taping down physical variables of ing the production under various frameworks. According to
the development at the rear of the drilling piece. The third the trial outcomes, the greatest precision was exhibited by the
was mechanics-while-drilling (MWD) sensors taping down LS-SVM model concerning the gas-to-oil proportion progno-
mechanical information from the assembly hole at the bot- sis. The authors deduced that the use of LS-SVM to various
tom. Every one of the sensors produced a time series that reservoir and production designing errands is favorable.
was utilized to deal with the process of drilling itself and Chaki et al. (2014) implemented the Modular ANN
refurbish the reservoir oilfields or geological model. The (MANN) way to deal with determining the part of reservoir
experiment resulted in a huge decrease of the non- sand. Three seismic properties, specifically, immediate ampli-
profitable time down to 20–50% on average with an exten- tude, seismic resistance, and frequency were utilized. The
sive decline of around 90% in failures with the growth and seismic and reservoir information for this examination was
execution of ML-helped drilling supportive networks oper- assembled from a hydrocarbon field in the western onshores
ating with ongoing drilling telemetry. of India. Three networks were intended for three different
Sneed (2017) tried to implement ML in producing reser- zones. The sand division and three seismic ascribes from sev-
voirs. Machine learning applications were applied to execute en wells were utilized to prepare and try out the ANNs. The
predictive maintenance on different pumps and select the ideal arrangement corresponding to the leftover all around was uti-
operational regimes relating to the production vs operational lized for blind prognosis. The proposed system gave a great
costs. A considerable lot of the pumps, including pumps for prognosis with an upgraded representation of the forecast sand
infusion wells, submersible pumps, hydraulic braking, and portion through post-processing and short program fruition
other pumps for well treatment, were outfitted with a huge time.
number of sensors estimating vibrations, temperatures, stream ML makes and uses calculations to make predictions on the
rates, pressure, and so on. Numerous models were run when information. The authors think the investigation stage, subject
an altogether driven on data or a physics-driven hybrid model to the understanding of layers of data filled with vulnerabil-
and information math-driven algorithm advanced the systems, ities, is an ideal fit for the ML approach. The quick identifi-
forestall sudden fiascos, and save money on upkeep on time. cation of examples working across different factors had been a
Koroteev and Tekic 2021) applied machine learning to time-concentrated process; this would now be able to be par-
apply safety in the OGI. Procedures on the oilfields can be tially computerized and improved with oversight from expe-
dangerous for staff as there are a few danger factors, including rienced experts. Reservoir modeling and attempting to discov-
weighty gear, instruments with non-covered rotary gear, high- er how development will respond to specific drilling methods
pressure factor, high-temperature tasks, and forceful can likewise be confirmed utilizing a mix of calculations and
chemicals. There were numerous IT frameworks dependent fuzzy logic, a strategy utilized for the forecast when data is
on ML aiding the security officials to spot safety concord either unreliable and incomplete. Looking back at past failures
savagery. Pattern identification using machine learning per- and applying the lessons learned is a crucial problem solving
mitted and video transfers taped with cameras to alert if a procedure. ML can accelerate this interaction and enlarge the
 Environ Sci Pollut Res

search net when looking for earlier cases of comparable issues By gathering information and using ML, the midstream
for a situation library. Fishing through a data set of logged oil and gas industry is improving operational productivity.
occasions for a depiction that coordinates with the current Prescient examination utilizing ML allows an exact sup-
issue, arrangements can be assessed to progressively give a port program that predicts when failures may happen and
fundamental manual for the workforce on the ground to use as give adequate opportunity for precaution activity.
an investigating tool. The authors think machine learning Midstream companies rapidly combine data from sensor
could save the O&G industry as much as $50 billion in the networks, operational sources, enterprise systems, and super-
coming decade. Since the cratering of the worldwide oil cost ficial providers to strengthen machine learning models that
in late 2014, organizations have progressively been taking a create predictive insights across compression, transportation,
gander at ML and other AI procedures to diminish costs, im- LNG, and other midstream operations. This addresses critical
prove proficiency, and limit downtime. issues such as reducing hydrocarbon losses, reducing down-
time, reducing the cost of transportation, and resulting in
$100s of millions in value annually.
Ahmed et al. (2017) presented a multi-specialist-based
Application of artificial intelligence methodology for forestalling and handling oil pipeline
in midstream wreckage. The authors utilized multi-agent system
(MAS) engineering based on the MAS algorithm to plan
One of the more unheralded aspects of machine learning in oil a framework including proactive and reactive agents
and gas involves simply getting the product from the field to cooperating to screen and turn away perils in a pipeline.
the refinery. As indicated by this MAS planning, the ML Agent distin-
guishes unusual pressure factors and alarms another agent,
Gathering called the Mobile Agent, with the subtleties and position
Transportation of the discovered anomaly. Simultaneously, any metallic
Logistics item in the pipeline is detected by the sensor agent and
Pipelines cautions the Master Agent, which constitutes the control
center, through the Mobile Agent. The Protective Agent
Again, because your algorithm can crunch numbers so gave direct insurance to the pipeline at whatever point the
quickly, it can provide specific recommendations for improv- existence of any metallic object is found before the entry
ing the efficiency of your delivery systems (Table 2). of safety officers. The presented submission has not been

Table 2 A comprehensive study on artificial intelligence in petroleum midstream

Optical technique Application Method Reference

1. Deep learning, Decision-making in gas networks Modeling and solving problems from discrete Anderson et al.,
MILP, ML optimization by integer variables 2021
2. Cloud Reduce seismic data processing time Streaming data into the Amazon Web Services cloud Barbaschow, 2020
computing, for analysis from few thousand sensors and then
ML adding all the 1 million sensors
3. ML Automated time series outlier detection Supports easy pipeline construction through implementation Lai et al., 2021
system of a function with hyperparameters
4. Auto ML Pipeline search Matrix and tensor factorization used as surrogate models Yang et al., 2020
to model combinatorial pipeline search space
5. DNN, ML Pipeline parallelism Derived a robust and simple training method called PipeMare Yang et al., 2020
that authorizes asynchronous updates during pipeline
parallel execution
6. Robots with ML Inspect and repair active oil and gas Robots used to inspect and repair pipelines and mitigate Mitchell, 2019
pipelines deaths, injuries, and damages worth billions
7. AI, ML Maximize recoveries and profits Mitigate unplanned equipment outages and resulting loss Blackmon, 2019
of production
8. AI, ML To reveal global oil storage Examine satellite images and recognize and quantify crude Myhr, 2017
oil storage tanks
9. ANN, ML Identify flow through pipelines Applying statistical analysis in predicting flow within each Aminian et al.,
region 2003
10. ML Identify flow regimes Applying RF algorithm giving the highest F-score of 90.8% Alhashem, 2019
Environ Sci Pollut Res

