PETROLEUM EXPLORATION AND DEVELOPMENT, 2021, 48(1): 1-14 doi: 10.1016/S1876-3804(21)60001-0

Application and development trend of artificial intelligence in petroleum exploration and development

KUANG Lichun1, LIU He2, REN Yili,2,*, LUO Kai1, SHI Mingyu1, SU Jian2, LI Xin2

1. Science and Technology Management Department of CNPC, Beijing 100007, China

2. PetroChina Research Institute of Petroleum Exploration & Development, Beijing 100083, China

Corresponding authors: *E-mail: renyili@petrochina.com.cn

Received: 2020-09-19   Online: 2021-01-15

Fund supported: National Natural Science Foundation of China72088101

Abstract

Aiming at the actual demands of petroleum exploration and development, this paper describes the research progress and application of artificial intelligence (AI) in petroleum exploration and development, and discusses the applications and development directions of AI in the future. Machine learning has been preliminarily applied in lithology identification, logging curve reconstruction, reservoir parameter estimation, and other logging processing and interpretation, exhibiting great potential. Computer vision is effective in picking of seismic first breaks, fault identification, and other seismic processing and interpretation. Deep learning and optimization technology have been applied to reservoir engineering, and realized the real-time optimization of waterflooding development and prediction of oil and gas production. The application of data mining in drilling, completion, and surface facility engineering etc. has resulted in intelligent equipment and integrated software. The potential development directions of artificial intelligence in petroleum exploration and development are intelligent production equipment, automatic processing and interpretation, and professional software platform. The highlights of development will be digital basins, fast intelligent imaging logging tools, intelligent seismic nodal acquisition systems, intelligent rotary-steering drilling, intelligent fracturing technology and equipment, real-time monitoring and control of zonal injection and production.

Keywords: artificial intelligence ; logging interpretation ; seismic exploration ; reservoir engineering ; drilling and completion ; surface facility engineering

PDF (179KB) Metadata Metrics Related articles Export EndNote| Ris| Bibtex  Favorite

Cite this article

KUANG Lichun, LIU He, REN Yili, LUO Kai, SHI Mingyu, SU Jian, LI Xin. Application and development trend of artificial intelligence in petroleum exploration and development. [J], 2021, 48(1): 1-14 doi:10.1016/S1876-3804(21)60001-0

Introduction

The concept of Artificial Intelligence (or AI for short) was first put forward at the Dartmouth Conference held in 1956. Scholars represented by John McCarthy defined the technology of “using machine to imitate human cognitive ability” as “artificial intelligence”[1], which has not been universally accepted until now. The authors believe that “artificial” implies artificial fabrication, while “intelligence” involves aspects such as consciousness, self-awareness, thinking and so on. In short, artificial intelligence is man-made consciousness and thinking capability. The key technologies of AI include machine learning, data mining, natural language processing, pattern recognition, computer vision, knowledge graph and so on. In recent years, with the development of big data, the emergence of deep learning and the significant boosting of computing power, AI technology has entered the stage of booming development. Artificial intelligence has been widely used in medicine, transportation, Internet, agriculture and other sectors, becoming the core driving force for the fourth generation of industrial revolution and the decisive factor to usher human society into the era of intelligence.

At present, the grade of domestic oil resources tends to become more inferior, and the major old oilfields generally enter the late development stage of ultra-high water cut. In order to maintain national economic stability and energy security, it is necessary to endeavor more in domestic exploration and development. The global science and technology is developing rapidly in the direction of digitalization, informatization and intelligentization. Logically, intelligent exploration and development has become the state of the art and growing trend of the industry, promising much higher efficiency and quality of exploration and development, lower costs and risks, and elevated level of exploration and development of complex reservoirs.

In this paper, the research progress of AI technology in terms of oil exploration and development is extensively investigated. Taking into consideration of the actual demands of exploration and development, we discuss the research progress and application of the technology in logging, geophysical exploration, drilling and completion (or D&C for short), reservoir engineering, and surface facility engineering, and make an outlook of the application highlights and growing trend of AI technology in the future.

1. R&D of artificial intelligence in oil companies

Strategies such as open innovation, in-depth integration of production, education and research, and cooperation with IT companies are universally adopted by oil companies to make the oil and gas industry more intelligent. Through joint efforts with IT companies, international oil companies have realized intelligent upstream operations, resulting in such crossover alliances as Total + Google Cloud, Chevron + Microsoft, Shell + HP, etc. For example, Shell's “Smart Field” program focuses on collaborative working environment, intelligent wells, optical fiber monitoring, real-time production optimization, intelligent water flooding and closed-loop reservoir management; Chevron's “i-Field” plan focuses on drilling optimization, production optimization and reservoir management; and BP's “Field of the Future” focuses on the application of real-time information system to optimize operations; and GASPROM's digital transformation (DT) scheme gives priority to 12 types of projects, including digital geological exploration, digital large-scale projects, digital production, midstream digital operation, digital HSE (health, safety, and environment), and digital facilities and equipment. Shown in Table 1 are AI strategies of some major oil companies and service companies globally.

Table 1   Comparison of AI strategies among global key oil and gas companies and service companies.

No.CompaniesOrientationAI PlatformPartners
1BPUpstream and downstream business to
realize decision automation
SandyBeyond Limits, Belmont Technology
2ShellHorizontal well trajectory control,
drilling data processing algorithm
GeodesicMicrosoft
3Exxon MobilData collection and integrated solutionsXTOMicrosoft
4TotalIntelligent solution for E&P,
intelligent seismic imaging processing
Cloud PlatformGoogle
5ChevronE&P, storage & transportation projectsDELFIMicrosoft,
Schlumberger
6SchlumbergerE&P, storage & transportation projectsDELFIMicrosoft, Chevron
7Baker HughesSeismic modeling, malfunction prediction
and supply chain optimization
Desktop Platform,
Azure
NVIDIA, Microsoft
8HalliburtonReservoir characterization and simulationAzureMicrosoft
9PetroChinaIntelligent basins, intelligent logging, intelligent geophysical
exploration, intelligent drilling & completion, intelligent oil
production, intelligent fracturing and intelligent equipment
Dream Cloud Platform,
Cognitive Computing
Platform
Huawei
10SinopecIntelligent factories, intelligent oilfields and
intelligent institutes
Oilfield Smart Cloud
Industrial Internet
Platform
Ali
11CNOOCIntelligent oilfields, E&D data managementIntelligent Oilfield
Technology Platform
Ali

New window| CSV


In order to expedite and automatize decision-making, and better manage operational risks, BP Ventures has invested 20 million US dollars in Beyond Limits to develop AI software that can locate and develop reservoirs, produce and refine crude oil, and sell refined products. In January 2019, BP also invested 5 million US dollars in Belmont Technology to further strengthen the digital transformation and upgrading of BP's upstream business.

Total signed an agreement with Google to jointly explore intelligent solutions for exploration and development, focusing on the intelligent processing and interpretation of downhole imaging, especially of seismic data, so as to improve the efficiency of engineers in exploration and evaluation of oil and gas fields.

