Reservoir Management Workflow Research Articles

A Hybrid Data-Physics Framework for Reservoir Performance Prediction with Application to H2S Production

Summary Model-based reservoir management workflows rely on the ability to generate predictions for large numbers of model and decision scenarios. When suitable simulators or models are not available or cannot be evaluated in a sufficiently short time frame, surrogate modeling techniques can be used instead. In the first part of this paper, we describe extensions of a recently developed open-source framework for creating and training flow network surrogate models, called FlowNet. In particular, we discuss functionality to reproduce historical well rates for wells with arbitrary trajectories, multiple perforated sections, and changing well type or injection phase, as one may encounter in large and complex fields with a long history. Furthermore, we discuss strategies for the placement of additional network nodes in the presence of flow barriers. Despite their flexibility and speed, the applicability of flow network models is limited to phenomena that can be simulated with available numerical simulators. Prediction of poorly understood physics, such as reservoir souring, may require a more data-driven approach. We discuss an extension of the FlowNet framework with a machine learning (ML) proxy for the purpose of generating predictions of H2S production rates. The combined data-physics proxy is trained on historical liquid volume rates, seawater fractions, and H2S production data from a real North Sea oil and gas field, and is then used to generate predictions of H2S production. Several experiments are presented in which the data source, data type, and length of the history are varied. Results indicate that, given a sufficient number of training data, FlowNet is able to produce reliable predictions of conventional oilfield quantities. An experiment performed with the ML proxy suggests that, at least for some production wells, useful predictions of H2S production can be obtained much faster and at much lower computational cost and complexity than would be possible with high-fidelity models. Finally, we discuss some of the current limitations of the approach and options to address them.

Technology Focus: Production Monitoring (March 2022)

We are seeing an uptrend in the instrumentation of legacy wells across the world, as costs lower and business cases become obvious. Several interesting technology applications have been showcased lately about retrofitting instrumentation and control technology in legacy wells. Advances on the digital front, where data science and engineering analytics are becoming more embedded in regular production monitoring and optimization processes, have been widespread. On the gas lift side, developments in surface controllable gas lift valves, which can be deployed during the completion phase, or retrofitted in existing mandrels through well intervention, have proceeded apace. This technology can bring significant improvement in the monitoring and management of wells because the operating envelope can be significantly expanded. Other technology developments involve the use of alternative data. As the saying goes, somebody’s noise is somebody else’s data. A use case of alternative data for rod pumps uses edge computing to process electric measurements and extract features using signal processing and machine learning. This can be used to create synthetic dynamometer cards used to optimize the wells in real time and predict failures ahead of time. Similar techniques could be applicable for electric submersible pumps. Distributed acoustic sensing (DAS) has many applications, but it is challenging to fully exploit the capability because of the sheer amount of raw data coming from these sensors. The data need to be compressed using feature-extraction algorithms. Each use case may require data over different regions of the frequency spectrum covered by the sensors, so how can feature extraction be set to account for all present and future use cases? Significant advances in inflow profiling have been achieved over the past few years, correlating DAS signals against flow-loop measurements and transient simulations. These technologies are very promising, especially now that wet connect and pumpdown technology for fiber optics is gaining more attention. Digital technology for orchestrating production-optimization and reservoir-management work flows has been increasingly embedding machine-learning functionality. Despite these advances, it remains a challenge to maintain these digital solutions over the long term. The most-successful companies in this area ensure the systems are tightly integrated with business-critical work flows, such as integrated activity planning, loss management, locked-in potential management, reservoir-surveillance planning, well-work-opportunity identification, production forecasting, and production back allocation. This involves a significant management of change process. It takes time and effort, but the rewards are worth it. Recommended additional reading at OnePetro: www.onepetro.org. SPE 207879 - Expert Advisory System for Production Surveillance and Optimization Assisted by Artificial Intelligence by Carlos Mata, ADNOC Upstream, et al. SPE 201313 - Production Rate Measurement Optimization Using Test Separator and In-Well Sound Speed by Ö. Haldun Ünalmis, Weatherford SPE 203119 - Wireless Completion Monitoring and Flow Control: A Hybrid Solution for Extended Capabilities by Marcel Bouman, Emerson Automation Solutions

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).

