Accurate geological parameterization is essential for reservoir history matching, yet existing methods face a persistent trade-off between computational tractability and reconstruction fidelity. Linear methods such as principal component analysis (PCA) are efficient and statistically well-behaved but produce over-smoothed fields that suppress fine-scale heterogeneity, while deep generative models offer richer representations at the cost of training instability and poorly conditioned latent spaces. This paper presents a hybrid framework, PCA-DDPM, which combines PCA-based dimensionality reduction with a denoising diffusion probabilistic model (DDPM) trained on residual fields. The DDPM is conditioned on the PCA reconstruction and learns to predict the difference between the ground-truth property field and its linear approximation, thereby stabilizing the training process. A patch-based inference strategy with geological mask-weighted merging enables scalable application to large three-dimensional reservoir volumes. Through testing on the COSTA carbonate reservoir model, PCA-DDPM achieves SSIM > 0.93 and R² > 0.87 for permeability reconstruction, and SSIM > 0.95 and R² > 0.92 for porosity, which represents an improvement of approximately 60% in SSIM over the PCA baseline. Full-physics two-phase flow simulations across 120 wells confirm that the reconstructed fields preserve well-level production responses, with median R² exceeding 0.97 for oil and water production rates and 0.99 for injection bottom-hole pressure. The framework operates within a compact 256-dimensional latent space, making it directly applicable to practical history matching workflows.

Abstract In this paper we demonstrate the application of state-of-the-art deep learning using hybrid neural networks (HNN) that generalize and scale to multi-million, structurally diverse reservoir model grids and generate long-term spatio-temporal predictions of fluid and pressure propagation. The HNN simulator (HNNS) is a surrogate framework that consists of a subsurface graph neural network (SGNN) to model the evolution of fluids, and a 3D-U-Net to model the evolution of pressure. We benchmark the HNNS with two conceptually different reservoir models: a) modified SPE-10 model, with approx. 1 million grid size and variable number and positioning of vertical producers and injectors, b) synthetic fractured model, 15+ million grid size and 100+ injector and producer wells with variable geometry. We construct the network graph, where graph objects (nodes), representing reservoir grid cells are encoded with tens of static, dynamic, computed (relative permeability, gradients) and control (well rates) features. The graph edges represent interactions between the nodes with encoded features like transmissibility, direction and fluxes. We implement sector-based training with multi-step rollout to avail for the use of large-scale models. To properly perform the sector-based training the masking of sector boundary effects, sector stride and mixing sectors were used. We present the comparative results between the HNNS and the full-physics simulation for up to 30-year prediction of the 3D flow dynamics.

Modeling variations in reservoir properties as a function of space is an important task in evaluating reservoir heterogeneity. While it provides considerable value to capture the relative change in properties in the entire reservoir, it is an overwhelming task to perform in 3D. Various scientific approaches exist that facilitate measuring static reservoir heterogeneity along the wellbore. However, they are very limited and confined to small static datasets that are very often not representative. This chapter introduces a new approach to quantify a metric of 3D static reservoir heterogeneity. It describes the methodology to calculate a spatial heterogeneity index that encapsulates complex features to benchmark formations, zones, reservoirs, and fields. The introduced workflow expands on Lorenz’s scientific approach as the basis for this methodology to develop the heterogeneity index. The systematic approach excels in rendering a universal systematic 3D spatial heterogeneity index where typically data are not available, representative of overall prolificacy for a particular region of interest, zone, or the entire reservoir. The chapter also envisions possible applications, where the spatial heterogeneity index is utilized as a 3D conditioning attribute in dynamic model calibration and optimization.

The continuous evolution of technologies in subsurface petroleum exploration and characterization makes imperative the creation of multidisciplinary up-to-date technical workflows aimed to represent complex reservoirs with robust reliability. The objective of this chapter is to promote an innovative and integrated collection of the latest technologies available for the study of naturally fractured hydrocarbon reservoirs. We assembled some of the present-day advanced technologies ranging from high-resolution digital core analysis, borehole fracture characterization, numerical deformation and geomechanics modeling, stochastic fracture modeling and the numerical simulation of dual-porosity, dual-permeability fractured reservoirs. The efficient application of robust fracture models in field development is also discussed. This chapter is intended to serve as a practical and structured handbook for professionals working on subsurface reservoir modeling, with emphasis on fractured reservoirs in the petroleum industry.

