DIGIRES is a joint Petromaks-2 and industry project that aims to develop the next-generation digital workflows for sub-surface field development and reservoir management. As such, DIGIRES addresses new challenges in the petroleum industry related to the processing and integration of a vast amount of data with models for reservoir characterization.
NORCE coordinates the project with industry support from Equinor, Aker BP, Vår Energy, Neptune Energy, DEA Wintershall, Petrobras, and Lundin. The University of Stavanger is a research partner in the project.
The project builds on an integrated reservoir-management philosophy for sub-surface modeling. We use multiple model realizations to characterize uncertainty together with sub-surface analytics and digitalization to handle big data. The project's objective is to "improve decision-making and uncertainty analysis for well-planning and field development by using a decision-driven ensemble-based approach."
Thus, an essential element of DIGIRES is the transition from data-driven to decision-driven workflows and the transformation into big-data analytics and digitalization. DIGIRES combines data analytics with model predictions and expert knowledge. A particular project outcome is the implementation and demonstration of ensemble-based probabilistic decision-making for reservoir management.
DIGIRES workflows and interfaces will allow for efficient processing of big sub-surface data sets. The project will improve reservoir understanding and decision-making to maximize future value creation.
DIGIRES integrates industrial experience and technology solutions, real field data, and forefront research by independent institutes and academia. As such, DIGIRES applies the most up-to-date technological solutions and methods to actual petroleum reservoirs with big data.
A significant delivery from DIGIRES is the full implementation of our newly formulated and published subspace EnRML (an iterative ensemble smoother) method in the Ensemble Reservoir Tool (ERT). Equinor has approved this method for general use in the company. We have published two high-impact papers documenting the work with the new subspace EnRML and its ERT implementation.
We have solved an outstanding problem of accounting for unknown model errors within history matching. Building on this work, we have reformulated the history-matching problem with consistent error statistics and illustrated its solution with the subspace EnRML. This study will likely change how we history-match the reservoir models and has lifted the formulation and solving of the history-matching problem to the next level.
We are now implementing these methods for operational use in our industry partners' software through ongoing separate industry projects.
We have developed and published new adaptive localization methods for ensemble history matching. We are now implementing these methods for operational use in our industry partners' software through ongoing separate industry projects.
We have also developed and published a "particle-flow" method for history matching (which avoids an approximation made in standard formulations). This research is a collaboration with Petrobras.
We have developed a public domain Python tutorial for ensemble history matching and optimization that is particularly suited for teaching and training reservoir engineers.
We have published a new method for optimizing discrete control variables (like drilling order), which we will now test on real field applications in collaboration with our industry partners. We have recently extended this work to account for the reservoir uncertainty to allow for robust optimization. We are now implementing these methods for operational use in our industry partners' software through ongoing separate industry projects.
We are now publishing new results using Bayesian optimization methods for optimizing well-controls and production strategies.
We have developed seismic data compression and feature extraction methods, and we used machine learning to replace a poorly known rock physics model. We have demonstrated the benefit of conditioning on compressed and denoised 4D data.
We have studied the impact of risk attitudes in production decisions and demonstrated robust ensemble optimization within decision-making.
DIGIRES has to lead to new state-of-the-art methods for ensemble history matching, robust optimization, and robust decision
making. "Robust" denotes that we take the reservoir uncertainty into account in the optimization and decision making.
DIGIRES has revised the formulation of the history matching problem and has led to a new formulation with consistent error
statistics that will be the standard in future history matching applications.
DIGIRES has provided and demonstrated a closed-loop reservoir management workflow. It uses an ensemble of models to represent
the reservoir uncertainty, ensemble history matching for recursive updating of the reservoir description, ensemble optimization
methods for finding the best production- or drilling strategy, and robust ensemble-based decision methods for decision making.
Better knowledge of the uncertainty reduces risks and leads to better decisions. The problem is first to create a consistent uncertainty basis and after that to use the knowledge about the uncertainty in a mathematically coherent and computationally efficient manner in the decision process.
This project builds on our previous experience from ensemble-based conditioning and optimization, from which we postulate the hypothesis that "it is possible to develop computationally efficient ensemble methods for probabilistic decision making in high-dimensional and nonlinear dynamical systems," taking the uncertainty into account.
The approach taken is to use multiple realizations and ensemble methods for generating the best possible uncertainty basis and then develop decision methods that use the ensemble of realizations as input in the decision-making process.
The project addresses the challenges:
1. How can we formulate an ensemble framework that provides accurate (in agreement with data) and consistent (in agreement with theory) knowledge about the probability of an event?
2. How can knowledge about the probability as represented by an ensemble of model states best be used to improve decision making and risk management, and create additional value?
If successful, this project will provide computationally affordable methods for decision making under uncertainty that integrates well with the ensemble modeling currently used in many fields, and particularly in reservoir modeling, for prediction of uncertainty and model conditioning.
We will perform several studies together with our industry partners where we use the ensemble decision system for real reservoir-management problems with real data.
The need for robust decision methods has become apparent when working with oil companies, which, in most cases, make decisions based on a single "best" model realization. The number of industry participants in the project documents the interest in DIGIRES.