carried out; however, the authors are convinced that the GOSP communicated in stages with the assistance of agents,
cooperation communicated by the agents is a decent indi- whose doled out ventures incorporate symptom discovery,
cation that the presented system is appropriate for holding failure identification, diagnosis, and repair pairing. Two as-
up the instance of the defacement in pipelines sessment measurements, specifically, the achievement pace of
Ramírez et al. 2009) presented an insightful administrative repair plan and adaptation to internal failure network pointer,
ML-optimized framework with numerous input controls for were utilized to gauge this execution. The authors planned to
mechanical operations. The authors consolidated internal con- implement the agents as a component of a work of the future.
trol model engineering, some fuzzy logic rules, and a MAS to They additionally featured the downsides of utilizing the wire-
achieve two O&G industrial activities: gas–oil partition and less network, the result of which can influence the exactness
oil heating. Local standard interdependent directing plans of information, along these lines making continuous control
were connected to shape a general chief to meet local objec- troublesome or unimaginable.
tives in every one of the cycles. The plan of the local directing In light of setup research toward an attainability investiga-
specialist comprised two levels: the administrative level with tion of a wise resource, the executive’s framework for the
an internal control model design and administrative level. The seaward OGI in Atlantic Canada (Sayda and Taylor 2008)
latter includes an event detector created utilizing fuzzy logic presented an ML and MAS called ICAM to command and
and a choice framework that utilized data gotten from the oversee resources in the OGI. To evaluate and approve the
event detector to settle on fitting choices. The global directing exhibition of the framework, the creators directed a real-time
specialist was planned dependent on a pre-established system simulation experiment on the ICAM framework model, which
to assure the achievement of the worldwide target of the whole demonstrated constructive enhancements in the plan and ad-
process. Computational simulations for the competition of the vancement of the MAS. Plant information was accumulated
local and worldwide overseeing agents were done, and the and inspected for each reactive agent during simulations, to
outcomes showed a chance of arriving at the whole investigate the conduct of reactive agents. The result from the
interaction worldwide target attributable to the participation simulation manifested that the ICAM framework model con-
between local administrative agents. ducts presented to its plan conditions. To investigate the
Oliveira et al. (2013) presented the Simple Gateway ICAM execution, the creators directed a constant simulation
Control Protocol (SGCP). The SGCP was utilized to recreate experiment dependent on the framework’s logical conduct
an oil field, situated in Bahia, Brazil. The oil field segments and its reaction to the elements of the external climate.
were observed by agents through association with an informa- Amazing logical conduct was noticed; the framework
tion base carrying genuine information from the field. The responded to basic malfunctioned actuators and sensors. The
accumulated information included readings from the sensors overall ICAM execution analysis result was acceptable.
for each field and a capable understanding of costs. The agents Sinha et al. (2009) put forward a mix of MAS, ML, and co-
implicated the data to assess the current situation with each oil developmental PSO dependent on the Cauchy circulation to
field part and the whole field. They all took measures by their deal with petroleum supply chain operations. In the frame-
allocated tasks, for example, operation, supply, and interven- work, supply agents were led by supporting agents who orga-
tion. Issues were recognized through a calculus approach by a nized themselves to optimize the predetermined subsystem
statistical method, which comprised examining the degree of cost, along these lines lessening the whole expense of the
oil creation. The creators coordinated the SGCP into an instru- inventory network. A statistical formula was utilized to ad-
ment that had earlier comprehension of the conditions of the vance the asset distribution to every agent. The simulation
oil field and helped dynamic in a clever manner to make result proved that putting in the MAS supported the upkeep
deductions dependent on rules and fuzzy logic. of ideal crude petroleum amounts in the inventories of ob-
The result demonstrated that utilizing ML in the oil field served refineries. The creators saw that the approach of
activity urged the agents to act collectively in dissecting is- Cauchy distribution genuinely affected the MAS asset re-
sues, subsequently chopping down the superfluous cost of quirement optimization.
delving into crude information in different arrangements in Julka et al., 2002a, b) presented a petroleum refinery incor-
the hunt for solutions. porated supply chain modeler and simulator (PRISMS) to help
A MAS for the network management system (NM-MAS) with offering choice help during acquisition activities with
containing agents with various capacities with finding a solu- crude oil. To construct the PRISMS for the whole network
tion to settle failover of network was presented by Ioniţə and of the supply chain, the authors utilized a brought together
Ioniţə 2014). For the trial, a gas–oil partition plant (GOSP) agent-based demonstrating system in an object aimed ap-
was plotted to some carried out industrial wireless organiza- proach. As indicated by this structure, data and assets are
tion segments. The GOSP administrator utilized significant displayed as objects, entities of the supply chain as agents,
level fulfilment data of organization components to distin- and the inner divisions of the system as sub-agents. Data flow
guish damages that might happen. NM-MAS, ML, and is addressed by the exchange of items among the agents.
 Environ Sci Pollut Res