Abstract History matching is an inverse problem in which a geological model needs to be calibrated under uncertainty in order to produce a representative reservoir simulation model. While there are many methods for computer-assisted history matching, most of them require substantial computational power and manifest slow convergence for large-scale simulation models. To tackle these challenges, we propose a new technique for history matching reservoir simulation models using generative diffusive learning (SimGDL) [1]. Our work consisted of several steps based on conditional denoising diffusion probabilistic models. First, we generated several thousands of geologic realizations covering a wide range of spatial subsurface uncertainties. Secondly, we added various amounts of gaussian noise to these realizations and trained a U-Net to predict the added noise based on the corresponding production response. When history matching, a certain amount of noise is added to one of the geologic realizations and then the trained U-Net, guided by the targeted production response, is used to iteratively remove the noise until it converges to a newly generated, dynamically calibrated geologic realization. To test the potential of this new approach, we created a synthetic simulation model. Several thousands of permeability realizations were prepared and used to train the diffusion U-Net. The simulation model combines a single water injector and an oil producer, where oil rate, water cut, and bottom-hole pressure were designated as production responses. Dynamically calibrated permeability realizations were generated by adding noise to some of the existing realizations and then passing them through our diffusion U-Net, conditioned by the selected target production data. The simulated dynamic responses using the generated geologic realizations showed great improvements in matching the target production data. In addition, the posterior realizations were able to retain main feature signatures from the prior distribution and the areas to be updated. Furthermore, the generated models incorporate the quantification of uncertainty, as captured with multiple geologic realizations that were used in training. This paper presents a new method for data-driven dynamic calibration of reservoir simulation models utilizing Generative Diffusive Learning. It incorporates a robust uncertainty quantification using an ensemble of representative subsurface reservoir models. By adding further complexity to conditioning parameters, this innovative method could be further extended and scaled-up to history match simulation models with diverse structural layouts and production profiles.

Abstract Sidetracking vertical wellbores into horizontal laterals to avoid or delay water coning is a common practice in managing oil reservoirs under water flooding. Accurately predicting if the sidetrack would improve the performance of the well (i.e., reduce water production, increase oil production, or both) represents a vital information to decide whether it is optimal to sidetrack or continue with current vertical hole and withstand the increasing water cut. In this paper, we employ algorithms of predictive analytics and machine learning (ML) on a large synthetic but realistic data set of vertical bore-holes that were converted later into sidetracks in order to predict if a sidetrack would be a success. We classify the sidetracks using ratios of cumulative oil production aggregated over different periods of time of wells before and after the sidetrack. As success criteria differ depending on business objective, multiple models were built to account for the pre-set possible scenarios. The data-driven models built were based on a binary supervised classification (successful/unsuccessful) with very good accuracy, measured with Area Under the Curve (AUC) and Receiver Operating Characteristics (ROC) scores. A large number of predictors is initially used in our model, such as well location, well configuration, well completion, normalized production data, pressure data, as well as assorted petro-physical attributes, such as reservoir quality and average saturation indices. Through an exercise of recursive feature elimination, only parameters that contribute to improve the quality of prediction are retained. Results of several classification algorithms such as decision trees and support vector machines are being studied and compared. Several developed models with cross-validation are introduced to reduce bias and variance and to render better prediction results. Applying the model on a set of test simulation data, ML model predictions of success and failure cases conform with simulated data with high level of accuracy. Additionally, predictive models were found useful in identifying parameters that highly correlate with the success or failure of sidetracks. Final results showed an accuracy as high as 80% and an AUC of the ROC curve of 0.79.

History matching is widely considered as the most time- and resource-consuming phase of reservoir simulation modeling. Even with the advent of modern, computer-assisted, history matching methods, the dynamic calibration of large-scale simulation models represents a considerable computational undertake. The challenges become even more pronounced with incorporation of subsurface and production uncertainty. This paper outlines a step forward in acceleration of reservoir simulation studies by applying a split/merge approach constrained by no-flow boundary drainage region. The method transforms the history-matching process into an accelerated progressive sequence of dynamic model updates in time and space. Each segment defined as distinctive drainage region, the boundaries of the drainage regions are mapped based on no-flow conditions. Each segment is dynamically calibrated and history matched simultaneously in parallel. Lastly, the segments are merged back to reconstruct the original model to run the prediction phase. The detail of the workflow is described, as well as the implementation of the workflow in a synthetic model. A comparison between the conventional approach and the new approach is discussed. Recommendation and a way forward are shared to capitalize on the accelerated method for future reservoir studies.