Shell and Microsoft are jointly developing the Geodesic platform to improve the accuracy of trajectory control of horizontal wells, so as to accurately hit the sweet spot. This solution can simplify the drilling data processing algorithm, and consequently enable real-time decision making and better result prediction.

ExxonMobil and Microsoft are working together to develop an integrated cloud platform that can safely and reliably collect real-time data from oil fields spreading hundreds of kilometers. These data will enable the company to make faster and better decisions in drilling and completion optimization and personnel deployment.

Chevron, Schlumberger and Microsoft are jointly developing DELFI cloud computing platform to integrate a large amount of information into one source, build an open and sharable data ecological environment, and reduce the time required for data analysis.

Halliburton and Microsoft formed a strategic alliance to combine Microsoft's Azure Intelligent Cloud solution with exploration and development, applying deep learning in reservoir characterization and simulation to create highly interactive application programs and realize digital E&D operations.

Baker Hughes, in cooperation with NVIDIA, uses artificial intelligence and GPU-accelerated computing to help extract industry data in real time, facilitating seismic modeling, machine malfunction prediction, supply chain optimization, data mining, and reduction of the costs of exploration, development, processing and transportation. In November 2019, Baker Hughes, C3.ai and Microsoft jointly announced their cooperation to introduce to the energy sector the enterprise AI solutions on Microsoft's cloud computing platform.

China National Petroleum Corporation (CNPC) has preliminarily constructed the Cognitive Computing Platform (E8) and deployed pilot applications in Dagang, Daqing, Changqing and other oilfields; established the Dream Cloud Platform based on the principle of "two unifications, one common" (unified data lake, unified technology platform, and common business application); and carried out the top-level design of artificial intelligence led by its Science and Technology Management Department, which all together comprehensively promoted the exploring implementation of artificial intelligence technology.

China Petrochemical Corporation (Sinopec) started the pioneering of intelligent manufacturing in 2012, and initiated the planning, designing and construction of intelligent factories, intelligent oilfields and intelligent institutes in succession. The corporation has built Oilfield Smart Cloud Industrial Internet Platform, which deeply integrated the new generation of IT and corporate businesses, promoting the upgrading and digitalization of enterprises.

In March 2020, China National Offshore Oil Corporation (CNOOC) officially issued the “Top-Level Design Scheme for Digital Transformation of CNOOC”, putting forward the overall blueprint for digital transformation, aiming at building an intelligent oilfield technology platform to construct intelligent oilfields and achieve data governance for exploration and development.

2. Application of artificial intelligence in petroleum exploration and development

2.1. Logging

Since its emergence in 1927, logging technology has experienced more than 90 years of advancement, from analog logging, digital logging, direct digital logging to imaging logging, and currently it is stepping into the era of intelligent logging.

2.1.1. Logging data collection

Due to the heterogeneity of reservoirs, the complexity of exploration objects, and the diversification and complexity of logging operation environments, it is of urgent necessity to study new measurement approaches and working modes involving especially downhole parameter acquisition and data transmission, and introduce artificial intelligence to achieve more accurate, more efficient and safer operation and geological information sensing.

Overseas companies have commercialized their products in data acquisition and remote logging, such as Schlumberger's Remote Logging Center, Intelligent Formation Tester and wellbore software Techlog capable of intelligent processing and interpretation. Eleven data server centers and 14 remote logging centers are deployed around the world, with 108 operation engineers, realizing remote collaborative work and decision-making of experts. Twenty percent of its logging operations are completed by the Remote Logging Center, amounting to over ten thousand jobs.

In China, some petroleum enterprises and research institutions have launched campaigns addressing key technologies related to networked surface facilities, intelligent winches, remote logging and so on, and have been applying the preliminary achievements in small-scale. R&D of intelligent downhole robots has also been initialized.

2.1.2. Logging processing and interpretation

Logging data is characterized by large data volume and multi-source heterogeneity, challenging the interpretation professionals with many difficulties, such as multiple solutions and uncertainty, which make it more difficult to identify pay zones. As a consequence, it is of urgent necessity to take advantage of artificial intelligence and other technologies to enhance work efficiency and interpretation coincidence rate. In recent years, the application of artificial intelligence in logging interpretation mainly focuses on automatic depth correction, automatic report generation, intelligent stratification, curve reconstruction, lithology identification, imaging logging interpretation, reservoir parameter estimation, oil and gas potential evaluation, shear wave velocity prediction, and fracture and its filler identification, etc.

Intelligent curve reconstruction uses deep learning, correlation analysis and other algorithms to find the correlation between various logging curves and reconstruct the erroneous, inappropriate and missing portions of logging curves. The AI algorithms used include neural network, combinatorial learning algorithm, clustering algorithm and so on. Zhang et al.[2] put forward a method to reconstruct logging curve based on recurrent neural network (RNN), namely long-term and short-term memory neural network (LSTM). Upon verification with actual logging curve, it is found that this method is more accurate than the conventional ones.

There are two approaches to identify lithology: The first one is coring, and using core sample analysis to determine lithology; the second is to identify lithology via logging curves. As for the first, with the continuous advancement of scanning equipment/instrumentation, tremendous numbers of core images such as thin section photos, CT images and SEM images have been accumulated in the sector of exploration and development. Currently, core image analysis software packages both domestic and abroad (such as Avizo, PerGeos, etc.) are capable of identifying lithology automatically. However, being mostly based on image processing algorithm, they rely much on too many human-computer interactions and are quite demanding in terms of professional experience. At present, most of the widely used thin section identification methods rely on manual identification, and the intelligentization level is low. The application of deep learning technology in core image processing has yet to be further studied. As for the second approach, the data processed and interpreted by professionals are taken as training specimens, and the intelligent lithology identification model based on logging curves is constructed using AI algorithm. Jiang et al.[3] built lithology prediction model with Boosting Tree, Decision Tree, Support Vector Machine and other algorithms, and taking logging data interpreted by professionals as training specimens. Taking the lithology resulted from mud logging as the benchmark, the prediction accuracy is higher than 80%.

Imaging logging basically transforms original logging curves into visual images that reflect the geological characteristics via the chromaticity calibration principle. As the application of deep learning technology in the sector of image analysis deepens, researchers begin to combine deep learning with image processing to realize the automatic interpretation of imaging logging. Ren et al.[4] used image segmentation algorithms such as U-Net to realize automatic recognition of geological feature boundaries with images generated by electrical imaging, and then used feature engineering to extract relevant features, and finally used machine learning algorithm to realize automatic classification of geological features. The research on artificial intelligence in processing and interpretation of imaging logging is just in its infancy. It is lack of label data for machine learning that restricts the further progress of the research.

The application of AI algorithm in reservoir parameter estimation got started earlier. Early scholars mainly used traditional machine learning algorithms (such as Support Vector Machine, linear regression, etc.) to estimate porosity, permeability, saturation and other parameters. In recent years, with the continuous progress of neural network, more and more scholars have begun to use BP (feedforward neural network), LSTM, Random Forest, GBDT (Gradient Boosting Decision Tree) and other combinatorial learning algorithms to estimate reservoir parameters.

