Quantifying the sensitivity of seismic facies classification to seismic attribute selection: An explainable machine learning study

During the past two decades, geoscientists have used machine learning to produce a more quantitative reservoir characterization and to discover hidden patterns in their data. However, as the complexity of these models increase, the sensitivity of their results to the choice of the input data becomes more challenging. Measuring how the model uses the input data to perform either a classification or regression task provides an understanding of the data-to-geology relationships which indicates how confident we are in the prediction. To provide such insight, the ML community has developed Local Interpretable Model-agnostic Explanations (LIME), and SHapley Additive exPlanations (SHAP) tools. In this study, we train a random forest architecture using a suite of seismic attributes as input to differentiate between mass transport deposits (MTDs), salt, and conformal siliciclastic sediments in a Gulf of Mexico dataset. We apply SHAP to understand how the model uses the input seismic attributes to identify target seismic facies and examine in what manner variations in the input such as adding band-limited random noise or applying a Kuwahara filter impact the models’ predictions. During our global analysis, we find that the attribute importance is dynamic, and changes based on the quality of the seismic attributes and the seismic facies analyzed. For our data volume and target facies, attributes measuring changes in dip and energy show the largest importance for all cases in our sensitivity analysis. We note that to discriminate between the seismic facies, the ML architecture learns a “set of rules” in multi-attribute space and that overlap between MTDs, salt, and conformal sediments might exist based on the seismic attribute analyzed. Finally, using SHAP at a voxel-scale, we understand why certain areas of interest were misclassified by the algorithm and perform an in-context interpretation to analyze how changes in the geology impact the model’s predictions.

Artificial Intelligence Aided Geologic Facies Classification in Complex Carbonate Reservoirs

Facies classification for complex reservoirs is an important step in characterizing reservoir heterogeneity and determining reservoir properties and fluid flow patterns. Predicting rock facies automatically and reliably from well log and associated reservoir measurements is therefore essential to obtain accurate reservoir characterization for field development in a timely manner. In this study, we present an artificial intelligence (AI) aided rock facies classification framework for complex reservoirs based on well log measurements. We generalize the AI-aided classification workflow into five major steps including data collection, preprocessing, feature engineering, model learning cycle, and model prediction. In particular, we automate the process of facies classification focusing on the use of a deep learning technique, convolutional neural network, which has shown outstanding performance in many scientific applications involving pattern recognition and classification. For performance analysis, we also compare the developed model with a support vector machine approach. We examine the AI-aided workflow on a large open dataset acquired from a real complex reservoir in Alberta. The dataset contains a collection of well-log measurements over a couple of thousands of wells. The experimental results demonstrate the high efficiency and scalability of the developed framework for automatic facies classification with reasonable accuracy. This is particularly useful when quick facies prediction is necessary to support real-time decision making. The AI-aided framework is easily implementable and expandable to other reservoir applications.

Machine Learning as a Silent Observer of Advanced Geoscience Interpretation

Machine learning is proving its successes in all fields of life including medical, automotive, planning, engineering, etc. In the world of geoscience, ML showed impressive results in seismic fault interpretation, advance seismic attributes analysis, facies classification, and geobodies extraction such as channels, carbonates, and salt, etc. One of the challenges faced in geoscience is the availability of label data which is one of the most time-consuming requirements in supervised deep learning. In this paper, an advanced learning approach is proposed for geoscience where the machine observes the seismic interpretation activities and learns simultaneously as the interpretation progresses. Initial testing showed that through the proposed method along with transfer learning, machine learning performance is highly effective, and the machine accurately predicts features requiring minor post prediction filtering to be accepted as the optimal interpretation.

