- Home
- Conferences
- Conference Proceedings
- Conferences
Petroleum Geostatistics 2019
- Conference date: September 2-6, 2019
- Location: Florence, Italy
- Published: 02 September 2019
21 - 40 of 108 results
-
-
Extraction of Petrophysical Information and Formation Heterogeneity Estimation from Core Photographs by Clustering Algorithms
By D. EgorovSummaryCore data is the most reliable and representative source of information, however it is very expensive. It is the only thing that can be used for precise description of geological situation in a target formation including its depositional environment, petrophysical properties and oil saturation distribution, reservoir heterogeneity and compartmentalization degree induced by secondary processes. Despite the fact that core retrieving operation can increase well cost in few times only a little amount of core information is utilized during field development projects as a huge part of it is represented by qualitative description which is hard to be implemented into digital quantitative geological modeling process.
-
-
-
Amplitude Supported Prospects, Analysis and Predictive Models for Reducing Risk of Geological Success
Authors I. Tishchenko and I. MallinsonSummaryDirect Hydrocarbon Indicators (DHI) are commonly used for exploration prospects. Amplitudes as an independent source of information could be used as conditional probability within Bayes Theorem to assess risk of geological success. Following research is aiming to construct predictive model for estimating probability of hydrocarbons observing DHI, P(dhi|hc). In order to build such model, we used Rose & Associates DHI Interpretation and Risk Analysis Consortium database, which contains extensive descriptions of 336 drilled prospects, with known results, across various categories: Geology, Data Quality, Amplitude Characteristics and Pitfalls. Multiple Logistic Regression was used for predicting probability P(dhi|hc). Three methods were considered within the study: two data-driven models - stepwise regression and lasso shrinkage method plus the third one, a combination of data-and expertise- driven approach - stepwise regression plus manual addition of predictors to the model. All three models with key predictors are described and give similar accuracy of prediction − 77%. Performed data analysis and calculated models reveal several insights into R&A DHI Consortium database and amplitude prospects characterisation. The best method to create such models is probably a combination of data and expertise driven approaches, while selection of most appropriate model is a question of company's strategy.
-
-
-
Application of Long Short-term Memory Algorithm for Prediction of Shale Gas Production in Alberta
More LessSummaryShale resources are developed with high-density drilling and differ to predict future production rate with conventional methods such as decline curve analysis (DCA). We developed a prediction model for shale production using machine learning (ML) algorithm. Long short-term memory algorithm, which is suitable for training time-series data, is applied to more than 300 shale gas wells at Duvernay formation in Alberta. To increase performance of the prediction model, an additional feature, shut-in (SI) period, is introduced for input layer. SI information is easily extracted from production history and it is highly correlated with production rate. The trained model can give reliable and stable prediction of shale gas production rates. Especially, the two-feature model (production and SI period) shows better performance than the one-feature model (production only). Therefore, a domain knowledge for petroleum industry has an important role to build reliable ML model. The developed model can dramatically reduce analysis cost comparing with DCA, which requires expert's judgement for each shale well. Also, this concept can be extended to other conventional and unconventional production wells.
-
-
-
Machine Learning-based Approach for Automated Identification of Produced Water Types from Conventional and Unconventional Reservoirs
Authors P. Birkle, M. Zouch, M. Alzaqebah and M. AlwohaibiSummaryProduced water represents the most commonly recovered fluids in oil and gas operations. They are composed of natural fluids from deep groundwater systems (formation water) or artificial fluids from operations, or a mixture between both fluid components.
-
-
-
Application of a Serial Denoising Autoencoder for Geological Plausibility of a Channelized Reservoir in History Matching
More LessSummaryDenoising autoencoder (DAE) is utilized to preserve and improve geological reality and plausibility in a channelized reservoir model during history matching by ensemble smoother with multiple data assimilation (ES-MDA). As one of history matching methods, ES-MDA calibrates reservoir properties such as rock facies corresponding to production history. While ES-MDA modifies reservoir parameters, it recognizes them only as figures not honoring to geological features. Thus, conservation of geological characteristics during calibration of reservoir parameters is challenging in ES-MDA. DAE is trained to restore lost connectivity and pattern of an original geological concept and it is applied to posterior reservoir models after an assimilation by ES-MDA. ES-MDA combined with DAE shows not only geologically enhanced channel models but also well-matched production prediction.
