- Home
- Conferences
- Conference Proceedings
- Conferences
EAGE/AAPG Digital Subsurface for Asia Pacific Conference
- Conference date: September 7-10, 2020
- Location: Kuala Lumpur, Malaysia
- Published: 07 September 2020
1 - 20 of 23 results
-
-
An integrated solution on cloud
Authors Farhan Ahmed Khan and Nik Nur Halim C SohSummaryIn the last decade, high performance computing (HPC) cloud has shown rapid growth in several industries. The cloud is known for its high-performance computing in handling of big data volume. The oil & gas industry is mostly using HPC for seismic imaging and reservoir simulation. Here we present the results of a pilot project to employ cloud as a common platform for exploration activities by venture team that resides at several locations on the globe and showcase special or prospect specific processing on cloud to cater the needs of business.
-
-
-
Computational approach for data-driven mining of upstream value-added structures
Authors N. Bukhanov, P. Trudkov, E. Vinogradova, I. Derevitskii, K. Balabaeva, A. Funkner and S. KovalchukSummaryAdoption of digital advances, machine learning and data science approaches opens up a range of opportunities to describe structure of industry processes and optimize their performance based on chosen objective function. We propose a computational approach which allow to derive processes interconnections structure from different corporate information storages.
-
-
-
Well placement strategies evaluation based on exploration game challenge
Authors E. Grishnyaev, Y. Paveleva, N. Klimenko, A. Volkova, N. Bukhanov, A. Orlov and M. OzhgibesovSummaryWe propose a method for well placement inspired by recent achievements in machine learning and data science which is able to offer options for well positions based on variational forecast with existing data perturbations. To benchmark our approach and compare it with expert choices in simplified and time constrained manner we summarize and evaluate in this paper the results of recent exploration game challenge and critically review the pros and cons of different well placement strategies.
-
-
-
Reducing turnaround by automating velocity model building using a big data approach
Authors T. Martin, M. Bell and *T. MassipSummaryMost steps in seismic processing require manual input, repetitive testing and significant quality control. Additionally, key steps like velocity model building for depth imaging can be time-consuming using workflows that are ‘stop-go’ linear processes. If model building can be automated, project timelines could be significantly reduced, enabling better decision making during evaluation, planning and development. Probability simulations, such as Monte Carlo methods use random sampling to resolve problems where the solution may be mixed or under determined. When using this type of ‘big data’ approach for velocity model building we need to understand how the data quality impacts the model. Following this, we create a population of models to solve with inverse theory. Using a simple statistics driven reinforcement loop for each population of inverted models enables automation for numerous repeated cycles. This continues until the data accurately converges to a global solution. The size of the population in each pass is data dependent, but over the lifecycle of the model building will be several orders of magnitude greater than conventional model building. This method transforms the model build from a
-
-
-
A Novel Subsurface Characterization Algorithm Parameterized in Terms Familiar to Geoscientists
Authors R. Ross, A. Jakobsen and H. HansenSummaryA novel algorithm for subsurface characterisation is presented. The algorithm shares many of the benefits of machine learning approaches but it has the advantage of being parameterised in terms familiar to a geoscientist.
-
-
-
Uncertainty quantification for deep learning in geoscience applications
Authors L. Mosser, S. Purves and E. Zabihi NaeiniSummaryInterpretability and uncertainty quantification is crucial in order to build machine learning models that are robust. One could argue there are at least two aspects where quantifying the uncertainty would help the practitioners. Firstly, it is useful to highlight where the machine learning model needs help, e.g. during the training process, so it has a better prediction accuracy. Secondly, it helps the practitioners to make an informed decision using the final predictions and the accompanying uncertainty measures. In this paper, we will show two geoscience applications, automatic fault prediction and 3D reservoir property prediction, where deep learning has been deployed and the corresponding uncertainty has been captured.
-
-
-
Missed pay identification using machine learning - a case study
SummaryThis paper demonstrates a methodology and workflow for the rapid identification of missed pay zones throughout many thousands of wells and crucially provides actual examples of missed pay in wells from the North Sea. Importantly, the methodology applied in this study and lessons learned are applicable to other basins and/or sub-basins throughout the world.
-
-
-
A semi-supervised learning framework for seismic acoustic impedance estimation
Authors H. Di, X. Chen, H. Maniar and A. AbubakarSummaryFor compensating the limited bandwidth in seismic data, one reliable approach for robust acoustic impedance estimation is to integrate 3D seismic data with 1D well logs by building an optimal non-linear mapping function between them. However, most of the existing mapping methods, including these by machine learning, are performed in 1D that utilizes only the single seismic trace corresponding to a well. Therefore, their performance is restricted within a small zone around the wells, while consistent prediction cannot be obtained throughout the entire seismic survey. In addition is the down-sampling of high-resolution well logs to the seismic scale, which fails to fully utilize the information available in the wells. For resolving both limitations, this work presents a semi-supervised learning framework of two components: (1) seismic feature self-learning and (2) seismic-well integration, each of which is formulated as a deep convolutional neural network. The performance of the proposed framework is evaluated through an application to the synthetic SEAM dataset. The good match between the machine prediction and the earth model demonstrates the capability of the proposed semi-supervised learning in reliable seismic and well integration, particularly in the zones of poor seismic signals due to the presence of geologic complexities.
