Results
Inversion Framework of Reservoir Parameters Based on Deep Autoregressive Surrogate and Continual Learning Strategy
Zhang, Kai (School of Petroleum Engineering, China University of Petroleum, Qingdao) | Fu, Wenhao (School of Civil Engineering, Qingdao University of Technology (Corresponding author)) | Zhang, Jinding (School of Petroleum Engineering, China University of Petroleum, Qingdao) | Zhou, Wensheng (School of Petroleum Engineering, China University of Petroleum, Qingdao) | Liu, Chen (State Key Laboratory of Offshore Oil Exploitation) | Liu, Piyang (CNOOC Research Institute Ltd) | Zhang, Liming (School of Civil Engineering, Qingdao University of Technology) | Yan, Xia (School of Petroleum Engineering, China University of Petroleum, Qingdao) | Yang, Yongfei (School of Petroleum Engineering, China University of Petroleum, Qingdao) | Sun, Hai (School of Petroleum Engineering, China University of Petroleum, Qingdao) | Yao, Jun (School of Petroleum Engineering, China University of Petroleum, Qingdao)
CNOOC Research Institute Ltd Summary History matching is a crucial process that enables the calibration of uncertain parameters of the numerical model to obtain an acceptable match between simulated and observed historical data. However, the implementation of the history-matching algorithm is usually based on iteration, which is a computationally expensive process due to the numerous runs of the simulation. To address this challenge, we propose a surrogate model for simulation based on an autoregressive model combined with a convolutional gated recurrent unit (ConvGRU). The proposed ConvGRU-based autoregressive neural network (ConvGRU-AR- Net) can accurately predict state maps (such as saturation maps) based on spatial and vector data (such as permeability and relative permeability, respectively) in an end-to- end fashion. Furthermore, history matching must be performed multiple times throughout the production cycle of the reservoir to fit the most recent production observations, making continual learning crucial. To enable the surrogate model to quickly learn recent data by transferring experience from previous tasks, an ensemble-based continual learning strategy is used. Together with the proposed neural network-based surrogate model, the randomized maximum likelihood (RML) is used to calibrate uncertain parameters. The proposed method is evaluated using 2D and 3D reservoir models. For both cases, the surrogate inversion framework successfully achieves a reasonable posterior distribution of reservoir parameters and provides a reliable assessment of the reservoir's behaviors. Introduction The goal of history matching is to calibrate the uncertain parameters of the reservoir using observed reservoir behaviors, so that the simulation results can reproduce the production history and subsequently provide a reliable prediction for reservoir management and optimization. This is typically an ill-posed inverse problem, which implies that more than one combination of parameters can lead to an acceptable match for reproducing past reservoir behaviors. Well production data are commonly used to represent the observed reservoir behavior. However, the information content in most sets of production data is fairly low because of the limited number of observation locations and the diffusive nature of the flow (Oliver and Chen 2011). This results in poorly constrained reservoir parameters between wells.
- Asia > China (1.00)
- North America > United States > Texas (0.28)
- Energy > Oil & Gas > Upstream (1.00)
- Government > Regional Government > Asia Government > China Government (0.34)
- Reservoir Description and Dynamics > Reservoir Simulation > History matching (1.00)
- Reservoir Description and Dynamics > Reservoir Fluid Dynamics > Flow in porous media (1.00)
- Reservoir Description and Dynamics > Reservoir Characterization (1.00)
- (4 more...)
- Information Technology > Modeling & Simulation (1.00)
- Information Technology > Artificial Intelligence > Representation & Reasoning (1.00)
- Information Technology > Artificial Intelligence > Machine Learning > Statistical Learning (1.00)
- (2 more...)