-
-
-
Automatic Recognition of Lithological Units in Gas-bearing Shale Complex with Neural Networks (the Baltic Basin, Poland)
By K. BobekSummaryThe study was performed to check the possibility of automatic recognition of lithological units of shale complex and testing their heterogeneity in different parts of the Baltic Basin. For this purpose, two separate neural network models have been trained based on geophysical logs datasets: one for perspective units recognition, and second for tuffite interbeds detection. Obtained results have compared with units distinguished by sedimentologists in continuous core profiles. The performance of trained networks has been checked for two boreholes B-7 and B-1 not included in the training process. In case of the borehole B-7, the units predicted by the model match well to those distinguished manually in borehole core. In contrary, the score of model performance for the B-1 borehole was unsatisfactory. Significant differences in the lithological profile of this borehole were also recognized by sedimentologists. Due to the location of borehole B-1 in the marginal part of the basin, most of the properties of the studied units differ relevantly from recognized in other investigated boreholes. The performance of the network trained for tuffite interbeds detection is considered satisfactory. The results imply that the trained models might be applied in the boreholes where the core profiles are modest or non-existent.
-
-
-
Geomodeling Using Generative Adversarial Networks and a Database of Satellite Imagery of Modern River Deltas
Authors E. Nesvold and T. MukerjiSummarySeveral studies on deep generative models for use in geomodeling show encouraging results with binary training data. An important question is what type of training data to use, since realistic 3D geology with natural variability is difficult to create. The advent of multiple types of remote sensing data of subaerial and subaqueous sedimentary patterns provides new possibilities in this context. Here, we train a Wasserstein GAN using 20,000 multispectral satellite images of subsections of 40 modern river deltas. The generated output has three facies and a background facies, and all quantitative evaluation methods of the unconditional output show a close overlap between the model and training data distributions. Standard MCMC sampling conditional on soft and hard data works well as long as the likelihood model is balanced against the prior model. Transfer learning, i.e. fine-training a small subset of the network parameters on smaller dataset of interest, such as highly non-stationary images of river deltas with similar characteristics, also shows promising results.
-
-
-
Ensemble-based Kernel Learning to Handle Rock-physics-model Imperfection in Seismic History Matching: A Real Field Case Study
Authors X. Luo, R. Lorentzen and T. BhaktaSummaryModel imperfection is ubiquitous in geophysical data assimilation problems. A common approach to accounting for model errors is to treat them as random variables following presumed distributions. While such a treatment renders certain algorithmic convenience, its underpinning assumptions may often be invalid in practice. In this study, we adopt an alternative approach, and treat the characterization of model errors as a functional approximation problem, which can be solved using a generic machine learning method, such as kernel-based learning adopted here. To enable a seamless integration of kernel-based learning into ensemble data assimilation, we also develop an ensemble-based kernel learning approach. We show that an existing iterative ensemble smoother can be naturally employed as the learning algorithm, which thus inherits all the practical advantages of ensemble data assimilation algorithms, such as derivative-free, fast implementation, and allowing uncertainty quantification and applications to large scale problems. To demonstrate the efficacy of ensemble-based kernel learning, we apply it to handle model errors in a rock physics model used in history matching real 4D seismic data from the full Norne field. Our experiment results indicate that incorporating kernel-based model error correction into 4D seismic history matching helps improve the qualities of estimated reservoir models, and leads to better forecasts of production data.
-
-
-
3D Geological Image Synthesis from 2D Examples Using Generative Adversarial Networks
Authors G. Coiffier and P. RenardSummaryRecently, Generative Adversarial Networks (GAN) have been proposed as a potential alternative to Multipoint Statistics (MPS) to generate stochastic fields from a large set of training images. A difficulty for all the training image based techniques (including GAN and MPS) is to generate 3D fields when only 2D training data sets are available. In this paper, we introduce a novel approach called Dimension Augmenter GAN (DiAGAN) enabling GANs to generate 3D fields from 2D examples. The method is simple to implement and is based on the introduction of a random cut sampling step between the generator and the discriminator of a standard GAN. Numerical experiments show that the proposed approach provides an efficient solution to this long lasting problem.
