Challenges remain in the upscaling and flow simulation of reservoir models from strong heterogeneity that may arise when representing complex patterns of connectivity and barriers. This is especially true in high contrast systems, e.g. for carbonate reservoirs, where statistical upgridding and upscaling approaches developed for clastic reservoirs perform less well. This has led to the development of a novel "Distance Based" upgridding technique which we combine with "Diffuse Source" upscaling to successfully simulate such models. We replace previously developed variance-based statistical sequential layer grouping reservoir coarsening analyses, with a novel distance-based calculation. It relies on the local errors in the interstitial velocity and the time-of-flight introduced when layers are grouped, as a measure of distance between reservoir models. The use of a distance measure allows for the inclusion of flow capacity in this calculation, and avoids the strong biases that arise from the previous variance-based approaches, especially with high contrast systems. We utilize the "Diffuse Source" (DS) upscaling approach to obtain the intercell transmissibility and well indices for the upscaled reservoir model. The DS approach is an extension of pseudo-steady-state (PSS) flow-based upscaling that utilizes the diffusive time of flight to identify well-connected sub-volumes in each adjacent pair of coarsened reservoir grid cells. DS upscaling retains the same localization advantage as the PSS approach. Unlike steady state upscaling, there is no coupling to a global flow field and local-global iterations are not required. DS upscaling was previously developed for more general reservoir coarsening, but the description now includes an extremely simple implementation for 1×1×N layer coarsening that naturally avoids the issues that arise in the use of the harmonic average for the vertical transmissibility. Starting from a high-resolution fine scale 3D geologic model, sequential layer grouping provides us with a series of increasingly coarsened reservoir models. Each model in turn minimizes the integrated distance between the fine and coarsened models, which is used as a measure of heterogeneity lost during coarsening. From these layer designs we apply a combined cost and heterogeneity objective function to determine an optimal layer design, which is then used for upscaling and flow simulation. We show that this distance-based optimal layer design does not experience the over-grouping of layers that arose from the previous variance-based approach. The new approach has been able to integrate the flow capacity similar to a Lorenz plot into the calculation of distance between reservoir models to include the impact of reservoir quality. This replaces the simple use of a net-to-gross cutoff utilized in previous work. The distance calculation uses a hyper-volume Lebesgue measure which provides a consistent means of combining different physical attributes: in this case interstitial velocity and the time-of-flight. The generalization of the hyper-volume to multiple properties, e.g., anisotropic permeability, is straight-forward. 1×1xN layer grouping is examined specifically to show the improvement in the optimal layer design compared to the previous statistical analyses. Areal coarsening is increased beyond 1×1 to show the general applicability of the DS upscaling approach. The combination of distance-based upgridding and DS upscaling is tested and performs extremely well on a series of sector, outcrop and full-field 3D reservoir models using a research simulation code and a commercial finite difference simulator. These models include the SPE10 reference model, the Amellago carbonate model, and additional full field examples. A novel distance-based measure of reservoir heterogeneity has been developed and applied to the design of an optimal reservoir simulation layering scheme, given a prior 3D geologic model. The distance measure combines elements of the Lorenz plot and previous variance-based analyses to avoid the strong biases and collapse of layering seen in the earlier approaches. Flow simulation based on Diffuse Source upscaled properties are shown to perform extremely well compared to the fine scale model.

Reliable and representative models and detailed characterization of a geologically complex reservoir are crucially important in having better understanding of the reservoir behavior and directly guiding to the most efficient implementation of the field development plan. This study focused on developing models of the Upper Cretaceous Waha carbonate and Bahi sandstone reservoirs and the Cambrian-Ordovician Gargaf sandstone reservoir in the Meghil field, Sirte Basin, Libya. The goals of the study were to characterize the vertical and lateral spatial continuity of each of the three formations and to calculate deterministic and probabilistic volumetrics. The Meghil Field, discovered in 1959, is located on the Zelten Platform and regionally classified as an extensional area of the larger Zelten field. Nineteen wells were drilled targeting the primary reservoir interval of interest at a drilled depth of around 8000 feet. A 3D seismic program was conducted to develop detailed structural maps for the Kalash and Waha/Bahi/Gargaf formations to evaluate future gas field development program. The field is characterized by three slightly asymmetrical anticlinal traps trending NW-SE. Major and minor faults that cut the interior of the structure were detected in the seismic block. The available drill stem DST and production tests were used to evaluate the level of communication between the structures. The structural framework for a 3D reservoir model is based on the interpretation and integration of the seismic volume and the available well logs. The well log data show that the net hydrocarbon bearing zone thickness is about 270 feet, the average porosity ranges from 4% in the Bahi/Gargaf sandstone to 13% in the Waha limestone. The average water saturation ranges from 15% to 32% in the Waha limestone and the Bahi/Gargaf sandstone respectively. Geostatistical models were developed using the well log and core data along with the structural model developed from the 3D seismic volume. The models suggest that porosity decreases towards the flanks and that separate flow units are likely present. The deterministic and stochastic give estimates of the original gas in place of about 830 Bscf and 732.2 Bscf for the upper and lower reservoir intervals, respectively. This study demonstrated the potential for significant additional hydrocarbon production from the Meghil field as well as the potential impact of heterogeneity on well placement and spacing.

