- Home
- Conferences
- Conference Proceedings
- Conferences
Second EAGE Digitalization Conference and Exhibition
- Conference date: March 23-25, 2022
- Location: Vienna, Austria
- Published: 23 March 2022
21 - 40 of 72 results
-
-
3D Two-Phase Computational Fluid Dynamics Modelling on Digital Models of Low-Permeable Rocks
Authors V. Krutko and A. AvdoninSummaryWe demonstrate the instruments of digital rock analysis and simulations developed specially for to low-permeable rocks. The results of the multiclass-multimineral 3D digital rock model construction workflow based on 3D x-ray mictomography, scanning electron microscopy (including focused ion beam annealing) and electron microbrobe analyzing are demonstrated. Pore-scale one- and two-phase immiscible flow simulations with and without influence of the unresolved microporosity have been performed on 3D multiclass-multimineral models of clastic rocks of the Achimov formation (Western Siberia, Russia). The results of the simulations reveal significant impact of the unresolved microporosity mineral composition on the resultant absolute and relative permeabilities calculations.
The developed algorithms of the 3D multiclass-multimineral rock models construction and the one- and two-phase fluid dynamics simulations with account of the subresolved microporosity have lied in the base of the integrated approach of the low-permeable Achimov formation rocks petrophysical analysis
-
-
-
An Optimized Injection Scenario in a Waterflood Increases Oil Production and Reduces Carbon Emission Intensity
Authors C. Calad, P. Sarma, J. Rafiee and F. GutierrezSummaryDigital Transformation technology application results in reduction in Carbon Intensity.
The paper introduces the application of a novel platform designed to Measure, Monitor, Model and Optimize Carbon Intensity leveraging the work of Stanford University and CARB (California Air Resources Board) on OPGEE (Oil Production Greenhouse gas Emissions Estimator). The tool address several limitations of OPGEE enabling collaboration, traceability and audit-ability and data integration from several sources and was used to estimate the current and future carbon emissions intensity in a mature field under waterflood. The field was then subject to optimization using a methodology than combines AI and reservoir modeling physics known as Data Physics and a target scenario to increase production by 7% was selected. The calculated emissions of this optimized scenario resulted in a reduction of 10% of the carbon intensity.
-
-
-
The proof is in the eating - tasting the future exploration workflow using ML
Authors D. Stoddart, A. Kvalheim, B. Alaei and D. OikonomouSummaryThe application of a ML driven workflow to hydrocarbon exploration has lead to the identification of additional hydrocarbon accumulations close to existing production infrastructure and has acted as an exploration accelerator.
-
-
-
Embedded Model Estimator (EMBER) Machine Learning Application in Complex Fluvial-Deltaic Reservoirs, Handil Field, East Kalimantan, Indonesia
Authors L. Nugrahadin, S. Courtade, A. Suardiputra, E.S. Erriyantoro, D.B. Prabowo and J.S. TrikukuhSummaryThis study presents comparative results of 3D reservoir property modeling done using the classic geostatistics method and machine learning approach (EMBER). Both of these methods were applied in complex fluvial-deltaic reservoirs in Handil Field, East Kalimantan, Indonesia. The results from each method were then cross-checked using blind well scenario in order to determine which can provide a better approach in terms of accuracy, robustness and time efficiency.
-
-
-
A comparison of deep learning paradigms for seismic data interpolation
Authors M. Fernandez, R. Durall, N. Ettrich, M. Delescluse, A. Rabaute and J. KeuperSummarySeismic data has often missing traces due to technical acquisition or economical constraints. A compete dataset is crucial in several processing and inversion techniques. Deep learning algorithms, based on convolutional neural networks (CNNs), have shown alternative solutions that overcome limitation of traditional interpolation methods e.g. data regularity, linearity assumption, etc. There are two different paradigms of CNN methods for seismic interpolation. The first one, so-called deep prior interpolation (DPI), trains a CNN to map random noise to a complete seismic image using only the decimated image itself. The second one, referred as standard deep learning method, trains a CNN to map a decimated seismic image into a complete one using a dataset of complete and artificially decimated images. Within this research, we systematically compare the performance of both methods for different quantities of regular and irregular missing traces using 4 datasets. We evaluate the results of both methods using 5 well-known metrics. We found that DPI method performs better than the standard method if the percentage of missing traces is low (10%) and otherwise if the level of decimation is high (50%).