Machine learning classification approach for formation delineation at the basin-scale

Machine learning and artificial intelligence approaches have rapidly gained popularity for use in many subsurface energy applications. They are seen as novel methods that may enhance existing capabilities, providing for improved efficiency in exploration and production operations. Furthermore, their integration into reservoir management workflows may shape the future landscape of the energy industry. This study implements a framework that generates predictive models using multiple machine learning classification-based algorithms which can identify specific stratigraphic units (i.e., formations) as a function of total vertical depth and spatial positioning. The framework is applied in a case study to 13 specific formations of interest (Upper Spraberry through Atoka / Morrow reservoirs) in the Midland Basin, West Texas, United States; a prominent hydrocarbon producing sub-basin of the larger Permian Basin. The study dataset consists of over 275,000 records and includes data fields like formation identifier, true vertical depth (in feet) of formations observed, and latitude and longitude coordinates (in decimal degrees). A subset of 134,374 data records were relevant to the 13 distinct formations of interest and were extracted and used for machine learning model training, validation, and testing. Four supervised learning approaches including random forest (RF), gradient boosting (GB), support vector machine (SVM), and multilayer perceptron neural network (MLP) were evaluated and their prediction accuracy compared. The best performing model was ultimately built on the RF algorithm and is capable of an overall prediction accuracy of 93 percent on holdout data. The RF-based model demonstrated high prediction accuracy for major oil and gas producing zones including the San Andres, Upper Spraberry, Lower Spraberry, Clearfork, and Wolfcamp at 98, 94, 89, 94, and 94 percent respectively. Overall, the resulting data-driven model provides a robust, cost-effective approach which can complement contemporary reservoir management approaches for multiple subsurface energy applications.

The Application of Flow Diagnostics for Reservoir Management

SummaryFlow diagnostics, as referred to herein, are computational tools derived from controlled numerical flow experiments that yield quantitative information regarding the flow behavior of a reservoir model in settings much simpler than would be encountered in the actual field. In contrast to output from traditional reservoir simulators, flow-diagnostic measures can be obtained within seconds. The methodology can be used to evaluate, rank, and/or compare realizations or strategies, and the computational speed makes it ideal for interactive visualization output. We also consider application of flow diagnostics as proxies in optimization of reservoir-management work flows. In particular, by use of finite-volume discretizations for pressure, time of flight (TOF), and stationary tracers, we efficiently compute general Lorenz coefficients (and variants) that are shown to correlate well with simulated recovery. For efficient optimization, we develop an adjoint code for gradient computations of the considered flow-diagnostic measures. We present several numerical examples, including optimization of rates, well placements, and drilling sequences for two- and three-phase synthetic and real field models. Overall, optimizing the diagnostic measures implies substantial improvement in simulation-based objectives.

Uncertainty assessment of Vapex performance in heterogeneous reservoirs using a semi-analytical proxy model