Using the agent-based methodology, the PRISMS simulated (UAVs) to take readings at sensitive pipeline utilizing thermal,
certain refinery departments like acquisition, logistics, sale, infrared, light location and ranging, and multispectral sensors
and storage activities. During the setup, specific activity in in real time. Although new pipelines are frequently pre-
the refinery was replicated for a specific number of days. outfitted with such advances, there is a critical chance to re-
Aissani et al., 2009) presented a multi-agent model and ML design the heritage framework and even sum up sensor
for the dynamic arrangement of maintenance tasks for the arrangements all through the pipeline network instead of
setup of oil industry production operations. As a case study, detecting just key intersections.
the authors utilized the ArzewRA1Z refinery in Algeria. Nozomi Networks (2017) developed SCADA solutions to
Considering the enterprising idea of the oil treatment facility, be better at spotting awful bad data, flagging deviations, and
the vulnerability, and the unanticipated complexion of the giving a superior establishment for production reporting. He
climate, the measured Markov decision process (MDP) utilized monikers, for example, intelligent SCADA and ML,
build-up was utilized in the framework. The authors demon- to denote the ability of these frameworks to track more data
strated numerous agents coordinated in a ranking fit from sources and spot deviations. He utilized rules and heuristic
experience learning utilizing the state-action-reward-state- methods to spot deviations affecting the precise comprehen-
action (SARSA) algorithm to guarantee effective maintenance sion of stream, production, and gas quality. Measurement data
arranging and continued advancement of the arrangement takes advantage of information from EFMs, storing the infor-
standard. The SARSA algorithm permits assessment by con- mation for examination and production reporting. Through the
sidering modifications during learning in the feedstream as a ability to set thresholds on deviations, estimation software
performance marker. The Contract Net Protocol was utilized helped spot issues like missing information, suspect data, or
for the agent interchange. Experimental results demonstrated uncollected information.
that the technique can give web-based scheduling answers for ENBRIDGE BLOG, 2018) worked with a leading tech or-
prescient and remedial maintenance undertakings and im- ganization and Finger Food Studios to initially incorporate 132
prove their tendency by limiting out-stream modifications discrete arrangements of data of pipeline (gathered by in-line
from the tanks. inspection apparatuses, sensors which can detect strain, LiDAR
Eze et al. (2017) put forward an integrated O&G monitor- remote sensors, ML algorithms, and so forth), alongside land-
ing of pipeline system and incident mitigation system scape data, to make a 3D rendering of the network of the pipe-
(IOPMIMS), which offered proactive security to the pipelines line by handling the tremendous amount of information in real
utilizing the circulated frameworks and case-based thinking time. This permitted him to forecast and analyze the zones
approach. To accumulate signals from all sensors, the authors where potential dangers, including dents, cracks, corrosion,
presented the architecture of wireless network sensor nodes. A and strain, may happen, just as tracking the boundaries (opera-
MAS implanted into Antronix Fiber Nodes (AFN) wires the tional and geological) answerable for dangers to integrity.
information dependent on the signs at the designated base In the author’s opinion, machine learning can be used ex-
station and sub-station to affirm if certain circumstances create clusively in the oil and gas industries (midstream) and serve
any danger to a pipeline. The MAS groups data from diverse the following functions:
wireless devices, at the base station which works with choice •Forecast long haul item information and item market cost
designing by the framework. Caution is being sent to the se- •Provide capital arranging and hazard assessment for better
curity workforce or engineers of the pipeline if there is a dan- long haul choices
ger or leak identification found. Information from the •Optimize item exchanging and supporting techniques
Concawe group’s website was composed and replicated to •Improve unwavering quality danger displaying for refin-
examine the exhibition of the suggested system. Three AI ing and handling resources
calculations (decision tree, SVM, and ANN) were utilized •Maximize work profitability and wrench time
for decision support and blending of data. Out of the three •Enhance resource scheduling for refining and preparing
calculations, SVM accomplished the finest execution. activities
ENBRIDGE BLOG, 2018) applied the buildout of sensors •Optimize pipeline booking for item streams
and correspondence networks for trunk lines, an activity
where computational observing using pressure, volume, and
temperature (PVT) examination, regulator checking through Application of artificial intelligence
supervisory control and data acquisition (SCADA) frame- in downstream
works, and booked line balance estimations through ML are
a customary issue. Of late, this fragment has quickly advanced There are numerous progressions and developments that oc-
to a high-level sensitizing stage in the DOT model as organi- curred in the OGI downstream up until now. Advancement in
zations are attempting to improve their reaction methodology innovation causes scaling down the costs in activities, expan-
during an occurrence. He conveyed unmanned aerial vehicles sion in nature of yield item, and decrease in adverse
Environ Sci Pollut Res

consequence of climate (Patel et al. 2020). Elatab (2016) in- Khvostichenko and Makarychev-Mikhailov (2018) build
vestigated that digitalizing oil fields and implementing ML up a plan to examine the impacts of the complete framework
techniques can reduce the expense in activities up to 10–25. on the efficiency of wells. They assembled the information
Mills (2013) has explored that the expense of shale oil pro- from 4500 wells that were under slickwater treatment. They
duction can be cut down somewhere in the range of $7 to $15 examined the impacts of two distinct chemicals, i.e.,
per barrel when using the new fracturing technology that uses surfactant-based flowback aids and linear gual gels. They ad-
hydraulic engineering. The effectiveness and profitability of ditionally assembled the month-to-month data of production
the shale fields of the USA expanded 200–300%. from the Energy database of IHS. The measurable methodol-
Padmanabhan (2014) have investigated that progression in ogy used to analyze the information was a t test.
ML examination assists with upgrading the oil production Anagnostopoulos, 2018) led an examination to apply ML
by 6–8% (Table 3). to enhance the execution of shipping. In his examination, he
Plate and Ag. (2016) utilized ML in refining. For this case intended to foresee the ability of propulsion to improvise the
study, a three-step strategy was designed to dissect and handle performances of boats and therefore to bring down the ozone-
the historical data to enhance petrochemical resource manage- depleting greenhouse substance outflows. The information as-
ment. For this operation, the gear of heed was a cracked gas sembled for this examination was gathered over time of a
four-stage compressor (CGC). The analytic examination be- quarter of a year through the sensors all through a Large Car
gan by first prognosticating the execution of CGC by exam- Truck Carrier (LCTC) M/V. In the subsequent stage, they
ining the historical and current working data. In the following utilized multi-layer perceptron (MLP) and extreme gradient
stage, given the gadget’s end-of-life measures and condition boosting neural organizations to lead the data analysis.
failure, the forecast of the CGC performance was additionally Park et al. (2018) utilized ML to build up a model with
tuned. Finally, the assessed execution of the CGC was intro- energy efficiency dependent on the data operation assembled
duced in an easy-to-use and visual report to be utilized for the during ship operations. In their investigation, an energy point-
board choices. These prescient reports which are created by er called energy efficiency operational indicator (EEOI) was
utilizing analysis of past data can altogether lessen the main- assessed dependent on openly accessible programmed recog-
tenance costs and significantly improve the downtime. nizable proof framework marine and data climate information.
In a recent undertaking by Brelsford (2018), ML was used The energy effectiveness was characterized as the fuel for ship
to conduct optimization of management for one of the orga- utilization by motor force vs the weight of the operation and
nization’s coordinated treatment facilities in Spain. For this distance. For executing ML, creators utilized the Hadoop
undertaking, Repsol is provided with products of data analyt- system and Apache Spark for ML errands.
ics from Google cloud and discourse just as machine learning Tarrahi and Shadravan (2016) utilized machine learning to
is given by google cloud administrations. improvise the O&G occupational security by dealing with the

Table 3 A comprehensive study on artificial intelligence in petroleum downstream

Optical technique Application Method Reference

1. AI, ML Model crude procurement Simulation-based decision support assisting in Julka et al., 2002a, b, c
process evaluation of refinery policies
in refinery
2. AI, ML Refinery planning and Proposition to supply chain optimization and Pinto et al., 2000
scheduling modeling analysis of supply chains capturing
interactions among enterprises
3. AI, ML, PRISMS Supply chain management KPIs utilized to analyze the effect of market fluctuations, Julka et al., 2002
procurement policy packet size, and impact of
configuration change on supply chain performance
4. AI, ML Improving transactional nature Specific agents embedded with detailed solution Reddy et al., 2004
of supply chain techniques like tank scheduling using LPs and crude
distribution to distillation units
5. AI, ML, nonlinear model Process transition monitoring Extended Kalman filters and open loop observers used Bhagwat et al., 2003a
to achieve estimation of process states and residuals
6. AI, ML, multilinear Fault detection For the purpose of residual generation and state estimation Bhagwat et al., 2003b
model based on linear models, multiple local linear regimes
with
Kalman filters and open loop observers were utilized.
7. Dynamic PCA, AI Using historical operating data Process data divided into modes and transitions Srinivasan et al., 2004
to classify process states
 Environ Sci Pollut Res