A Study on Well Placement and Performance Forecasting in Uinta Basin Considering Geological Uncertainty
Eltahan, Esmail (The University of Texas at Austin) | Fiallos-Torres, Mauricio (The University of Texas at Austin / SLB) | Ganjdanesh, Reza (The University of Texas at Austin) | Sepehrnoori, Kamy (The University of Texas at Austin)
Abstract Reservoir uncertainty can be represented by an ensemble of physics-based models, each producing a satisfactory match to production history. Deterministic assumptions are often made for a subset of the uncertain parameters to make the problem more feasible. A key issue is that, even if the models accurately reproduce past production, different assumptions can lead to vastly different predictions for future production. Using three different modeling scenarios, we quantified the uncertainty in the reservoir/completion properties and initial conditions for two horizontal wells in the Uinta Basin. The obtained uncertainty is propagated further to report probabilistic production forecast and assess the well-spacing impact on oil recovery both for individual wells and for the whole field. We built a 1-by-2-mile reservoir-simulation model for the two wells of interest. We hypothesized three scenarios: (1) non-planar fracture geometry obtained by a fracture-propagation model (FPM), (2) planar fractures with uniform geometry, and (3) stimulated rock volume (SRV) region surrounding uniform fractures. The remaining uncertain parameters include fracture properties, initial saturations, relative permeability, and matrix and SRV permeability if applicable. We obtained multiple realizations using an assisted-history-matching method, and then production forecast is recorded over a 20-year period. The impact of well spacing is studied considering two cases: singly bounded, by placing only a pair of wells in the section, and fully bounded, by placing as many wells as possible. We showed that the modeling approach has a profound impact on the recovery estimates. We see a 30% decrease in oil recovery as we implement SRV permeability. Such a decrease in recovery is caused by the less permeable far-acting zone. Since fracture size is important for spacing effects, we consider the cases with longest half-length. Oil recovery starts to degrade only when spacing is less than 600 ft for scenario 3, or 700 ft for the others. We explain that the high permeability SRV regions dampen the pressure depletion, and hence interference effects. The degradation is observed to be 1.3 to 1.9 more severe if a well is fully bounded as opposed to singly bounded. Implementing a staggered (also known as modified zipper) configuration results in considerably better performance, particularly for short-term recovery.
- North America > United States > Texas (1.00)
- North America > United States > Colorado (0.84)
- North America > United States > Utah (0.70)
- Geology > Geological Subdiscipline > Geomechanics (1.00)
- Geology > Petroleum Play Type > Unconventional Play > Shale Play (0.46)
- North America > United States > Wyoming > Uinta Basin (0.99)
- North America > United States > Utah > Uinta Basin > Altamont-Bluebell Field > Altamont Field (0.99)
- North America > United States > Texas > West Gulf Coast Tertiary Basin > Eagle Ford Shale Formation (0.99)
- (33 more...)
- Well Completion > Hydraulic Fracturing (1.00)
- Reservoir Description and Dynamics > Unconventional and Complex Reservoirs (1.00)
- Reservoir Description and Dynamics > Reservoir Simulation > History matching (1.00)
- (5 more...)
- Information Technology > Modeling & Simulation (1.00)
- Information Technology > Artificial Intelligence > Representation & Reasoning (1.00)
- Information Technology > Artificial Intelligence > Machine Learning > Statistical Learning (0.67)
_ This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 200807, “Closing the Loop on a History Match for a Permian EOR Field With Relative Permeability Data Uncertainty,” by Usman Aslam, Emerson, and Jorge Burgos, SPE, and Craig Williams, Occidental Petroleum, et al. The paper has not been peer reviewed. _ Reservoir production forecasts are inherently uncertain because of the lack of quality data available to build predictive reservoir models. Traditionally, a best estimate for relative permeability data is assumed during the history-matching process despite significant uncertainty. Performing sensitivities around the best-estimate relative permeability case will cover only part of the uncertainty space. In the complete paper, the authors present an application of a Bayesian framework for uncertainty assessment and efficient history matching of a Permian carbon-dioxide (CO2) enhanced oil recovery field for reliable production forecasting. Field Details Regional and Structural Geology. The Central Basin Platform (CBP) is a positive tectonic feature that separates the Delaware and Midland sub-basins of the Permian. The study field is located on the northeastern end of the CBP, with the Permian (Guadalupian) -aged reservoir composed