Unconventional tight reservoirs currently make up more than 60% of domestic oil and gas production in the United States. However, developing unconventional formations requires intensive drilling and completion campaigns to maintain steady production of a field. Therefore, the prediction of estimated ultimate recovery, which measures the producible reserve from a well, is demanding, particularly as operators becomes more rational under the current volatile market conditions. Despite unconventional reservoirs being considered a resource play with low geological risks, their economic appraisal is challenged by unknown stimulation outcomes and intricate producing mechanisms. Therefore, this work aimed to leverage machine-learning techniques with big data to analyze the multivariant relationship of geological and engineering parameters with unconventional reservoir production and to improve the prediction of estimated ultimate recovery in unconventional formations. In this case study, a multiscale machine-learning workflow was deliberated and applied to a big data set from the Eagle Ford shale. First, quality control and feature selection were performed on a data set consisting of 4,067 wells with 30+ geophysical, petrophysical, drilling and completion, and production features. Then, a regional inferencing model, based on a K -nearest neighbor with bagging algorithm, was trained to obtain the spatial trend of estimated ultimate recovery across the Eagle Ford formation. The last part of the analysis was to build a local-scale prediction model. With the study area confined to East Texas, a random forest regression was performed to rigorously predict oil and gas estimated ultimate recoveries. The selected training features were finalized based on the results of a higher-dimension regression, as well as domain knowledge. Overall, the data-driven model trained with physically controlled data captured the production behavior of the Eagle Ford shale. The application of the proposed workflow on the Eagle Ford shale demonstrates a progressive building of the machine-learning model. The quality control of data allows global inspection of the data set and, more importantly, confirms the statistical distribution of training data. This study emphasizes the philosophy of multiscale data analytics. The large-scale model portraying sweet spots using location variables grants direct guidance for acreage acquisition and development across the basin; the small-scale model trained with reduced dimensionality generates quantitative prediction of oil and gas estimated ultimate recoveries for an area of interest. Compared with our previous work using higher dimensionality and extensive spatial interest, this progressive learning maintains similar explained variance in out-bag model check, but grants 26% and 52% reductions in mean square error for predicting oil and gas estimated ultimate recoveries in the Eagle Ford formation. In the end, prediction validation is performed by revisiting the data set. Overall, the proposed workflow demonstrates successful application in the Eagle Ford formation such that it can be directly implemented for other unconventional resource plays.

We propose a novel approach for rate optimization during a waterflood under geologic uncertainty in reservoir properties such as permeability and porosity. The traditional approach typically involves several runs of the forward simulator. This may not scale well when the optimization is to be performed at the full field-level and over multiple geologic realizations. A machine-learning (ML) based approach which is quick and scalable for rate optimization over multiple geologic realizations is proposed instead. The training data for the model is generated by running the forward simulator with randomly assigned well rates using multiple geologic realizations. A reduced order representation of the permeability heterogeneity in each of the realizations is derived using a grid connectivity transformation (GCT). This step involves finding basis functions corresponding to the different modal frequencies of the grid connectivity represented by the grid Laplacian. The projection of the heterogeneous property field along these basis functions gives the basis coefficients that form the reduced order representation. Subsequently, for each training datapoint, streamlines are traced and the minimum time of flight (TOF) representing the tracer breakthrough time at each producer is recorded. The basis coefficients and well rates are fed to a machine learning model as input and the minimum TOF at the producers forms the output of the model. This trained model can then be used along with an optimizer for computing the optimal injection rates to maximize the injection sweep efficiency. This corresponds to minimizing the variance in the minimum TOF within each well group. Different architectures of neural network are tested using 5-fold cross validation to decide the best ML model to compute the streamline time of flight. The trained model is used to perform well rate optimization over multiple realizations of geology by using a risk tolerance penalty. The optimal well rates thus obtained are compared with two cases: a) equal well rates assigned to all injectors and producers and b) well rates obtained by optimizing over a single realization without considering the uncertainty in geology. The optimal well rates are seen to offer better oil recovery and sweep efficiency than both cases. The workflow is tested for a 50x50 two-dimensional (2D) heterogenous permeability field and for the SPE benchmark Brugge field, and is seen to result in significant improvement in oil recovery and sweep efficiency. A single forward run of the trained ML model is faster than the conventional simulator by about 3 orders of magnitude, making the approach suitable for large scale field application accounting for geologic uncertainty. The parsimonious representation of geologic heterogeneity and the use of ML for forward modeling makes the approach highly scalable and well-suited for full field applications.