-
-
-
A geological regional case study for pressure, temperature, and salinity for the GoM using machine learning technology on unstructured data
SummaryTraditionally explorationist working on a new area are given a huge amount of the data and will start with a regional study to identify plays and reservoirs on a basin scale. Once opportunities are identified, an area will be selected, and the study will move into a block scale. Drilling locations will then be defined, and further studies will be conducted on existing wells, looking at logs, cores and even samples. In this paper, we are going to investigate how the ingestion of previously conducted studies from unstructured data and Machine Learning (ML) can help to reverse this traditional workflow from basin-to-samples to extract valuable regional geological maps related to pressure, temperature and salinity from samples and production tests themselves.
-
-
-
Seismic Processing Workflow Improvements with AI
Authors Paul Zwartjes, Rolf Baardman and Rob HeggeSummaryThe application of artificial intelligence techniques in seismic data processing holds a promise to significantly reduce project turn-around time. This can be achieved by a reduction of time spent on manual labor-intensive task such as workflow parameterization, quality control of result and picking and editing a variety of attributes (from first breaks on fields gathers to horizons on stacks). In addition, we expect this will lead to an increase in consistency of results between projects and processors.
-
-
-
A case study of fully automated machine learning petrophysical interpretation using unstructured data
More LessSummaryMachine Learning (ML) has become widely used for regional studies and prediction of petrophysical interpretation and facies classification. In recent years, application of ML has shown significant benefit to improve productivity, decision making and success rate for regional exploration campaigns. However, while ML petrophysical prediction and classification provides expert based performance, it relies heavily on curated and validated raw logs for the algorithm to be trained on ( Pham and al., 2019 ). In this paper we are going to present and investigate the broader usage the unstructured data for knowledge extraction and the use of it for better and more reliable ML petrophysical interpretation. This includes (1) The definition of wells as geological analogues and/or outliers using knowledge graph. (2) The automated extraction of regional parameters. (3) Validation of ML prediction using unstructured data.
-
-
-
Deep convolutional autoencoder for unsupervised seismic facies classification
Authors V. Puzyrev and C. EldersSummaryWe present an application of a deep convolutional autoencoder for unsupervised seismic facies classification. This approach is entirely data-driven, requires no labelled data, and provides the results almost instantaneously, thus opening possibilities to analyse geological patterns in real-time without human intervention.
-
-
-
An optimization method for the assisted history matching (AHM) process using the gradient boosting approach
Authors M. Melnikov, G. Shishaev, I. Matveev, G. Eremyan, V. Demyanov, N. Bukhanov and B. BelozerovSummaryIn this article, we propose a machine-learning approach, which is aimed to filter out the redundant reservoir models prior to simulations during the assisted history matching (AHM). The redundant models may be generated during AHM due to arbitrary switch of the flow simulation well control for a particular combination of reservoir model parameters. This aproach allows to save CPU time and increase efficiency of AHM process. Optimization algorithms used in AHM to iterate through possible combinations of model parameters trying to minimize an objective function may lead to unrealistic parameters combinations [ 2 ]. Specifically, some parameters combinations may results in reservoir models automatically switch from under bottom hole pressure control due to insufficient productivity properties of the model. As a result, AHM will generate many simulations, which are initially wrong and have no chance to match the history. We created an approach which is able to classify if the model parameters will lead to switching under bottom hole pressure control or not before running a flow simulations
-
-
-
Running Reservoir Simulations in the public cloud; A case study of a cost-controlled method, running tNavigator and Eclipse in an Azure HPC environment
Authors M. Stordalen Flister and K. HopstakenSummaryRunning HPC workloads in the public cloud is a very good example of how we can utilize available compute power on demand without planning, buying, installing and operating a lot of servers in your own data centers.
-
-
-
Well Log data augmentation influence at accuracy of machine learning interpretation
Authors A. Semenikhin, A. Shchepetnov, A. Karavaev, D. Egorov and O. OsmonalievaSummaryStudy provides a comprehensive analysis of data issues on quality of machine learning model for well logs interpretation problem. This work presents results of data augmentation experiments in order to make model familiar with different types of issues and overall recommendation on its potential impact to final f1-score of the model.
-
-
-
Deep Machine Learning Application for Supervised Facies Classification
More LessSummaryApplications of Machine Learning (ML) algorithms to solve problems in seismic reservoir characterization (SRC) are drawing widespread attention in the last few years. One of the challenging problems to solve is doing facies classification based on seismic amplitude/ Amplitude Versus Offset (AVO) attributes only without employing commonly used results from seismic inversion. The objective of this study is to leverage the new developments in Deep Learning (DL) techniques to perform a supervised facies classification based on available post/pre-stack seismic attributes. This simple ML workflow will help geoscientists perform a reconnaissance of the available seismic data and formulate a program for detailed analysis using more advanced tools. Results of the facies classification were then validated by comparison of the facies volume with Lithofacies logs from both training and blind wells, and additionally compared with pre-stack seismic inversion results.
-
-
-
Case Study: Predicting Missing Recorded Well Logs using Classification and Regression Methodology
Authors R. Thiyagu, M.A. Mohammad Nazmi, A.A. Amirsaman, T. Shaur and M.Y. Mohamad YusmanSummaryWell Logs is been a key data to interpret the reservoir and rock properties. This data used by petrophysicist, geoscientist, reservoir engineers for their respective interpretation, due diligence. Often this well logs have gaps due to drilling, logging operational challenges. Several other factors influence the quality of well log data is tool types & sensitivity, mud types, or geological condition of the subsurface. The team work on a solution during APGCE Geo Hackhathon 2019, on how to predict the missing sections in well logs using classification and regression methodology.
-