Advances On Sparse Representation of Reservoir Properties, Fluid Characterization Methods for Steam-Solvent Co-Injection, and Full-Resolution Simulations of Multi-Scale Processes
Recorded on July 10, 2013 (90 minutes)

Select a Webinar Format:
 On-Demand Archive
       $0.00 for Registrant

Topic One

High-resolution reservoir imaging is critical for predicting reservoir response to alternative development strategies and thereby optimizing reservoir production performance. Inference of high-resolution heterogeneous rock properties from low-resolution production measurements leads to a challenging nonlinear inverse problem with many non-unique solutions. The goal of this research is to combine recent advances in signal processing, applied mathematics and computational inverse theory with insights gained from geosciences and multiphase flow in porous media to formulate effective, robust, and geologically consistent frameworks for solving subsurface characterization and imaging inverse problems.

In this presentation, I first motivate low-rank (sparse) representations for describing reservoir properties and review some of the conventional reduced-order parameterization techniques in reservoir engineering. I then introduce a flexible and robust sparse representation approach for reservoir property description that is inspired by recent developments in sparse signal processing. I present some of our recent research work in sparse history matching and discuss the “sparse geologic dictionaries” as a novel conceptual framework for reservoir model selection and updating under geologic uncertainty.

Results from several numerical history matching examples illustrate that geologically-inspired sparse representations of rock properties offer important advantages over conventional reduced-order parameterization techniques. We are currently working on extending our formulation to a stochastic formulation for uncertainty quantification. Our future research in this direction will be focused on developing computationally efficient sparse learning algorithms for estimating facies distribution in large-scale realistic history matching problems.

Topic Two

Design of steam-solvent coinjection requires accurate representation of multiphase behavior that results from interaction of fluid and energy flow in porous media.  Although PVT measurements provide important data at selected pressure-temperature-composition conditions, it is difficult to measure actual phase behavior encountered during in-situ processes.  Thus, reliable fluid characterization is as important as reliable experimental data for coinjection simulation. 

The research project proposed for this SPE award is to develop a fluid characterization method for steam-solvent coinjection.  It aims to accurately model multiphase behavior of heavy-oil/solvent/water mixtures using the Peng-Robinson equation of state (PR EOS).  At most four equilibrium phases are considered; the oleic (L1), gaseous (V), solvent-rich liquid (L2), and aqueous (W) phases. 

A new fluid characterization method is developed based on the concept of perturbation from n-alkanes (PnA).  The focus in the first year is on L1-V two phases.  The PnA method initially gives pseudo components critical parameters that are optimized for n-alkanes in terms of liquid densities and vapor pressures predictions using the PR EOS.  Critical pressures, critical temperatures, and acentric factors for pseudo components are then perturbed in well-defined directions to match PVT data available.  The perturbation considers the size of composition space (i.e., the carbon-number range) and the distribution of components in that space.

Consequently, a lighter oil tends to have a greater sensitivity of the energy density to molecular weight in the PnA method.  The energy density is defined as ai/bi, where ai and bi are the attraction and covolume parameters for pseudo component i, respectively.  This energy density modeling is the key to development of a universal characterization method that is applicable to a wide range of reservoir fluids.

The extensive testing of the PnA method uses 77 different reservoir fluids, including near-critical fluids, gas condensate, volatile oil, and heavy oil.  Challenging cases include predictions of liquid dropout curves for near-critical gas condensates, where accurate prediction of both volumetric and compositional properties is crucial.  Simulation case studies are presented for gas injection processes.  Results show that the PnA method requires much fewer pseudo components to correctly model phase behavior during oil displacements, compared to the conventional characterization methods.  The PnA method uses only three adjustment parameters in its regression process, regardless of the number of components used.  With the PnA method, the PR EOS does not need to correct volumetric predictions using volume shift parameters for the fluids tested.

In the next year, the PnA method will be extended to L1-L2-V three hydrocarbon phases and L1-L2-V-W four phases.  To this end, we have examined prior methods for modeling water-hydrocarbon mixtures, and collected experimental data for such mixtures.  Effects of fluid characterization will be also investigated on flash and stability calculations in solvent injection simulation.

Topic Three

By revisiting classic analytical methods that were known to Newton, this talk will answer a quintessential question of modern simulation research. For any subsurface process, what, precisely, is the instantaneous, spatiotemporal support of change?

Coupled nonlinear flow and transport is inherently a multiscale process; flow evolves on a global scale thereby dictating the extent of the system, whereas transport waves travel with a distinctly local spatial support, dictating resolution. This disparity in scales continues to challenge modern numerical simulation strategies which invariably rely on the orthogonal notions of resolution and extent. Through this work, we can develop full-resolution simulations of multiscale processes while only performing computation when and where it is absolutely necessary.


Behnam Jafarpour, Assistant Professor, Petroleum and Electrical Engineering, University of Southern California

Behnam Jafarpour is an assistant professor of petroleum and electrical engineering at USC Viterbi School of Engineering. From 2008 to 2011, he served as an assistant professor of petroleum engineering at Texas A&M University. He holds a PhD in environmental engineering and a SM degree in electrical engineering from MIT.

Jafarpour currently leads the Subsurface Energy and Environmental Systems lab at USC focusing on systems-based approaches to characterization and development of subsurface energy and environmental resources.


Ryosuke Okuno, Assistant Professor, Petroleum Engineering, University of Alberta
 

Ryosuke Okuno has served as an Assistant Professor of Petroleum Engineering in the Department of Civil & Environmental Engineering at the University of Alberta since 2010. His research and teaching interests include enhanced oil recovery, thermal oil recovery, numerical reservoir simulation, thermodynamics, and multiphase behavior.

Okuno has seven years of industrial experience as a reservoir engineer with Japan Petroleum Exploration Co., Ltd., and is a registered Professional Engineer in Alberta, Canada.    He holds BE and ME degrees in Geosystem Engineering from the University of Tokyo, and a PhD degree in Petroleum Engineering from the University of Texas at Austin.


Rami M. Younis, Assistant Professor, University of Tulsa

Rami M. Younis received his bachelor's degree with Honors in Mechanical Engineering from McGill University, Canada. In 2002 and in 2005 respectively, he received Masters Degrees in both, Petroleum Engineering, and in Scientific Computing and Computational Mathematics from Stanford University, California. He received his Ph.D. in Petroleum Engineering from Stanford University in 2011.

He has served on the faculty of The University of Tulsa since 2012 and is presently the founder and director of the Future Reservoir Simulation Systems and Technology FURSST industrial affiliates consortium at the McDougall School of Petroleum Engineering. His research interests are at the intersection of Applied Mathematics, Computer Science, and Reservoir Engineering. His work on simulation software is enabling a future where simulators will literally write themselves; he is the original developer of the Automatically Differentiable Expression Templates Library ADETL software. On the numerical solution algorithms front, his work is pioneering iteration-free and globally convergent nonlinear solution algorithms.

Dr. Younis has served as technical editor for several journals and is currently associate editor for the SPE Journal. He is the organizer of several SIAM minisymposia. He is the recipient of the Ramey, Brigham, and Centennial Teaching Awards from Stanford University.