The lacustrine delta sandbody deposited in the north of Albert Basin is unconsolidated due to the shallow burial depth, which leads to an ultra-high permeability (up to 20 D) with large variation and poor diagenesis. Log derived permeability differs greatly with DST results. Thus, permeability simulation is challenging in 3D geomodeling. A hierarchical geomodeling approach is presented to bridge the gap among the ultra-high permeability log, model and DST results. The ultimate permeability model successfully matched the logging data and DST results into the geological model. Based on the study of sedimentary microfacies, the new method identifies different discrete rocktypes (DRT) according to the analyis of core, thin section and conventional and special core analysis (e.g., capillary pressure). In this procedure, pore throat radius, flow zone index (FZI) and other parameters are taken into account to identify the DRT. Then, hierarchical modeling approach is utilized in the geomodeling. Firstly, the sedimentary microfacies model is established within the stratigraphic framework. Secondly, the spatial distribution model of DRT is established under the control of sedimentary microfacies. Thirdly, the permeability distribution is simulated according to the different pore-permeability relation functions derived from each DRT. Finally, the permeability model is compared with the logging and testing results. Winland equation was improved based on the capillary pressure (Pc) data of special core analysis. It is found that the highest correlation between pore throat radius and reservoir properties was reached when mercury injection was 35%. The corresponding formula of R35 is selected to calculate the radius of reservoir pore throat. Reservoirs are divided into four discrete rock types according to parameters such as pore throat radius and flow zone index. Each rock type has its respective lithology, thin section feature and pore-permeability relationship. The ultra-high permeability obtained by DST test reaches up to 20 D, which belongs to the first class (DRT1) quality reservoir. It is located in the center of the delta channel with high degree of sorting and roundness. DRT4 is mainly located in the bank of the channels. It has a much higher shale content and the permeability is generally less than 50 mD. Through three-dimensional geological model, sedimentary facies, rock types and pore-permeability model are coupled hierarchically. Different pore-permeability relationships are given to different DRTs. After reconstructing the permeability model, the simulation results are highly matched with the log and DST test results. This hierarchical geomodeling approach can effectively solve the simulation problem in the ultra-high permeability reservoir. It realizes a quantitative characterization for the complex reservoir heterogeneity. The method presented can be applied to clastic reservoir. It also plays a significant positive role in carbonate reservoir characterization.

The 3D geological model is an important tool, which is used for decision-making process at each step of reservoir lifecycle. In turn the quality of the model as well as field operating efficiency is directly dependent on the geologist's expertise level and requires continuous improvement of professional skills. In the digital technology era the software functions continuously enhance leading to increased number of trainings in order to implement new algorithms in the working process. The aim of the paper is to share the best practices of geological and petrophysical modelling on-job trainings applied in Scientific Technology Centre. There are several web-resources used in the company in order to store operational regulations, corporate educational trainings, original workflows, created plug-ins etc. By dint of special module in the working software the above mentioned data sources were implemented into the integrated knowledge platform. Now new software functions can be explored on-the-job and without any charges for external trainings. The module affords the development of interactive training cases for new functions study. The results of case accomplishment can be saved and available for competency assesment. Another key aspect of the module is the automatization of model quality control provided by application of guided workflows. Additionally, the solutions for some cases can be found during the crossfunctional co-operation among geologists, petrophysicists and petroleum engineers in online question platform, which is the part of the professional groups. The unique integrated approach for increasing the quality of geological models through training and education, which is developed and successfully implemented in Scientific Technology Centre, will be beneficial not only for other companies, which plan to create and apply the equivalent employee training system, but also for oil and gas industry in general.

Geostatistical-based models provide a considerable improvement for predictive reliability of dynamic models and the following reservoir management decisions. This study focuses on geostatistical modeling the Paleocene Zelten Carbonate reservoir in the Meghil field. The field was discovered in 1959 and production operations began in 1961. Nineteen wells have been drilled to date. The structural framework consists of three slightly asymmetrical anticlinal structures trending NW-SE with steeper dip on the SW flanks. Each of the structures are separated by major normal faults. Seismic interpretation suggests that carbonate build-ups are most likely present on the three separate structures. Edge detection was used to clarify the structural geometries and the presence of additional minor faults. Pillar gridding technique was used to develop the structural framework including four major faults that are partially sealed based on analysis of the available DST and production test data. Stratigraphic analysis indicates a local presentation of dolomitic limestone in the northern portion of the main and the western structures caused considerable litho-facies variation that impacted the distribution of the petrophysical properties. Basic and advanced formation evaluation the net reservoir thickness of about 15 feet with an average porosity of 17% and average water saturation of 35%. Geostatistical-based applications that combine the spatial statistics (e.g. the semivariogram) and the available well and core data were used to populate the reservoir model with porosity, permeability, facies (lithology), net/gross, and water saturation. A conceptual facies model was also used to constrain the reservoir property distributions. Sequential Gaussian Simulation (SGS) was used to populate the model with porosity and water saturation and Sequential Indicator Simulation (SIS) was used to populate the facies model with permeability. The modeling parameters (e.g. semivariogram, correlation coefficients) were significantly constrained by the limited number of wells. Based on the limited number of wells available the semivariogram analysis resulted in a spherical semivariogram model with major axis range of 1435 meters for porosity and 1800 meters for water saturation. Minor axis ranges were about 50% of the major axis ranges. Given the limited well data, a significant effort was made to document the potential impact of the semivariogram parameters on the original hydrocarbon in place (OHIP) estimates and the lateral stratigraphic continuity of reservoir properties. The deterministic approach resulted in place volume estimates of 60 MMBBL and the stochastic approach provided an estimate of 45 MMBBL.