danger and upgrading the safety. The experiment was contin- Plate and Ag (2018) used a lot of conditions and data pro-
ued dependent on a case by the labor statistics bureau which cesses which accounts to more readily gain profits with RUL
included 846 sources of injuries from 1278 enterprises some- equipment, with the goal that client chooses Cassantec prog-
where in the range of 2010 to 2015. The initial phase in their nostic solutions and implements them on a cracked gas four-
examination was a grouping of information and handling. For stage compressor. The fundamental advantages of utilizing
this reason, they filtered the raw data dependent on the nature prognostic arrangements were a decrease in the cost of down-
of recordings and they disposed of the anomalies from sets of time and a decrease in the overall maintenance cost of the
data by relative standard error estimation. At that point, they gear.
built up the organized information by design change and
decoding the data. In the following stage, the authors led the
clustering of data and planning to recognize the underlying Challenges
hidden patterns. Toward the end, to introduce an effectively
justifiable result, they utilized multi-dimensional statistical While some oil and gas organizations, for example, Saudi
analysis. Aramco, Gazprom Neft, BP, and Shell, are kicking off their
Pettinger (2014) gathered information from safety assess- AI drives by aggressively investing in new companies and
ments that can be utilized to create analytics of safety predic- R&D, a few difficulties are forestalling them to greatly and
tion through machine learning. It is pivotal to assemble the quickly carry out AI in the investigation and creation of oil
indication safety information inside the organization cease- and gas. That is not an oil and gas explicit issue, yet an ordi-
lessly and fuse them in the predictive investigation. The secu- nary one in applying AI at this phase of its expansion (Duan
rity markers which will give the necessary information incor- et al., 2019).
porate evaluating behaviors and surveying. The accomplishment of computerized reasoning relies up-
Cadei et al., 2019) utilized machine learning to create pre- on human insight. Man-made intelligence arrangements are
diction software to figure risk occasions and failures in the not nonexclusive—they cannot be simply purchased. In any
operations during production tasks in the OGI. The marker that event, when created by third parties (and disposed of for free
they utilized as a risk occasion for prognostication was the of cost, similar to TensorFlow by Google), AI arrangements
concentration of H2S. They accumulated information from dif- must be tweaked to the business setting and information an
ferent sources like real-time series, past historical data, accounts organization has (Ng, 2016).
of maintenance, administrator data, and analytic study of chem- Subsequently, to effectively utilize AI in products and pro-
istry. The work process of their investigation incorporates in- cesses, organizations should form in-house groups made out
formation collection, definitive problem solving, information of information and AI subject matter experts. These groups
handling, modeling (utilizing random forest, artificial neural ought to have the option to help the improvement of AI frame-
network (ANN)), and lastly validation of the model. work (calculations and datasets) and, in any event, to redo
Azadeh et al. (2008) reviewed and researched an ML mod- instruments that organizations will later use in their tasks.
el for energy examination in a refinery called the data devel- Indeed, that implies that oil and gas organizations will turn
opmental analysis. DEA strategy is utilized to measure the out to be (mostly) information-driven organizations and that
effectiveness in a given informational collection of decision- AI experts will get indispensable in supporting practically all
making units (DMUs). For the refinery, they utilized input- development endeavors in oil and gas organizations in the
situated DEA models since they needed to assess the efficacy following 10 years. Be that as it may, finding and holding
of refineries under a rigid construction. The DEAs input- AI ability are difficult undertakings. There is a huge deficien-
oriented model gives productivity refinery scores that assist cy of AI ability in the work market (Global AI Talent Report
us with assessing how proficient we can utilize fuels and elec- 2019, jfgagne).
tricity to manufacture products of the refinery. In DEA, they Artificial intelligence instruments require great quality in-
utilized Anderson and Peterson model. AP model is consider- formation of a reasonable amount to be prepared and after-
ably more productive if the effective decision-making unit ward to perform appropriately in the functioning mode. While
score is equivalent to or greater than 1. utilizing more clever calculations that might help in improving
Tokarek et al. (2018) utilized one of the ML models called outcomes from datasets of restricted size, no control can assist
PCA investigation to measure the contamination of air of oil with bad information (Ransbotham et al., 2017).
sands of Athabasca. To elucidate 28 variables including or- Hence, admittance to huge and quality information is an
ganic volatile compounds (VOCs) and intermediate volatile essential empowering agent and boundary for AI applications’
organic compounds (IVOCs), they utilized varimax rotation effective turn of events. Oil and gas fields produce a lot of
alongside PCA in this examination, although ten of them were crude information. All things considered, it is anything but
mottled and characterized dependent on sort of source insurance for progress as there are familiar issues with the
difference. quality and exactness of field information and the general
Environ Sci Pollut Res