A successful field development plan relies heavily on comprehending various subsurface complexities and their impact on the petroleum system. Hence, reservoir modeling plays an essential role in aggregating the complex systems and incorporating uncertainties. In the past decades, fracture characterization and modeling has advanced significantly in providing a high-resolution depiction of the subsurface. This paper focused on examining the impact of major fractures on reservoir connectivity and well productivity. Fracture modeling and parameterization were performed using a synthetic dual porosity, dual permeability (DPDP) simulation model that was based on SPE10 model. Various deterministic fracture realizations were generated by incorporating multiple scenarios of different features such as fractures, high-permeability and high-flow feature planes, and layer bound fractures. The positioning of major longitudinal and lateral fractures was examined to analyze their role on reservoir connectivity and well productivity. Special focus was devoted to evaluate the impact of limited vertical and horizontal fracture extensions compared to more conventional workflows where fractures are modeled as planes across the entire reservoir. Furthermore, additional operational conditions, such as variable water injection schemes and depletion strategies, were applied to assess their impact on the model's response. The evaluation provided in-depth analysis of various subsurface flow scenarios with DPDP systems. The characterization and modeling of complex subsurface features enhanced the understanding of spatial flow dynamics and their impact on reservoir performance. For example, layer-bound fractures result in different dynamic behavior compared to the ones crossing the entire reservoir. Additionally, high-flow features like high permeability streaks alter subsurface flow dynamics by accelerating fluid movement. Therefore, production analysis was performed for every scenario independently to benchmark the impact of various fracture parameters. The paper provides comprehensive evaluation of fracture parameterization on subsurface dynamics by utilizing a DPDP SPE10 model to determine the distinct fracture signatures of reservoir performance and their influence on the overall hydrocarbon recovery.

Presence of paleo zone, which frequently exists below Free-Water-Level surface, can impact dynamic reconciliation of reservoir simulation models. The process is even more challenging with embedded complex representations of reservoir connectivity (conductive fractures) and inherent uncertainty associated with geological and flow modeling. We present a rigorous approach that integrates characterization of paleo zone, parameterization of paleo zone conductivity and application of flow profiles as a guide in accelerated history matching study of large-scale Dual Porosity-Dual Permeability model. The presence of immobile oil within paleo zone can cause permeability reduction and inherently limit aquifer support to oil zone. Accordingly, such occurrence can be represented as a low permeability streak or region in the simulation model and leveraged for more accurate calibration of model injection wells located inside the paleo zone. We performed probabilistic sensitivity analysis and parameterization of paleo zone conductivity using Design of Experiments on a synthetic simulation model with optimized aquifer size and strength as the basecase. The outcome of the synthetic sensitivity scenarios using dynamic model strongly indicates that paleo zone is partially sealing. Multiple scoping runs were performed to identify appropriate permeability values required to calibrate the model. The use of multipliers in porositypermeability transform reproduces blocking or baffling effect of the paleo zone, considering this fluid will behave as part of the rock framework. Porosity and permeability were recomputed inside the paleo zone based on Bulk Volume of Water (BVW) data assessment. The higher the BVW the higher the chance to have effective communication between the oil leg and aquifer. These multipliers represent the probability of the sealing character of the paleo zone and reflect on the non-uniform distribution of accumulated hydrocarbons. Above methodology was used to define the initial set of paleo zone petrophysical property modifiers, rendering multiple model realizations within optimistic-pessimistic range. Flow profiles can be used to guide segmentation of paleo zone with preferential well injectivity to further improve the efficiency of history matching process. Our paper demonstrates a successful application of multi-variate characterization and modeling of paleo zone geometry and properties for a history match of a conceptual, complex reservoir simulation model under reservoir uncertainty. An innovative approach to probabilistic parameterization of paleo zone conductivity has contributed to a model with exceptionally high quality and rendered a reservoir simulation model with reliable predictive capability in accelerated time.