of San Andres Formation dolostone. The total thickness of the unit is approximately 1,500 ft, with the main reservoir within the middle 600 ft. Structural Framework. The area of interest features 270 wells that have a total depth in the San Andres formation or deeper. Of these, 234 have digital open- or cased-hole logs that were used to correlate formation tops. After reviewing all well logs, it became clear that, historically, the zones were primarily identifiable from porosity picks, which are more subjective than markers identified with gamma ray methods. A new sequence stratigraphic framework was developed based on core descriptions and outcrop analogs. This correlation framework was then extrapolated to well logs. After a grid of north/south and west/east cross sections was correlated across the field and a series of loop ties was made for quality-control purposes, structure and isopach maps were created for each zone. Reservoir Description. At the time of discovery, natural gas was trapped at the structural high point of the study field. Above the gas/oil interface is the gas cap. Below the gas was an oil accumulation, which extended to the producing oil/water contact (POWC). The POWC was defined by early drilling as the maximum depth where water-free oil was produced. The base of the transition zone is the top of the residual oil zone (ROZ); this reservoir interval extends to the free-water level and is an interval believed to have been waterflooded naturally. The ROZ is the target for CO2 injection. Grid Construction and Property Modeling. A full-field detailed 3D reservoir characterization was modeled at 50-ft × 50-ft grid increments and an approximate 2-ft average layer thickness. The reservoir model consists of 114,222,528 cells. A total of 235 wells in the unit area was correlated and used to constrain the structural framework of the model. Quantitative porosity log suites were available in 233 of the wells in the model area, which were used to create the full-field porosity property in the 3D model. The logs were normalized to the core data and upscaled.
- North America > United States > Texas > Permian Basin > Yeso Formation (0.99)
- North America > United States > Texas > Permian Basin > Yates Formation (0.99)
- North America > United States > Texas > Permian Basin > Wolfcamp Formation (0.99)
- (25 more...)
- Reservoir Description and Dynamics > Reservoir Simulation > History matching (1.00)
- Reservoir Description and Dynamics > Reservoir Fluid Dynamics > Flow in porous media (1.00)
- Reservoir Description and Dynamics > Reservoir Characterization (1.00)
- Reservoir Description and Dynamics > Formation Evaluation & Management (1.00)
- Information Technology > Artificial Intelligence > Representation & Reasoning > Uncertainty > Bayesian Inference (1.00)
- Information Technology > Artificial Intelligence > Machine Learning > Learning Graphical Models > Directed Networks > Bayesian Learning (0.87)
Robust Optimization Technique Using Modified Net Present Value and Stochastic Simplex Approximate Gradient
Fortaleza, Eugenio L. F. (Universidade de Brasília (Corresponding author)) | Sanchez, William Humberto Cuellar (Universidade de Brasília) | Neto, Emanuel Pereira Barroso (Universidade de Brasília) | Miranda, Marco Emilio Rodrigues (Universidade de Brasília) | Munerato, Fernando Perin (REPSOL Sinopec Brasil)
Summary This article aims to combine, from previous works, a modified objective function and the stochastic simplex approximate gradient (StoSAG) to provide a robust technique that optimizes reservoir production on the basis of a sequence of short-term optimizations. Usually, in reservoir optimization, the main goal is to maximize the net present value (NPV); this work used a modified NPV (MNPV) function that introduces reservoir parameters into the objective function. This MNPV analyzes the relation between cash flow and the reduction of the produced oil fraction, which is an indicator of the reduction of the well production life. On the other hand, the StoSAG is a well-established algorithm for robust optimization, and it was used to perform a constructive optimization with the MNPV cost function. The proposed technique (MNPV), together with StoSAG, is compared with other techniques from the literature using a regular base of all reservoir life cycle and the same proposed short-term optimizations, but using the classical NPV. These comparisons were made based on two benchmarks, SPE9 and Egg reservoir models, with an ensemble of 5 and 100 realizations, respectively. As a result, the MNPV StoSAG presents strong cash flow at the beginning of the reservoir production, a competitive NPV along the entire life cycle, and fast simulation time.
- Financial News (0.55)
- Research Report (0.46)
- Reservoir Description and Dynamics > Reservoir Characterization (1.00)
- Reservoir Description and Dynamics > Improved and Enhanced Recovery > Waterflooding (1.00)
- Reservoir Description and Dynamics > Formation Evaluation & Management (1.00)
- (4 more...)