Palaeogeography and Palaeoenvironments – A Multifield Examination of the Devonian-Permian Evolution of the Dneipr-Donets Basin

Context is everything. Not all thick sands pay out and not all thin sands are poorly productive. It is important to understand a basin's palaeogeographical drivers, the resultant palaeoenvironments and their constituent sedimentary architecture. Development of a depositional model can be predictive with respect to the magnitude of accessible pore space for potential development. We present a multi-field study of the Dneipr-Donets basin. Over 600 wells were studied with >4500 lithostratigraphical picks being made. Over 7500 sedimentological picks were made allowing mapping of facies bodies and charting shifts in facies types. A facies classification scheme was developed and applied. The Devonian-Permian sedimentary section records the creation, fill, and terminal closure of the Dneipr-Donets Basin:Syn-rift brittle extension (late Frasnian-Famennian): intracratonic rifting between the Ukrainian Shield and Voronezh Massif formed a NW-SE orientated trough, with associated basaltic extrusion. Basin architecture consists of rotated fault blocks forming graben mini-basins. Sedimentation is dominantly upper shoreface but sand packages are poorly correlatable due to the faulted palaeotopography.Early Post-rift thermal subsidence (Visean-Lower Bashkirian): the faulted palaeotopography was filled and thermal subsidence drove basin deepening. Cyclical successions of offshore, lower shoreface and upper shoreface dominate. Sands are typically thin (<10m) but can be widely correlated and have high pore space connectivity.Mid Post-rift: the Bashkirian (C22/C23 boundary), paralic systems prograde over the shoreface. Changes in vertical facies are abrupt due to a low gradient to basin floor. Deltaic and fluvial facies can produce thick amalgamated sands (>30m), but access limited pore space because they are laterally restricted bodies.Terminal post-rift (Mykytivskan): above the lower Permian, the convergence of the Kazahkstanian and Siberian continents began to restrict the Dnieper-Donets basin's access to open ocean. The basin approached full conditions and deposition was dominated by evaporite precipitation, with periodic oceanic recharge. Ultimately, this sediment records the formation of Pangea. The successions examined were used to construct a basinal relative sea level curve, which can be applied elsewhere in the basin. This can be used to help provide palaeogeographical context to a field, which in turn controls the sedimentary architecture.

The Influence of Lithofacies Features of a Deposit on the Efficiency of Reserves Recovery

The main task in petroleum engineering is to achieve the maximum possible production of hydrocarbon reserves with low expenditures. Many reasons influence the economics of the project. And one of them is related to choosing the right location for drilling a well in order to produce unrecovered hydrocarbons. The choice of this place has a direct correlation with the geological aspects of an oil field. This paper showed that different facies have a great influence on reserves recovery on the example of the South-Vyintoyskoye field. The classification of facies involved the study of production data. This study was presented by analysis of core sample, application of models by Muromtsev, reading of well logs, and build-up of geological models. The construction of geological models involved work in IRAP RMS TEMPEST, CorelDRAW, Geoglobe, Roxar Program package. The subject of study was the reservoir rock BV7/3-4that is a part of the Barremian age formation. It was concluded that this reservoir rock is composed of mainly argillaceous sandstones, interbedded with siltstones and shales. Authors identified that this formation belongs to three different depositional facies. The facies classification showed that the A1 zone is associated with well-graded fine to the fine-grained size of sediments. The reservoir rock of this zone is composed of sediments that belong to the distribution channel of deltas. And as a result, it is characterized by its high flow rate of production wells. The reservoir rock of the A2 facies zone is composed of sediments that are related to the collapse of mouth bars and branches of deltas. This zone has difficulties in fluid migration due to the presence of heterogeneities and clay material as well as the drilling of new wells close to the given zone lead to the medium flow rate of production wells. The reservoir rock of the A3 facies zone is composed of sediments that belong to turbidity flows. In this zone, organic matter has been recognized as a group of ichnofacies such as "Cruziana". This zone is characterized by the low flow rates of the production wells.