Subsurface simulations use computational models to predict the flow of fluids (e.g., oil, water, gas) through porous media. These simulations are pivotal in industrial applications such as petroleum production, where fast and accurate models are needed for high-stake decision making, for example, for well placement optimization and field development planning. Classical finite difference numerical simulators require massive computational resources to model large-scale real-world reservoirs. Alternatively, streamline simulators and data-driven surrogate models are computationally more efficient by relying on approximate physics models, however they are insufficient to model complex reservoir dynamics at scale. Here we introduce Hybrid Graph Network Simulator (HGNS), which is a data-driven surrogate model for learning reservoir simulations of 3D subsurface fluid flows. To model complex reservoir dynamics at both local and global scale, HGNS consists of a subsurface graph neural network (SGNN) to model the evolution of fluid flows, and a 3D-U-Net to model the evolution of pressure. HGNS is able to scale to grids with millions of cells per time step, two orders of magnitude higher than previous surrogate models, and can accurately predict the fluid flow for tens of time steps (years into the future). Using an industry-standard subsurface flow dataset (SPE-10) with 1.1 million cells, we demonstrate that HGNS is able to reduce the inference time up to 18 times compared to standard subsurface simulators, and that it outperforms other learning-based models by reducing long-term prediction errors by up to 21%.

Carbonate reservoirs typically exhibit very complex geological structures and are characterized by flow dynamics primarily occurring in fractures. The intricate network of fractures as well as their interconnectedness may lead to unexpected flow patterns and uneven sweep efficiency. Determining reservoir properties of both matrix and fracture channels is quintessential for accurately tracking the fluid front movement in the reservoir, optimizing sweep efficiency, and maximizing hydrocarbon production. In this study, we showcase the application of a feature-oriented ensemble-based history matching workflow to a complex fractured carbonate reservoir box model, focusing on the use of formation resistivity tomography data that are usually inferred from deep crosswell electromagnetic (EM) surveys. Compared with the production data that are commonly used in history matching, deep EM measurements provide additional information about the spatial distribution of subsurface reservoir properties in the interwell volumes by exploiting the strong resistivity contrast between water and hydrocarbons. A hybrid parameterization approach is used to represent the multiscale fracture distribution in which the spatial distribution of small-scale fractures is modelled by a truncated Gaussian simulation method. A large number (over one million) of uncertain model parameters including reservoir matrix and fracture properties as well as Archie's parameters are identified and updated by an iterative ensemble smoother. For an efficient integration of the high-dimensional and noisy EM tomography data, the boundary or contour information extracted from the EM resistivity field is instead assimilated through a distance parameterization approach. A modified bootstrap-based localization is proposed to regularize the model updates adaptively during the iteration to reduce sampling errors. Especially, to improve the computational efficiency in dealing with the large dimensions of both data and model parameters, the localization is implemented in a projected low-dimensional data subspace. Experimental results demonstrate the applicability and efficiency of the developed workflow for reservoir history matching in more realistic model settings. The comparative case study also illustrates the significance of jointly incorporating multiple sources of data for better quantification of model uncertainty, and the great potential of deep EM data for enhancing the characterization of complex fractured carbonate reservoirs.

