A Multiscale Workflow for History Matching in Structured and Unstructured Grid Geometries

This paper discusses the application of an efficient streamline-based inversion method to a large multi-well multi-tracer Partitioning Interwell Tracer Test (PITT) in the McClesky sandstone, Ranger Field, Texas, to characterize both permeability and oil saturation distribution. During a typical PITT, a conservative and a partitioning tracer are injected into the reservoir. The partitioning tracer gets partially absorbed into the oil phase, leading to a separation in the tracer responses that can be used to infer oil saturation distribution in the tracer-swept area. Our approach is extremely efficient because it relies on analytic computation of the sensitivity of the tracer response to reservoir parameters such as permeability and saturation using a single streamline simulation. We follow a two step procedure whereby we first match the conservative tracer response to determine the permeability distribution and then match the partitioning tracer response to obtain oil saturation distribution in the reservoir. The entire history matching took less than 6 hours in a PC as opposed to several months typically required for a manual history matching. We compared our results to a manual history match obtained using a finite-difference simulator. Both the manual history matching and the streamline-based inversion identified similar large-scale trends in permeability and saturation distribution. However, well-specific matches were significantly improved over those obtained through the manual history matching. Our approach is much more efficient in terms of computation time and effort and the results are less sensitive to personal bias compared to manual history matching. Finally, we discuss a procedure to assess the results in terms of resolution of the estimates of permeability and saturation distribution.

High resolution sequence stratigraphical methods have been used in the detailed geological modeling of an element of the Ness formation of a North Sea field. This updated geological description has been used as input to the stochastic geological model MOHERES. 14 realizations of facies architecture and petrophysical properties have been generated. The stochastically generated realizations have been scaled up to a refined element of an existing deterministic, history matched reservoir simulation model for the full Upper Brent reservoir. The upscaled versions of the realizations have been connected to the full field model, and reservoir simulations have been performed to compare twelve years of production history with simulated results for well production, RFT- and PLT-data. Six of the realizations gave good to very good results when compared to measured production data. These realizations were used for simulation of the future production performance to the year 2010, making it possible to estimate its uncertainty.

A particularly efficient flow solver can be obtained by combining a recent mixed multiscale finite-element method for computing pressure and velocity fields with a streamline method for computing fluid transport. This multiscale-streamline method has shown to be a promising approach for fast flow simulations on high-resolution geologic models with multimillion grid cells. The multiscale method solves the pressure equation on a coarse grid while preserving important fine-scale details. Fine-scale heterogeneity is accounted for through a set of generalized, heterogeneous basis functions that are computed numerically by solving local flow problems. When included in the coarse-grid equations, the basis functions ensure that the global equations are consistent with the local properties of the underlying differential operators. The multiscale method offers a substantial gain in computation speed, without significant loss of accuracy, when the multiscale basis functions are updated infrequently throughout a dynamics simulation. In this paper we propose to combine the multiscale-streamline method with a recent ‘generalized travel-time inversion’ method to derive a fast and robust method for history matching high-resolution geologic models. A key point in the new method is the use of sensitivities that are calculated analytically along streamlines with little computational overhead. The sensitivities are used in the travel-time inversion formulation to give a robust quasilinear method that typically converges in a few iterations and generally avoids much of the subjective judgments and time-consuming trial-and-errors in manual history matching. Moreover, the sensitivities are used to control a procedure for adaptive updating of the basis functions only in areas with relatively large sensitivity to the production response. The sensitivity-based adaptive approach allows us to selectively update only a fraction of the total number of basis functions, which gives a substantial savings in computation time for the forward flow simulations. We demonstrate the power and utility of our approach using a simple 2D model and a highly detailed 3D geomodel. The 3D simulation model consists of more than one million cells with 69 producing wells. Using our proposed approach, history matching over a period of 7 years is accomplished in less than forty minutes on an ordinary workstation PC.