Unravel Carbonate Reservoir Heterogeneities Through 3D Seismic Neural Network Application: A Machine Learning Based Approach and Workflow

The availability of 3D seismic data undoubtedly plays an important role in reservoir characterization. Currently seismic technology continues to advance at a rapid pace not only in the acquisition but also in processing and interpretation domain. The advance on this is well supported by the digitalization era which urges everything to run reliably fast, effective and efficient. Thanks to continuous development of IT peripherals we now have luxury to process and handle big data through the application of machine learning. Some debates on the effectiveness and threats that this process may automating certain task and later will decrease human workforce are still going on in many forums but still like it or not this machine learning is already embraced in almost every aspect of our life including in oil & gas industry. Carbonate reservoir on the other hand has been long known for its uniqueness compared to siliciclastic reservoir. The term heterogeneous properties are quite common for carbonate due to its complex multi-story depositional and diagenetic facies. In this paper, we bring up our case where we try to unravel carbonate heterogeneity from a massive tight gas reservoir through our machine learning application using the workflow of supervised and unsupervised neural network. In this study, we incorporate 3D PSTM seismic data and its stratigraphic interpretation coupled with the core study result, BHI (borehole image) log interpretation, and our regional understanding of the area to develop a meaningful carbonate facies model through seismic neural network exercises. As the result, we successfully derive geological consistent carbonate facies classification and distribution honoring all the supporting data above though the limitation of well penetration in the area. This result then proved to be beneficial to build integrated 3D geomodel which later can explain the issue on different gas compositions happens in the area. The result on unsupervised neural network also able to serves as a quick look for further sweetspot analysis to support full-field development.

Rock-physics machine learning toolkit for joint litho-fluid facies classification and compaction modeling

Correct lithofacies interpretation sourced from wireline log data is an essential source of prior information for joint seismic inversion for facies and impedances, among other applications. However, this information is difficult to interpret or extract manually due to the multivariate and high dimensionality of wireline logs. Facies inference is also challenging for traditional clustering-based approaches because pervasive compaction trends affect a number of petrophysical measurements simultaneously. Another common pitfall in automated clustering approaches is the inability to account for underlying diagenetic processes that correlate with depth. Here, we address these challenges by introducing a rock-physics machine learning toolkit for joint litho-fluid facies classification. The litho-fluid types are inferred from the borehole data within the objective framework of a maximum-likelihood approach for latent facies variables and rock-physics model parameters, explicitly accounting for compaction and depth effects. The inference boils down to an expectation-maximization (EM) algorithm with strong spatial coupling. Each litho-fluid type is associated with an instance of a particular rock-physics model with a unique set of fitting parameters, constrained to a physically reasonable range. These fitting parameters in turn are inferred using bound-constrained optimization as part of the EM algorithm. Outputs produced by the toolkit can be used directly to specify the necessary prior information for seismic inversion, including per-facies rock-physics models and facies proportions. We present an example application of the tool to real borehole data from the North West Shelf of Australia to illustrate the method and discuss its characteristic features in depth.

STUDY ON THE PARAMETERIZATION RESPONSE OF PROBABILISTIC NEURAL NETWORKS FOR SEISMIC FACIES CLASSIFICATION IN THE GULF OF MEXICO

Nowadays, there are many unsupervised and supervised machine learning techniques available for performing seismic facies classification. However, those classification methods either demand high computational costs or do not provide an accurate measure of confidence. Probabilistic neural networks (PNNs) overcome these limitations and have demonstrated their superiority among other algorithms. PNNs have been extensively applied for some prediction tasks, but not well studied regarding the prediction of seismic facies volumes using seismic attributes. We explore the capability of the PNN algorithm when classifying large- and small-scale seismic facies. Additionally, we evaluate the impact of user-chosen parameters on the final classification volumes. After performing seven tests, each with a parameter variation, we assess the impact of the parameter change on the resultant classification volumes. We show that the processing task can have a significant impact on the classification volumes, but also how the most geologically complex areas are the most challenging for the algorithm. Moreover, we demonstrate that even if the PNN technique is performing and producing considerably accurate results, it is possible to overcome those limitations and significantly improve the final classification volumes by including the geological insight provided by the geoscientist. We conclude by proposing a new workflow that can guide future geoscientists interested in applying PNNs, to obtain better seismic facies classification volumes by considering some initial steps and advice.