The Gulf of Mexico, and more precisely the Wilcox trend, has long been considered as challenging area for developing profitable hydrocarbon fields. In fact, the safe drilling of deep offshore wells needs to take into account the geological and geomechanical complexities, generated by the different sedimentological and tectonic events that accompanied the development of the Wilcox trend. In the case of Buckskin field, located in Keathley Canyon protraction ( Figure 1 ), and in order to overcome those challenges, we developed a workflow that ranks all the parameters related to the geometry, the geology, the rock quality and the geomechanics characteristics of the reservoir. The core of the workflow is articulated around a probabilistic method that will assess the uncertainty of the productivity index, based on experimental design and Monte Carlo simulation. The proposed workflow allowed the optimization of the PI of the well thanks to a highly deviated reservoir section at a depth below 24,000’, combined with an optimal fracking job. Figure 1 Location of Buckskin field After a brief introduction of the general context of the Buckskin field, we will review the notion of the PI, then we will review the adopted process prior to the geological and geomechanical processes for modeling the rock properties and stresses and their corresponding uncertainties. The selected design of experiment approach will then be described then we will discuss the outcome.

Reservoir model calibration against dynamic response data is often constrained by a prior conceptual model of geologic scenario that specifies the expected types of spatial variability and features in the solution. However, geologists have significant uncertainty in developing a conceptual model, e.g., due to limited data, process-based modeling assumptions, and subjectivity. Therefore, it is prudent to consider the uncertainty in the geologic scenario when solving the model calibration problem as it will provide an opportunity to utilize the response data in supporting or rejecting the proposed scenarios. We present a new approach for geologic scenario identification based on dynamic response data by combining gradient-based inversion for feature extraction and a convolutional neural network for feature recognition. To compactly represent each scenario while ensuring data sensitivity, the approach relies on extremely low-rank parameterization of individual geologic scenarios based on the Principal Components Analysis (PCA). The PCA basis elements of each scenario are then combined to capture the salient features in any of the scenarios, or their possible combinations. An iterative least-squares formulation is then formulated to detect scenarios that are supported by the observed data. The inversion results in an approximate (smooth) spatial solution that contains the dominant spatial patterns. A pre-trained convolutional neural network (CNN) is then used to identify the relevant geologic scenarios based on the reconstructed spatial solution. Two main advantages of the workflow include: (i) the ability to combine different scenarios if supported by data, instead of evaluating individual scenarios, and (ii) efficient gradient-based implementation that does not require extensive forward simulation runs. In addition, the training of CNN is implemented using only geologic realizations without requiring additional reservoir simulation. The performance of the workflow is evaluated using tomographic inversion and model calibration in fluvial reservoirs.

The spectral simulation approach is a recently developed geostatistical method of stochastic reservoir property simulation. The method has theoretical advantages over classical methods and the present work studies its practical effectiveness applied to synthetic data and real case of oil reservoirs. The study provides analysis of simulations results and describes strengths and limitations of the spectral method as well as application domain where it is most efficient. The spectral method is based on Fourier analysis of well log data and consists of three major steps: decomposition of the well logs into Fourier series, simulation of Fourier coefficients in the interwell space and reconstruction of synthetic logs at every lateral point as sum of the Fourier series. The method was implemented in a software application and was used by authors for simulation of three-dimensional reservoir properties. The effectiveness of the spectral method was studied based on simulations of continuous variables on a synthetic model. Simulations performed by the spectral method were compared to simulation results obtained on the same data by the more traditional method of sequential Gaussian simulation (SGS) provided by a commercial software. Analysis of the results from quantitative and qualitative points of view showed that the spectral method performed better at reproducing vertical non-stationarities observed in well data. Practical applicability of the spectral method was demonstrated by simulation of porosity on a real oil field model, characterized by increase of porosity along depth with higher values of porosity closer to the bottom of the simulated layer. It was shown that the spectral method handled better this type of heterogeneity and reproduced distribution of porosity along the vertical axis closer to that of well data.

