Reservoir Simulation Model Research Articles

Sensitivity and Uncertainty Analysis for Natural Gas Hydrate Production Tests in Alaska

The Japan Research and Development Consortium for Pore Filling Hydrate in Sand (MH21-S), the U.S. Department of Energy National Energy Technology Laboratory (DOE NETL), and the U.S. Geological Survey (USGS) are planning a long-term production test from gas hydrate (methane hydrate) bearing reservoirs in the Prudhoe Bay Unit in northern Alaska. Prediction of gas and water production using 2D reservoir simulation models can be very useful for decision-making for long-term production testing. However, the input parameters applied to the simulation model, which are essentially based on the most likely values obtained from the well-log interpretation and core sample measurements of the Hydrate-01 stratigraphic test well (STW), have some degrees of uncertainty. In this study, to evaluate the impact of uncertainties in the input model on the prediction results of gas and water production as well as to analyze how each input parameter contributes to the uncertainties of prediction results, we conducted uncertainty analysis and parameter sensitivity analysis based on two different methods (the regression-based method and the Sobol’ method). The sensitivity analyses reveal that the parameters regarding relative permeability (exponential of the gas relative permeability (Ng) and residual gas saturation (Sgcr)) and the intrinsic/effective permeability of the lower B1 sand significantly contribute to the uncertainties of the gas and water production predictions. In addition, the uncertainty analyses reveal that the P50 value of cumulative gas production can result in less than the reference case with the most likely values for the input parameters due to the uncertainty of Ng. These results indicate that reducing the uncertainties of gas relative permeability and residual gas saturation is essential for more reliable numerical simulations both for prediction of test well performance and subsequent history matching of field test results.

Forecasting Technique Bridges Gap Between Material Balance and Reservoir Simulation

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 203941, “Bridging the Gap Between Material Balance and Reservoir Simulation for History Matching and Probabilistic Forecasting Using Machine Learning,” by Nigel H. Goodwin, Essence Analytics. The paper has not been peer reviewed. Currently, building and maintaining a reservoir simulation model is the bottleneck for reservoir decision support. In this paper, the authors describe an approach for creating a digital twin that is informed by the data itself based on established physics modeling and uses a fully probabilistic Bayesian approach. Statistical Time-Series Modeling and Physics Models Differences Between Statistical Models and Physics Models. The statistical model includes an error term for each timestep. In practice, this means that the predicted fit to historic data is very good and the error terms are small, but forecasting can have a major funneling effect. Many methods exist for forecasting of statistical models, but the increasing uncertainty of prediction over time, and the large magnitude of uncertainty, is a common feature. The error term implies a future random walk behavior. When forecasting, multiple single-step integrations are being performed in which each step has its own error. Where known values of exogenous variables exist in the future, prediction uncertainty can be greatly ameliorated. In contrast, the differential algebraic equations (DAE) for a physical model have no error term for each timestep. Each time the model is run with the same parameter values, the same results are obtained. The error in reservoir simulation lies in measurement error and model error. Errors exist in individual historical measurements (or maybe systemic errors), and the model and its parameters are known to be wrong. For forecasting, a range of models and model parameter values are used—ideally, based around a robust Monte Carlo approach. An ensemble of forecasts can then be generated that represent uncertainty.

Evaluation of Hydropower Generation and Reservoir Operation under Climate Change from Kesem Reservoir, Ethiopia

Climate changes significantly cause the precipitation deficiency and in turn reduce the inflow amount in reservoir affecting hydroelectric power generation. The primary objective of this study was to evaluate hydropower generation and reservoir operation under climate change from Kesem reservoir. Recent Representative Pathway (RCP) scenarios were used to evaluate the impact of climate change on power generation. Power transformation equation and variance scaling approach were amalgamated to adjust the bias correction of precipitation and temperature, respectively. Bias, root mean square error, and coefficient of variation were used to check the accuracy of projected rainfall. The base and future precipitation, temperature, and evaporation trend was analysed using the Mann–Kendall test. The flow calibration and validation were carried out by the Hydrologic Engineering Center-Hydrologic Modelling System (HEC-HMS), and hydropower generation was evaluated with reservoir simulation model (MODSIM 8.1) under climate scenarios. The performance of the model was found good with Nash–Sutcliffe coefficient (NSE) of 0.72 and coefficient of determination (R2) of 0.73 for calibration and NSE of 0.74 and R2 of 0.75 for validation. Projected future climate scenarios predicted increasing and decreasing trend of temperature and precipitation, respectively. For RCP4.5 climate scenario, the average energy generation is likely to decrease by 0.64% and 0.82% in both short-term (2021–2050) and long-term (2051–2080), respectively. In case of RCP8.5 climate scenario, the average energy generation will be decreased by 1.06% and 1.35% for short-term and long-term, respectively. Remarkable reduction of energy generation was revealed in RCP8.5 with relation to RCP4.5 scenario. This indicates that there will be high energy fluctuation and decreasing trend in the future energy generation. The research finding is crucial for decision-makers, power authorities, governmental and nongovernmental organizations, and watershed management agencies to take care for sustainability in the future hydropower generation in the Kesem reservoir.