2.1.3. Integrated software

Internationally, represented by Schlumberger, and with more than 10 software packages including Petrel, Techlog, Eclipse as the kernel, a digital collaborative intelligent workflow has been set up to reduce the uncertainty and risk in exploration and development. The integrated platform of cognitive exploration and development software, DELFI, establishes the workflow of intelligent processing and interpretation, offering the functions of data standardization, data cleaning, intelligent interpretation and result submission. The wellbore software Techlog is comprised of a series of intelligent function modules such as analysis, prediction and classification of curve sensitive factors, and curve reconstruction, etc., powerfully supporting intelligent interpretation.

In China, the application platforms such as the CNPC Dream Cloud Collaboration Platform, the integrated logging processing and interpretation software LEAD, and the new generation of multi-well formation evaluation software CIFlog have been built. The applications of artificial intelligence in reservoir description and simulation, and in multi-well logging interpretation have achieved preliminary results, and the geo-steering system for horizontal drilling is taking shape.

2.2. Geophysical prospecting

The research on “AI + Geophysical Exploration” is expanding rapidly worldwide. Geophysical exploration has always been a prominent application sector of information technology including high-performance computing, three-dimensional visualization, computer network and so on. It is the sector where digital acquisition, processing and analysis were realized at an earlier time.

2.2.1. Equipment for geophysical prospecting

The application of artificial intelligence in geophysical prospecting equipment centers on vibroseis, UAVs, seismic instrumentation and so on. With intelligent vibroseis the output, frequency range, scanning time, phase and other parameters can be adjusted according to the specific surface conditions and deep seismic geological conditions of the work area, making it safe and environmentally friendly. With intelligent UAVs for geophysical data acquisition, high-precision terrain detection, risk assessment, node monitoring, data recovery, material delivery, rescue and other tasks can be achieved. As for seismic instrumentation, G3i (wired), Hawk (node), eSeis (node) and other products have been developed. The technological campaign for OBN (ocean bottom node) has overcome the limitations of poor obstacle-crossing ability, narrow range of observation azimuth, strong offshore noise and uni-component reception.

2.2.2. Geophysical acquisition

With the continuous advancement of cloud computing, artificial intelligence, robotics, communication and other technologies, geophysical acquisition has stepped into the intelligent stage from the digital stage, featured by noninductive digitization, highly closed loop automation, “robotization” of essential equipment, integrated procedures of operation, prediction of production performance, and even some big data edge computing capabilities.

For geophysical acquisition the transformation from traditional seismic crews into digital seismic crews has come true. With digital seismic crews IT technologies such as Internet of Things and cloud computing are integrated with geophysical acquisition methods, which enables wireless and visual digital management of field tasks, personnel, equipment and HSE, optimization of operation procedures, and realization of intelligence stimulation, real-time quality control and remote technical support, command and dispatch.

2.2.3. Seismic data processing and interpretation

For seismic data processing and interpretation, artificial intelligence technology is mainly applied in such aspects as seismic structure interpretation (including fault identification, horizon interpretation, dome top and bottom definition, river channel or cavern interpretation, etc.), noise suppression/signal enhancement, seismic facies identification, reservoir parameter estimation, seismic wave field forward modeling, seismic inversion, seismic velocity picking and modeling, first break picking, seismic data reconstruction and interpolation, seismic attribute analysis, micro-seismic data analysis, comprehensive interpretation etc. The essential technologies used are target detection, segmentation, image classification and prediction, all belonging to the category of computer vision. The application of AI technology greatly improves the efficiency of seismic data processing and interpretation on the basis of guaranteeing accuracy.

In recent years, automatic fault recognition based on deep learning has gradually become a typical application highlight. Many scholars use convolution neural network to train with synthetic seismic data set or actual seismic data set, build intelligent fault recognition models, and automatically define the probability of existence and dip angle of faults. Wu et al.[5] developed a convolutional neural network model based on encoding and decoding, which can realize fault detection and slope estimation simultaneously. In order to train the network, thousands of 3D synthetic noise seismic images and corresponding fault images, clean seismic images and seismic normal vectors were automatically generated. Several application cases indicated that the network is superior to the traditional method in fault detection and reflection slope calculation.

There is more and more research on seismic facies recognition based on deep learning. The traditional recognition of seismic facies is to cluster seismic attributes first, and then classify seismic waveforms to do the recognition. With the advancement and application of AI technology such as machine learning, more and more researchers directly apply convolutional neural network (CNN), recurrent neural network (RNN), probabilistic neural network (PNN), deep neural network (DNN), generative adversarial network (GAN) etc. to the classification and recognition of seismic waveforms. Zhang et al.[6] proposed an enhanced codec structure DeepLabv3+ developed relying on Google. Compared with CNN model and simple semantic segmentation model (such as deconvolution neural network), this codec structure can achieve higher accuracy and efficiency in extracting multi-scale semantic information and recovering more pixel-level details in prediction results, which is promising in improving the accuracy and efficiency of seismic facies recognition.

Seismic inversion is basically to transform the conventional interface reflection profile into the rock stratum logging profile, so that the seismic data can be directly correlated to the logging data. In recent years, great progress has been made in application of AI technology in this field, and the main algorithms used are CNN, RNN, DNN, Boltzmann machine and GAN. Phan et al. [7] combined the cascade method with convolution neural network, constructed a deep learning model by minimizing an energy function similar to the least square solution of the inverse problem, and then used the network to carry out pre-stack seismic inversion, and predicted impedance by training the network to learn the nonlinear relationship between rock properties and seismic amplitude. The inversion algorithm requires that the inputs be normalized before training the network, and the result be converted to absolute value after network application. The results show that the algorithm can capture all the features in the training data set, reconstruct the input logging curve of well points accurately, and generate the geologically reasonable impedance profile.

First break picking is the basis of subsequent seismic processing and imaging. With the rapid increase of seismic data, manual acquisition is extremely time-consuming. In order to address these challenges, it is necessary to develop robust automatic picking methods. Ma et al.[8] proposed a method to extract the first break data automatically using an improved 2D pixel convolution network. The event-picking problem was transformed into a problem of binary image segmentation, and the signals before and after the first break were marked with 1 and 0, respectively. The effectiveness of this method and its superiority over the traditional automatic picking method were confirmed by analysis of actual borehole seismic data.

2.3. Drilling and completion

Upon evolving from being conceptual, to experiential, to scientific, and to automatic, a technology system has been well established for drilling engineering which takes technological principles and methods as its guidance, and takes equipment, tools and materials as its means. D&C technology has basically progressed into the scientific stage, and to be more specific, it is currently in the stage of fusing automation and intelligence, with the latter being the emphasis.

Intelligent D&C, a brand new D&C mode, takes intelligent software system as a link or tie, relies on surface intelligent equipment and downhole intelligent tools, and integrates them into a closed-loop system with calculation model and intelligent decision-making technology. Downhole intelligent tools are mainly intelligent drill bits and drill pipe with embedded chips, and rotary steering system etc.; surface intelligent equipment is mainly rig floor robots with industrial control core system, automatic tripping control mechanisms, and automatic bit-feeding mechanisms etc.; as a link or tie, intelligent software integrates the them into a whole, and realizes efficient and automatic drilling to the sweet spot based on the geological conditions, thus secures the maximum productivity. In China this technology has just been started and is in the early stage of single technology R&D. There is still a big gap between domestic and overseas level in terms of the overall automation and intelligence degree of equipment and tools.