-
-
-
Interpretability of geological relationships within fluid flow simulations
Authors M. Subbotina, A. Voskresenskiy and N. BukhanovSummaryReservoir simulation based on numerical solvers of fluid flow within porous media remains the main approach for reservoir management and decision-making process in exploration and production. It is important to develop potential approaches to handle reservoir complexity along with the ability to make rapid and reliable forecasts, based on graphical networks for example. Our approach of delineating the relevant dependencies between parameters of conventional workflow using explainable AI methods is one step along this way. We have implemented hierarchical workflow proposed by authors of Watt benchmark field and have used explanatory and causal AI methods to establish relationships between different uncertainty sources: top structure, fault models, fault transmissibility and relative permeability.
-
-
-
AI-driven 2D conceptual geological modeling
Authors K. Chirkunov, S. Zaytsev, Z. Filippova, A. Shohin, A. Gorelova and O. PopovaSummaryExploiting new modern AI tools and methods for geological modeling might help to achieve a new level of decision quality in oil exploration. In our work we developed new automated pipeline for building 2D conceptual model as set of alternative scenarios, which takes as input low-level geological and geophysical properties, using as data source filter-and-capacity properties and seismic spatial properties over a surface, and transforms it to top-level concepts by applying DLNN and probabilistic reasoning, taking into account non-trivial spatial dependencies.
-
-
-
An automatic pipeline for 3D seismic geometry QC using Convolutional Neural Networks
Authors D. Semin, N. Kalashnikov, D. Podvyaznikov and A. KuvaevSummaryWe have developed an approach to automate the QC of the geometry assignment of seismic data using CNN. The developed approach allows to reduce the time spent by a specialist at the QC stage of geometry by up to 10 times
-
-
-
Ai Driven Core Description Using White Light Images and Sedimentological/Coredna Data
Authors W. Hujer, J. Peisker, K. Dasgupta, C. Germay, A. Metz and T. KuffnerSummaryDrill cores obtained from oil and gas wells are one of the three pillars of subsurface information (others being seismic and well-log data) available in the E&P industry. Cores plus core analyses will furthermore be irreplaceable for subsurface projects in course of the energy transition. The E&P industry as a whole is becoming more and more digital and automatized, using increasingly artificial intelligence (AI) and machine learning (ML). Therefore, traditional techniques of core description must adapt to fit into the new technology landscape. One opportunity to utilize AI is for reducing time and increasing automatization of core descriptions and interpretations. The principle usability of image recognition methods using AI has been proven in several publications (Pires de Lima et al. 2019; Baraboshkin et al. 2020). However, the dependency only on one data domain ignores the vast amounts of usually existing additional measurements. Therefore, we selected an approach where core image data is combined with a suite of core analyses (sedimentology, geochemical, mineralogical, geomechanical). Aim of our approach is to develop an automatized ML based workflow that delivers a core description including lithology, texture, physical rock properties and lithofacies.
-
-
-
Cloud-Based Exploration Collaboration Tools for Managing Nation-wide Upstream Oil and Gas Exploration Activities
Authors S. Damayanti, S.E. Saputra, Andre, C. Nurwani, F.T. Amir, M.Z. Rubianto, T. Wibowo, R. Affandi, F.F. Rahmawati, S. Sitompul, R. Marbun, R. Musa and W. SeptiningrumSummarySKK Migas, the Indonesia Oil and Gas Regulatory Body, has set the ambitious goal to achieve 1 million BOPD oil and 12 Bscf/D gas production in 2030. To achieve this, one of the aggressive transformation plans is to enhance exploration activity to find the next giant discoveries. A cloud-based exploration planning solution is seen as the best solution to execute this plan swiftly and efficiently. The exploration planning solution powered by artificial intelligence/machine learning (AI/ML) capabilities enables high-quality exploration planning in less time with better-informed decision-making through an open, collaborative, cloud-based environment that enables transparency and focus on shared business objectives. For operators, this cloud-based exploration planning solution will help them streamline the exploration planning process and allow them to be more agile in adapting to market changes, achieving better efficiency and larger returns on investment in future exploration. For SKK Migas, better exploration planning means that a consistent approach is applied across the entire exploration portfolio. This enables SKK Migas to carry out their responsibilities in managing and monitoring all exploration projects or activities more effectively. The first stage of the cloud-based exploration planning solution has successfully been implemented for three basins (North East Java, Kutai, and South Sumatra).