Water Weakening of Artificially Fractured Chalk, Fracture Modification and Mineral Precipitation during Water Injection—An Experimental Study

This experiment was designed to study the water-weakening effect of artificially fractured chalk caused by the injection of different compositions of brines under reservoir conditions replicating giant hydrocarbon reservoirs at the Norwegian Continental Shelf (NCS). NaCl, synthetic seawater (SSW), and MgCl2, with same ionic strength, were used to flood triaxial cell tests for approximately two months. The chalk cores used in this experiment originate from the Mons basin, close to Obourg, Belgium (Saint Vast Formation, Upper Cretaceous). Three artificially fractured chalk cores had a drilled central hole parallel to the flooding direction to imitate fractured chalk with an aperture of 2.25 (±0.05) mm. Two additional unfractured cores from the same sample set were tested for comparison. The unfractured samples exposed a more rapid onset of the water-weakening effect than the artificially fractured samples, when surface active ions such as Ca2+, Mg2+ and SO42− were introduced. This instant increase was more prominent for SSW-flooded samples compared to MgCl2-flooded samples. The unfractured samples experienced axial strains of 1.12% and 1.49% caused by MgCl2 and SSW, respectively. The artificially fractured cores injected by MgCl2 and SSW exhibited a strain of 1.35% and 1.50%, while NaCl showed the least compaction, at 0.27%, as expected. Extrapolation of the creep curves suggested, however, that artificially fractured cores may show a weaker mechanical resilience than unfractured cores over time. The fracture aperture diameters were reduced by 84%, 76%, and 44% for the SSW, MgCl2, and NaCl tests, respectively. Permeable fractures are important for an effective oil production; however, constant modification through compaction, dissolution, and precipitation will complicate reservoir simulation models. An increased understanding of these processes can contribute to the smarter planning of fluid injection, which is a key factor for successful improved oil recovery. This is an approach to deciphering dynamic fracture behaviours.

Mangala Polymer Flood Performance: Connecting the Dots Through In-Situ Polymer Sampling

Summary The Mangala field contains medium-gravity viscous crude oil. Notably, it is the largest polymer flood in India and 34% of the stock tank oil initially in place (STOIIP) has been produced in 11 years. Mangala was put on full field polymer flood in 2015, 6 years after the start of field production on waterflood in 2009. Polymer flood added nearly 93 million barrels above the anticipated waterflood recovery in 6 years. Reservoir simulation models could replicate the initial Mangala polymer flood performance. However, the performance of the lower layers of Mangala (FM-3 and FM-4) continued to progressively deviate from modeling estimates. Importantly, the observed polymer breakthrough deviated significantly from predictions. As the polymer flood matured, the trend of field water cut with time indicated that in-situ polymer viscosity was equivalent to only 50 to 60% of the surface polymer viscosity. For better predictions and corrective actions, it was necessary to understand the nature of degradation, the progressively deteriorating field performance, especially of the lower layers, and the deviation of polymer breakthrough trends from predictions. Carefully designed in-situ polymer sampling, laboratory studies, and reservoir modeling studies helped connect the dots to understand the field performance. There are several excellent publications on accelerated aging studies and some on polymer sampling. This paper offers an opportunity to directly compare experimental results with field data. The procedures used and lessons learned during field sampling can be useful for other operators for management of polymer floods.