absence of huge volumes of marked information in the OGI country). Some AI apparatuses will be acknowledged as valu-
(Hajizadeh, 2019). able guides, and the concentration inside AI improvements
To upgrade the worth of information oil and gas organiza- would be moved toward grey box crossover models, where
tions have or can get, they should update and change their the physically driven science part will make up for the short-
authoritative designs and activities. Oil and gas organizations fall of admittance to a decent lot of the genuine field informa-
are not known for their coordinated, lean, and base up im- tion. The authors anticipate this as the most sensible case, with
provement attitude; however, for exacting locale division a general effect on the upstream, midstream, and downstream
and cascade cycles and strategies, that must be altered. sectors (Nishant et al., 2020).
Additionally, information stockpiling ought to be incorporat- The capacity to address everything and proficiently explore
ed into one or few information stockrooms to permit individ- different avenues regarding information permits AI-conceived
uals and AI programming simple access and usage (Ng, new companies to stand out and record speculations. In 2019,
2016). AI-affiliated organizations in the USA raised $18.5 billion,
Artificial knowledge is brought into the world in an open right around 2 billion more than in 2018 (O’Brien, 2020).
and cooperative climate as an outcome of the scholarly com- The biggest OGIs are dynamic in obtaining AI new com-
munity being the main power in AI research for quite a long panies. For instance, Statoil and GE mutually put resources
time, nearly with no business impacts. This made the free into Ambient (General Electric n.d.).
sharing culture (for example GitHub) and open distributing Saudi Aramco puts resources into a startup named Earth
(for example, arXiv), which organizations across ventures Science Analytics, building up the up and coming age of oil
(and globally) needed to accept as a norm to prevail in the geoscience AI software (Saudi Aramco energy ventures in-
period of AI once they appended the race. vests AI company earth science).
While open advancement is getting common in the tech BP puts resources into a startup named Belmont
area, oil and gas organizations are not well known for their Technology expecting to support the organization’s AI and
projects in joint industries, particularly among contenders and computerized abilities in its offshore upstream business
particularly not in essential spaces like AI (Hajizadeh, 2019). (Umar, 2019).
Chevron, Saudi Aramco, and Shell put together in
“Maana” an AI startup, which collaborated with Microsoft
Future scope to utilize its distributed computing stage azure (Luck and
Chronicle, 2019).
The positive situation depends on a universally proliferated
arrangement that cross-organization and cross-line informa-
tion sharing are essential. Accepting solid and submitted ini-
tiative in the key organizations and that great information Conclusion
stages are set up, one may expect quick development of AI
capacities for the upstream, midstream, and downstream The authors examined the advancement of practical instru-
implementations followed by AI instruments for dynamic at ments dependent on man-made brainpower for oil and gas
different levels. Here, the authors can fearlessly predict tre- upstream, midstream, and downstream. Unmistakably even
mendous prospects in the development of performability of though man-made consciousness is an arising pattern in
the OGI businesses because of extremely critical investment OGI, some implementations have effectively brought enu-
funds on costs and adapted misfortunes because of non-ideal merable worth. They have given a few instances of how
choices coming to half of the current levels. There is likewise man-made brainpower assists speed with increasing and
a major segment of the natural impact at hand here. de-hazard numerous business measures related to the in-
If AI infiltrates the business along these lines, numerous vestigation of production assets, well logging, drilling,
chances limit the negative impression of the upstream, mid- pipelines, transportation, and production of fuels in the
stream, and downstream equipment innovations, for instance, refinery. There is an ongoing analysis of the versatility of
one can utilize a legitimate AI model to limit dangerous parts man-made reasoning across the entire business. Here, they
at occupations using well treatment processes or re-use the have talked about the specialized drivers of such versatility
created water such that retains the recuperation factor on a as well as the non-specialized factors. They assessed the
fitting level. impact of schooling, organizational demeanor, and infor-
The sensible (impartial) situation is the point at which the mation accessibility on the pace and bearing of computer-
IT stages are set up, yet the advancement of information shar- ized reasoning entrance to OGI. Basing on this examina-
ing arrangements is restricted. Restricted implies that there are tion, they inferred potential situations on how man-made
some extra sharing changes regarding what we have now (like brainpower could proliferate inside the OGI in the coming
dividing among gatherings of the organizations inside a 10 to 20 years.
 Environ Sci Pollut Res

Abbreviations AI, artificial intelligence; OGI, oil and gas industry; ML, Azadeh A, Ghaderi SF, Asadzadeh SM (2008) Energy Efficiency model-
machine learning; IoT, Internet of things; O&G, oil and gas; HPC, high- ing and estimation in petroleum refining industry-a comparison
performance computing; ANN, artificial neural network; MANN, modular using physical data. 10.24084/repqj06.242
artificial network; LNG, liquefied natural gas; MAS, multi-agent system; Baraboshkin EE, Ismailova LS, Orlov DM, Zhukovskaya EA, Kalmykov
SGCP, simple gateway control protocol; GOSP, gas–oil separation plant; GA, Khotylev OV, Baraboshkin EY, Koroteev DA (2019) Deep
PSO, particle swarm optimization; PRISMS, petroleum refinery incorpo- convolutions for in-depth automated rock typing. In arXiv (Vol.