Summary Network graphs represent a general language for describing complex systems and a framework for knowledge discovery. Graph learning is a new concept with applications emerging in biomedicine, pharmacology, smart mobility, and physical reasoning. When applied to petroleum systems, such as reservoir models, graphs provide unique differentiators for the abstraction of reservoir connectivity to facilitate “reservoir-centric” machine learning (ML) applications. In this paper, we demonstrate, for the first time, the application of geoscience-based deep interaction networks (GeoDIN) to learn complex physics relationships from 3D reservoir models for fast and accurate prediction of subsurface spatio-temporal flow dynamics. We build the network graph with embedded subsurface and physics representations and train the ML model to “act like the reservoir simulator.” We use a simulation benchmark model for two-phase incompressible flow, with approximately 1.1 million grid size, one central injector, and four corner producers. Static 3D grid properties include porosity and permeability. We use full-physics simulation output to construct the interaction network (IN) graph, where graph nodes objects (nodes) represent reservoir grid cells. We embed the feature vector combining pore, oil and water volumes, and pressure and relative permeability. The graph objects representing wells are connected with well completion factors. The producing wells have embedded oil and water production rates, while the objects representing injecting wells have embedded water injection rates. We represent graph relations (edges) with bidirectional transmissibility of the source cell. To preprocess the data for ML, we scale the graph object attributes using “min-max” normalization and we normalize the graph relation attributes using Box-Cox transformation. We train the GeoDIN framework to predict oil and water saturation dynamics in space and time. When benchmarked with full-physics simulation, the INs ran on two V100 graphics processing units and substantially accelerated the prediction phase compared to the physics-based simulator running on 70 Intel Xeon E5 CPU cores. On average, the error in GeoDIN predicted spatio-temporal distribution of oil saturation remains within 5% of full-physics simulation for 90% of model grid cells, while the error in water saturation remains within 2.5% of full-physics simulation. The spatio-temporal propagation of pressure is more sensitive to local embeddings of INs, which communicate on node-to-node information transfer. This results in a larger prediction error of the GeoDIN model when benchmarked to full-physics simulation. On average, the error distribution suggests that the great majority (90 to 95%) of grid cells fall within 10 to 30% error bound relative to full-physics simulation. The presented GeoDIN approach to network learning carries a game-changing potential for the prediction of subsurface flow dynamics. As the way forward, we will investigate the implementation of graph neural networks with automated feature learning, generalization, and scaleup.

Abstract This paper presents an approach to optimize field water injection strategies using stochastic methods under uncertainty. For many fields, voidage replacement was the dictating factor of setting injection strategies. Determining the optimum injection-production ratio (IPR) requires extensive experience taking into consideration all the operational facility constraints. We present the outcome of a study, in which several optimization techniques were used to find the optimum field IPR values and then elaborate on the techniques? strengths and weaknesses. The synthetic reservoir simulation model, with millions of grid blocks and significant numbers of producers and injectors, was divided into seven IPR regions based on a streamline study. Each region was assigned an IPR value with an associated uncertainty interval. An ensemble of fifty probabilistic scenarios was generated by experimental design, using Latin Hypercube sampling of IPR values within tolerance limits. Scenarios were used as the main sampling domain to evaluate a family of optimization engines: population-based methods of artificial intelligence (AI), such as Genetic algorithms and Evolutionary strategies, Bayesian inference using sequential or Markov chain Monte Carlo, and proxy-based optimization. The optimizers were evaluated based on the recommended IPR values that meet the objective of minimizing the water cut by maximizing oil production and minimizing water production. The speed of convergence of the optimization process was also a subject of evaluation. To ensure unbiased sampling of IPR values and to prevent oversampling of boundary extremes, a uniform triangular distribution was designed. The results of the study show a clear improvement of the objective function, compared to the initial sampled cases. As a direct search method, the Evolutionary strategies with covariance matrix adaptation (ES-CMA) yielded the optimum IPR value per region. While examining the effect of applying these IPR values in the reservoir simulation model, a significant reduction of water production from the initial cases without an impact on the oil production was observed. Compared to ESCMA, other optimization methods have dem

Simulations of fractured reservoirs are usually performed by dual porosity, dual permeability models. Traditional deterministic workflows that model prominent fracture lineaments often fail to integrate quantification of uncertainty, inherent in fracture spatial distribution and properties. This poses significant challenges on history matching and frequently requires extensive manual beyond geological consistency to achieve the match. We present a method for assisted history matching (AHM) that calibrates models of fractured reservoirs by dynamically updating matrix properties and discrete fracture networks (DFN), while retaining the highest levels of geological consistency and model predictability. The new workflow simultaneously interfaces between applications for building the geo-model and the DFN model "on the fly" and integrates them into a Closed-Loop framework for global stochastic optimization using evolutionary algorithms. Rigorous uncertainty quantification is performed with sensitivity analysis and variability refinement, using the multi-level design of experiments (DoE). We deploy the workflow on the model of a fractured and faulted reservoir developed under natural aquifer drive. Geo-modeling uncertainty workflow generates multiple realizations of seismic-inverted acoustic impedance, used as a 3D trend for populating porosity, with varying variogram parameters. Uncertainty in porosity-permeability correlation coefficients is leveraged to generate multiple, spatially diverse permeability models. Realizations of porosity and permeability are used to generate corresponding realizations of water saturation. By sampling probability distributions of fracture density, geometry, aperture and orientation, 3D realizations of fracture porosity, fracture horizontal and vertical permeability and matrix-fracture transfer parameters are generated. The workflow produces statistically and geologically diverse ensemble of matrix and DFN model realizations that results in excellent variability in dynamic simulation response and confines the observed data. The multi-objective misfit function (OF), subject to minimization in the AHM process, incorporates static well pressures and was evaluated with a reservoir simulator that employs Massive Parallel Processing to achieve practical computation times, even with large-scale simulation grids. The presented AHM workflow demonstrates a unique functionality that enables the integration of DFN geo-mechanical properties (e.g. paleo-stress, pore-pressure) as predictors for fracture network attributes in the process of Closed-Loop model inversion and optimization. The method enables a robust, multivariate reservoir uncertainty quantification and dynamic calibration and delivers geologically consistent fractured reservoir models for reservoir forecasting under uncertainty.