The paper describes reservoir management challenges during the first 5 years of the production from the Terra Nova oil field, located offshore Newfoundland and Labrador, Canada. The purpose of the paper is to share lessons learned in the appraisal strategy, data monitoring, subsurface modeling and production management of this subsea/ floating, production, storage, offloading (FPSO) development. The material covers appraisal history, reservoir description, production history, reservoir monitoring, reservoir management, lessons learned and future challenges. The paper will discuss the need for a balanced approach regarding development and delineation of the resources. Since there is no gas export from Terra Nova or any of the Grand Banks fields, surplus gas has to be re-injected. The Terra Nova depletion plan requires that the production from the water injection region and the gas flood regions be balanced for optimum pressure management. The relative richness of the injection gas from the Terra Nova production process has the potential to create miscible gas-oil conditions. We will discuss how slim tube experiments and compositional tracking of the fluids have demonstrated benefits to the oil recovery by operating the gasflood at a reasonable pressure. The Terra Nova field lies not only in a harsh environment but also in an immature, low infrastructure area. This has led to additional challenges regarding reservoir management and future development. Production behavior and history matching have shown that seismically defined fault blocks are transmissible rather than sealing. The presence of leaky faults is beneficial for oil recovery, but presents challenges for precise reservoir simulation modeling and history matching. Terra Nova has extensive amount of data from bottom hole pressure gauges, wireline formation pressures (e.g. RFT/RCI), water cut trends GOR trends, gas tracers, etc. The paper discusses how these data have been integrated in the reservoir management process. Various history matching approaches will be discussed; both manual history matching and assisted techniques (experimental design). This is to the best of our knowledge the first SPE paper from the Grand Banks oil province that describes integrated reservoir management challenges particular to this region.

Model calibration or history matching has commonly been conducted on a single deterministic model by a "manual" trial-and-error approach. With recent advances in the application of design of experiments for using numerical models in probabilistic forecasting, the feasibility of the manual history match process used for a single deterministic model has become questionable. In manual history matching, a structured approach is used in which the sequence of adjustments has been from global, then to flow units, followed by local changes in model properties. However, the applicability of this approach to probabilistic history matching has not been demonstrated. Moreover, the industry's lack of experience in using assisted history-matching tools and the careless application without using a structured logic that is based on engineering judgment can lead the users to several potential pitfalls resulting in unrealistic solutions. This work shows that in using evolutionary algorithms, a structured (staged) assisted history-matching methodology can be applied by considering probabilistic ranges of relevant input parameters and tailored objective functions for each stage of the process. The key "heavy-hitter" parameters with the highest impact on the history-match process are identified and introduced through conducting sensitivity runs. A physically-sound and proper set of parameters with realistic ranges are introduced at each stage of the history-match process in a logical order. This paper shows that a workflow can be designed so that the ranges of selected parameters that are used to attain the best solutions at each stage can be carried over to the next stage to continue the history-match process. Since this workflow is conducted in a probabilistic semi-automatic manner, a diverse set of solutions can be obtained with the flexibility to guide the process as is traditionally done in manual history matching. The workflow is demonstrated through its application to history match production data from the Tengiz super-giant carbonate oil field located on the shores of the Caspian Sea in the Republic of Kazakhstan.

Ganna field is a part of North Bahariya concession which is located in Abu El-Gharadig basin, Western Desert, Egypt. The field was discovered and started production in 2004 through the well Ganna-1 from Middle Abu Roash “G” reservoir. Five main reservoirs were tested and produced during the field life namely Abu Roash “E”, Upper Abu Roash “G”, Middle Abu Roash “G”, Lower Abu Roash “G”, and Upper Bahariya. A 3D reservoir simulation study was performed to aid in field future development. Major challenges appeared during history matching process. The main challenge was the commingled production wells producing from multiple reservoirs with different characteristics. Even most of the static pressure points were measured commingle as well. Post-production RFT data was the key utilized to calibrate the history match and reach the most accurate reservoir characterization for each reservoir individually. The history match strategy depended only on global modifications using the new commercialized optimization techniques. Previously, model calibration or history matching has commonly been conducted on a single deterministic model by “manual” or “Trial and Error” approach. With recent advances in the application of uncertainty and optimization in numerical reservoir simulation it became a must to assess the effect of different parameters not only during history matching but also its effect on the forecast through producing a probabilistic forecasting (3). The feasibility of using the manual history matching with a single deterministic model has become questionable. A Physically-sound proper set of parameters with realistic ranges are introduced at each stage of the process in logical order and introduced to optimizers to get the best history matched case with a possibility of considering more than one history matched case for a probabilistic forecasting scenarios (4). This paper will present the approach utilized for calibrating the comingled production wells throughout post-production RFT data using the new uncertainty & optimization module to obtain the best global matched model.