rated supply chain modeler and simulator; SVM, support vector machine; 135, p. 104330). arXiv. https://doi.org/10.1016/j.cageo.2019.
SCADA, supervisory control and data acquisition; DOT, designated order 104330
turn around; EFM, enterprise feedback management; LIDAR, light detec- Barbaschow Asha. (2020) Woodside uses cloud to reduce seismic data
tion and ranging; DEA, data envelopment analysis; PCA, principal com- processing time down to 90 minutes | ZDNet. https://www.zdnet.
ponent analysis; RUL, remaining useful life; BP, British petroleum; com/article/woodside-uses-cloud-to-reduce-seismic-data-
GRNN, general regression neural network; BPNN, backpropagation neu- processing-time-down-to-90-minutes/#ftag=RSSbaffb68
ral network; DRIFT, diffuse reflectance infrared Fourier transform; ROP, Barker JW, Thibeau S (1997) Permeabilities for Upscaling. SPE
rate of penetration; MILP, mixed-integer linear programming; DNN, deep Reservoir Engineering (Society of Petroleum). A critical review of
neural network; RF, random forest; KPI, key performance indicator; LP, the use of pseudorelative engineers 12(2):138–143. https://doi.org/
linear programming; NM-MAS, network management multi-agent system 10.2118/35491-PA
Acknowledgements The authors are grateful to the Department of Bello O, Teodoriu C, Yaqoob T, Oppelt J, Holzmann J, Obiwanne A
Chemical Engineering, Manipal Institute of Technology, and (2016) Application of artificial intelligence techniques in drilling
Department of Chemical Engineering, School of Technology, Pandit system design and operations: a state of the art review and future
Deendayal Energy University, for the permission to publish this research. research pathways. Society of Petroleum Engineers - SPE Nigeria
Annual International Conference and Exhibition. https://doi.org/10.
Author contribution All the authors make a substantial contribution to 2118/184320-ms
this manuscript. DG and MS participated in drafting the manuscript. DG Bhagwat A, Srinivasan R, Krishnaswamy PR (2003a) Fault detection
wrote the main manuscript. All the authors discussed the results and during process transitions: a model-based approach. Chem Eng Sci
implication on the manuscript at all stages. 58(2):309–325. https://doi.org/10.1016/S0009-2509(02)00520-1
Bhagwat A, Srinivasan R, Krishnaswamy PR (2003b) Multi-linear mod-
el-based fault detection during process transitions. Chem Eng Sci
Data and materials availability All relevant data and material are present-
58(9):1649–1670. https://doi.org/10.1016/S0009-2509(03)00008-3
ed in the main paper.
Blackmon David. (2019) Technology is a huge driver of the U.S. oil and
gas boom. Retrieved May 22, 2021, from https://www.forbes.com/
Declarations sites/davidblackmon/2019/03/25/technology-is-a-huge-driver-of-
the-u-s-oil-and-gas-boom/?sh=6a4cdc2a5ac5
Ethics approval and consent to participate Not applicable. Brelsford: Repsol launches big data AI project at... - Google Scholar.
(n.d.) Retrieved May 22, 2021, from https://scholar.google.com/
scholar_lookup?title=Repsol launches Big data%2C AI project at
Consent for publication Not applicable.
tarragona refinery&publication_year=2018&author=R. Brelsford#
d=gs_qabs&u=%23p%3DkNNp5HXmqQIJ
Competing interests The authors declare no competing interests. Brelsford (2018) Repsol launches Big Data, AI project at Tarragona re-
finery. https://www.ogj.com/refining-processing/refining/
operations/article/17296578/repsol-launches-big-data-ai-project-at-
References tarragona-refinery
Burtescu E, Cooper J, Technology Ltd H, Marian Dardala U,
Dusmanescu D, Fotache M, Garlasu D, Romania Marius Guran O,
Ahmed M, Bhattarai R, Hossain SJ, Abdelrazek S, Kamalasadan S (2017) Ticiano Costa Jordão L, Kahanwal B, Konstantas D, Kumar Sharma
Coordinated voltage control strategy for voltage regulators and volt- H, Nithchi S, Paraschiv C, Popescu D, Gheorghe Sabau I, Wrembel
age source converters integrated distribution system. 2017 IEEE R (2014) Database Systems Journal BOARD. In Database Systems
Industry Applications Society Annual Meeting, IAS 2017, 2017- Journal (Issue 1)
January, 1–8. https://doi.org/10.1109/IAS.2017.8101780 Buterin V (2014) A next generation smart contract & decentralized ap-
AI transformation playbook - landing AI. (n.d.). Retrieved May 22, 2021, plication platform
from https://landing.ai/ai-transformation-playbook/ Cadei L, Montini M, Landi F, Porcelli F, Michetti V, Origgi M, Tonegutti
Aissani N, Beldjilali B, Trentesaux D (2009) Dynamic scheduling of M, Duranton S (2019) Big data advanced analytics to forecast op-
maintenance tasks in the petroleum industry: a reinforcement ap- erational upsets in upstream production system. Society of
proach. Eng Appl Artif Intell 22(7):1089–1103. https://doi.org/10. Petroleum Engineers - Abu Dhabi International Petroleum
1016/j.engappai.2009.01.014 Exhibition and Conference 2018. ADIPEC 2018. https://doi.org/
Alhashem M (2019) Supervised machine learning in predicting multi- 10.2118/193190-ms
phase flow regimes in horizontal pipes. Society of Petroleum Chaki S, Verma AK, Routray A, Mohanty WK, Jenamani M (2014) Well
Engineers - Abu Dhabi International Petroleum Exhibition and tops guided prediction of reservoir properties using modular neural
Conference 2019. ADIP 2019. https://doi.org/10.2118/197545-ms network concept: a case study from western onshore, India. J Pet Sci
Aminian K, Ameri S, Oyerokun A, Thomas B (2003) Prediction of flow Eng 123:155–163. https://doi.org/10.1016/j.petrol.2014.06.019
units and permeability usina artificial neural networks. SPE Western Chen L-P (2019) Mehryar Mohri, Afshin Rostamizadeh, and Ameet
Regional/AAPG Pacific Section Joint Meeting:299–305. https://doi. Talwalkar: Foundations of machine learning, second edition. Stat
org/10.2118/83586-ms Pap 60(5):1793–1795. https://doi.org/10.1007/s00362-019-01124-9
Anagnostopoulos A (2018) Big data techniques for ship performance de Oliveira VLC, Tanajura APM, Lepikson HA (2013) A Multi-agent
study. OnePetro System for Oil Field Management. 11th IFAC Workshop on
Anderson L, Turner M, Koch T (2021) Generative deep learning for Intelligent Manufacturing Systems. The International Federation of
decision making in gas networks. http://arxiv.org/abs/2102.02125 Automatic Control. 1-6
Environ Sci Pollut Res