Summary Carbonate reservoirs represent strongly complex geological structures whose main feature is that the flow dynamics primarily occurs in fractures. The complexity of the network of fractures as well as their interconnectedness may lead to unexpected flow patterns and uneven sweep efficiency. Determining the fracture distribution and reservoir properties of both matrix and fracture channels is quintessential for accurately tracking the fluid front movement in the reservoir, optimizing sweep efficiency, and maximizing hydrocarbon production. A feature oriented ensemble-based history matching workflow was introduced previously to enhance the characterization of petroleum reservoirs through the assimilation of time-lapse electromagnetic (EM) data in combination with other available measurements. Compared with seismic measurements, which provide effective information related to reservoir structure, deep EM measurements in the interwell volumes are more sensitive to distinguish between hydrocarbon fluids and water. The developed workflow calibrates model variables of interest utilizing the information of formation resistivity that is usually made available through geophysical inversion of raw EM data. Archie’s law is typically used to build a relation between formation porosity, fluid properties (e.g., water saturation and salt concentration) and formation resistivity. Instead of integrating directly the inverted EM resistivity data, which is usually of high dimensions and noisy in amplitude, the boundary or contour information extracted from the EM resistivity field is utilized through an image oriented distance parameterization combined with an iterative ensemble smoother. We are showcasing this framework on a realistic carbonate reservoir box model with a complex fracture channel network. Time-lapsed cross-well EM data was assimilated to update fracture and matrix reservoir properties, ensuring that the heterogeneity in the properties is maintained. The framework exhibited strong performance in the history matching of the complex carbonate reservoir structure. In comparison with conventional ensemble-based history matching techniques, this innovative developed approach led to significantly more accurate sweep efficiency maps, while maintaining the heterogeneity in the parameters between the fractures and the matrix. Finally, uncertainty in the saturation maps could be significantly reduced with the assistance of deep EM reservoir tomography. Carbonate reservoirs represent highly complex geological structures and are characterized by flow dynamics dominated by natural fractures. The complexity of the network of fractures as well as their interconnectedness may lead to unexpected flow patterns and uneven sweep efficiency. Determining the fracture distribution and reservoir properties of both matrix and fracture channels is quintessential for accurately tracking the fluid front movement in the reservoir, optimizing sweep efficiency, and maximizing hydrocarbon production. A feature-oriented ensemble-based history matching workflow was introduced previously to enhance the characterization of petroleum reservoirs through the assimilation of time-lapse electromagnetic (EM) data in combination with other available measurements. Compared with seismic measurements, which provide effective information related to reservoir structure, deep EM measurements in the interwell volumes are more sensitive to distinguish between hydrocarbon fluids and water due to the difference in electrical conductivity. The developed workflow calibrates model variables of interest utilizing the information of formation resistivity that is usually inferred through geophysical inversion of raw EM data. Archie’s law is typically used to describe the relation between formation porosity, fluid properties (e.g., water saturation and salt concentration) and formation resistivity. Instead of integrating directly the inverted EM resistivity data, which is usually of high dimensions and noisy in amplitude, the boundary or contour information extracted from the EM resistivity field is utilized through an image-oriented distance parameterization combined with an iterative ensemble smoother. We are showcasing this framework using a realistic carbonate reservoir box model with a complex fracture channel network. We history matched time-lapsed crosswell EM data to update fracture and matrix reservoir properties, by preserving the heterogeneity in the properties. The framework exhibited strong performance in the history matching of the complex carbonate reservoir structure. The developed innovative approach led to significantly more accurate sweep efficiency maps, while maintaining the heterogeneity in the fractures and the matrix parameters. Uncertainties in the saturation maps were also significantly reduced with the history matching of deep EM reservoir tomography data.