-
-
-
Deep Stochastic Inversion
Authors L. Mosser, O. Dubrule and M.J. BluntSummaryNumerous geophysical tasks require the solution of ill-posed inverse problems where we seek to find a distribution of earth models that match observed data such as reflected acoustic waveforms or produced hydrocarbon volumes. We present a framework to create stochastic samples of posterior property distributions for ill-posed inverse problems using a gradient-based approach. The spatial distribution of petrophysical properties is created by a deep generative model and controlled by a set of latent variables. A generative adversarial network (GAN) is used to represent a prior distribution of geological models based on a training set of object-based models. We minimize the mismatch between observed ground-truth data and numerical forward-models of the generator output by first computing gradients of the objective function with respect to grid-block properties and using neural network backpropagation to obtain gradients with respect to the latent variables. Synthetic test cases of acoustic waveform inversion and reservoir history matching are presented. In seismic inversion, we use a Metropolis adjusted Langevin algorithm (MALA) to obtain posterior samples. For both synthetic cases, we show that deep generative models such as GANs can be combined in an end-to-end framework to obtain stochastic solutions to geophysical inverse problems.
-
-
-
Integrated Geo-modelling and Ensemble History Matching of Highly Faulted Turbiditic Reservoir Model
Authors V. Zaccardi, A. Abadpour, N. Haller, P. Berthet, D. Rappin and J. Grange-PraderasSummaryData assimilation methods and multi-realisation geomodelling workflow have been applied to history match a heavily faulted turbidic reservoir. With this approach it was possible to take into account a large number of uncertainties related to seismic data interpretation, geological modelling and dynamic data.
-
-
-
Ensemble History Matching Enhanced with Data Analytics - A Brown Field Study
Authors E. Tolstukhin, E. Barrela, A. Khrulenko, J. Halotel and V. DemyanovSummaryThis paper presents a methodology used in a subsurface uncertainty study for redevelopment of an oil field in the North Sea. A fractured chalk reservoir was depleted for more than 30 years with limited water injection. The uncertainty study aims to find an ensemble of geologically consistent scenarios that would honor production history. The scenarios then serve as input for the redevelopment concept selection, well placement and economic evaluation. The challenge in this study was that the field has long production history that must be respected. In addition, the uncertainty that may not be resolved by HM must be preserved in the scenarios in order to estimate all the risks and capture all the potential associated with the remaining oil pockets and future well targets. For the brown field, it is difficult to analyze all the information and utilize its full potential. In this work we use data analytics can improve efficiency of ensemble history matching by analyzing links between the static and dynamic model ensemble update: screening of the initial ensemble, model localization based on spatial analysis dynamic observations to the parameter update and identification of conflicts between groups of production observations that prevent balanced model update.
-
-
-
Updating MPS Facies Realizations Using the Ensemble Smoother with Multiple Data Assimilation
Authors A. Thenon, A. Abadpour and T. ChugunovaSummaryThis paper is presenting the results of a study focused on the update of facies models generated by multipoint statistics (MPS) within an ensemble history matching workflow. The tested parameterization consists in updating the uniform random numbers used by MPS to generate the facies realizations with an ensemble method as proposed in the work of Hu et al. (2012). The novelty of this study lies in use of the ensemble smoother with multiple data assimilation (ES-MDA) to update the random number realizations. Tests on synthetics cases show that increasing the iteration number within ES-MDA can significantly improve the quality of the history-match achieved by this approach. The workflow is briefly detailed in the first part of the paper. Then, we present and discuss the history-matching results for two synthetic cases: an inverted five-spot 2D model with channels and a more complex 3D model with dunes.
-
-
-
Improving Local History Match Using Machine Learning Generated Regions from Production Response and Geological Parameter Correlations
Authors T. Buckle, R. Hutton, V. Demyanov, D. Arnold, A. Antropov, E. Kharyba, M. Pilipenko and L. StulovSummaryWe present a data driven workflow to improve local history match quality by identifying model regions from correlation between production response and geological modelling parameters for use in an assisted history matching framework. This paper outlines the implementation and results from a large mature field case study. Regions are identified by calculating the partial correlation between individual well production misfits and uncertain geological modelling parameters across 500 models. Wells are then categorised into three groups based on their correlations: positive, negative and insignificant. A probabilistic neural network (PNN) is trained on the location of each well and its group. A map of regions can then be calculated using the PNN. The parameters used to define the region map are then varied separately in each region in an assisted history matching loop. In the full field case study, an 8.8% improvement in oil rate misfit within the positively correlated well group was achieved by regional modification of the net-to-gross multiplier, with no detrimental effect on the other groups match quality. This case study demonstrates the effective identification and utilisation of geologically and dynamically inferred regions which improve the local history match
-
-
-
Fluid Flow Consistent Geostatistical History Matching of an Onshore Reservoir
Authors E. Barrela, L. Azevedo, A. Soares and L. GuerreiroSummaryThis paper shows the application to a real field case of an iterative geostatistical history matching technique, integrating geological and engineering consistency. Current trends reflect a growing interest on developing workflows that simultaneously integrate petrophysical modeling with dynamic calibration of reservoir models to historical production data. Contrary to manual history matching techniques, where model perturbation often disregards geological or physical realism leading to poor production forecast, this example introduces geological consistency through geostatistical simulation and physical realism by using streamline regionalization, while holding the predictive capability of resulting petrophysical models. This is achieved by iteratively updating the reservoir static properties using stochastic sequential simulation and co-simulation, constrained to production data, while using streamline information for electing preponderant flow production regions of the model, focusing property perturbation. In order to capture the complex subsurface heterogeneities of the reservoir, petrophysical property realizations are obtained using the direct sequential simulation and co-simulation with multi-local distribution functions. The location and proportion of reservoir facies is also automatically updated throughout the iterative procedure, using Bayesian Classification. The technique was successfully applied to a real case study, located in North-East onshore Brazil, resulting in multiple history matched models that better reproduce historic data.
