A reservoir engineering Grid computing toolkit, ResGrid and its extensions, were developed and applied to designed reservoir simulation studies and continuous reservoir model updating. The toolkit provides reservoir engineers with high performance computing capacity to complete their projects without requiring them to delve into Grid resource heterogeneity, security certification, or network protocols. <p align=left>Continuous and real-time reservoir model updating is an important component of closed-loop model-based reservoir management. The method must rapidly and continuously update reservoir models by assimilating production data, so that the performance predictions and the associated uncertainty are up-to-date for optimization. The ensemble Kalman filter (EnKF), a Bayesian approach for model updating, uses Monte Carlo statistics for fusing observation data with forecasts from simulations to estimate a range of plausible models. The ensemble of updated models can be used for uncertainty forecasting or optimization. <p align=left>Grid environments aggregate geographically distributed, heterogeneous resources. Their virtual architecture can handle many large parallel simulation runs, and is thus well suited to solving model-based reservoir management problems. In the study, the ResGrid workflow for Grid-based designed reservoir simulation and an adapted workflow provide tools for building prior model ensembles, task farming and execution, extracting simulator output results, implementing the EnKF, and using a web portal for invoking those scripts. <p align=left>The ResGrid workflow is demonstrated for a geostatistical study of 3-D displacements in heterogeneous reservoirs. A suite of 1920 simulations assesses the effects of geostatistical methods and model parameters. Multiple runs are simultaneously executed using parallel Grid computing. Flow response analyses indicate that efficient, widely-used sequential geostatistical simulation methods may overestimate flow response variability when compared to more rigorous but computationally costly direct methods. <p align=left>Although the EnKF has attracted great interest in reservoir engineering, some aspects of the EnKF remain poorly understood, and are explored in the dissertation. First, guidelines are offered to select data assimilation intervals. Second, an adaptive covariance inflation method is shown to be effective to stabilize the EnKF. Third, we show that simple truncation can correct negative effects of nonlinearity and non-Gaussianity as effectively as more complex and expensive reparameterization methods.

<p>Reservoir architecture may be inferred from analogs and geologic concepts, seismic surveys, and well data. Stochastically inverted seismic data are uninformative about meter-scale features, but aid downscaling by constraining coarse-scale interval properties such as total thickness and average porosity. Well data reveal detailed facies and vertical trends (and may indicate lateral trends), but cannot specify intrawell stratal geometry. Consistent geomodels can be generated for flow simulation by systematically considering the precision and density of different data. Because seismic inversion, conceptual stacking, and lateral variability of the facies are uncertain, stochastic ensembles of geomodels are needed to capture variability. <p>In this research, geomodels integrate stochastic seismic inversions. At each trace, constraints represent means and variances for the inexact constraint algorithms, or can be posed as exact constraints. These models also include stratigraphy (a stacking framework from prior geomodels), well data (core and wireline logs to constrain meter-scale structure at the wells), and geostatistics (for correlated variability). These elements are combined in a Bayesian framework. <p>This geomodeling process creates prior models with plausible bedding geometries and facies successions. These prior models of stacking are updated, using well and seismic data to generate the posterior model. Markov Chain Monte Carlo methods sample the posteriors. Plausible subseismic features are introduced into flow models, whilst avoiding overtuning to seismic data or conceptual geologic models. Fully integrated cornerpoint flow models are created, and methods for screening and simulation studies are discussed. The updating constraints on total thickness and average porosity need not be from a seismic survey: any spatially dense estimates of these properties may be used.