2.3.1. Key technologies of intelligent drilling and completion

The key technologies of intelligent D&C include intelligent optimization of well trajectories, intelligent steering, intelligent optimization of ROP, etc.

The intelligent optimization of well trajectories based on multi-source data of geological engineering generally uses AI algorithms such as genetic algorithm and neural network to realize the optimization of hole azimuth and other related parameters. The essence of the intelligent steering technology is to realize prediction and automatic control during drilling operations by taking advantage of AI algorithm, real-time monitoring and analysis of the trajectory, seismic-while-drilling, near-bit measurement and other new technologies. As for ROP intelligent optimization, in most cases it uses big data and intelligent optimization algorithm to optimize multiple parameters globally, so as to achieve the best matching among the formation, bit and parameters, realizing the dynamically optimized design of parameters such as hole inclination and azimuth. Via linkage with the rig, it automatically issues control instructions, hence intelligently optimizing the ROP. Commonly used algorithms include random forest, artificial neural network, ant colony algorithm, particle swarm algorithm, etc.

2.3.2. Intelligent drilling and completion equipment

As for surface equipment, overseas oil companies have been applying on a large scale such technologies as rig floor robots, automatic tripping control, automatic bit-feeding, automatic pressure management, on-line monitoring of drilling fluid, etc. In terms of downhole tools, intelligent drill bits and intelligent drill pipe make unmanned operation and precise control possible, substantially enhancing drilling efficiency and reducing drilling risk and labor cost. The wide commercialization of the rotary steering system realizes the functions of measurement and control while drilling, ensuring intelligent steering as well as efficient rock breaking. Based on information on geological conditions and reservoir characteristics, the parameter-optimization model, which takes the maximum oil/gas productivity as the objective function, is established by using neural network and other methods; and the best parameters resulting from calculation are compared with the real-time data collected while drilling, so as to automatically seek for the best trajectory.

There are some automatic drilling rigs manufactured in China, which basically realizes automatic control of tubular strings. However, the reliability and effectiveness of sensors in sensing some scenarios, and the accuracy of online warning and diagnosis of equipment have yet to be improved. The operation accuracy of hydraulic driving equipment is relatively low, and the same is true for the overall degree of intelligence. The pressure management drilling equipment is basically automatized, but the abilities of industrial control software to sense the hole conditions and to recognize formation have yet to be improved. In terms of downhole tools, MWD and other tools are basically domestically produced, but the comprehensive sensing ability has yet to be enhanced, and the precise geo-steering, fine geological evaluation, intelligent navigation speed-up and other functions have also to be further perfected.

2.3.3. Intelligent drilling and completion software

In terms of applied software, digital twin system, automatic control system, drilling emulation and remote decision-making software have been commercialized and continuously perfected. Based on the tremendous amount of data on drilling and completion (D&C), overseas companies have introduced machine learning, big data, cloud computing and other state-of the-art technologies to establish the method for predicting borehole environment and the technical system for characterizing the mechanical characteristics of the formation. They have developed platform for integrating and analyzing D&C big data, the well construction engineering design and intelligent optimization system based on cloud platform, and the integrated fracturing optimization software, which together greatly improved the D&C design, forecast of downhole complexes, and precision of analysis and control, leading to maximum automation, efficiency and intelligence of D&C engineering.

Halliburton Well Construction Project 4.0 makes use of big data analysis and drilling intelligent optimization platform etc. to build Digital Twin Wellbore, which is capable of conducting the whole cycle of pre-drilling simulation rehearsal, real-time decision-making during drilling, and post-drilling playback analysis. DrillPlan, an integrated drilling design module in Schlumberger's DELFI platform, can shorten the duration of drilling design and planning from a few weeks to a few days. ConocoPhillips D&C big data analysis platform (IDW) can simplify the process of data collection and processing, and effectively analyze the data, so it can be applied to reduce rig time, optimize completion design, and improve the understanding of the formation.

The R&D of intelligent D&C software is just in its infancy in China, which basically possesses the functions of D&C design, monitoring and optimization, but the data norms are not unified, the information sharing is not smooth, the fusion degree of physical model and machine learning algorithm is low, and the accuracy, comprehensiveness and field applicability have to be improved.

2.3.4. Intelligent D&C integration platform

Schlumberger's new generation of onshore “Drilling System of the Future” organically combines digital technology, equipment, tools and software into an entire system, equipped with automatic drill pipe handling device and 1000+ built-in sensors, which can monitor more than 350 different drilling activities and continuously boost the level of automation and intelligence.

The integration control platform (eVolve) of National Oilwell integrates NOVOS (a surface equipment control software system), Intelliserv (information drill pipe), BlackSteam (a measurement while drilling tool) and DrillShark (analysis and optimization software) all together. It processes and analyzes the downhole dynamic data and the surface data in combination, interacts with the comprehensive drilling simulation model, and realizes closed-loop control of whole drilling operation by virtue of NOVOS. It has been applied to 6 horizontal wells drilled into Eagle Ford shale formation, and the net drilling time was reduced by 37%. However, it is bothered with some problems such as high cost and to-be-improved system reliability.

2.4. Reservoir engineering

The major task of reservoir engineering is to study the migration pattern and displacement mechanism of oil, gas and water during reservoir development on the basis of fluid mechanics in porous media and petrophysics, so as to take corresponding engineering measures to reasonably increase the production rate and enhance final recovery. In the era of Industry 4.0, intelligent reservoir engineering has become an inevitable trend, the basic principle of which is to fully understand the reservoir and the laws governing porous flow with the aid of computer algorithms and software tools, so as to realize intelligent dynamic management and performance prediction.

2.4.1. Reservoir performance analysis and simulation

Reservoir engineering is a highly inclusive discipline; only the application of artificial intelligence in reservoir performance analysis and simulation is briefly addressed in this paper. Generally speaking, dynamic analysis and simulation are conducted mainly via reservoir numerical simulation and reservoir engineering methods. At present, the application of artificial intelligence mainly focuses on real-time control of waterflooding development, production prediction, saturation prediction, picking of production measures, numerical simulation and so on.

As for real-time control of waterflooding development, production parameters are optimized mainly by optimization means and data mining approach. Jia et al.[9], under the constraint of dynamic data, adopted the traditional numerical simulation and optimization algorithm to compute the flow relationship between injectors and producers by automatically identifying the flow pattern in zonal injection. At the same time, they applied the multi-layer and multi-directional production splitting technology to compute liquid and oil production from individual zones and individual directions, hence quantifying the water injection effect metrics. Upon evaluating the zonal injection effect in multiple wells and analyzing the requirement for injection adjustment by machine learning algorithm, a series of fine injection scheme optimization approaches driven by big data was proposed. Supported by field tests, the integrated technology of reservoir and production engineering, which features injection scheme design, intelligent optimization and synchronous adjustment, was preliminarily established. Taking carbonate reservoirs as their concern, Jia et al.[10] used streamline clustering method to distinguish streamlines with different water phase driving abilities, and further distinguished the streamlines between a specific injector-producer pair, thus proposed a flow field identification method for streamline simulation results. As a consequence, flow field identification for waterflooded reservoirs based on streamline clustering artificial intelligence method materialized, laying a foundation for decision-making on injection optimization, well pattern and zonation adjustments, and deep profile control schemes.