Numerical effects on the simulation of surfactant flooding for enhanced oil recovery

Numerical simulation of surfactant flooding using conventional reservoir simulation models can lead to unreliable forecasts and bad decisions due to the appearance of numerical effects. The simulations give approximate solutions to systems of nonlinear partial differential equations describing the physical behavior of surfactant flooding by combining multiphase flow in porous media with surfactant transport. The approximations are made by discretization of time and space which can lead to spurious pulses or deviations in the model outcome. In this work, the black oil model was simulated using the decoupled implicit method for various conditions of reservoir scale models to investigate behavior in comparison with the analytical solution obtained from fractional flow theory. We investigated changes to cell size and time step as well as the properties of the surfactant and how it affects miscibility and flow. The main aim of this study was to understand pulse like behavior in the water bank, which we report for the first time, Our aim was to identify their cause and associated conditions. The pulses are induced by a sharp change in relative permeability as the interfacial tension changes. Pulses are diminished when adsorption is modeled, and ranged from 0.0002 kg/kg to 0.0005 kg/kg. The pulses are absent for high-resolution model of 5000 cells in x direction with a typical cell size as used in well-scale models. The growth or dampening of these pulses may vary from case to case but in this instance was a result of the combined impact of relative mobility, numerical dispersion, interfacial tension and miscibility. Oil recovery under the numerical problems reduced the performance of the flood, due to large amounts of pulses produced. Thus, it is important to improve existing models and use appropriate guidelines to stop oscillations and remove errors.

A comparative machine learning study for time series oil production forecasting: ARIMA, LSTM, and Prophet

It is challenging to predict the production performance of unconventional reservoirs because of the sediment heterogeneity, intricate flow channels, and complex fluid phase behavior. The traditional oil production prediction methods (e.g., decline curve analysis and reservoir simulation modeling forecasting) are subjective. This paper presents a machine learning-based time series forecasting method, which considers the existing data as time series and extracts the salient characteristics of historical data to predict values of a future time sequence. We used time series forecasting because of the historical fluctuations in production well and reservoir operations. Three algorithms were studied and compared to address the limitations of traditional production forecasting: Auto-Regressive Integrated Moving Averages (ARIMA), Long-Short-Term Memory (LSTM) network, and Prophet. This study starts with the representative oil production data from a well located in an unconventional reservoir in the Denver-Julesburg (DJ) Basin. 70% of the data was used for model training, whereas the remaining 30% of data was used to evaluate the performance of the above-mentioned methods. Then, the decline curve analysis and reservoir simulation modeling forecasting were applied for comparison. The advantages of the machine-learning models include a simple workflow, no prior assumption about the reservoir type, fast prediction, and reliable performance prediction for a typical fluctuating declining curve. More importantly, the ‘Prophet’ model captures production fluctuation caused by winter impact, which can attract the operator's attention and prevent potential failures. This has rarely been explored and discussed by previous studies. The application of ARIMA, LSTM, and Prophet methods to 65 wells in the DJ Basin show that ARIMA and LSTM perform better than Prophet—probably because not all oil production data include seasonal influences. Furthermore, the wells in the nearby pads can be studied using the same parameter values in ARIMA and LSTM for predicting oil prediction in a transferred learning framework. Specifically, we observed that ARIMA is robust in predicting the oil production rate of wells across the DJ Basin.

The combining different-scale studies in a reservoir simulation model of a deposit with a fractured-cavernous type of carbonate reservoir

When constructing reservoir simulation models of complex reservoirs, it is necessary to take into account the zonal heterogeneity of the filtration and reservoir properties and the influence of secondary processes on the filtration characteristics of the deposit. The underestimation of complex mining and geological conditions negatively affects the forecast for technological and economic indicators of the field development. The paper presents an approach to creating a reservoir simulation model of a complex reservoir based on the integration of different-scale research that allow us to judge the alternation of different types of reservoirs along the lateral and vertical lines. As a result, a refined reservoir permeability cube is constructed which allows us to take into account the influence of the fracturing and cavernous reservoir on the technological indicators of development.