Abstract Performance evaluations of oil and gas assets are crucial for continuously improving operational efficiency in the mainstream petroleum industry. The success of such evaluations is largely driven by the analysis of the data accumulated during the asset's operational cycle. Usually, the amount of data stored in the databases dramatically exceeds the ability to approach the analysis with traditional spreadsheet-based tools or linear modeling. In this study we use data mining with multivariate predictive analytics and monetize on the value of data by transforming the inferred information into knowledge and further into rigorous business decisions. With the expansion of the Digital Oil Field and transformation into the 4th Industrial Revolution, the oil and gas industry is acquiring tremendous amounts of data that come from disparate sources in a variety of origins, time scales, structures and quality. The underlying variable root-cause relationships are highly non-linear and non-intuitive, and simplistic linear regression methods are suboptimal. We approach the challenge by developing a data-driven workflow that integrates components of artificial intelligence, machine learning and pattern recognition to enhance quantitative understanding of complex data. The sanitized aggregated data set combines 470 horizontal wells, covering 15 numerical (e.g., stimulation interval length, production rates) and categorical (e.g., target zone, proppant type) predictors and the total produced BOE, as the response variable. The objective is to predict an optimal set of variables that maximize the production. We utilize an integrated analytics platform that enables a variety of sophisticated statistical operations on large-scale data: a) comprehensive data QA/QC for outliers, consistency and missing entries; b) Exploratory Data Analysis and visualization; c) feature selection, screening and ranking; d) building and training of multiple machine learning (ML) models for multi-variate regression (e.g. generalized linear model, deep learning, decision tree, random forest and gradient boosted machine); and e) response optimization of an identified "best-performing" ML model for highest prediction accuracy. Our study introduces the initiative to establish concepts best practices for predictive and prescriptive analytics in domains of reservoir simulation, description and asset management. Given the unique volume and information richness of operational data, acquired over decades of production history, the anticipated applications of predictive analytics could expand to drilling optimization, smart data aggregation, well stimulation and equipment maintenance.

Abstract Considering carbonate oil reservoirs, a rock fracture is a planar-shaped void filled with oil, water, gas and/or rock fines. These fractures vary in scale forming connected and complex networks of fractures. They have an effect on deliverability of fluids depending on their geometrical complexity, extent, matrix-fracture interaction, wettability, and orientation. In fractured reservoir rocks, relative to the rock matrix, fractures form highly permeable flow pathways that dominate fluid flow and transport in the reservoir which might have favorable or non-favorable effects on hydrocarbon production. It is crucial to characterize the fluid flow in the fracture networks to examine the root-cause relationships, the impact on hydrocarbon recovery and quantify the efficiency of enhanced recovery mechanisms. This work describes the development of a machine learning model for history matching and predicting two-phase relative permeability. Capitalizing on the main principles of the 4th Industrial Revolution (IR 4.0), the development of this model was achieved by training machine learning (ML) algorithms and using advanced predictive data analytics on data collected from lab experiments as input. The model derived from the analysis describes two-phase flow of oil and water in a single discretized fracture taking into account fracture aperture, wall roughness, orientation and, flow rates and direction. It also accommodates fluids and fracture characteristics to match laboratory SCAL experimental of co-current oil and water flow in a mixed-wettability single fracture modeled as narrow gap in a Hele-Shaw cell. The experimental data exhibit variations in shape and end-points that mainly reflect the effects of fracture aperture, roughness, inclination, and hysteresis effects. This in turn demonstrate the effects of phase interference, saturation changes, and major forces acting on two-phase flow in fractures like capillary and viscous forces. The empirical relationship showed an acceptable match to the experimentally derived relative permeability in most of the cases as well as good predictive capabilities against the blind tests on other sets of experimental data and numerical simulation models. Having both fracture relative permeability data (describing the fluids flow) and detailed fracture characterization improves our understanding of the reservoir dynamics and fractured network impact on hydrocarbon recovery.