SummaryIt is well documented that history matching is a problem with possibly nonunique solutions. In the past few years, several automated or semiautomated history-matching algorithms have been proposed. Depending on the algorithm used, it is possible that the final estimated reservoir-property distribution that allows for a good history match may not be geologically realistic. Therefore, there is a need to include other constraints to generate multiple, geologically realistic history-matched realizations. These constraints might, for example, include the variogram, a training image, the distribution of net-to-gross, pore volume, or other geostatistical information about the reservoir. This inclusion is particularly useful because it introduces uncertainty information in the reservoir description when we have limited history from existing wells in the field and intend to drill infill wells or implement a secondary-recovery process.The algorithm proposed in this paper uses multiresolution wavelet analysis to integrate history data with the geostatistical information contained in the variogram proposed for the reservoir. Wavelets allow the representation and manipulation of property distributions at various resolutions at the same time. Using wavelets, information from different sources such as production history and seismic surveys (that would be at different resolutions) can be incorporated directly at the appropriate resolution level. In the first step, we fix the wavelet coefficients sensitive to the history-match data. This has the effect of fixing the field history without fixing individual gridblock properties. In the second step, the remaining free wavelet coefficients are modified to integrate variogram information into the reservoir description. Generating multiple realizations of only the second set of wavelets coefficients results in multiple history-matched, variogram-constrained descriptions of the reservoir. The computational investment is very modest because the history match is done only once.In a number of example cases, different areal Gaussian fields with varying amounts of available production-history data were studied to test the algorithm. It was found that the wavelet coefficients constraining the history can be decoupled from those constraining the variogram. The implication of this observation is that the history data and variogram can be integrated sequentially into the reservoir model—that is, after the initial history match, new information can be added to the model without disturbing the original match to yield multiple history-matched and geostatistically constrained realizations.

Monument Butte field produces oil from 22 commingled sandstone reservoirs across a 2200-ft interval of the Green River formation. Total (primary and secondary) recovery is less than 10% of the OOIP. Green River formation sands are lenticular, lacustrine fluvial-deltaic units that have limited aerial extent and vertical communication, resulting in poor interwell connectivity. Moreover, the reservoirs sands have low permeability. To determine interwell connectivity and to identify methods to improve oil recovery, we conducted detailed reservoir characterization and simulation studies in a pilot area of Monument Butte field. We built a 3D geostatistical model and ranked geological realizations using production history matching with streamline simulation. Detailed petrophysical and geological modeling were used to quantify interwell connectivity of sands as a function of well spacing. Studies of interwell connectivity were also used to identify flow barriers that we applied in upscaled reservoir simulation models to better represent interwell connectivity. A combination of manual and assisted history matching methods was used to calibrate the reservoir model. Primary, secondary, and tertiary development strategies were then evaluated for future optimization of oil recovery for new developing areas in Monument Butte field. This study yielded a reservoir model much improved over those of previous studies. A good history match was obtained for most wells, despite the complexity of this stacked, heterogeneous reservoir system. Simulation results indicate that interwell connectivity has the greatest impact on reservoir performance. Interwell connectivity was demonstrated for only major sands, and interwell connectivity increases as well spacing decreases for these sands. A pilot 20-acre infill drilling study is needed to verify this conclusion. The overall connectivity for the combined 22 zones is 0.056 (connected adjacent well pairs divided by total adjacent well pairs). The connectivities for major sand zones evaluated by history matching are 0.65, 0.16, 0.23, and 0.01 for the D1 sand, C sands, A sands, and LODC sands, respectively. Simulation results indicate that reservoir performance can potentially be improved with various reservoir management strategies, including infill drilling. For new development areas, water flooding should start as soon as possible (within 6 months). Tertiary recovery methods may have potential in the Monument Butte field, given the low recovery efficiencies for primary and secondary recovery. Water-alternating-CO2 injection appears to be superior to other tertiary recovery methods. Further evaluation and piloting of tertiary methods is required, given the connectivity risk. The target area study results can be used to guide the development of other parts of the large Monument Butte field.