For production prediction, some scholars use recurrent neural network to predict cumulative oil/liquid production by taking reservoir static and dynamic parameters and production parameters as the inputs. Based on the historical data of field production, and taking into account the relationship between the production and its influencing factors, as well as the variation trend of production with time, Wang et al.[11] used the long-term and short-term memory neural network (LSTM) in the territory of deep learning to build the corresponding production prediction model, so as to achieve the goal of predicting the production in the ultra-high water cut period. Compared with the traditional waterflooding chart method and FCNN (fully connected neural network) model, their model delivers more accurate prediction results. Kubota et al.[12] enabled oil production prediction of mature onshore fields developed by water or steam injection upon extracting hidden patterns and potential relationships form large amounts of historical data, and utilizing two machine learning algorithms, namely linear regression and recurrent neural network. In this approach neither geological model nor reservoir numerical simulation was necessary; injection history, production history and number of producers were the all inputs they needed. Bao et al.[13] applied recurrent neural network to the analysis of control parameters (flow rate and bottom hole flowing pressure) and historical production data, directly linked control parameters with expected production data (such as productivity and water cut), and thus established end-to-end production prediction workflow. Reservoir characterization and production prediction can be better carried out with their methodology, which is useful for guiding the development deployment expediently when the reservoir is put into production.

For saturation prediction, Tariq et al.[14] optimized the Functional Network model with differential evolution (DE), particle swarm optimization (PSO) and covariance matrix adaptive evolution strategy (CMAES), and established a water saturation prediction model with petrophysical logging data as the input and water saturation as the output. Benchmarked with the core experimental value, the accuracy of the model was 97%. Shahkarami et al.[15] developed and validated an intelligent agent model for reservoir simulation history matching, sensitivity analysis and uncertainty evaluation with neural network technology. Via the analysis of two field cases, it was verified that the model delivers good prediction results for production, reservoir pressure and phase saturations, and accelerates computation speed as well.

For optimization of production measures, based on the fuzzy reasoning system, Artun et al.[16] transformed all relevant parameters into fuzzy variables with low, medium and high class membership functions, and thus architected a decision-making method based on artificial intelligence, which can be applied to identify candidate wells suitable for repeated fracturing in tight sandstone gas reservoirs. In order to predict the future production performance of the target reservoir and seek for the scheme of enhancing oil recovery, Sengel et al.[17] proposed an AI method which eliminates the inherent uncertainty in addressing highly complex and heterogeneous carbonate reservoirs with insufficient data. This method enables more accurate reservoir description, simplifies the calibration of dynamic models, and improves the quality of history matching.

For numerical simulation, some scholars try to build intelligent models based on the existing historical data, realize automatic history matching, and speed up the simulation. Zhang et al.[18] developed a neural simulation protocol that integrated the internal numerical simulation package with the expert system based on artificial neural network, and established the neural simulation workflow to empower the expert system to update its knowledge base automatically based on the data generated by the numerical simulation model. Costa et al.[19] successfully carried out field history matching with neural network model and genetic algorithm. In this application, the artificial neural network expert system was trained to imitate the high fidelity numerical model, thus predicted the field production data.

2.4.2. Integrated analysis software

There are some integrated analysis software packages on reservoir performance analysis and prediction overseas, including Eclipse and INTERSECT (upgraded) of Schlumberger, VIP (for reservoir numerical simulation) of Landmark, and tNavigator of RFD (a Russian company), etc. These software packages all use machine learning, optimization and other AI technologies for automatic history matching to accelerate the simulation speed, hence improving the intelligent level of the software.

In China, HiSim, which was independently developed by RIPED, PetroChina, is being quickly integrated with deep learning, machine learning and other AI technologies to further boost the simulation efficiency and intelligent level. The reservoir analysis software IRes, developed based on computational geometry, morphology, optimization and other methodologies, enables real-time monitoring and intelligent control of zonal injection and zonal production in waterflooding development.

2.5. Surface facility engineering

An oil and gas lease is a “fenceless factory”, and the construction of the digital twin of field surface facilities is crucial to its intelligentization. The intelligentization here is to construct a digital surface system via 3D design and digital delivery, implant dynamic production data and intelligent application modules into the digital twin, and realize perception, analysis, prediction, optimization and decision-making for the real operation in the virtual environment, thereby achieving the prediction, analysis and dynamic optimization of the operation status of the real surface facility system.

2.5.1. Key kechnologies for surface facility engineering

Field surface facility engineering has experienced three stages, i.e., manual, automatic and digital, and currently is being integrated with AI technology to build intelligent oil and gas fields capable of “perception, analysis, prediction, optimization, decision, and execution”. The key technologies consist of digital delivery, digital metrology, flow assurance and operation optimization of pipeline networks, operation optimization of gathering and storage stations, etc.

The key to realize intelligentization of surface facilities is to realize the digital delivery based on 3D design and construct the digital twin. 3D design facilitates multi-disciplinary collaboration, real-time review and joint optimization of all participants, seamless connection of all stages, and highly unified and harmonized management. It is capable of optimizing construction scheme to the maximum degree, speeding up construction progress, and improving project quality. Digital delivery lays a foundation for forging digital assets, realizing digital transformation, and optimizing production operations. But at present, the inclusions and extent of digital delivery are not consistent; the platform upon delivery has only construction engineering data, with production data to be yet loaded; the ecology of co-generation and co-growth of assets has not taken shape, and the value of data has yet to be mined.

For digital metrology in oil and gas fields, the introduction of digitalization and automation into the metering station is necessary to collect oil and gas from producers, measure the liquid volume of individual wells and boost the metrology efficiency. At present, the mainstream technology is basically digital single-well metering based on principles of digital simulation and reinforcement learning. In addition, in some oilfields indicator diagram metrology is applied instead of metering stations, which generally cuts down the investment and simplifies the gathering pipeline network. However, in the case of low production, high GORs, deep wells, heavy oil and other complex scenarios, the adaptability is poor, the coincidence rate of diagnostic is low, and the metering error is big. Concerning gas fields, some software packages developed by overseas companies are capable of substituting physical metrology with single-well digital metrology, hence eliminate wellhead separators and gas gathering stations, realize multi-well tandem connection, and simplify the gathering process.

Oil and gas gathering networks play an important role in field development. The composition and flow regime of the produced fluids are quite complex, probably giving rise to risks of wax precipitation, corrosion/erosion, pipeline failures and the like. There are too many and complex process links, which greatly impair the safe and stable operation of gathering pipeline networks. Some overseas software packages are essentially capable of offering flow assurance, monitoring corrosion/erosion rate, pigging, and managing suspension/restart. Some domestic software packages are capable of simulating the complex flow regime in gathering networks, but not of carrying out flow assurance; we will wait and see what will happen in research and application of related technologies.