Eastern Siberia is characterized by the extremely complex geological structure. The main factors include multiple faults, trappean and salt tectonics, the complex structure of the upper part of the section (0–1200 m) and its high-velocity characteristic (5000–6000 m/s), the high degree of rock transformation by secondary processes, low formation temperatures (10–30°C), the mixed fluid composition (gas, oil and water), and low net thicknesses (5–7 m) of productive layers. The fields of the region are among the most complex ones in the world according to the BP Company's statistics. New seismic and geologic model based on complex analyses of core, well logs, well tests, seismic and electromagnetic data allowed the Gazpromneft-GEO company to drill a series of successful wells. Gazpromneft-GEO, LLC.holds three oil and gas exploration and production licenses within the Ignyalinsky, Vakunaisky and Tympuchikansky (Chona field) subsurface blocks (Russia, Eastern Siberia, Irkutsk Region and Republic of Sakha (Yakutia)). The area of the blocks is 6,855 sq.km, 3,050 sq.km of which are covered by the 3D seismic and high-density electric prospecting ( Fig. 1 ). 70 exploration wells were drilled. The geological oil reserves of the Chona field are about 500 Mt. Nonetheless, the level of study of the blocks is extremely irregular, despite such significant reserves. To transfer from the exploration stage to production the capital-intensive scope of exploration is required, which expands the exploration program in time and can influence the final economic profitability of the project. The optimization of drilling costs by minimization of drilling wells and the efficiency of their drilling is the key purpose of the project. The work was carried out within the frames of scientific research and field works at the Gazpromneft-GEO, LLC. fields in Eastern Siberia. The high-density full-azimuth ground-based seismic using the UniQ technology was performed in Russia for the first time. The electric exploration with the near-field time-domain electromagnetic method was carried out along the same lines for the first time in Russia as well. This allowed to form the high-density cube of geoelectric properties. Model based on the wells (Facies model, Petrophysics model) and field geophysical data (3D seismic survey, 3D electric exploration, gravimetric survey, magnetic survey) complexation was made. The use of the approach allows to reduce the number of wells required for exploration of fields by 40%.

A common pitfall in history matching is to falsely generalize the learning from local data to the entire field, which can lead to radical over-estimation of uncertainty reduction and bad reservoir management decisions. This problem is referred to as the local-global problem and in this paper a methodology is proposed to quantify and correct for the error arise from this problem. Most performance metrics in an oil field, such as estimated ultimate recovery (EUR), are field-level ob jective functions that depend on properties (e.g., porosity) over the entire field. On the other hand, most measurement data (e.g., BHP) are sensitive only to a local area around the wells and are thus susceptible to local variation of geological properties. Calibrating field-level objective functions and multipliers of global properties (e.g., porosity) to local well data over-estimates the reduction of global uncertainties. In this paper, we derived the formula to quantify error in the calibrated posterior distribution (S-Curve) resulted from the local-global problem, as well as a correction factor to recover the true posterior S-Curve. Through theoretical derivation, it is shown that the model error arise from the local-global problem is dependent on the magnitude of the global and local variation of the uncertain properties (e.g., porosity). The larger the local variation relative to the global variation, the larger the error is in the estimated posterior S-Curve. The error also depends on the variogram of the local variation, and the detection range of the data. The error is larger for case with long variogram for the local variation and short data detection range. In addition, this model error can be highly correlated for different measurement data points even when the measurement error for these data were independent. To address this local-global modeling error, a series of analytical and empirical formula is proposed, which has successfully corrected the error and greatly improve the posterior S-Curve for a series of cases. To the best of our knowledge, this is the first time the error from the local-global problem is quantified and corrected. The methodology proposed could help improve the reliability of the result from probabilistic history matching.

Reservoir modeling and the derived fluid production over time curves are a key part of the workflows associated with major capital project decisions. These models may be very complex and use a variety of geological constraints in an effort to develop the porosity, permeability, and saturation distributions used in dynamic models (with or without upscaling). Over time and partially in response to increased computing capability as well as the need for more realistically heterogeneous models, model size as measured by number of model cells and model complexity has increased but model-derived production forecasts remain optimistic. This paper, one of a series that now stretches back over a decade, addresses a number of modeling issues with the goal of (1) better understanding how modeling workflows may contribute to forecast optimism and (2) what reservoir modelers, both geologists and engineers, may do to reduce forecast optimism derived from their subsurface models by improved understanding of how model parameters such as grid size, number of grid cells, semivariogram parameters (e.g. the range), and number of geological/stratigraphic "control" surfaces used to constrain models. Adequate modeling of reservoir heterogeneity appears to require very to extremely large models (e.g. large number of small cells). Many of the parameters used to "control" heterogeneity including the semivariogram range parameter, the number of "detailed" stratigraphic layers, and the number of rock/facies "containers" or model regions appears to have only a small impact on forecast recovery.