Using commercial numerical reservoir simulators to build a full field reservoir model and simultaneously history match multiple dynamic variables for a highly complex, offshore mature field in Malaysia, had proven to be challenging, manpower intensive, highly expensive, and not very successful. This field includes almost two hundred wells that have been completed in more than 60 different, non-continuous reservoir layers. The field has been producing oil, gas and water for decades. The objective of this article is to demonstrate how Artificial Intelligence (AI) and Machine Learning is used to build a purely data-driven reservoir simulation model that successfully history match all the dynamic variables for all the wells in this field and subsequently used for production forecast. The model has been validated in space and time. The AI and Machine Learning technology that was used to build the dynamic reservoir simulation and modeling is called spatio-temporal learning. Spatio-temporal learning is a machine-learning algorithm specifically developed for data-driven modeling of the physics of fluid flow through porous media. Spatio-temporal learning is used in the context of Deconvolutional Neural Networks. In this article Spatio-temporal Learning and Deconvolutional Neural Networks will be explained. This new technology is the result of more than 20 years of research and development in the application of AI and Machine Learning in reservoir modeling. This technology develops a coupled reservoir and wellbore model that for this particular oil & gas field in Malaysia uses choke setting, well-head pressure and well-head temperature as input and simultaneously history matches Oil production, GOR, WC, reservoir pressure, and water saturation for more than a hundred wells through a completely automated process. Once the data-driven reservoir model is developed and history matched, it is blind validated in space and time in order to establish a reliable and robust reservoir model to be used for decision making purposes and opportunity generation to maximise the field value. The concepts and the methodology of history match of multiple wells, individual offshore platforms, and the entire field will be presented in this article along with the results of blind validation and production forecasting. Results of using this model to perform uncertainty quantification will also be presented. A case study of a highly complex mature field with large number of wells and years of production has been used to be studied and simulated by this data-driven approach. Time, efforts, and resources required for the development of the dynamic reservoir simulation models using AI and Machine Learning is considerably less than time and resources required using the commercial numerical simulators. It is validated that the TDM developed model can make very reasonable prediction of field behavior and production from the existing wells based on modification of operational constraints and can be a prudent complementary tool to conventional numerical simulators for such complex assets.

We explore the use of efficient streamline-based simulation approaches for modeling partitioning interwell tracer tests in hydrocarbon reservoirs. Specifically, we utilize the unique features of streamline models to develop an efficient approach for interpretation and history matching of field tracer response. A critical aspect here is the underdetermined and highly ill-posed nature of the associated inverse problems. We have investigated the relative merits of the traditional history matching ('amplitude inversion') and a novel travel time inversion in terms of robustness of the method and convergence behavior of the solution. We show that the traditional amplitude inversion is orders of magnitude more non-linear and the solution here is likely to get trapped in local minimum, leading to inadequate history match. The proposed travel time inversion is shown to be extremely efficient and robust for practical field applications. The streamline approach is generalized to model water injection in naturally fractured reservoirs through the use of a dual media approach. The fractures and matrix are treated as separate continua that are connected through a transfer function, as in conventional finite difference simulators for modeling fractured systems. A detailed comparison with a commercial finite difference simulator shows very good agreement. Furthermore, an examination of the scaling behavior of the computation time indicates that the streamline approach is likely to result in significant savings for large-scale field applications. We also propose a novel approach to history matching finite-difference models that combines the advantage of the streamline models with the versatility of finite-difference simulation. In our approach, we utilize the streamline-derived sensitivities to facilitate history matching during finite-difference simulation. The use of finite-difference model allows us to account for detailed process physics and compressibility effects. The approach is very fast and avoids much of the subjective judgments and time-consuming trial-and-errors associated with manual history matching. We demonstrate the power and utility of our approach using a synthetic example and two field examples. We have also explored the use of a finite difference reservoir simulator, UTCHEM, for field-scale design and optimization of partitioning interwell tracer tests. The finite-difference model allows us to include detailed physics associated with reactive tracer transport, particularly those related with transverse and cross-streamline mechanisms. We have investigated the potential use of downhole tracer samplers and also the use of natural tracers for the design of partitioning tracer tests. Finally, we discuss several alternative ways of using partitioning interwell tracer tests (PITTs) in oil fields for the calculation of oil saturation, swept pore volume and sweep efficiency, and assess the accuracy of such tests under a variety of reservoir conditions.