Gathering and storage stations are the hubs of oil and gas production. Optimization of the station operation can reduce energy consumption and labor, and enhance the system safety. At present, some overseas software packages are able to preliminarily realize energy consumption control of some production units, optimization of operation parameters, control of production metrics, and diagnosis of operating condition. There are no widely accepted process-simulating software packages domestically for large-scale stations, so we have to generally rely on overseas software.

2.5.2. Intelligent equipment for surface facilities

UAVs/robots are commercially used in routing inspection overseas, with UAV patrolling pipelines regularly, and robots patrolling well sites, compression stations, gathering stations, multi-purpose stations and gas plants. They reveal risks via image recognition and big data analysis, offer intelligent early warning and alarm, with accuracy up to 95%. Meanwhile, they alleviate labor intensity and upgrade the working environment of employees. However, in China these applications have been initialized just recently.

2.5.3. Intelligent software for surface facilities

Overseas there are already online simulation editions of commercial software addressing large-scale processes, with which process twins of production systems can be built to capture dynamic production data and conduct online simulation. The twin can sense and predict the operation scenario, discover anomaly tendencies, and optimize the operation.

A giant decision-making software package for production engineering, PetroPE, developed by RIPED, PetroChina based on Web, allows diagnosis, optimization and decision-making on operation of various well types and various lifting modes on the basis of integrated analysis of reservoir productivity, rod/tubing string stress and pumping unit movement pattern. In more than 30 000 well tests, the coincidence rate of diagnosis was 93%, with the production efficiency and field management greatly improved.

In general, there is still a gap between China and leading countries in terms of intelligentization of field surface facility engineering, which has to be urgently perfected and commercialized to overcome the bottleneck.

3. Summary and prospect

Digital transformation of enterprises is to achieve automatic data acquisition, data automatic transmission and real-time perception at the front end, secured storage, real-time monitoring and centralized control in the middle, and intelligent analysis, data sharing and technical support at the back end, by utilizing the information technologies like Internet of things, cloud computing, big data, artificial intelligence, and block-chain. The integration, collaboration, efficient linkage and data sharing among the front, the middle and the back strongly facilitate the reconstruction of traditional business operation flow and revolution of work mode. Artificial intelligence plays an essential role in this transformation.

AI technology is expected to break the bottlenecks in exploration and development, transform management mode from being vertical and isolated (traditional) to being integrated, collaborative and flat, thus reconstructing business operation flow, improving quality, lowering cost and boosting efficiency, and finally enhancing the digital transformation of enterprises. The reconstruction is implemented mainly with the following aspects: (1) automatic data acquisition equipment to provide real-time dynamic data; (2) intelligent data analysis and processing software to improve the efficiency of processing and interpretation, reduce the dependence on expert experience, optimize human resource deployment, and cut down labor costs; (3) UAVs and electronic routing inspection to substitute manual work so as to raise employee happiness index; (4) early warning against malfunction to take proactive measures so as to shorten the time for malfunction detection and information transmission and reduce the cost of production maintenance; and (5) dynamic management of production operation to strengthen the ability of emergency response and minimize production losses.

3.1. Problems and challenges in the application of AI technology in exploration and development

Data has become a newly recognized resource, and not only promotes social and economic development, but also facilitates continuous progress of AI technology. However, the application of the technology in the oil E&D sector is often biased towards constantly upgrading equipment and software, which eventually leads to off-line machines, fragmented software and isolated data. In order for artificial intelligence to be applied in the industry, high-quality data, clear application scenarios, scientific and appropriate algorithm model and other provisions are necessary. It is relatively easy to carry out exploratory research, but many difficulties have to be overcome in practical applications at industrial scale.

Objectively, reservoir heterogeneity leads to multiple solutions and uncertainty in solving petroleum geological problems, and it is difficult to obtain the “textbook” (tag data) for machine learning. However, high quality tag data is the key to realize industrial application of AI technology. The cost of collecting geological data is often high, so the data obtained is mostly a “small sample”, and data volume cannot meet the requirements of deep learning. Due to the strong specialty and particularity of oil E&D data, the general AI algorithm cannot be used directly. When using the transfer learning technology to improve the training accuracy, it is necessary to invoke the existing relevant pre-training model. It is impossible to find an appropriate pre-training model or priori knowledge in the existing resource library because of the particularity of oil E&D application scenarios. All these hinder the progress of AI application in some degree.

Subjectively, constrained by the management regime and data status, the practical application of artificial intelligence is bothered with many difficulties. At present, the research of AI technology in the sector of exploration and development is explosively expanding, but there is shortage of systematic coordination, leading to waste of resources and overlapping investment to a certain extent. E&D data generally bears the characteristics of big data, such as large volume and multi-source heterogeneity. However, “big volume of data” is not equal to “big data”. At present, oil E&D data standards are inconsistent, data quality is uneven, and data sharing is not realized, which together leads to the lack of data foundation for AI applications. Furthermore, AI application scenarios are not clear and systematic, its advancement goals and technical routes are not clear, and the key fundamental theories and technical equipment for “oil and gas + intelligence” are absent. Therefore, in AI applications, how to reconstruct the management process and give play to the facilitating effect of artificial intelligence on improving quality, increasing efficiency and reducing cost will be a great challenge confronted by enterprises in the future.

3.2. Development direction of AI application

AI technology will certainly provide new momentum for scientific breakthroughs in the whole industrial chain of oil and gas. Considering the demand of petroleum E&D and the status of research on AI technology, the future development direction will cover the following three aspects:

(1) Intelligent equipment. With the continuous successful integration of deep learning, natural language processing, speech recognition, reinforcement learning and other technologies into robots, industrial robots become gradually matured. More and more oil companies have been using robots in place of human beings for dangerous operations. Up to now robots have been successfully used in pipeline patrol, deep-water work, and other high-risk operations. UAV technology is gradually put into application in E&D operations, especially in the sector of geophysical exploration, where UAVs can conduct geological detection, data acquisition, video surveillance, material delivery, engineering rescue and so on. Meanwhile, thanks to the implantation of professional software into the equipment, the equipment becomes increasingly more intelligent. In the future, intelligent equipment, with Internet of Things, machine vision, deep learning and other technologies embedded, will greatly reduce operation costs and boost operation efficiency.

(2) Automatic data processing and interpretation. The application of data mining and mathematical statistics in petroleum E&D is quite successful, demonstrated by the wide uses in logging curve interpretation, reservoir parameter prediction and other work. In recent years, with the continuous advancement of deep learning, ensemble learning, transfer learning and other technologies, prominent superiority of artificial intelligence in image processing, analysis and prediction has been manifesting. Deep learning, ensemble learning, transfer learning, strong memory learning and other technologies are expected to get in-depth application in automatic processing and analysis of petro-physics, seismic images, logging curves, digital core, production operation and other data in the future.