-
-
-
Revising the Method of Ensemble Randomized Maximum Likelihood
Authors P.N. Raanes, G. Evensen and A.S. StordalSummaryA popular (iterative ensemble smoother) method of history matching is simplified. An exact relationship between ensemble linearizations (linear regression) and adjoints (analytic derivatives) is established.
-
-
-
Theoretical and Practical Aspects of Stein Variational Gradient Descent with Applications to Data Assimilation
Authors A. Storksen Stordal, R.J. Moraes, P. Nima Raanes and G. EvensenSummaryWe have introduced and applied the Stein variational gradient descent (SVGD) algorithm to the reservoir history matching problem. The method has been extended with a Monte Carlo approximation of the gradient of the measurement function using the reproducible property of the reproducing kernel Hilbert space in order to avoid adjoint implementation. Furthermore we have implemented SVGD with p-kernels in order to extend the applicability to higher dimension than the standard Gaussian kernels used in the literature. The gradient approximation was validated both theoretically and on a toy model. Implementation on larger problems will be addressed in the future. We also showed using a reservoir model that the SVGD with p-kernels provided better estimates of the uncertainty than the standard Gaussian kernels. Further comparison with ensemble-based methods and other gradient-based methods is left for future research.
-
-
-
Well Logs Inversion Into Lithology Classes: Comparing Bayesian Inversion and Machine Learning
More LessSummaryLithology classification is a crucial challenge in geological research. Lithologies at thelocations without cores need to be predicted by using indirectly geophysics measurementssuch as well logs and seismic data. In this study, we use the spatial dependency of sedimentsand well logs data for inversion into lithologies by a kernel-based hidden Markov model(HMM) and a gated recurrent unit (GRU) model. We estimate model parameters from twotraining wells and predict lithologies from well log observations for a blind test well. Thepredictions are also compared to results from a deep neural network (DNN) model, whichassumes spatial independence. Results indicate that the lithology predictions supplied by theHMM and GRU models are more reliable than the ones from DNN in term of classificationaccuracy and make more senses in geological interpretation. Moreover, the HMM providesquantifications of the classification uncertainty.
-
-
-
Well Log Data Standardization, Imputation and Anomaly Detection Using Hidden Markov Models
SummaryThe main goal of this work is to develop a systematic approach to work with raw well log data. Toaccomplish this goal, we propose to fit a simple unsupervised generative model to the input data and au-tomate the preprocessing step using the generative model. This approach allows to detect the anomaliesin the data as the regions that the model struggles to explain (i.e., samples with extremely low like-lihood), infer approximations to the missing features using the Bayes rule and incorporate additional expert knowledge in the design of the model.
-
-
-
Seismic Estimation of Sub-resolution Reservoir Properties with Bayesian Evidential Learning: Application to an Offshore Delta Case
Authors A. Pradhan and T. MukerjiSummaryWe present a Bayesian framework that facilitates estimation of and uncertainty quantification of sub-resolution reservoir properties from seismic data. Our workflow exploits learning the direct relation between seismic data and reservoir properties by the evidential learning approach to efficiently estimate sub-resolution properties. The major focus of this paper is on a real case study of seismic characterization of reservoir layers significantly below the seismic resolution. We employ deep neural networks as summary statistics within Approximate Bayesian Computation to estimate posterior uncertainty of reservoir net-to-gross from 3D pre-stack seismic data without an explicit inversion.
-