- Information Technology > Artificial Intelligence > Representation & Reasoning > Optimization (1.00)
- Information Technology > Artificial Intelligence > Machine Learning (1.00)
Summary Rock composition can be related to conventional well logs through theoretical equations and petrophysical endpoints. Multimineral analysis is a formation evaluation tool that uses inversions to quantify rock composition from well logs. However, because of data errors and the multivariate selection of petrophysical endpoints, solutions from the multimineral analysis are nonunique. Many plausible realizations exhibit comparable data misfits. Therefore, the uncertainties in rock composition and petrophysical endpoints must be quantified but cannot be fulfilled by deterministic solvers. Stochastic Bayesian methods have been applied to assess the uncertainties, but the high run time, tedious parameter tuning, and need for specific prior information hinder their practical use. We implement Markov chain Monte Carlo with ensemble samplers (MCMCES) to assess the uncertainties of rock composition or petrophysical endpoints in the Bayesian framework. The resultant posterior probability density functions (PDFs) quantify the uncertainties. Our method has fewer tuning parameters and is more efficient in convergence than the conventional random walk Markov chain Monte Carlo (MCMC) methods in high-dimensional problems. We present two independent applications of MCMCES in multimineral analysis. We first apply MCMCES to assess the uncertainties in volume fractions with a suite of well logs and petrophysical endpoints. However, defining the petrophysical endpoints can be challenging in complex geological settings because the values of standard endpoints may not be optimal. Next, we use MCMCES to estimate petrophysical endpoints’ posterior PDFs when the endpoints are uncertain. Our methods provide posterior volume-fraction or petrophysical-endpoint realizations for interpreters to evaluate multimineral solutions. We demonstrate our approach with synthetic and field examples. Reproducible results are supplemented with the paper.
- North America > United States > North Dakota (1.00)
- North America > United States > Texas (0.68)
- Geology > Rock Type > Sedimentary Rock (0.69)
- Geology > Geological Subdiscipline (0.67)
- Geology > Mineral > Silicate (0.50)
- North America > United States > South Dakota > Williston Basin > Bakken Shale Formation (0.99)
- North America > United States > North Dakota > Williston Basin > Bakken Shale Formation (0.99)
- North America > United States > Montana > Williston Basin > Bakken Shale Formation (0.99)
- North America > United States > North Dakota > Williston Basin > Three Forks Group Formation (0.98)
- Reservoir Description and Dynamics > Reservoir Simulation > History matching (1.00)
- Reservoir Description and Dynamics > Reservoir Simulation > Evaluation of uncertainties (1.00)
- Reservoir Description and Dynamics > Reservoir Characterization (1.00)
- Reservoir Description and Dynamics > Formation Evaluation & Management > Open hole/cased hole log analysis (1.00)
- Information Technology > Artificial Intelligence > Machine Learning > Statistical Learning (1.00)
- Information Technology > Artificial Intelligence > Representation & Reasoning > Uncertainty > Bayesian Inference (0.68)
- Information Technology > Artificial Intelligence > Machine Learning > Learning Graphical Models > Directed Networks > Bayesian Learning (0.68)
"In offshore and coastal engineering, metocean refers to the syllabic abbreviation of meteorology and (physical) oceanography" (Wikipedia). Metocean research covers dynamics of the oceaninterface environments: the air-sea surface, atmospheric boundary layer, upper ocean, the sea bed within the wavelength proximity (~100 m for wind-generated waves), and coastal areas. Metocean disciplines broadly comprise maritime engineering, marine meteorology, wave forecast, operational oceanography, oceanic climate, sediment transport, coastal morphology, and specialised technological disciplines for in-situ and remote sensing observations. Metocean applications incorporate offshore, coastal and Arctic engineering; navigation, shipping and naval architecture; marine search and rescue; environmental instrumentation, among others. Often, both for design and operational purposes the ISSC community is interested in Metocean Extremes which include extreme conditions (such as extreme tropical or extra-tropical cyclones), extreme events (such as rogue waves) and extreme environments (such as Marginal Ice Zone, MIZ). Certain Metocean conditions appear extreme, depending on applications (e.g.
- Europe > United Kingdom > England (1.00)
- Asia > Middle East > Saudi Arabia (1.00)
- Asia > Japan (1.00)
- (16 more...)