(3) Professional software platforms. In terms of petroleum E&D, professional software and information system for oil and gas are the vehicle and core of AI technology. Professional software is the most important research instrument, the essential accomplishment of experts’ wisdom, and the core competitiveness of oil companies and service companies. With the application of AI algorithms in automatic data acquisition, and intelligent processing and analysis, some professional software packages use machine learning, machine vision, data mining and other algorithms to further boost the level of intelligence, and are committed to the realization of collaborative research on the basis of data sharing. Professional software packages such as Petrel, Techlog, and Eclipse continuously absorb AI technology to be more intelligent and realize the integration of simulation and design. In the future, R&D of AI technology will be intensified pertinent to the existing well-known professional software packages in the industry, making them more intelligent, and concurrently, giving birth to some new professional software packages to satisfy various demands.

3.3. Development focuses of AI application

The application of artificial intelligence should be gradually promoted by using the successful experiences of individual cases to promote dissemination. Aiming at the requirements of E&D operations, the development focuses of the application in the future shall include intelligent basins, intelligent logging, intelligent geophysical prospection, intelligent D&C, intelligent fracturing, intelligent oil recovery, etc. The priorities in the next five years shall include digital basins, swift intelligent image-logging tools, intelligent nodal seismic acquisition systems, intelligent rotary steering, intelligent fracturing technology and equipment, real-time monitoring for zonal injection and production, etc.

With the help of the concept of "Digital Earth" in 1998, the overseas petroleum industry has fabricated many digital basins. However, there is no unified mode or standard for digital basin fabrication in China. There are many theoretical researches and few practical applications. In the next five years, by virtue of big data and AI technology, and based on the E&D achievements of mature basins home and abroad, Chinese professionals will analyze the whole E&D life-cycle, in order to establish an intelligent decision-making system for exploration, guide the prediction of spatial distribution of remaining quality oil and gas resources, and clearly define the exploration focus and objectives.

For intelligent logging, the overseas Scanner 3D scanning imaging series are all-inclusive and widely used. Domestically, EILog, the homemade swift imaging system is applied at large-scale, and there have been prototypes of whole domain imaging and imaging while drilling; but there is still a gap in comparison with the overseas counterparts in terms of stability, reliability, practicality and the like, and the AI systems cannot meet the needs of industrial scale application. The future advancement shall be focusing on R&D of stable and reliable swift intelligent imaging logging tools and large-scale application of them, and improve the products to catch up the world-class level.

For intelligent geophysical exploration, the acquisition technology with strong band channel, low cost, wide band and high efficiency is the key to achieve high-precision geophysical exploration. At present, the nodal acquisition system, both home and abroad, is based on blind acquisition of local storage, and uses analog circuit geophones, resulting in a limited frequency band. The future advancement shall be focusing on the construction of digital nodal acquisition systems and integrated acquisition systems (combining vibroseis and electronics), to fabricate intelligent nodal acquisition systems with ~1000 000 channels for onshore applications and systems for deep sea (~1000 m) applications.

For intelligent D&C, multi-size steering tools with various steering modes and various angle-building capacities are well developed overseas, which satisfy D&C operations in complex geology and harsh engineering conditions. The tools are applicable in large-scale operations, proving to be the “chip” technology for shale oil and gas development. Domestic products do not yet meet the needs of scaled industrial application in terms of stability, reliability, practicability and service life. The future development focus shall be R&D of intelligent rotary steering MWD technology and equipment with the advantages like high angle-building capacity, high ROP, high trajectory control accuracy and high operation reliability.

For intelligent fracturing, there is a gap between domestic and overseas intelligent electric-drive fracturing equipment. The maximum delivery capacity of the overseas Model 2500 fracturing pump truck is 4.9 m3/min; but that of the domestic counterpart is only 2.8 m3/min, which can neither meet the requirements of high-intensity fracturing operations with high pressure, great delivery, high proppant concentration and continuous operations in unconventional exploitation, nor of the operations in mountainous regions and loess plateaus in China. The future development focus shall be R&D of a whole set of high-power electric-drive fracturing equipment, intelligent life cycle management systems and smart fracturing systems, so as to perform “small volume, high- power and intelligent” fracturing operations.

For intelligent oil recovery, domestic oil fields are mainly developed by water flooding, and this technology is advanced in comparison to the overseas counterpart. Due to the strong heterogeneity of terrestrial sedimentary layered reservoirs, the overall recovery is low up to the high water cut stage. The implementation of fine and intelligent zonal injection and production is an important approach to enhance the recovery. The future development focus shall be R&D of a series of real-time monitoring and control technologies for intelligent zonal injection and production, as well as intelligent integrated reservoir-engineering optimization systems.

3.4. Suggestions on AI development and application

Short-term and long-term strategies shall be combined together, and the successful experiences of individual cases shall be used to promote dissemination in the application of artificial intelligence. Top-level design, data management, deployment of R&D forces, talent training, value promotion and other aspects should be put into overall consideration to achieve the collaborative innovation. The short-term strategy is to strengthen the understanding and catch-up learning, especially for the managements at all levels. In order to build a scientific and perfect AI application environment, we should focus on business application, fundamental research, gradual dissemination of individual successes, and establishment of supporting management systems. Specifically, the following aspects shall be involved:

(1) Strengthening top-level design. At the perspective of the whole industry, academicians, senior managers and experts should jointly make an official proposal, to convince major oil companies to unify the understanding and coordinate efficiently, and give full play to the institutional advantages of our country with the socialist market economy; at the perspective of enterprises, company managements should adhere to business orientation, problem orientation, and goal orientation, and conduct integrated design, integrated organization, and integrated dissemination, to smooth data flow, reconstruct business process, and realize innovation, revolution and transformation of the management mode; and finally at the perspective of disciplines, we should pay equal attention to software and hardware, be guided by application, and facilitate mutual promotion of research and application.

(2) Boosting data management. “Big volume of data” is not equal to “big data”. Standard or normative data and sample library are the basis of AI application. Data management should be thought of as the first priority in AI application. We have to unify data labeling, promote data interoperability, and strengthen data management, hence further establishing a data trust mechanism and management mode, and booting the normalization and compliance of data sharing.

(3) Enhancing talent training. Currently, AI algorithm engineers and oilfield engineers can neither communicate nor understand each other smoothly. In addition, in the transformation from digitalization to intelligentization, the problem of "building more, but using less" exists in varying degrees. At the same time, it is difficult and takes a long time to cultivate interdisciplinary talents because the two sectors, petroleum E&D and artificial intelligence, both involve a wide range of disciplines. We should strengthen the in-depth cooperations between colleges and petroleum enterprises, and between petroleum enterprises and IT enterprises to cultivate the talents

(4) Promoting collaboration and sharing. We should try to establish innovation consortiums of “cross-industry, cross-enterprise, cross-discipline” to promote the diversified fusion between oil enterprises, between oil enterprises and IT enterprises, and between different disciplines, so as to fabricate a perfect AI technology R&D system for China's petroleum industry.