The purpose of this project is to develop a pore pressure and fracture gradient prediction strategy for the Ewing Banks 910 (EW 910) area. Petrophysical and measured pressure data for eight wells previously drilled in the EW 910 area will be examined and reviewed. This strategy will help design future drilling and completion operations in the aforementioned area. Two pore pressure prediction strategies and one fracture gradient prediction strategy will be reviewed and applied to the available data. The first pore pressure prediction strategy reviewed was developed by W. R. Matthews. This strategy utilizes a geologic age specific overlay which indicates the normally pressured compaction trendline for the appropriate geologic age. After plotting the observed resistivity/conductivity data on the geologic age specific overlay, formation pore pressures can be predicted. A simple calibration of the data is required to implement this method. The second pore pressure prediction strategy reviewed was developed by Ben Eaton. Eaton developed a simple relationship that predicts the formation pore pressure knowing the normally pressured compaction trendline, the observed resistivity/conductivity data and a relationship for formation overburden stress. The fracture pressure prediction strategy reviewed was also developed by Ben Eaton. The data required for this prediction strategy is formation overburden stress, pore pressure and formation Poissons ratio. A relationship for the overburden stress and Poissons ratio can be developed or one of Eatons published relationships can be used. Ultimately, the Eaton fracture gradient prediction strategy results in a simple and accurate relationship provided an accurate estimate of pore pressure is available. The two formation pore pressure prediction strategies were applied to the petrophysical data. The resulting formation pore pressure prediction was compared to the measured pressure data obtained from the eight offset wells. After analyzing each pore pressure model against the available pressure data, the Eaton pore pressure prediction strategy was chosen as the best model to implement in future operations. The fracture gradient prediction strategy was implemented using the formation pore pressures estimated by the Eaton pore pressure prediction strategy. The fracture gradients predicted were within range of the fracture gradients suggested by the offset data.

<p>This study derives from observations made in petroleum research and practices of chemical industry that efficient mixing takes place in segregated immiscible fluid flow in granular packs and static mixers. A hypothesis was formulated that transverse mixing (TM) across oil-water interface may occur in segregated inflow to wells resulting in progressive transition zone, more water production, and reduced oil productivity. Mixing is broadly interpreted here to address the entire range of stirring, splitting, dispersion and diffusion processes between two fluids.</p> <p>Initial study showed that a commercial reservoir simulator would not model any transition zone in segregated oil-water flow at high pressure gradient as it lacks a mathematical description of the phenomenon. Initial analysis identified two major effects contributing to transverse mixing: shear mixing due to velocity contrast and momentum transfer due to tortuosity and streams collisions.</p> <p>The shear mixing effect was studied in the Hele-Shaw (H-S) flow cell, and TM of oil and water above unstable interface were observed. However, considering wavelength reduction caused by H-S model gap size corresponding to rocks pore size, the mixing zone appeared to be negligible.</p> <p>The momentum transfer (collision) effect has been studied by considering ratio of size of pore and throat. TM criterion was developed using modified Richardson number.</p> <p>Only early TM has been confirmed with granular-pack flow cell experiments due to dimensional restrictions. The results showed only water invading oil layer above the initial water/oil interface. Also, TM increased for higher pressure gradients and larger grain sizes, and reduced for more viscous oil.</p> <p>A mathematical model of early TM has been derived by solving a diffusion equation with constant flow velocity and water saturation at the initial W/O interface. The model reasonably matches experimental results thus enabling determination of the transverse dispersion coefficient, similar to miscible dispersion.</p> <p>The TM effect in wells was qualified by converting the linear TM model to radial flow model and integrating within the wells inflow zone. The results showed TM would increase water production by 2.5%, and reduce oil rate by 8.3% thus reducing wells productivity.</p> <p>Limitations and shortcoming of the study are discussed together with recommendations.</p>

The use of foam for mobility control is a promising means to improve sweep efficiency in subsurface applications such as improved/enhanced oil recovery and aquifer remediation. Foam can be introduced into geological formations by injecting gas and surfactant solutions simultaneously or alternatively. Alternating gas and surfactant solutions, which is often referred to as surfactant-alternating-gas (SAG) process, is known to effectively create fine-textured strong foams due to fluctuation in capillary pressure. Recent studies show that foam rheology in porous media can be characterized by foam-catastrophe theory which exhibits three foam states (weak-foam, strong-foam, and intermediate states) and two strong-foam regimes (high-quality and low-quality regimes). Using both mechanistic foam simulation technique and fractional flow analysis which are consistent with foam catastrophe theory, this study aims to understand the fundamentals of dynamic foam displacement during gas injection in SAG processes. The results revealed some important findings: (1) The complicated mechanistic foam fractional flow curves (fw vs. Sw) with both positive and negative slopes require a novel approach to solve the problem analytically rather than the typical method of constructing a tangent line from the initial condition; (2) None of the conventional mechanistic foam simulation and fractional flow analysis can fully capture sharply-changing dynamic foam behavior at the leading edge of gas bank, which can be overcome by the pressure-modification algorithm suggested in this study; (3) Four foam model parameters (¤Po, n, Cg/Cc, and Cf) can be determined systematically by using an S-shaped foam catastrophe curve, a two flow regime map, and a coreflood experiment showing the onset of foam generation; and (4) At given input data set of foam simulation parameters, the inlet effect (i.e., a delay in strong-foam propagation near the core face) is scaled by the system length, and therefore the change in system length at fixed inlet-effect length requires the change in individual values Cg and Cc at the same Cg / Cc. This study improves our understanding of foam field applications, especially for gas injection during SAG processes by capturing realistic pressure responses. This study also suggests new fractional flow solutions which do not follow conventional fractional flow analysis.