- Summary/Review (1.00)
- Research Report > New Finding (1.00)
- Research Report > Experimental Study (1.00)
- (3 more...)
- Geology > Geological Subdiscipline > Geomechanics (1.00)
- Geology > Sedimentary Geology > Depositional Environment (0.67)
- Geophysics > Electromagnetic Surveying (0.65)
- Geophysics > Seismic Surveying > Seismic Modeling (0.45)
- Transportation > Passenger (1.00)
- Transportation > Marine (1.00)
- Transportation > Infrastructure & Services (1.00)
- (36 more...)
- Europe > Denmark > North Sea > Danish Sector > Central Graben > Block 5504/12 > Tyra Field (0.99)
- Europe > Denmark > North Sea > Danish Sector > Central Graben > Block 5504/11 > Tyra Field (0.99)
- North America > United States > Colorado > Ice Field (0.98)
- (18 more...)
- Well Drilling > Well Planning > Trajectory design (1.00)
- Well Drilling > Drillstring Design > Drill pipe selection (1.00)
- Well Drilling > Drilling Operations (1.00)
- (53 more...)
Abstract Geochemical data in produced water contain important reservoir information but are seldomly exploited, especially for the nonconservative chemicals. Some conservative chemical data have been integrated in history matching workflow to obtain better knowledge of reservoirs. However, assuming reservoir chemicals being conservative is impractical because most chemicals are involved in interactions with other chemicals or reservoir rock, and mistakenly regarding nonconservative chemicals as being conservative can cause large error. Nevertheless, once the interactions can be accurately described, nonconservative chemical data can be used to obtain more reservoir information. In this work, a new physicochemical model is proposed to describe the transport of natural nonconservative chemicals (barium and sulfate) in porous media. Both physical reactions, such as ion adsorption and desorption, and chemical reactions, such as barite deposition, are integrated. Based on the new model, the ensemble smoother with multiple data assimilations (ES-MDA) method is employed to update reservoir model parameters by assimilating oil production rate, water production rate, and chemical data (barium and sulfate concentration). Data assimilation results show that integrating geochemical data in ES-MDA algorithm yields additional improvements in estimation of permeability. Besides, clay content distribution, which is critical in injection water breakthrough percentage calculation, can be accurately estimated with relative root mean square error (rRMSE) being as small as 0.1. However, mistakenly regarding nonconservative chemicals as conservative can cause large errors in reservoir parameters estimation. Accurately modeling the chemical interactions is crucial for integrating chemical data in history matching algorithm.
- Research Report > New Finding (0.48)
- Research Report > Experimental Study (0.34)
- North America > United States > Texas > Fort Worth Basin > Ranger Field (0.99)
- Europe > Norway > North Sea > Northern North Sea > North Viking Graben > PL 190 > Brent Group > Tarbert Formation (0.99)
- Europe > Norway > North Sea > Northern North Sea > North Viking Graben > PL 104 > Block 30/9 > Oseberg Field > Tarbert Formation (0.99)
- (5 more...)
- Reservoir Description and Dynamics > Reservoir Simulation > History matching (1.00)
- Reservoir Description and Dynamics > Reservoir Fluid Dynamics > Flow in porous media (1.00)
- Reservoir Description and Dynamics > Improved and Enhanced Recovery > Waterflooding (1.00)
- (5 more...)