History matching is an integral part of field development planning for oil and gas reservoirs. It can be viewed as an inverse modeling technique which utilizes information contained in the observed flow response variables like flow rate, well bottom hole pressure etc. to better quantify the spatial distribution of reservoir model parameters like permeability or porosity. As the reservoir parameters and flow response variables are often related by non-linear relationships, the solution to the inverse problem of history matching is often non-unique. This makes it a problem that can be better handled by stochastic approaches than deterministic approaches. The proposed approach called ‘Indicator-based Data Assimilation’ (InDA) is suitable for such problems. To initiate the process, multiple realizations of the reservoir model parameters are generated based on the available information related to the reservoir description. The ensemble of realizations serves as a measure of initial uncertainty in the spatial distribution of model parameters. Next, flow simulations are run on the ensemble of models and the difference between the observed and simulated flow response variables are used to update the reservoir parameters using InDA. With successive updates, the uncertainty in the model parameters is reduced and the spatial distribution approaches the "true" distribution. Considering the residual uncertainty of the final updated models, reliable field development planning decisions can be made. The proposed method is validated using a realistic reservoir with complex channel-like features emulating a reservoir formed in a fluvial depositional environment. Liquid rate data from existing wells are used for updating reservoir parameters for several time steps using InDA. A comparison of the spatial distribution of final model parameters with the reference model used for validation shows a good match. After the history matching period, existing and new infill wells are run in a forecast mode where the observed and simulated flow responses show a good match. As majority of oil reservoirs comprise of high permeability oil-bearing zones in form of channels passing across low permeability zones, the statistical permeability distribution is bimodal making it non-Gaussian. It is shown that ‘Indicator Transformation’ of variables used in InDA preserves the non-Gaussian structure of the permeability field in comparison to methods like ‘Ensemble Kalman Filter’ (EnKF) that are sub-optimal in such cases.

The uncertainties of fluvial reservoir geologic model are notably high due to complicated geological conditions and unknown strong heterogeneity. However, previous uncertainty analysis approaches mainly focus on qualitative evaluation. In this work, we proposed a novel workflow to quantify and optimize the geologic model uncertainties using the virtual outcrops. First, the 3D geologic model is built by the virtual outcrops and geology database is established by the fine description of modern sedimentations, geologic outcrops and dense spacing areas. Next, geologic modeling algorithm is optimally selected based on the complexities of target reservoirs, computational speed and the shape of sand bodies. In this work, three indicators are proposed to evaluate the accuracy of geologic models, including matching coefficient of digital grey images to virtual outcrops, consistency of braided stream facies morphology and connectivity of inter-well effective sand bodies. For the braided channel in Sulige field examples, the width of channel belt is 1500~3500m and the average sand thickness is 5m. For channel bar, the width is 350~650m, the length is 800~1500m and the thickness is 3~6m. The virtual outcrops help determine the vertical sequence and planar characteristics of sedimentary facies and sand bodies. The comparisons between established model and virtual outcrops indicate that the accuracy of geologic models increases as the denseness of hard data becomes smaller and the optimal well spacing and row spacing match up with the sand body size and average well spacing of studied area. The evaluation system proposed in this work demonstrates the degree of geologic reproduction, reasonability and partial uncertainty of the models to the real reservoir. The value of this work is to provide a novel practical approach to optimize and quantify the uncertainties of geologic model. Furthermore, the established workflow can further be applied to identify the most significant controlling factor to determine geologic modeling in unconventional reservoirs.