(5) Securing IPR on algorithm. Upon experiencing informatization construction, the massive data, which has been accumulated and is continuously generated all the time in the oil and gas industry, is basically manageable. Besides, the network and nodes support the computing power to some extent. It is necessary to launch a research campaign on the essential algorithm, and to secure an algorithm system with independent intellectual property rights, so as to provide fundamental support for the intelligentization of the oil industry.

4. Concluding remarks

The application of artificial intelligence in the sector of petroleum E&D has just got on track and has not yet delivered disruptive achievements, but it has exhibited great potential. The results delivered can be summarized into three aspects: First, preliminary application of intelligent equipment. UAVs and robots are used to replace human beings for patrolling, covering pipelines, unattended platforms and other occasions. Secondly, application of big data, machine learning and other IT technologies in data processing and analysis for exploration and development. Unfortunately, currently in most cases there are only “sparse flowers” instead of a “blossoming garden”. And thirdly, most enterprises have been aware of the importance of data sharing, and initialized the R&D of integrated analysis platforms, integrated software, etc.

The application of artificial intelligence in the sector of petroleum E&D has been mainly concentrated on logging processing and interpretation (such as lithology identification, curve reconstruction, etc.), seismic processing and interpretation (such as first break picking, fault identification, etc.), real-time control of waterflooding development, and production prediction. The application of intelligent algorithms has boosted the intelligent level of integrated analysis software, and the embedded chips have offered intelligent equipment. The mapping relationship between the input and output for the algorithm has to be clear and unambiguous because AI algorithms rely on big data. However the subsurface conditions of reservoirs are complex and diversified, and E&D operations are bothered with problems of multiple solutions, small samples, etc., making it highly difficult to disseminate the application of artificial intelligence. Application of artificial intelligence in petroleum E&D should be gradually disseminated, from individuals to industry-wide scale, instead of being attempted at full-scale right now. The future development of artificial intelligence in the sector shall be focused on technologies such as digital basins, swift intelligent imaging logging tools, and real-time monitoring of zonal injection and production.

Reference

MCCARTHY J, MINSKY M L, ROCHESTER N, et al.

A proposal for the Dartmouth summer research project on artificial intelligence

AI Magazine, 2006,27(4):12-14.

[Cited within: 1]

ZHANG Dongxiao, CHEN Yuntian, MENG Jin.

Synthetic well logs generation via Recurrent Neural Networks

Petroleum Exploration and Development, 2018,45(4):598-607.

[Cited within: 1]

JIANG Kai, WANG Shoudong, HU Yongjing, et al.

Lithology identification model by well logging based on boosting tree algorithm

Well Logging Technology, 2018,42(4):395-400.

[Cited within: 1]

REN Yili, GONG Renbin, FENG Zhou, et al.

Valuable data extraction for resistivity imaging logging interpretation

Tsinghua Science and Technology, 2020,25(2):281-293.

DOI:10.1109/TST.5971803      URL     [Cited within: 1]

WU X, LIAN, L, SHI Y, et al.

Deep learning for local seismic image processing: Fault detection, structure-oriented smoothing with edge- preserving, and slope estimation by using a single convolutional neural network

San Antonio: 2019 SEG Annual Meeting, 2019.

[Cited within: 1]

ZHANG H, LIU Y, ZHANG Y, et al.

Automatic seismic facies interpretation based on an enhanced encoder-decoder structure

San Antonio: 2019 SEG Annual Meeting, 2019.

[Cited within: 1]

PHAN S, SEN M.

Deep learning with cross-shape deep Boltzmann machine for pre-stack inversion problem

San Antonio: 2019 SEG Annual Meeting, 2019.

[Cited within: 1]

MA Y, CAO S, RECTOR J W, et al.

Automatic first arrival picking for borehole seismic data using a pixel-level network

San Antonio: 2019 SEG Annual Meeting, 2019.

[Cited within: 1]

JIA Deli, LIU He, ZHANG Jiqun, et al.

Data-driven optimization for fine water injection in a mature oil field

Petroleum Exploration and Development, 2020,47(3):629-636.

[Cited within: 1]

JIA Hu, DENG Lihui.

Oil reservoir water flooding flowing area identification based on the method of streamline clustering artificial intelligence

Petroleum Exploration and Development, 2018,45(2):312-319.

DOI:10.1016/S1876-3804(18)30034-X      URL     [Cited within: 1]

WANG Hongliang, MU Longxin, SHI Fugeng, et al.

Production prediction at ultra-high water cut stage via Recurrent Neural Network

Petroleum Exploration and Development, 2020,47(5):1009-1015.

[Cited within: 1]

KUBOTA L K, REINERT D.

Machine learning forecasts oil rate in mature onshore field jointly driven by water and steam injection

SPE 196152-MS, 2019.

[Cited within: 1]

BAO Anqi, GILDIN E, HUANG Jianhua, et al.

Data-driven end-to-end production prediction of oil reservoirs by EnKF-enhanced Recurrent Neural Networks

SPE 199005-MS, 2020.

[Cited within: 1]

TARIQ Z, ABDULRAHEEM A.

An artificial intelligence approach to predict the water saturation in carbonate reservoir rocks

SPE 195804-MS, 2019.

[Cited within: 1]

SHAHKARAMI A, MOHAGHEGH S.

Applications of smart proxies for subsurface modeling

Petroleum Exploration and Development, 2020,47(2):372-382.

[Cited within: 1]

ARTUN E, KULGA B.

Selection of candidate wells for re-fracturing in tight gas sand reservoirs using fuzzy inference

Petroleum Exploration and Development, 2020,47(2):383-389.

DOI:10.1016/S1876-3804(20)60055-6      URL     [Cited within: 1]

SENGEL A, TURKARSLAN G.

Assisted history matching of a highly heterogeneous carbonate reservoir using hydraulic flow units and artificial neural networks

SPE 200541-MS, 2020.

[Cited within: 1]

ZHANG Jian.

Development of automated neuro-simulation protocols for pressure and rate transient analysis applications

University Park, PA, USA: The Pennsylvania State University, 2017.

[Cited within: 1]

COSTA L A N, MASCHIO C, SCHIOZER D J.

Application of artificial neural networks in a history matching process

Journal of Petroleum Science and Engineering, 2014,123:30-45.

DOI:10.1016/j.petrol.2014.06.004      URL     [Cited within: 1]

Reservoir simulation is an important tool for reservoir studies because it enables the testing of production strategies and to perform forecasts. To obtain a reliable production prediction, the reservoir model must reproduce results similar to the observed data. This is accomplished through a history matching process, which basically consists of modifying the reservoir parameters until this condition is reached. Usually the process is complex, demanding great time and computational effort and, thus, has been the object of several studies, such as the use of proxy models to substitute the flow simulator in some stages of the history matching process to reduce the number of simulations required to achieve an acceptable match. In this work, the application of proxy models generated through Artificial Neural Networks (ANN) as a substitute for the flow simulator in the history matching process was assessed, showing that the ANN can efficiently capture the nonlinearities of the problems. A synthetic reservoir with real characteristics was used to test the methodology. The results showed that the application of the ANN as a proxy model is promising and that a good match can be achieved with fewer simulations. (C) 2014 Elsevier B.V.

/