Dietterich TG (1997) Machine-Learning Research. In AI Magazine (Vol. Koroteev D, Tekic Z (2021) Artificial intelligence in oil and gas up-
18, Issue 4). https://doi.org/10.1609/AIMAG.V18I4.1324 stream: trends, challenges, and scenarios for the future. Energy and
Duan Y, Edwards JS, Dwivedi YK (2019) Artificial intelligence for de- AI 3:100041. https://doi.org/10.1016/j.egyai.2020.100041
cision making in the era of Big Data – evolution, challenges and Lai K-H, Zha D, Wang G, Xu J, Zhao Y, Kumar D, Chen Y,
research agenda. Int J Inf Manag 48:63–71. https://doi.org/10.1016/ Zumkhawaka P, Wan M, Martinez D, Hu X (2021) TODS: an au-
j.ijinfomgt.2019.01.021 tomated time series outlier detection system. https://github.com/
Elatab M (2016) 5 Trends in oil & gas technology, and why you should datamllab/tods. Accessed 7 Jul 2021
care. https://venturebeat.com/2012/03/28/5-trends-in-oil-gas- Li H, Yu H, Cao N, Tian H, Cheng S (2020) Applications of artificial
technologyand-why-you-should-care/. Accessed 18 Nov 2019 intelligence in oil and gas development. Archives of Computational
ENBRIDGE BLOG. (2018) Virtual reality project goes underground for Methods in Engineering 28(3):937–949. https://doi.org/10.1007/
pipeline safety - Enbridge Inc. https://www.enbridge.com/stories/ s11831-020-09402-8
2018/march/virtual-reality-3d-holographic-pipeline-digital-twin. Lu H, Huang K, Azimi M, Guo L (2019) Blockchain technology in the oil
Accessed 5 Jul 2021 and gas industry: a review of applications, opportunities, challenges,
Eze J, Nwagboso C, Georgakis P (2017) Framework for integrated oil and risks. IEEE Access 7:41426–41444. https://doi.org/10.1109/
pipeline monitoring and incident mitigation systems. Robot Comput ACCESS.2019.2907695
Integr Manuf 47:44–52. https://doi.org/10.1016/j.rcim.2016.12.007 Luck Marissa, Chronicle Houston (2019) Backed by Big Oil, tech startup
Feblowitz J (2013) Analytics in oil and gas: the big deal about big data. launches new partnership with Microsoft. https://www.chron.com/
Society of Petroleum Engineers - SPE Digital Energy Conference business/energy/article/Backed-by-Big-Oil-A-I-startup-launches-
and Exhibition 2013:286–291. https://doi.org/10.2118/163717-ms new-13682135.php. Accessed 11 Jun 2021
Fletcher Philip, Peter Coveney, TL Hughes, CM Methven (1995) Brief: Mao B, Han LG, Feng Q, Yin YC (2019) Subsurface velocity inversion
Predicting quality and performance of oilfield cements with artificial from deep learning-based data assimilation. J Appl Geophys 167:
neural networks and FTIR spectroscopy | Request PDF. https:// 172–179. https://doi.org/10.1016/j.jappgeo.2019.04.002
www.researchgate.net/publication/255020930_Brief_Predicting_ Mills MP (2013) Big data and microseismic imaging will accelerate the
quality_and_performance_of_oilfield_cements_with_artificial_ smart drilling oil and gas revolution. https://venturebeat.com/2012/
neural_networks_and_FTIR_spectroscopy. Accessed 7 Apr 2021 03/28/5-trends-in-oil-gas-technologyand-why-you-should-care/.
Gasda SE, Celia MA (2005) Upscaling relative permeabilities in a struc- Accessed 18 Nov 2019
tured porous medium. Adv Water Resour 28(5):493–506. https:// Mitchell Oliver (2019) Robots can play key roles in repairing our infra-
doi.org/10.1016/j.advwatres.2004.11.009 structure | Robohub. 2019. Retrieved May 22, 2021, from https://
Gidh Y, Purwanto A, Ibrahim H (2012) Artificial neural network drilling robohub.org/robots-can-play-key-roles-in-repairing-our-
parameter optimization system improves ROP by predicting/ infrastructure/Overview of blockchain for energy and commodity
managing bit wear. Society of Petroleum Engineers - SPE trading
Intelligent Energy International 2012(1):195–207. https://doi.org/ Mohammadpoor M, Torabi F (2020) Big Data analytics in oil and gas
10.2118/149801-ms industry: an emerging trend. In Petroleum (Vol. 6, Issue 4, pp. 321–
Hajizadeh Y (2019) Machine learning in oil and gas; a SWOT analysis 328). KeAi Communications Co. https://doi.org/10.1016/j.petlm.
approach. J Pet Sci Eng 176:661–663. https://doi.org/10.1016/j. 2018.11.001
petrol.2019.01.113 Myhr Gunnar. (2017). Artificial intelligence (ai) to reveal global oil stor-
Hanga KM, Kovalchuk Y (2019) Machine learning and multi-agent sys- age | LinkedIn. https://www.linkedin.com/pulse/artificial-
tems in oil and gas industry applications: a survey. Computer intelligence-ai-reveal-global-oil-storage-gunnar-myhr/. Accessed
Science Review 34:100191. https://doi.org/10.1016/j.cosrev.2019. 10 Jun 2021
08.002 Ng A (2016) What artificial intelligence can and can’t do right
Hegde C, Gray KE (2017) Use of machine learning and data analytics to now. https://hbr.org/2016/11/what-artificial-intelligence-can-and-
increase drilling efficiency for nearby wells. Journal of Natural Gas cant-. 20. Accessee 7 Jul 2021
Science and Engineering 40:327–335. https://doi.org/10.1016/j. Nishant R, Kennedy M, Corbett J (2020) Artificial intelligence for sus-
jngse.2017.02.019 tainability: challenges, opportunities, and a research agenda. Int J Inf
How do you monitor your crude oil pipeline system? - Enbridge Inc. Manag 53:102104. https://doi.org/10.1016/j.ijinfomgt.2020.102104
(n.d.) Retrieved May 22, 2021, from https://www.enbridge.com/ Nozomi Networks (2017) Improving ICS cyber security for pipeline sys-
Your-questions/Enbridge-FAQs/How-do-you-monitor-your-crude- tems. Nozomi Networks. https://www.nozominetworks.com/blog/
oil-pipeline-system.aspx improving-ics-cybersecurity-for-pipeline-systems/. Accessed 9
Ioniţə L, Ioniţə I (2014) NM-MAS: a multi-agent system for network Aug 2021
management in oil industry. Proceedings - RoEduNet IEEE O’Brien Chris (2020) AI startups raised $18.5 billion in 2019, setting new
International Conference. https://doi.org/10.1109/RoEduNet- funding record | VentureBeat. https://venturebeat.com/2020/01/14/
RENAM.2014.6955302 ai-startups-raised-18-5-billion-in-2019-setting-new-funding-record/
Jordan MI, Mitchell TM (2015). Machine learning: trends, perspectives, Padmanabhan, V., 2014. Big Data analytics in oil and gas. Bain and
and prospects. In Science (Vol. 349, Issue 6245, pp. 255–260). Company, 1-8
American Association for the Advancement of Science. https://doi. Panja P, Velasco R, Pathak M, Deo M (2018) Application of artificial
org/10.1126/science.aaa8415 intelligence to forecast hydrocarbon production from shales.
Julka N, Karimi I, Srinivasan R (2002a) Agent-based supply chain man- Petroleum 4(1):75–89. https://doi.org/10.1016/j.petlm.2017.11.003
agement - 2: a refinery application. Comput Chem Eng 26(12): Park S-W, Roh M-I, Oh M-J, Kim S-H, Lee W-J, Kim I-I, Kim C-Y
1771–1781. https://doi.org/10.1016/S0098-1354(02)00151-5 (2018) Estimation model of energy efficiency operational indicator
Julka N, Srinivasan R, Karimi I (2002c) Agent-based supply chain man- using public data based on big data technology. OnePetro
agement - 1: framework. Comput Chem Eng 26(12):1755–1769. Patel H, Prajapati D, Mahida D, Shah M (2020) Transforming petroleum
https://doi.org/10.1016/S0098-1354(02)00150-3 downstream sector through big data: a holistic review. J Pet Explor
Khvostichenko D, Makarychev-Mikhailov S (2018) Effect of fracturing Prod Technol 10(6):2601–2611. https://doi.org/10.1007/s13202-
chemicals on well productivity: avoiding pitfalls in big data analysis. 020-00889-2
Proceedings - SPE International Symposium on Formation Damage Pettinger CB (2014) Leading indicators, culture and big data: using your
Control, 2018-February. https://doi.org/10.2118/189551-ms data to eliminate death. OnePetro
 Environ Sci Pollut Res