One major goal in reservoir engineering is to improve ultimate recovery factor. In reservoir models with complex geological settings, e.g., river channel reservoir models, the quality of estimation of the patterns of channel connectivity is critical in designing a reservoir development plan to improve ultimate recovery factor. However, geostatistical techniques based on two-point statistics, such as variogram model based methods, are not suitable to reproduce those complex geological connectivity patterns. Moreover, traditional history matching methods have difficulties in estimating and preserving channel connectivity patterns. Recently, multiple point statistics (MPS) methods have been developed to model the complex geological connectivity and generate discrete facies models (e.g., channel facies). In MPS, facies models, e.g., channel models, are sampled based on empirical facies probabilities that are estimated from a geological conceptual model, known as a training image (TI). A training image can be regarded as a databased of spatial structures that contain higher order (higher than 2 nd order) multivariate geostatistical information. As a result, MPS can generate facies models that are geologically consistent with the geological features contained in the training image. An advantage of MPS facies simulation is that it is convenient to condition facies models on hard data (e.g., well measurements) and soft data (e.g., seismic data). By taking this advantage, I have developed two MPS based history matching methods, which are able to reproduce geological consistent channel connectivity patterns while matching dynamic flow data. The adapted pilot point method strategically places pilot points in a reservoir model based on both sensitivity and uncertainty information of the model. Then facies type values at the pilot points are estimated and considered as hard data to condition facies simulations. This method enables conditioning MPS facies simulations on flow data, however, it only uses limited information from the reservoir via pilot points that have limited spatial coverage. The second method is a facies probability conditioning method, which infers a facies probability map (the same size as the reservoir model) from flow data and then considers the probability map as soft data to condition the facies simulations. Although this method has a much better coverage of the reservoir model, it suffers from generate facies models that are not consistent with the facies probability map. In this paper, I will show the results from the investigation on this inconsistency issue and explain the reason responsible for the issue: the intrinsic behavior of MPS facies simulation algorithm tends to disregard the facies probability map at the later stage of the simulation. More specifically, the facies patterns in the output of MPS facies simulation are primarily controlled by the first few grid cells along the corresponding random path and the facies simulations at the rest of the grid cells are dominated by the conditional probability estimated from the TI, while the facies probability map has a minimal impact on the outcome. Then, I proposed a new approach for conditioning MPS facies simulations on flow data by ensuring gaining more control on the simulations a t the early grid cells on the random path of a MPS simulation. I will also present numerical experiments to show that this approach leads to generation of more geologically consistent facies connectivity patterns.

Basin modelling encompasses a broad range of related geological disciplines used to represent and describe how sedimentary basins were formed and evolved over time. It is often employed, but not solely, to appraise their potential for hydrocarbons exploration and production. Regarding petroleum systems, discerning and possessing the ability to precisely quantify the uncertainties associated to the generation and accumulation of oil and gas are the keys to deliver relevant and exhaustive analysis since no model or interpretation can at present perfectly mirror reality. To achieve this, classical techniques, as Monte Carlo, rely on statistical concepts necessiting important computing power as they require thousands of simulations. Modern optimized algorithms combining statistical concepts such as Latin Hypercube Sampling and machine learning methods can significantly downplay the number of simulations and thus the computing time required to generate realistic alternative probabilistic outcomes. They turn out to be particularly effective when mastered but require expertise across numerous scientific fields further to an extensive basin modelling experience. Thence, they remain used by a limited community of experts, often limited to R&D centers, rather than being widely adopted in exploration practices. In addition, plenty of these experts will retire in the coming years without necessarily being able to pass the torch and too few geoscientists are adequately trained to master these methods, facing the industry with a dilemma. In this paper, the objective is to present an innovative workflow for risk and uncertainty analysis in basin modelling. By using already developed advanced statistics optimization approaches and combining it with a new parametrization procedure based on well-known geological concepts instead of independent variables, this new workflow is designed to reach a wider and less experienced community while downsizing computing time and remaining agile and pertinent. In illustration of the general parametrization procedure, three new specific methodologies to assess the richness, the reactivity and the thickness of a source-rock are described following the philosophy of the workflow. They have been tested on a real case study over the Levantine Basin and present promising results as they required 50 times fewer simulations than Monte Carlo approach to provide results coherent with expected geological contexts and physical behaviors of the basin.

This paper presents the development of a static model for a naturally-fractured High-Pressure-High-Temperature (HPHT) gas sand reservoir located in the Tarim basin, Western China. The study is part of a well placement optimization study. It is motivated by the big challenges of drilling a well at depths ranging from 6800m-8000m[AG1] in a HPHT environment. A detailed fine-scale model is required as input for the dynamic model. The static model is developed through an integration [AG2]process. It consists of both matrix and fractures. The matrix modeling started by integrating 3D seismic and log data to build the structural model. A new rock type scheme was developed by reconciling log and core data, including capillary pressures. Additionally, permeabilities are estimated at each uncored location using a two-step approach, namely trend estimation by regression analysis and variability simulation by 1D Gaussian simulation. The 3D modeling was executed in the order of least dependent to most dependent variable (i.e., from facies, to rock type, then followed by porosity, permeability and saturation respectively). From the geology, the sand bodies were interpreted to be continuous throughout the field. Discontinuous mudstone layers are sandwiched in-between the sand bodies. This information, together with outcrop data, is used to guide the spatial relationships in the model. Facies, rock type, porosity and permeability are simulated using geostatistical procedures. Meanwhile, saturation is generated based on the Leverett J-Function. To quantify the uncertainty in the various data, especially in the capillary pressure data, the porosity-permeability relationship, the gas-water contact and the surface tension of the gas-water system, a probabilistic model of the Gas Initially in Place (GIP) is created through uncertainty and sensitivity analysis. The origin of the fracture system was analyzed by developing a prototype of a conceptual model. The understanding from the prototype model is coupled with the 3D seismic, outcrops, drilling information, rock mechanics, image log, core, and dynamic data, to develop fracture characteristics and correlations. The discrete fracture system is modelled using a stochastic simulation approach, constraining it to the seismically-inverted fracture density map for each zone through well-seismic correlations and a nonlinear inversion [AG3]to build the Discrete Fracture Network (DFN). Finally, the fracture model is integrated with the matrix model by upscaling the DFN model into the grid system. Following the creation of the static model, a dual porosity model was prepared for dynamic modeling by maintaining consistency between the fine scale and upscaled models throughout the upscaling process. The methodology described above has produced a detailed fine scale model that shows consistency between properties and geology. This is a direct consequence of the new rock type system and the order in which the simulation was conducted. The facies model shows the continuity of the sand bodies, and the discontinuity of the mudstone, as indicated by the geological interpretation. The 3D Poro-Perm relationship shows the variability which is a reflection of the variability of the core data. The probabilistic distribution of the GIP is in agreement with the results of conventional reservoir engineering analyses, namely Material Balance and Rate Transient Analysis. Furthermore, the fracture distribution confirms the information both at the wells, as well as in-between the wells as given by the seismic interpretation. This study demonstrates that a reliable fine scale model can be developed to match the available data and interpretation by properly preparing the pre-requisite inputs and following the order of dependency in the reservoir attributes.