Application of Harris Hawks Optimization with Reservoir Simulation Model Considering Hedging Rule for Network Reservoir System

This research aims to apply the Harris hawks optimization (HHO) technique connected with a reservoir simulation model to search optimal rule curves of the network reservoir system in Thailand. The downstream water demand from the network reservoir that required shared water discharge, hydrological data, and physical data were considered in the reservoir simulation model. A comparison of the situation of water shortage using optimal rule curves from HHO technique, genetic algorithm (GA), and wind-driven optimization (WDO) is presented. The results showed that the new rule curves derived from the HHO technique with network reservoir searching were able to alleviate the water shortage and over-flow situations better than the current rule curves. The efficiency of using rule curves from HHO technique compared to GA and WDO techniques showed that the HHO technique can provide a better solution that reduced water scarcity and average over-flow compared with the current rule curves by up to 4.80%, 4.70%, and 4.50%, respectively. In addition, HHO was efficient in converging rule curve solutions faster than GA and WDO techniques by 15.00% and 54.00%, respectively. In conclusion, the HHO technique can be used to search for optimal network reservoir rule curves solutions effectively.

Fiber-Optic Sensors Evaluate Well Performance in Polymerflooding Pilot

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 199023, “Evaluation of Well Performance Using Data From Fiber-Optic Sensors in Wells of a Pilot Polymerflooding Test, Orinoco Basin, Venezuela,” by Andres Ramirez, Edgar Vasquez, and Luis Quevedo, PDVSA, et al. The paper has not been peer reviewed. The complete paper describes a pilot test using polymerflooding in which evaluating the performance of injector wells was a main priority. The authors chose to complete the wells using a distributed temperature sensor. The paper describes how a better understanding of the connection between the well and the reservoir can improve reservoir simulation models, the decision-making process, and well-performance evaluation. The authors also stress the importance of determining the possibility of unswept areas in the reservoir, which could result in low volumetric sweep efficiency and, consequently, low recovery factor. Introduction The field is in the Orinoco Belt, mainly unconsolidated Miocene sandstone, and has produced heavy oil since 1999. At the time of writing, more than 711 producing wells have been drilled since the beginning of the exploitation of the field. After reaching a production plateau of 200,000 B/D, enhanced-oil-recovery technologies have been considered. One of these is polymerflooding. A pilot test was conducted in an area of the field in which conditions did not indicate application of any thermal recovery technique. The reservoir pressure at project startup was 315 psia, with a reservoir temperature of 115.5°F. Average porosity was 0.32, reservoir absolute permeability was estimated to be between 25 and 15 darcies, and reservoir thickness was 25 ft. Oil mobility was very low because of high viscosity. The pilot test began in 2016. Polymer at a concentration of 900 ppm was injected in a deltaic sedimentary environment composed of distributary channel fills and mouth bars. Polymer was injected through three injector wells. The objective of the test is to provide rapid answers to questions about injection performance, polymer stability, and the effect of heterogeneities. The results will determine if the technology represents a feasible opportunity for future development of the field. The objectives of the control plan in this phase include the following: - Establish how much of the horizontal section contributes to the well injectivity - Estimate an injection profile distribution - Determine if important changes occur during injection

Knowledge-based machine learning techniques for accurate prediction of CO2 storage performance in underground saline aquifers

Carbon dioxide storage in underground saline aquifers is considered a promising technique for decreasing atmospheric CO2 emissions. The CO2 residual and solubility in deep saline aquifers are crucial processes for improving CO2 storage security. In this study, the trapping efficiency of CO2 sequestration in saline formations was predicted by developing three supervised machine learning (ML)-based models: random forest (RF), extreme gradient boosting (XGBoost), and support vector regression (SVR). A diverse field-scale simulation database of 1509 samples were collected from the literature, and the proposed models were examined for training and testing. To verify the prediction accuracy of the three ML models, the prediction results were analysed and compared using graphical and statistical indicators. From the prediction results, the proposed ML models were ranked based on their accuracy: XGBoost > RF > SVR. The XGBoost-based predictive model achieved an extremely low root mean square error (RMSE = 0.0041) and high correlation factor (R2 = 0.9993) for both residual and solubility trapping efficiency. However, RF and SVR exhibited RMSEs of 0.0243 and 0.074 and R2 values of 0.9781 and 0.9284, respectively. Furthermore, the applicability of the XGBoost model was validated and only 15 suspected data points were detected across the entire database. Therefore, the proposed model can be a valuable and viable template for predicting the CO2 trapping index in other saline formations worldwide. Utimately, the XGBoost model has been tested against reservoir simulation models in comprehensive blind testing and may be used as a robust screening and process planning tool for the uncertainty assessment of carbon storage projects.