The understanding of geological characteristics and heterogeneity of a reservoir enables better decisions for reservoir development. Statistical methods use universally available injection and production rate data to help evaluate reservoir characteristics and behavior.In this research project, statistical methods typically used to infer communication between injector-producer well pairs in a waterflood reservoir using only production and injection rate data are applied to a CO2 flood. The multivariate linear regression (MLR) technique computes weighting coefficients possibly related to the fraction of the flow in a producer that comes from each of the injectors (Albertoni and Lake, 2002). MLR was applied to the Phase 2 portion of the Little Creek field, Mississippi CO2 flood. Albertoni and Lake use diffusivity filters to model the time lag and attenuation between the stimulus (injection) and the response (production), and further modify the model by successive elimination of negative weighting coefficients (SEN) and successive elimination of positive coefficients larger than 1 (SEP). Diffusivity filters do not improve the results for the Little Creek Field. The statistical implications of the SEN and SEP procedures were compared with a less complex simple linear model (SLM) which eliminates the need to make ad hoc assumptions. A statistical hypothesis test (P-Value test) was carried out to determine the significance of each injector-producer well pair relationship. Well pairs with non-significant relationships are then eliminated from the model. This avoids making statistically questionable assumptions to eliminate injector-producer well pairs with connection strengths (i.e., connections not in the range [0,1]). Recommendations to improve sweep were made using results from the Simple Linear Model with the application of the statistical significance test. Suggestions for future work are also presented.

The advent of the digital oil eld and rapidly decreasing cost of computing creates opportunities as well as challenges in simulation based reservoir studies, in particular, real-time reservoir characterization and optimization. One challenge our eorts are directed toward is the use of real-time production data to perform live reservoir characterization using high throughput, high performance computing environments. To that end we developed the required tools of parallel reservoir simulator, parallel ensemble Kalman lter and a scalable work ow manager. When using this collection of tools, a reservoir modeler is able to perform large scale reservoir management studies in short periods of time. This includes studies with thousands of models that are individually complex and large, involving millions of degrees of freedom. Using parallel processing, we are able to solve these models much faster than we otherwise would on a single, serial machine. This motivated the development of a fast parallel reservoir simulator. Furthermore, distributing those simulations across resources leads to a smaller total time to completion by making use of distributed processing. This allows the development of a scalable high throughput work ow manager. Finally, with thousands of models, each with millions of degrees of freedom, we end up with a super uity of model parameters. This translates directly to billions of degrees of freedom in the reservoir study. To be able to use the ensemble Kalman lter on these models, we needed to develop a parallel implementation of the ensemble Kalman lter. This thesis discusses the enabling tools and technologies developed to address a speci c problem: how to accurately characterize reservoirs, using large numbers of complex detailed models. For these characterization studies to be helpful in making production decisions, the time to solution must be feasible. To that end, our work is focused on developing and extending these tools, and optimizing their performance.