Abstract While compositional simulators are available for modeling the recovery performance of Vapex process, computational constraints often preclude detailed numerical solution of the flow and transport differential equations, as implemented in traditional flow simulators, using an entire suite of geostatistical realizations that represent the uncertainty due to reservoir heterogeneity occurring at different scales. Proxy models that are based on analytical formulations provide efficient alternatives to expensive detailed flow simulations. A novel semi-analytical proxy has been developed to model solvent transport in Vapex at isothermal conditions. Detailed analytical formulations were derived and implemented in a calculation procedure to advance the solvent–oil interface and estimate producing oil rate with time. A mass penetration parameter was formulated, and its change with time was tracked. Results obtained from the proxy were validated against experimental data available in the literatures as well as detailed compositional simulation studies ( Shi and Leung, 2014a , Shi and Leung, 2014b ). Later, a suite of geostatistical realizations are ranked based on their cumulative oil production using this proxy, and the results demonstrate good agreement with those based on detailed compositional simulations but with significant savings in computational costs. Finally, this proxy is employed to assess impacts of uncertainty in subscale heterogeneity. An important contribution from this work is that process physics are built directly into this proxy; it represents an advantage over other alternative modeling approaches (i.e., regression) that are driven only by data. This proposed proxy can be easily integrated in existing reservoir management workflows to optimize production scenarios in a quick and robust manner. It can be applied to rank numerous geostatistical realizations and quickly identify a smaller, more manageable subset of realizations for further simulation analysis. It presents an important tool for assessing the uncertainty in reservoir properties on effective mass transfer and the ensuing recovery performance, as well as assisting decisions-making for future pilot and field development planning.

Ekofisk life-of-field seismic: Operations and 4D processing

In 2010, the world's largest optical permanent reservoir monitoring (PRM) system was installed in the southern part of the Norwegian North Sea at Ekofisk. The life-of-field seismic (LoFS) system consists of 3966 seabed multicomponent sensors along 200 km of mostly trenched fiber-optic seismic cables and covering about 60 km2 of Ekofisk field. Seismic data are acquired via a topside recording unit and a containerized source operated on a supply vessel. Six vintages of data were acquired between the end of 2010 and fall 2013. Different aspects of seismic operations at Ekofisk include seismic source, recording system, data transfer, quality control, and processing. One of the key factors in achieving the full value of a PRM system is to handle such operations in a safe, integrated, and efficient manner to deliver high-quality seismic volumes for interpretation with rapid turnaround. Key aspects of the 4D processing sequence include robustness and optimal turnaround. Integration of the different operational phases of the LoFS project and integration of expertise between client and contractor play a key role in delivering clean, well-resolvable 4D signals and low residual 4D noise with NRMS as low as 5%. The high-quality data delivered by operations and processing are now routinely used in well planning and reservoir-management workflows.

Advanced Reservoir-Management Workflow - EnKF-Based Assisted-History-Matching Method

This article, written by Senior Technology Editor Dennis Denney, contains highlights of paper SPE 118906, ’Advanced Reservoir-Management Workflow Using an EnKF-Based Assisted-History-Matching Method,’ by A. Seiler, G. Evensen, J.-A. Skjervheim, J. Hove, and J.G. Vabo, StatoilHydro, prepared for the 2009 SPE Reservoir Simulation Symposium, The Woodlands, Texas, 2-4 February. The ensemble Kalman filter (EnKF) is a tool to assist history matching. It uses sequential processing of measurements, is capable of handling large parameter sets, and it solves the combined state- and parameter-estimation problem. The proposed workflow was applied to a complex North Sea oil field. The EnKF provided an ensemble of history-matched reservoir models. The updated ensemble was used to predict the uncertainty in future production. Introduction Reservoir modeling and history matching aim to deliver integrated reservoir models for reservoir-management purposes. These reservoir models must reproduce the historical field performance and must be consistent with all available static data (e.g., core, well-log, and seismic) and dynamic data (e.g., well-production, tracer-concentration, and 4D-seismic). Also, they should integrate current information about the reservoir and the associated uncertainty to enable real-time decisions. The EnKF was introduced for updating nonlinear ocean models and is a Monte Carlo approach in which errors are represented by an ensemble of realizations. Model parameters and state variables are updated sequentially as new measurements become available. The result is an updated ensemble of realizations, conditioned to all production data, that provides an improved estimate of the model parameters and state variables and their uncertainty. Previously, the focus was on estimating porosity and permeability fields in the simulation models. Here, parameters such as initial fluid contacts and fault- and vertical-transmissibility multipliers are included as additional uncertain parameters to be estimated. The full-length paper details an EnKF method for history matching reservoir-simulation models and discusses its relation to traditional methods. Assisted History Matching The EnKF method is an alternative to traditional assisted history matching. Conditioning reservoir stochastic realizations to production data is described generally as finding the optimal set of model parameters that minimizes the misfit between a set of measurements and the corresponding responses calculated on the realization of the stochastic model. However, it also can be formulated in a Bayesian framework as finding the posterior probability-density (PD) function of the parameters and the model state, given a set of measurements and a dynamic model with known uncertainties. The Bayesian formulation is the common starting point for traditional minimization methods and sequential-data-assimilation methods (detailed in the full-length paper). EnKF History-Matching Workflow The history-matching workflow, independent of the method used, involves three major steps. First, parameterization of the uncertain parameters, which are a characterization of the major uncertainty of the model solution, is identified. Then, a prior-error model is specified for the selected parameters on the basis of an initial uncertainty analysis. Finally, a solution method must be selected. All three steps may be equally important, and the selections and choices made will depend on the problem at hand.