Active Learning for Well Control Optimization with Surrogate Models

SummaryOptimization of well controls in waterflooding operations has been established as an effective strategy to increase oil recovery efficiency and to reduce operating costs. Optimization problems, however, entail hundreds of full-scale reservoir simulation runs, which can be computationally expensive and impractical for field-scale applications. Fit-for-purpose surrogate models present more efficient alternatives that can be deployed for faster implementation and to enable real-time applications. In recent years, machine learning (ML)-based surrogate models have gained traction in building efficient proxy models to facilitate reservoir engineering and management workflows. For high-dimensional control variables that are frequently encountered in life cycle optimization, gradient-based local search methods have proven to be the optimization algorithms of choice, especially when adjoint models are available to provide the required gradient information in a computationally efficient manner. The use of local search optimization techniques implies that the surrogate models should be locally accurate and adapted to the optimization path that is traversed during the iterations. The adaptation can be accomplished through active learning, where the training of the surrogate model is performed online (throughout the iterations) and tied to the optimization iterations. This paper presents an active learning framework that is tuned to the local search nature of the optimization problem by updating an initially trained surrogate model based on new samples that are generated during the optimization iterations. Specifically, at each iteration, the full reservoir simulation model is performed to evaluate the true objective function and provide the new simulated data to update the surrogate model, thereby enhancing its local prediction accuracy. As the optimization progresses, older samples that become less relevant to the solution at the current iteration are dropped to preserve the local nature of the surrogate model. Various aspects of designing an active learning-based proxy model for well control optimization are discussed and the framework is applied to waterflooding problems to demonstrate its efficiency and superior performance compared to the traditional offline training approach.

Factoring in Scale Effect of Core Permeability at Reservoir Simulation Modeling

Summary This paper presents approaches for qualitative and quantitative estimation of the scale effect for the plug (30 mm diameter) and full-diameter (100 mm diameter) core samples based on statistical methods and data analysis including student and Pearson’s tests, stepwise regression, correlations, and other statistical approaches. Different methods were used to analyze carbonate cores from Elets reservoirs D3(el1) and D3(el3) located at Lambeyshorskoye field. The analysis has identified the core data scale effect. In the course of the study, a matrix of correction scale factors was constructed for the plug samples to reduce the scale effect. The work results open up the potential for using plug core samples on a par with full-diameter ones for addressing the problems of permeability array refinement for geologic (cells size 100×100 m) and dynamic models (cells size 50×50 or 25×25 m) and upscaling.

Feasibility Investigation on the N2 Injection Process to Control Water Coning in Edge Water Heavy Oil Reservoirs

N2 injection process is a potential technique to control the water coning behavior in heavy oil reservoirs. In this paper, by using the methods of experiment and numerical simulation, the N2 injection process for controlling the edge water coning behavior is investigated. First, through a visual fluid flow experimental device, the flow behavior of N2-water in porous media is discussed. Also, the effects of temperature, pressure, and injection rate were studied. Then, based on the experimental results, aiming at an actual edge water heavy oil reservoir, a reservoir simulation model is developed. Thus, the water coning behavior of edge aquifer is systematically studied. Also, two novel indicators are proposed to evaluate the water coning behavior. Then, a series of numerical models are developed to investigate the performance of N2 injection process in edge water heavy oil reservoirs after water coning, and the adaptability and the optimal operation parameters are analyzed. Results indicate that under the effect of porous media, N2 can cut into a series of small gas bubbles. It is a typical dispersed phase and can effectively plug the water coning path. Compared with pressure and injection rate, temperature is a more sensitive factor to affect the plugging performance of N2. From the simulation results, it is observed that the permeability, water/oil ratio, and distance between well and aquifer can significantly affect the performance of water coning behavior. N2 injection process can effectively control the edge water coning and improve the CSS performance. Furthermore, from the simulation results, it is found that the optimal operation parameters for a N2 injection process is that the total N2 injection volume should be higher than 6,000 m3 within one operation cycle and the optimal N2 injection rate should be lower than 700 m3/day. This investigation further clarifies the mechanisms of N2 injection process to control the water coning behavior in heavy oil reservoirs. It can provide a useful reference for the EOR process of the heavy oil reservoirs with edge water.