Managed Pressure Drilling (MPD) is an emerging drilling technology that utilizes mud weight, surface backpressure and annular frictional pressure loss (AFP) to precisely control the wellbore pressure. The goal of this project is to identify the most appropriate initial response and kick circulation method for the kicks that result from complications specific to MPD. These complications that can cause a reduction in bottomhole pressure were classified as surface equipment failures and unintended equivalent circulating density (ECD) reductions. Rotating control device (RCD) and pump failures are the examples of surface equipment failures. Pump efficiency loss and BHA position change represent the unintended ECD reductions. Shut-in (SI), MPD pump shut down, increasing surface backpressure, increasing pump rate, starting a new pump with surface backpressure and increasing pump rate with surface backpressure responses were simulated on a transient drilling simulator for kicks taken due to the pump efficiency loss, and the simulation results were evaluated. Shut-in and starting a new pump with a surface backpressure were simulated for a pump failure, which led to a loss of total AFP, and the simulation results were evaluated. A shut-in response was simulated for surface pressure loss (RCD failure), and its results were evaluated. Shut-in, MPD pump shut down, increasing surface backpressure pressure, increasing pump rate and increasing pump rate with surface backpressure responses were simulated, and the simulation results were evaluated for the kick taken due to BHA position change. Kick circulation was also simulated after the influx was stopped by the initial responses. The kicks were circulated using drillers method at normal, half, and increased circulating rates depending on the initial response. The results of circulating simulations were also evaluated. SI was concluded to be applicable for all kicks caused by bottomhole pressure fluctuations. However, increasing casing pressure is the most effective response if it is practical given the surface equipment and its condition. Normal rate circulation following these responses is generally better than using an increased or slow pump rate for these kinds of kicks. It reduces the surface backpressure and non productive time (NPT) required versus slower pump rates.

The Thesis is an analytical and numerical analysis of a new method for completing and producing oil wells affected by water coning. The method enables producing oil with no or minimal water cut while keeping the water subsurface with downhole water loop (DWL) installation. Typically, a DWL well is triple-completed in the oil and water zones with the three completions separated by parkers. The top completion produces oil to the surface while the middle and bottom completions drain from and inject into the bottom water zone, respectively. Segregated-inflow operation of DWL well requires keeping the production and drainage-injection rates below their critical values. Therefore, the theory of water coning is re-visited and examined using analytical modeling of critical height and dynamic stability of water cone. The analytical model employs transformation from anisotropic to equivalent isotropic radial flow system. Also, considered are the effects of partial penetration and capillary-pressure transition zone. The analytical model is used to determine operational domain of DWL for different well-reservoir systems. The results are then compared with data from commercial simulator and real field showing good match. Also investigated is the effect of the distance between water drainage and injection completions (D/I spacing), which is the most important design parameter for DWL wells. The results show that DWL wells could successfully work in reservoirs with relatively small aquifer as the DWL operational domain is only sensitive to small values of D/I spacing. A commercial simulator is employed to build a numerical model of DWL operations outside the segregated-inflow domain where the top completion produces oil with water. The steady demonstrates the flexibility of DWL in controlling water cut. Then, the model is used to study DWL performance with controlled water production using a modified nodal analysis approach that includes the D/I spacing constraint. The results show that DWL could improve critical oil rate and reduce water cut before and after water breakthrough, respectively. Nodal analysis is used to seek the possible production operations of DWL which would help to design the D/I spacing and decide if one or two downhole pumps were needed for the system.

Managed pressure drilling (MPD) is an adaptation of conventional drilling that has been developed to manage and control subsurface pressures in the well in order to minimize specific drilling problems. The constant bottom hole pressure approach (CBHP) is a versatile method of MPD, where a closed annulus allows initial responses to kicks other than simply shutting in the well. The objective of this research was to identify and evaluate the best initial response to gas kicks taken during drilling as a basis for developing reliable well control procedures for CBHP operations. Nine non-circulating and circulating responses (NCRs and CRs) were defined, and their application to kicks in two different wellbore geometries was studied through the use of computer simulations. Two different kick sizes, two different formation permeabilities, and three different kick intensities were considered. NCRs included a rapid shut in (SI) and four different MPD pump shut down schedules ending in SI. CRs included stepwise and rapidly increasing the casing pressure until the mud flow out equaled mud flow in, increasing casing pressure to a pre-defined limit and increasing the ECD by increasing mud pump rates. The initial responses were compared, based on the ability to stop an influx, determine whether the influx was stopped assuming intact wellbore, minimize risk of lost returns, minimize additional kick influx, and minimize excessive pressure at the surface and casing shoe. The results of over 150 simulations revealed that no single best initial response to all kicks could be identified. Three initial responses showing broad applicability include a rapid increase of casing pressure until flow rates are equal, shutting the well in and an adaptation of the MPD pump shut down schedule that allowed confirmation of a low rate kick. Increasing mud pump rate also showed advantages, but has limited application. Potential advantages and limitations of each were also explained. A method to confirm that the influx stopped during the application of CRs was also proposed. The best initial response was dependent on well conditions and the equipment used. Therefore, a simple decision tree was developed to plan an appropriate response.