Pinto JM, Joly M, Moro LFL (2000) Planning and scheduling models for Journal of Petroleum Science and Engineering (Vol. 173, pp.
refinery operations. Comput Chem Eng 24(9–10):2259–2276. 1173–1186). Elsevier B.V. https://doi.org/10.1016/j.petrol.2018.
https://doi.org/10.1016/S0098-1354(00)00571-8 10.050
Plate MV, Ag C (2018) SPE-181037-MS big data analytics for prognos- Sinha AK, Aditya HK, Tiwari MK, Chan FTS (2009) Multi-agent based
tic foresight new dimension of petroleum asset management, pp 6–8 petroleum supply chain coordination: a co-evolutionary particle
Plate: SPE-181037-MS big data analytics for prognostic... - Google swarm optimization approach. 2009 World Congress on Nature
Scholar (n.d.) Retrieved May 22, 2021, from https://scholar. and Biologically Inspired Computing, NABIC 2009 - Proceedings,
google.com/scholar_lookup?title=SPE-181037-MS Big Data 1349–1354. https://doi.org/10.1109/NABIC.2009.5393741
Analytics for Prognostic Foresight New Dimension of Petroleum Slaughter Andrew, Mittal Anshu. (2021) The Internet of Things in the oil
Asset Management&publication_year=2016&author=M Von and gas industry | Deloitte Insights. https://www2.deloitte.com/us/
Plate&author=C. Ag#d=gs_qabs&u=%23p%3DnIW3bKHmMZ4J en/insights/focus/internet-of-things/iot-in-oil-and-gas-industry.html
Popa AS., Popa CG, Malamma M, Hicks J (2008 Case-based reasoning Sneed J (2017) Predicting ESP lifespan with machine learning. SPE/
approach for well failure diagnostics and planning. 10.2118/ AAPG/SEG Unconventional Resources Technology Conference
114229-mspwc. (2013). Economic impacts of the oil and natural 2017. 10.15530/urtec-2017-2669988
gas industry. https://www.api.org/~/media/Files/Policy/Jobs/ Srinivasan R, Wang C, Ho WK, Lim KW (2004) Dynamic principal
Economic_Impacts_ONG_2011.pdf. Accessee 7 May 2021 component analysis based methodology for clustering process states
Qiao Y, Peng J, Ge L, Wang H (2017) Application of PSO LS-SVM in agile chemical plants. Ind Eng Chem Res 43(9):2123–2139.
forecasting model in oil and gas production forecast. In: https://doi.org/10.1021/ie034051r
Proceedings of 2017 IEEE 16th International Conference on Strantzali E, Aravossis K (2016) Decision making in renewable energy
Cognitive Informatics and Cognitive Computing, ICCI*CC 2017, investments: a review. In Renewable and Sustainable Energy
pp 470–474. https://doi.org/10.1109/ICCI-CC.2017.8109791 Reviews (Vol. 55, pp. 885–898). Elsevier Ltd. https://doi.org/10.
Ramírez Miguel Angel, Quintero Jacinto Alfonso Dávila, Morles Eliezer 1016/j.rser.2015.11.021
Colina (2009) Intelligent supervision based on multi-agent systems: Tarrahi M, Shadravan A (2016) Advanced big data analytics improves
application to petroleum industrial processes. https://www. HSE management. Society of Petroleum Engineers - SPE Bergen
researchgate.net/publication/228746098_Intelligent_supervision_ One Day Seminar. https://doi.org/10.2118/180032-ms
based_on_multi-agent_systems_application_to_petroleum_ Temirchev P, Simonov M, Kostoev R, Burnaev E, Oseledets I,
industrial_processes. Accessed 17 Jul 2021 Akhmetov A, Margarit A, Sitnikov A, Koroteev D (2020) Deep
Ransbotham S, Kiron D, Gerbert P, Reeves M (2017) Reshaping business neural networks predicting oil movement in a development unit. J
with artificial intelligence closing the gap between ambition and Pet Sci Eng 184:106513. https://doi.org/10.1016/j.petrol.2019.
action RESEARCH REPORT In collaboration with 106513
Reddy PCP, Karimi IA, Srinivasan R (2004) A new continuous-time
Tokarek TW, Odame-Ankrah CA, Huo JA, McLaren R, Lee AKY,
formulation for scheduling crude oil operations. Chem Eng Sci
Adam MG, Willis MD, Abbatt JPD, Mihele C, Darlington A,
59(6):1325–1341. https://doi.org/10.1016/j.ces.2004.01.009
Mittermeier RL, Strawbridge K, Hayden KL, Olfert JS, Schnitzler
Robertson AW (2014) Securities and Exchange Commission.
EG, Brownsey DK, Assad FV, Wentworth GR, Tevlin AG, Worthy
Encyclopedia of U.S. political history. https://doi.org/10.4135/
DEJ, Li SM, Liggio J, Brook JR, Osthoff HD (2018) Principal
9781608712380.n519
component analysis of summertime ground site measurements in
Rooki R, Doulati Ardejani F, Moradzadeh A (2014) Hole cleaning pre-
the Athabasca oil sands with a focus on analytically unresolved
diction in foam drilling using artificial neural network and multiple
intermediate-volatility organic compounds. Atmos Chem Phys
linear regression. Geomaterials 04(01):47–53. https://doi.org/10.
18(24):17819–17841. https://doi.org/10.5194/acp-18-17819-2018
4236/gm.2014.41005
Umar Ali (2019) BP invests in artificial intelligence for upstream busi-
Sadiq T, Gharbi R (1998) Prediction of frictional drag and transmission of
ness. https://www.offshore-technology.com/news/bp-artificial-
slack-off force in horizontal wells using neural networks.
intelligence-investment/. Accessee 28 Jan 2021
Proceedings - SPE Annual Western Regional Meeting, 259–266.
Wang Yangzihan, Menglan Duan, Deguo Wang, Yinhuan Dong (2011)
https://doi.org/10.2118/51083-ms
A model for deepwater floating platforms selection based on BP
Sadiq T, Nashawi IS (2000 Using neural networks for prediction of for-
artificial neural networks | Request PDF. Retrieved May 22, 2021,
mation fracture gradient. https://doi.org/10.2118/65463-ms
from https://www.researchgate.net/publication/290599141_A_
Saudi Aramco Energy Ventures Invests AI Company Earth Science
model_for_deepwater_floating_platforms_selection_based_on_
Analytics - Algorithm-X Lab (n.d.) Retrieved May 22, 2021, from
BP_Artificial_Neural_Networks
https://algorithmxlab.com/blog/saudi-aramco-energy-ventures-
invests-ai-company/ Weiss WW, Balch RS, Stubbs BA (2002) How artificial intelligence
Sayda AF, Taylor JH (2008) A multi-agent system for integrated control methods can forecast oil production. Proceedings - SPE
and asset management of petroleum production facilities - part 1: Symposium on Improved Oil Recovery, 212–227. https://doi.org/
prototype design and development. IEEE International Symposium 10.2118/75143-ms
on Intelligent Control - Proceedings, 162–168. https://doi.org/10. Wood DA (2020) Predicting porosity, permeability and water saturation
1109/ISIC.2008.4635950 applying an optimized nearest-neighbour, machine-learning and
Shadizadeh Seyed Reza, F Karimi, Mansoor Zoveidavianpoor (2010) data-mining network of well-log data. J Pet Sci Eng 184:106587.
Drilling Stuck pipe prediction in Iranian oil fields: an artificial neural https://doi.org/10.1016/j.petrol.2019.106587
network approach. https://www.researchgate.net/publication/ Yang C, Fan J, Wu Z, Udell M (2020) Efficient AutoML pipeline search
229040683_Drilling_Stuck_Pipe_Prediction_in_Iranian_Oil_ with matrix and tensor factorization. ArXiv. http://arxiv.org/abs/
Fields_An_Artificial_Neural_Network_Approach. Accessee 17 2006.04216. Accessee 9 Feb 2020
Jan 2021
Shafiee M, Animah I, Alkali B, Baglee D (2019) Decision support Publisher’s note Springer Nature remains neutral with regard to jurisdic-
methods and applications in the upstream oil and gas sector. In tional claims in published maps and institutional affiliations.