- Information Technology > Artificial Intelligence > Machine Learning > Statistical Learning (1.00)
- Information Technology > Artificial Intelligence > Representation & Reasoning (0.93)
Conditioning Model Ensembles to Various Observed Data (Field and Regional Level) by Applying Machine-Learning-Augmented Workflows to a Mature Field with 70 Years of Production History
Vanegas, Gisela (OMV Exploration & Production) | Nejedlik, John (OMV Exploration & Production) | Neff, Pascale (OMV Exploration & Production) | Clemens, Torsten (OMV Exploration & Production)
Summary Forecasting production from hydrocarbon fields is challenging because of the large number of uncertain model parameters and the multitude of observed data that are measured. The large number of model parameters leads to uncertainty in the production forecast from hydrocarbon fields. Changing operating conditions [e.g., implementation of improved oil recovery or enhanced oil recovery (EOR)] results in model parameters becoming sensitive in the forecast that were not sensitive during the production history. Hence, simulation approaches need to be able to address uncertainty in model parameters as well as conditioning numerical models to a multitude of different observed data. Sampling from distributions of various geological and dynamic parameters allows for the generation of an ensemble of numerical models that could be falsified using principal-component analysis (PCA) for different observed data. If the numerical models are not falsified, machine-learning (ML) approaches can be used to generate a large set of parameter combinations that can be conditioned to the different observed data. The data conditioning is followed by a final step ensuring that parameter interactions are covered. The methodology was applied to a sandstone oil reservoir with more than 70 years of production history containing dozens of wells. The resulting ensemble of numerical models is conditioned to all observed data. Furthermore, the resulting posterior-model parameter distributions are only modified from the prior-model parameter distributions if the observed data are informative for the model parameters. Hence, changes in operating conditions can be forecast under uncertainty, which is essential if nonsensitive parameters in the history are sensitive in the forecast.
- Europe > Austria (0.46)
- North America > United States > Texas (0.29)
- Asia > Middle East > UAE (0.28)
- Geology > Geological Subdiscipline (0.88)
- Geology > Rock Type > Sedimentary Rock > Clastic Rock (0.66)
- Geology > Sedimentary Geology > Depositional Environment > Marine Environment (0.46)
- North America > Canada > Alberta > Flood Field > Adamant Masters Flood 6-6-85-24 Well (0.99)
- Europe > Austria > Vienna Basin > Matzen Field (0.99)
- Information Technology > Artificial Intelligence > Representation & Reasoning (1.00)
- Information Technology > Artificial Intelligence > Machine Learning > Statistical Learning (1.00)
Summary Reservoir management in offshore fields is a challenging task, particularly for mature fields because of a typical excessive production of water and/or gas. Because of several constraints on facilities capacity, an assisted reservoir management process can deliver solutions to optimally operate offshore fields, seeking to increase oil production with better assessment of water and gas production and injection. Optimal reservoir management (ORM) can be applied aiming at maximizing reservoir performance and to deliver well controls applicable to field operations. In this work, we implemented an assisted optimization procedure to maximize overall oil production for a field offshore Brazil in Campos Basin. We applied our ORM technique in an important field offshore Brazil, where cumulative oil production is maximized by optimally controlling water rates through injection wells. Injection rates can vary with time, honoring operational requirements of smoothness. Geomechanical limits on injection pressures are considered to avoid loss of rock integrity, and platform constraints on overall production and injection are imposed at all times. Our approach deals with reservoir uncertainties described within a large set of calibrated simulation models to decide on optimal injection rates, taking into account possible risks. The model-based ORM under uncertainty that we developed showed gains in total oil production over 20 years of operation up to 7.2% with respect to the base strategy currently applied. On average, results show an increase of near 4% in oil production, with concomitant reduction in total water production and in overall water injection. To guarantee that the gains forecast by our study are feasible, a pilot test in the actual field has been implemented to verify the consistency between modeling and reality (data observation). We have chosen an area in the field to proceed with the optimal injection control pilot, aiming to check the quality of the uncertain models in comparison to the observed data in practice. The pilot area has been selected on the basis of aspects related to geological description, connectivity expected in the reservoir, and operational constraints. The results of 8 months of the pilot show clear coherence between models and reality that is well within the uncertainty range accepted at the reservoir of interest. To the best of our knowledge, it is the first time that an offshore field is actually operated on the basis of a set of controls obtained through an assisted ORM procedure, although it was performed at a pilot scale. Results suggest robust benefits under reservoir-uncertainties consideration, and large-scale application will take place soon, but that is outside the scope of this work. The pilot provided more confidence in field applications, leading to a broader perspective for full-field implementations.
- South America > Brazil (1.00)
- Europe > United Kingdom > North Sea > Central North Sea (0.41)
- North America > United States > Wyoming > Sweetwater County (0.40)
- (3 more...)