-
-
-
Expert System for Selection of Technical Solutions for Development of Western Siberia Low Permeable Turbidite Reservoirs.
SummaryThe paper represents the results of expert system MVP for technology screening and approaches to reservoir development, drilling, completion, reservoir stimulation. The work considers low permeable turbidite reservoirs within the West Siberia region of Russia. On current stage, the following tusks were in scope:
- Clustering of objects based on reservoir properties, PVT, sedimentation features.
- Reservoir development analysis.
- Data base gathering.
- Prototyping of clustering module.
- Uncertainty calculations.
- Prototyping of technology screening module.
-
-
-
Reservoir Complexity Index evaluation for the Pannonian Basin oil fields
Authors M. Naugolnov, M. Pilipenko and S. PerunicicSummaryIn this paper the authors described the experience of Reservoir Complexity Index evaluation and adaptation it for the Pannonian field conditions.
-
-
-
Prediction of Missing Compressional Sonic logs using Ensemble Learning technique in Gandhar oilfield, Cambay basin, India.
Authors S. Dabi, S. Zia and N. VedantiSummaryReservoir characterization requires the assimilation of different types of data to extract the subsurface rock properties. To integrate multiple geophysical well log types into reservoir characterization studies, the prediction of missing well logs is a crucial step. We have deployed a machine learning ensemble technique called Gradient Boosting Regression (GBR) to estimate missing sonic logs in Gandhar oilfield, Cambay Basin, India. Nevertheless, an extremely complicated relationship exists between the non-linear input and output well logs due to heterogeneity and complex conditions in the reservoir, creating bias and variance problems. The GBR has unique functionality to reduce bias and variance problems by converting weak learners to strong ones. Also, the GBR can optimize on different loss functions and provides several hyperparameter tuning options that make the function fit very flexible. Thus, this study deploys the GBR algorithm to the complex well log datasets of the Gandhar region. We utilized the input training features as neutron porosity, gamma-ray, density, and resistivity logs of multiple wells, which correlate better with the observed data. Also, we correlated prediction with the conventional PNN method to check the reliability of models over complex datasets. Eventually, the P-sonic predictions were made in 3 blind wells.
-
-
-
UAV Technologies: from exploration of deposits to digital twins
Authors V. Gulin, G. Grigoriev, A. Nikitin and N. SivoySummaryUnmanned aerial vehicles (UAV) have a great potential for geological exploration optimization at all stages. This study considers UAV implementation at different exploration stage. Integrated approach using unmanned aerial systems shows great effectiveness based on the completed surveys. Low-depth electrical exploration using the shallow electrical exploration method is one of the possible UAVs technologies with great potential. In addition, UAV are actively developing in the field of monitoring work on a deposit, allowing us to create digital twins of processes. In this study there are several cases describing main field data, processed models and cross-sections.
-
-
-
Digital twin: Mobile Projects Management
By A. GavrilovSummaryObjectives/Areas of application: Mobile Office is a unified digital platform that is a set of interconnected tools for operational management of geological exploration projects, and available for smartphones, personal computers, and tablets. The platform is designed to provide a unified environment for bringing together geographically distributed project teams, consisting of representatives of Gazprom Neft and their business partners. We see good prospects for using the Mobile Office platform by any project teams that value transmission data speed, convenience, immersion, and security.