Enhancing model parameterization with linearly constrained deep generative network for ensemble-based history matching

As the role of reservoir flow simulation increasingly impacts existing operations and field development decisions, it follows that rigor, fitness and consistency should be imposed on the calibration of reservoir flow models to dynamic data through history matching. Although a wealth of history matching techniques exist in the petroleum literature that propose novel algorithms or share case studies, seldom does the content guide the modeler in fit-for-purpose reservoir model calibration for an operating asset. To evaluate the applicability of these diverse techniques against standards required for reservoir management, an internal study was performed to benchmark four assisted history matching (AHM) techniques commonly promoted in the oil and gas industry. The techniques were vetted against a comprehensive suite of modeling requirements for multiple asset classes, integrating a variety of historical dynamic data types through the calibration of reservoir properties that control flow behavior from the field to inter-well scale. The methods benchmarked were: (1) Design of Experiments (DoE)-based, (2) Ensemble Kalman Filter and Ensemble Smoother, (3) Genetic Algorithm and (4) Generalized Travel Time Inversion. This manuscript focuses solely on the DoE-based technique.In order to consistently benchmark the techniques, a set of standards was defined against which each was evaluated to determine its suitability for widespread history matching applications. The standards involve: the capacity to parameterize (and therefore calibrate) a diversity of reservoir flow model attributes, the capacity to integrate different types of dynamic data, the level of independence from the flow simulator and the capability to provide probabilistic outcomes for predictive uncertainty assessment. Of the four techniques, the DoE-based approach uniquely satisfied all requirements. Its history matching workflow has the flexibility to incorporate any form of reservoir model parameter and to assimilate a history matching error metric for any individual or group of historical data types; therefore, benchmarking established DoE-based techniques as unambiguously the most compliant with generic asset modeling requirements. The approach was also identified as the most straightforward, both theoretically and in practical computation, and therefore applicable to the broadest range of practitioners. Perhaps most importantly, the approach demonstrated the capacity for accurate quantification of uncertainty (or non-uniqueness) in reservoir quality resulting from an exhaustive, although approximate, exploration of model parameter space and the associated history matching error metric(s).This manuscript compiles the results and insights gained from benchmarking of the DoE-based techniques through the proposal of a comprehensive assisted history matching workflow. The workflow is designed for generality while providing best practices that guide the modeler in fit-for-purpose application. Limitations of the workflow are also recognized. Key components include: selection and screening of calibration parameters based on statistical significance, development of surrogate models to characterize the relationship between parameters and the simulated historical data being integrated into the reservoir model, use of the surrogate models for exhaustive yet efficient exploration of parameter space to identify (non-unique) history matched models, and (deterministic or probabilistic) discrete reservoir model selection for use in forecast-based decision making. Each step of the AHM workflow is presented from a conceptual and applied perspective, and field applications are provided to demonstrate key concepts. Although the applications presented include two deepwater Gulf of Mexico assets, the workflow and insights provided are developed from benchmarking across a diverse suite of asset types.

Surrogate modeling is essential in reducing computational costs for history-matching applications. Yet, traditional deep learning-based surrogate models cannot cope with high dimensional input parameters, such as the permeability field. This work introduces a robust method to automate the history matching process utilizing the Bayesian inversion assisted by a hybrid convolutional neural network and long short-term memory (CNN-LSTM) model and principal component analysis (PCA) method. The method includes five main steps. Step 1: Generate a high-spatial permeability field using a geostatistical approach. Step 2: use the PCA to reduce the dimensionality of the permeability fields, followed by using PCA to generate permeability fields and perform simulations. Step 3: construct the CNNLSTM to map the nonlinear relationship between the extracted features from PCA and the sequential outputs, such as the pressure response. Here, Bayesian optimization is employed to automate hyperparameter tuning. Step 4: perform the Bayesian inversion to inverse the high dimensional inputs, e.g., permeability field, in which the CNN-LSTM serves as the forward model to reduce the computational cost. The inversed PCA features are then fed into the PCA to recover the high dimensional inputs. Step 5: check convergence and if the errors are significant between the inversed high dimensional permeability field and the ground truth, revisit the construction of the CNN-BiLSTM and the prior information for the uncertainty parameters. A 2D reservoir model demonstrates the proposed history-matching method. We can inverse the high dimensional inputs (e.g., permeability field) with minor errors between the prediction and ground truth. We propose a Bayesian inversion assisted by a hybrid CNN-LSTM model and PCA method for high-dimensional parameter inversion, which is superior to the traditional models regarding accuracy and efficiency. This method enables us to perform history matching for reservoir simulation with high dimensional inputs and significant uncertainties.