Substituting petro-elastic model with a new proxy to assimilate time-lapse seismic data considering model errors

Dynamic data from oil and gas reservoirs (such as well-production or 4D seismic data) have been used often to reduce uncertainty in reservoir simulation models. These data are assimilated in the simulation models separately or jointly. Assimilation of 4D seismic data conventionally involves with a petro-elastic model (PEM) to transform outputs from the simulation models to elastic properties. The PEM is a set of different equations with uncertain parameters and its inclusion in assimilation algorithms calls on multidisciplinary teams of geoscientists and engineers. Moreover, PEM requires extraction of different outputs from the simulation models for the seismic forward model calculations. The extraction process can be costly for large-scale simulation models of giant reservoirs. This research presents a new petro-elastic proxy model (named DAI-Proxy) with a novel formulation to substitute the PEM and integrate 4D seismic data. DAI-Proxy relates time-lapse acoustic impedance to a summation of saturation and pressure changes with two coefficients which are functions of porosity. As the proxy is an approximation of the PEM, its application is affected by model error. We introduce two approaches to account for the proxy model error: (1) considering uncertain coefficients for the DAI-Proxy and (2) using fixed coefficients in the proxy while estimating model errors statistics from the prior ensemble of models. We incorporate these two approaches with a data assimilation algorithm to assimilate simultaneously 4D seismic and well-production data. A benchmark case is used with different cases of data assimilation to compare the DAI-Proxy and the PEM applications. Results show that data match quality for 4D seismic and well-production have similar responses for the PEM and DAI-Proxy implementations. In terms of production forecast, using fixed coefficients in the proxy with its model error treatment create a data assimilation framework comparable to the PEM case. Our results indicate that the traditional PEM application to integrate jointly 4D seismic and well-production data can be replaced with our new DAI-Proxy application. Given the degree of uncertainty in the PEM, related to the rock and fluid models, our proxy provides similar results with fewer uncertain inputs. The proxy offers further advantage as it needs less outputs from the simulation models for seismic forward model calculations. In addition, it helps petroleum engineers to use a computationally less expensive model (light model) as a substitute for the PEM to assimilate 4D seismic data.

Fast diagnosis of reservoir simulation models based on 4D seismic similarity indicators

In the last ten years, 4D seismic (4DS) data acquisition is evolving to permanent reservoir monitoring (PRM) systems where sensors are installed at the ocean-bottom, collecting seismic data according to the project's monitoring demand. Simultaneously, reservoir management workflows evolved to include uncertainties, where multiple reservoir models may be considered. Model-based approaches for developing and managing a field rely on error minimizations between modelled and measured data, traditionally from well production data, and eventually adapted to 4DS data. This study presents a fast, robust and unsupervised workflow to provide a comprehensive diagnosis of multiple reservoir simulation models using similarity indicators with observed 4DS. The methodology comprises a seismic forward modelling to convert hundreds of models from the reservoir engineering domain to the seismic domain. The diagnosis includes a novel region-by-region approach to compare the predicted synthetic seismic response with the observed 4DS anomalies using ternary maps generated from Gaussian mixture models (GMM), in addition to a magnitude metric. The methodology is tested in an ultra-deep turbidite field from the Campos basin in Brazil with a PRM system that captured various 4DS anomalies of different polarity, magnitude, shapes and sizes over several production and injection years. The contributions of this work are demonstrated in four applications: (1) feedback on various iterations of geomodelling, (2) feedback on well and seismic data assimilation, (3) quick evaluation of a new seismic monitor, and (4) ranking models for further decision-making studies. The workflow advantages are proved along the model-based reservoir management outline. For application (1), we successfully flagged which 4DS anomalies were being honored in the simulation models and quantified the impact of introducing features interpreted from seismic monitors in the geomodelling. For application (2), we quantified simulation model improvements provided by data assimilation. For application (3), we rapidly evaluated the quality of the seismic monitor against the existing simulation models as soon as the new seismic acquisition and processing was complete, validating the requirement to re-visit the simulation models. Finally, the workflow was crucial to select best models, out of hundreds, for the decision-making process, in application (4).