-
-
-
One-Shot learning for seismic fault identification
Authors P. Morozov, R. Miftakhov, I. Efremov, A. Bazanov and P. AvdeevSummaryWe considered fault detection as a binary segmentation problem using Convolutional Neural Networks (CNN). We proposed a pre-trained CNN on a vast corps of synthetically generated data which can be adapted to any seismic by providing a few slices of manually interpreted data.
In this paper, we described the theoretical background on the neural network and provide an example of shot-learning for fault delineation on a real seismic from the Adele Field on the NW Shelf of Australia.
-
-
-
Aggregation and Assimilation of Complex Laboratory Data to Fuel Corporate Innovation
By C. HantonSummaryAdoption of a digital-first business model is seen as a key driver of increased innovation and efficiency in upstream oil and gas. However, the path to increased efficiency requires redesigning and overhauling processes which have existed in place for decades.
A key common driver for digital transformation is the promotion of actionable and accessible subsurface data. Providing a consistent, open, and trusted platform of information allows engineers and sub-surface scientists to problem solve and generate innovative new workflows without being constrained by an imperfect data ecosystem.
This paper details how, by working with subject matter experts at a major European operator, a rich and varied, yet historically under-utilized dataset was able to power new insights into subsurface understanding. Key aspects of this project focused on framing the business problem, developing the underlying technical processes to overcome the problem, and deploying the technical solution to allow increased access and actioning of this data type.
Savings were implemented through the retirement of legacy systems and infrastructure, with earnings generated through increased speed of access to information higher levels of data trust.
-
-
-
Capturing Tabular Data in Documents Related to the Mining Domain.
By H. BlondelleSummaryThe mining companies store millions of documents on file systems. Unfortunately, the information contained in the documents is frequently locked in unstructured formats as PDF or graphic scans. A major mining company has trained an AI solution to capture assay tables from their subsurface documents.
The presentation details the open-source technology developed to train an hybrid model able to detect and to segment assay tables at scale in subsurface documents. It also describes the way this technology has been implemented to solve this important mining challenge.
According to the project manager, accessing with confidence massive amount of legacy geochemical data reduces the number of necessary drill holes and trenches, hence impacts the cost and environmental aspects of ore exploration.
-
-
-
Highly Robust Supervised Deep Learning-Based Model for Faster History Matching and Forecasting
Authors M. Pal, P. Patil, J. Chahar, U. Kumar and L. VeščiūnasSummaryAdvances in new technologies particularly artificial intelligence (AI) and machine learning (ML) have now made it possible to use AI and ML-based approaches for building reservoir simulation models, which do not rely on conventional simulation tools for history matching and production forecasting. These new methods have the capabilities to improve speed, efficiency and potential to eventually will replace numerical reservoir simulators. The advantages of AI and ML-based simulators are significant time and money savings to achieve statistically accurate and physics-assisted history matching and forecasting results.
AI and ML-based deep learning approach involve using subsurface reservoir static, dynamic, and well production data to form a fully independent deep learning-based reservoir simulator with capabilities to perform history matching and production forecasting. The results show that AL and ML-based deep learning algorithm-based models are very good and give close to 85% accuracy in history matching well patterns. Models are evaluated on 3 different industrial datasets from the middle east and the North Sea fields. The models are also extended to generate long-term forecasts. The work printed in this paper will demonstrate that AI and ML-based models have the potential to replace the conventional reservoir simulation workflow if an exhaustive data set is available.
-
-
-
Logging curve prediction based on regression analysis
More LessSummaryIn oil and gas exploration, it is often necessary to study the relationship between two or more variables, in view of this, regression analysis uses one or more (independent) variables to determine another (dependent) variable, so using regression analysis methods can solve most of the problems we encounter. Using different regression analysis models, we predicted log curve values in geophysics by sampling in geochemical logging. For multiple linear regression models, we used the following two linear regression methods: standard equation (SE) and ridge regression (RR); for nonlinear regression models, we used the gradient descent (GD), kernel ridge regression (KRR) and support vector regression (SVR). The KRR achieved satisfactory results in terms of predictive effect on the porosity log curve of Well A when predicting the log curve. Finally, our results show that the regression analysis method is highly scientific and comprehensive, and it has wider application prospects than traditional methods.
-