From Real-Time Data to Production Optimization

Summary A new way of reservoir management is dawning on the horizon: intelligent-reservoir management using continuous data from intelligent wells and/or smart fields. Even though there are many different buzz words for this new technology, they all lead to the same thing—managing a reservoir in real time or close to real time. Real time usually means reacting to an event as it happens or within a short time lag. In the petroleum industry, however, real time is different. This short time lag can be hours, days, or even weeks, which highly depends, of course, on the objective of a project. Integrating real-time data into a reservoir-management workflow, and turning the data into value, is a complicated task. The bottleneck for the data flow, right now, is the transfer of the realtime data, measured with time increments by the second and/or minute and stored on a real-time server, to the engineers' desktops in a clean, timely, and useful fashion. This paper shows ways to provide a continuous (i.e., 24-hours-a-day/7-days-a-week) flow of clean data to the engineers' desktops as a first step in the intelligent-reservoir management. It shows that the implementation of a smart field rises or falls with its ability to provide the data to the information specialist—the petroleum engineer. Because the data are coming into the database very frequently (e.g., at times, every hour or every other day), the engineer is not able to check these data for discrepancies. Therefore, intelligent-reservoir management needs an alarm system that informs the engineers of any underperformance or critical condition of a well or a reservoir. Another important aspect to the intelligent-reservoir management system is the integration of the standard petroleum engineering tools [e.g., decline-curve analysis, material balance, inflow performance relationship (IPR) curves, and reservoir simulation] into this work process. Currently, an IPR curve not only gets data every other month but every other day. This gives the engineer completely new opportunities for closer observation of the workflow (e.g., monitoring the permeability impairment over time). Well tests are usually a snapshot in time, but with continuous surveillance of the reservoir parameters, the development of the skin, for example, can be followed over time and preventative actions can be taken (i.e., predictive maintenance). Neural networks and genetic algorithms are other powerful tools in the real-time environment for handling large amounts of data. A neural network learns from the data gathered and detects underlying relationships—the more data, the better the network can detect these relationships. Once the underlying relationships have been established, the neural networks can be used for predictions (predictive data mining) such as predicting sand production. This approach gives the engineer time to react and prevents the equipment from getting damaged. This work provides a straightforward way of integrating realtime data into a reservoir-management process, and its methodology implies how to gain value from the information provided by a continuous data stream.

Closed-loop reservoir management

The article discusses concepts and computational tools to enable 'closed-loop' model-based reservoir management. Also known as 'real-time' reservoir management, this involves the use of (uncertain) reservoir and production system models in combination with production measurements and other data, such as time-lapse seismics, to continuously update the models. The key sources of inspiration for our work are measurement and control theory as used in the process industry and data assimilation techniques and, in particular, their integrated application in a reservoir management workflow. As an example the authors present early results that illustrate the scope for model-based optimal control of waterflooding using real-time production data under uncertain reservoir conditions.