- Research Report > New Finding (1.00)
- Research Report > Experimental Study (0.86)
- Energy > Oil & Gas > Upstream (1.00)
- Water & Waste Management > Water Management > Lifecycle > Disposal/Injection (0.38)
- South America > Brazil > Campos Basin (0.99)
- Europe > Norway > Norwegian Sea > Halten Terrace > PL 128 > Block 6608/10 > Norne Field > Tofte Formation (0.99)
- Europe > Norway > Norwegian Sea > Halten Terrace > PL 128 > Block 6608/10 > Norne Field > Not Formation (0.99)
- (5 more...)
- Reservoir Description and Dynamics > Reservoir Simulation > History matching (1.00)
- Reservoir Description and Dynamics > Reservoir Characterization (1.00)
- Reservoir Description and Dynamics > Improved and Enhanced Recovery > Waterflooding (1.00)
- Reservoir Description and Dynamics > Formation Evaluation & Management (1.00)
- Information Technology > Modeling & Simulation (1.00)
- Information Technology > Artificial Intelligence > Representation & Reasoning > Optimization (1.00)
- Information Technology > Artificial Intelligence > Machine Learning (0.93)
Calibrating Field-Scale Uncertainties to Local Data: Is the Learning Being Overgeneralized?
He, Jincong (Chevron Energy Technology Company) | Reynolds, Albert C. (University of Tulsa) | Tanaka, Shusei (Chevron Energy Technology Company) | Wen, Xian-Huan (Chevron Energy Technology Company) | Kamath, Jairam (Chevron Energy Technology Company)
Summary A common pitfall in probabilistic history matching is omitting the local variation of spatial uncertainties and falsely generalizing the learning from local data to the entire field. This can lead to radical overestimation of uncertainty reduction and bad reservoir‐management decisions. In this paper, we propose a methodology to quantify and correct for the error that arises from the omission of local variation in probabilistic history matching. Most performance metrics in an oil field, such as the original oil in place (OOIP) and the estimated ultimate recovery (EUR), are field‐scale objective functions that depend on properties (e.g., porosity) over the entire field. On the other hand, many measurement data from wells [e.g., bottomhole pressure (BHP)] are mainly sensitive to the reservoir properties near the locations where they are measured, and thus they are susceptible to local variations of reservoir properties. Calibrating field‐scale objective functions to local well data without properly characterizing the local variation can overestimate the uncertainty reduction of field‐scale objective functions. In this paper, we derived formulas to quantify errors in the posterior cumulative distribution functions (CDFs) of the objective functions resulting from the omission of local variation. We also provide a way to correct for the error and to recover the true posterior CDFs. Through theoretical derivation, we show that the modeling error that arises from the omission of local variation is dependent on the magnitude of the global and local variations of the uncertain properties (e.g., porosity). The larger the local variation relative to the global variation, the larger the error in the estimated posterior distributions. The error also depends on the variogram of the local variation and the detection range of the data. The error is larger for cases with a long variogram for the local variation and a short data‐detection range. In addition, the modeling errors for different measurement data points can be highly correlated even when the measurement errors for these data are independent. To correct for this modeling error, analytical and empirical formulas are proposed that have been shown to greatly improve the accuracy of the posterior distributions in a number of cases. To the best of our knowledge, this is the first time that the modeling error from the omission of local variation in the probabilistic history‐matching process has been quantified and corrected. The methodology proposed could help improve the reliability of the result from probabilistic history matching.
- Europe (1.00)
- Asia (1.00)
- North America > United States > Texas (0.68)
- Asia > Kazakhstan > Mangystau Oblast > Precaspian Basin > Tengiz Field > Tengiz Formation (0.99)
- Asia > Kazakhstan > Mangystau Oblast > Precaspian Basin > Tengiz Field > Korolev Formation (0.99)
- Reservoir Description and Dynamics > Reservoir Simulation > History matching (1.00)
- Reservoir Description and Dynamics > Reservoir Characterization > Geologic modeling (1.00)
- Reservoir Description and Dynamics > Reserves Evaluation > Estimates of resource in place (1.00)
- (2 more...)
- Information Technology > Modeling & Simulation (0.93)
- Information Technology > Artificial Intelligence > Representation & Reasoning (0.93)
- Information Technology > Artificial Intelligence > Machine Learning (0.67)