Abstract Effective management of Voidage Replacement Ratio (VRR) throughout the producing life of an oil reservoir is essential for achieving optimal oil recovery. VRR is quantitatively defined as injection/production fluid volume ratio at reservoir conditions. The primary goal in managing voidage replacement is to replenish the energy in a reservoir to a degree that the producing wells yield hydrocarbons at economical rates. The determination of VRR, however, becomes more complicated when reservoirs are significantly affected by fluid influxes. This paper presents a method developed to optimize VRR calculations using streamlines, traced from finite-difference reservoir simulation model outputs. Good reservoir management practice necessitates that conventional VRR should be maintained at or above unity. Maintaining appropriate injection performance is therefore an essential requirement for achieving optimal oil recovery in secondary recovery processes. This can be achieved through effective VRR surveillance, water breakthrough monitoring, and reservoir pressure maintenance. This paper presents a new technique and associated workflow for rigorous VRR determination that resolves a number of shortcomings inherent in conventional VRR analysis. This rigorous VRRR determination methodology was applied to an existing field with considerable operating history including multiple displacement and recovery processes: primary depletion, aquifer influx, gas re-injection, gravity water injection, and power water injection. This new methodology utilizes finite difference reservoir simulation models to generate streamlines from the pressure field and fluxes. Streamlines represent flow paths between injectors and producers. The streamline trajectories with associated time-of-flight values thus obtained take into account geologic complexity, external fluxes, well locations, phase behavior, and reservoir flow behavior. Rigorous VRR estimates are obtained by accounting for the influxes and well allocation factors (WAF), which represent a measure of connectivity between specific injector/producer pairs with associated fluxes. The fluxes and WAF values are calculated automatically from the history-matched reservoir simulation model during streamline tracing for associated time steps. Traditionally, the well VRR values are calculated via the formulation of well inflow performance relationship (IPR), which may result in suboptimal estimations by not accounting for external sources of energy, such as influx from neighboring zones. The presented approach allows for improved optimisation of waterflood injection efficiency, where the off-set oil production can be derived directly from reservoir material balance (MB) calculations and streamline-generated well allocation factors. In order to facilitate VRR calculations with dynamic simulation regions, we propose a workflow for streamline (SLN) based VRR calculations using the time-dependent flow-based SLN-conditioned drainage volumes, automatically extracted from the simulation grid and iteratively incorporated into simulation model constraints as a function of simulation run time-steps.

AbstractIn this paper we present the framework and results of a benchmarking study to validate performance of conceptually different methods for computer-assisted history matching (AHM) under reservoir uncertainty on an example of complex waterflooding process.Traditionally reservoir models were manually reconciled with production data, relying mostly on workflows based on engineering judgment and established best practices. The main disadvantage of the manual HM process is that the reservoir simulation disengages from the geological model and fails to quantify reservoir uncertainty. The oil and gas industry has advanced in developing methods for AHM that enable producing geologically-consistent reservoir simulation models with robust uncertainty quantification and high predictive value. In our study we build distinctive, computationally-intensive AHM workflows covering global multi-objective stochastic optimization and streamline sensitivity-based inversion to perform dynamic real-field model update and history matching.We use the model of faulted reservoir under a waterflooding improved oil recovery regime. A comprehensive geo-modeling workflow incorporates facies, petrophysical properties, and saturation height function with uncertainty quantification performed on rock types, porosity, permeability, Kv/Kh, water saturation and fault transmissibility. During dynamic model update, the uncertainty workflows were executed as data assimilation on the sufficiently diverse prior ensemble of geomodel realizations, and as a Closed (Big) Loop process, where geomodel realizations are parameterized and updated simultaneously (on-the-fly) using global optimization. The misfit function, subject to minimization, incorporates oil and water rates and static wellhead pressure.The study concludes that different AHM techniques demonstrate different convergence performance. The overall success dramatically depends on the uncertainty quantification of the initial geomodel. Practical reservoir simulation times were achieved through utilization of a Massive Parallel Processing deployed on an HPC cluster. However, most importantly we demonstrate that, by deploying advanced AHM workflows, the updates of geologically and structurally complex models, with tens of millions grid-cells and large number of matched wells can render geology well-conditioned realizations, without necessity of introducing non-geological features in order to achieve a satisfactory match. Thus, the validated practices add a tangible business value to the process of integrated reservoir modeling by delivering a robust simulation model with high predictive value for future field development planning.