Study on recovery factor and interlayer interference mechanism of multilayer co-production in tight gas reservoir with high heterogeneity and multi-pressure systems

The coupling effects of complex pore structure and poor reservoir properties generally lead to relatively low production of single layer in tight gas reservoir. The multilayer or multilayer co-production increasingly gain popularity in exploitation of tight gas reservoirs to achieve high recovery factor and economic profit. The multilayer co-production of tight gas reservoir with high heterogeneity or multi-pressure systems is more complex than that of the single-layer production. It is still not very clear about the mechanism of co-production. To fill this gap, a set of numerical simulations are intentionally designed and implemented in this paper to systematically understand production characteristics, interlayer interference, and recovery factor of multilayer co-production in tight gas reservoirs. Based on the established numerical gas reservoir simulation model, sensitivity analysis of influencing parameters of interlayer interference and recovery factor, including interlayer permeability contrast, interlayer pressure difference, thickness of gas layer, gas saturation, and producing rate, has been carried out. Subsequently, a type of machine-learning method, i.e., random forest method (RFM), is use to quantitatively characterize the main influencing parameters on interlayer interference and recovery factor. In the end, the predicted models of recovery factor and interference time are proposed by using the linear correlation and nonlinear correlation. The numerical experiment results reveal that the gas of high-pressure layer will flow back to the low-pressure layer in the multilayer co-production scenario and the influencing parameters have a great effect on daily gas production, recovery factor, backflow time, formation pressure and contribution rate of each layer. The thickness of gas layer and producing rate have the strongest and weakest influences on the gas recovery, respectively. However, the influence of interlayer pressure difference on interlayer interference is highest and the gas saturation has a lower effect on interference time than other parameters. The recovery factor and interference time of multilayer co-production is about 54.09%–63.82% and 55–565 days respectively. The correlation results of recovery factor and interference time have a good agreement with the results of numerical simulation, and the uncertainties are controlled within 10%. The research results of numerical simulation, RFM and predicted model are applied to a gas well of the tight gas and the validity and practicability of the results are verified, which can provide a solid foundation for the exploitation of tight gas reservoirs with high heterogeneity and multi-pressure systems.

GeoDIN - Geoscience-Based Deep Interaction Networks for Predicting Flow Dynamics in Reservoir Simulation Models

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

Novel Self-Adaptive Shale Gas Production Proxy Model and Its Practical Application.

Recently, production optimization has gained increasing interest in the petroleum industry. The most computationally intensive and critical part of the production optimization process is the evaluation of the production function performed by the numerical reservoir simulator. Employing proxy models as a substitute for the reservoir simulator is proposed for alleviating this high computational cost. In this study, a new approach to construct adaptive proxy models for production optimization problems is proposed. An adaptive difference evolution algorithm (SaDE) optimized least-squares support vector machine (LSSVM) is used as an approximation function, while training is performed using a self-adaptive response surface experimental design (SaRSE). SaDE selects the optimal hyperparameters of LSSVM during the training process to improve the prediction accuracy of the proxy model. Cross-validation methods are used in the recursive training and network evaluation phases. The developed method is used to optimize the production of block gas reservoir models. Computational results confirm that the developed adaptive proxy model outperforms traditional regression methods. It is further verified that when the experimental data are updated, the alternative model still has high prediction accuracy when performing the objective function evaluation. The results show that the proposed proxy modeling approach enhances the entire optimization process by providing a fast approximation of the actual reservoir simulation model with better accuracy.

Impact of water saturation on diffusion coefficients determined by constant volume diffusion method

In recent decades, carbon dioxide (CO2) injection has become a promising technique for enhanced oil recovery (EOR). To design and model the CO2-EOR process in a petroleum reservoir, knowledge of the molecular diffusion of CO2 in oil is required. In this research, the constant volume diffusion (CVD) method is utilized to determine the diffusion coefficient of CO2 in a core saturated with synthetic oil. In addition, the effect of the water saturation level in the core on the resulted diffusion coefficients is examined. This method includes an oil-saturated core, which is in direct contact with a CO2 chamber. CVD tests were conducted at initial pressure and temperature of 410 psi and 64°C for water saturation of 0, 39, and 53%. CO2 is injected from the top of the core and oil is produced from the bottom. Once the CVD initial condition is established, the injector and producer valves are shut-in and the CVD test starts. In the CVD test, the pressure drop data are collected and used to tune the diffusion coefficients using a compositional reservoir simulation model equipped with a developed equation of state. History matching of the pressure decay data is conducted to determine the tuned diffusion coefficients. Experimental results show that the existence of water in the core causes a smaller pressure drop in the system. We find that diffusion coefficients decrease linearly with increasing water saturation in the core operating at the same initial pressure and the system temperature. The presence of water in the core reduces the rate of CO2 diffusion into the synthetic oil. Finally, we present an empirical relationship of the diffusion coefficient and the irreducible water saturation at the proposed lab conditions.