To reliably quantify changes in the in-situ stresses due to oilfield exploitation, it is necessary to start from a representative description of the subsurface and simulate both fluid movement and geomechanical effects. For this purpose, a new 3D "hybrid" grid is presented. This grid accurately honors geological features, with no unwarranted simplifications, while being ideally suited for geomechanical simulators and the new generation of flow simulators. The new 3D Hybrid Grid is dominated by hexahedrons arranged in a structured manner except around faults, where it is made up of tetrahedrons. The grid is constructed from the underlying geological model and the sealed fault network. Both structured (hexahedrons) and unstructured (tetrahedrons) parts follow stratigraphic deposition information. For geomechanical simulations, each compartment is considered as an independent mesh part allowing sliding effects along the faults. At the interfaces between the hexahedrons and tetrahedrons in a fault block, shared nodes are used to ensure stress equilibrium and displacement continuity. Uncertain states of stress and unforeseen changes in the integrity of the subsurface can have grave economic and environmental consequences. Understanding these helps mitigate development risks, and optimally develop the field. Today, geomechanical studies are not routinely performed and are often based on simplified descriptions of the subsurface. With this new Hybrid Grid, we propose to combine a reliable representation of subsurface with state-of-the art rock mechanics to assess how reservoirs respond to drilling, completion and production. Simplifications in either can lead to incorrect assessment of risks or production forecasts. The grid presented in this paper aims at honoring geology accurately while also being optimal for numerical computations. The zones of tetrahedrons enable the inclusions of even the most complex faulting systems, while the structured hexahedrons precisely follow the stratigraphy and are most efficient for geomechanical simulations. The gridding technology presented enables a coherent representation of the subsurface for constructing geological models for simulating both flow and geomechanics. Although such meshing schemes exist for modeling manufactured objects, these are difficult to apply to geological formations; our approach now enables them by guiding the meshing using the chronostratigraphic parameterization of the subsurface. It will allow engineers to routinely include the effects of stress changes during production and will build confidence in development plans.

The work deals with a conceptual geological model creation that is necessary for further geocellular modelling and field development scheme optimization. The object of the study is Lower Jurassic terrigenous formations U15 and U14 of Urman-Archin fields group located in south-eastern part of West Siberian basin. Low quality of these reservoirs requires detailed geological analysis for allocation of the most perspective zones. This work integrates the results of seismic, paleogeographic, palaeotectonic, logging, well testing, macroand microscopic core analyses. It was discovered that the formations were deposited in fluvial conditions. Moreover, the evolution of the river system is observed across the interval of U15-U14 formations – braided streams change to meandering ones. The sedimentary processes determine vertical and lateral lithological heterogeneities as well as the primary distribution of grains of different size characterizing the activity of depositional agent. However, it was found that lithology and grain size is not directly related to the presence of reservoirs, so sedimentary conditions are not the key factor of reservoir continuity. Analysis of diagenetic alternations of the formations’ rocks showed that their compaction influenced reservoir quality and its lateral heterogeneity. In some parts of the studied area the rocks are more compacted, and, consequently, they have lower porosity and permeability. Tectonic movements caused non-uniform rock compaction across the area. In addition, they led to reservoir compartmentalization. There are hydrodynamically isolated blocks, which may influence fluid distribution within the formations and should be considered during development scheme planning. Significant influence of diagenetic alternations on the reservoir properties is not typical for the deposits in terrigenous rocks within West Siberian basin, so this conclusion should be taken into account when analyzing similar reservoirs in this region.