Reservoir model history matching is a complex, time-consuming, and resource intensive process that needs to be carried out carefully for building reliable predictive tools to manage Oil & Gas assets. Reservoir models encompass detailed geological description representing subsurface heterogeneities that influence its dynamics. To intelligently manage and preserve the complexity of the reservoir models, an artificial intelligence, Progressive-Recursive Self-Organizing Maps (PR-SOM), algorithm was developed. PR-SOM is an unsupervised artificial intelligence neural network algorithm that classifies the reservoir grid cells into progressive reservoir parameters to identify similarly adjoining regions. The algorithm explores and identifies model geo-bodies with similarities and dissimilarities in a progressive and recursive manner. This allows history matching to be conducted on much smaller subsets of the reservoir model of similar geological features. In this work, an artificial intelligence (AI) algorithm was applied, first, to guide the reservoir-wide history match processes. Next, the algorithm was applied to fine-tune well performance using information form well testing and historical data. The algorithm uses both static properties (permeability, rock quality indices, porosity, flow zonation … etc.) and dynamic properties (pressure or saturation) to construct similarities matrix. The results show that the clusters’s growth is progressive, controlled and quality assured by accounting for the controlling reservoir parameters. The number of mapped regions (clusters) is determined by optimizing the similarity matrix recursively. The quality of the global reservoir history match shows the effectiveness of the algorithm, better quality matching for historical production data, and fewer iterations (i.e. less simulation runs). The process is repeated to calibrate the reservoir model near wellbores by limiting the AI algorithm to only the drainage regions seen from well tests and historical data. The results show that employed AI-guided history matching revealed similarities and dissimilarities in the reservoir model. That not only enhance field and well match, but also allowed us to maintain the heterogeneity contrasts inherited from the Earth model. The advanced algorithm was successfully used to assess the extent of geological heterogeneity and its impact on reservoir dynamics, to enhance history match quality, minimize human interaction, and to reduce computational requirements.

This paper describes the results of an integrated study to assess redevelopment options in a mature field with over 40 years of production. The field has been produced under gas cap drive and partly under an internal gas injection scheme to increase oil recoveries. The objective of the study was to confirm the effectiveness and possible expansion of the gas injection scheme and to assess merits of waterflooding. A two phased modeling approach was adopted and a coarse Phase 1 model for the entire reservoir stack was built using existing interpretations and data available from previous studies while seismic reinterpretation and petrophysical reevaluation to feed the detailed Phase 2 models was ongoing. The Phase 1 model helped to understand the plumbing between the sands and fault blocks and the key parameters impacting field level pressure and cumulative fluid history matches. In Phase 2, two separate models were built for groups of reservoirs at a much finer vertical scale to address sand specific saturation changes and individual well performance. A structured experimental design (ED) methodology using SUM (Shell proprietary software) was used to explore the full uncertainty space for potential history matches and use calibrated models with the remaining uncertainties to generate a range of forecasts. With the large number of reservoirs involved even a modest number of uncertainties resulted in a large number of ED parameters. The number of parameters and their associated uncertainties was partly narrowed down and constrained in Phase 1 allowing focusing on detailed reservoir specific parameters in Phase 2. In parallel with and complementing the simulation work, conventional contact mapping integrating the historical pressure, water cut and GOR data was carried out to locate potential areas with remaining oil. This paper will highlight how the phased modeling approach facilitated early integration of reservoir and well performance data for detailed static model construction, ways of managing uncertainty in a brown field with multiple reservoirs, large number of wells and more than 40 years of production history, and finally how traditional reservoir assessment techniques can complement reservoir modeling work.