Application of Grid-enabled technologies for solving optimization problems in data-driven reservoir studies
Introduction
The ultimate goal of reservoir modeling is to generate both good estimates of reservoir parameters and reliable predictions of oil production to optimize return on investment from a given reservoir. This process is performed by the use of numerical simulators that represent the multiphase fluid flow phenomenon under the subsurface. However, little use has been made of reservoir simulations coupled with systematic optimization techniques. The main advantage of applying these mathematical tools to the decision-making process is that they are less restricted by human imagination than conventional case-by-case comparisons.
A key issue is to come up with reliable prediction models, which operate by searching a large space of oil production and reservoir parameters. The dynamic, data-driven application systems (DDDAS) paradigm provides a viable mechanism to address this issue. A main feature of DDDAS is the on-the-fly interaction between and integration of numerical models and data from simulations or field measurements. The integration of data and numerical models through DDDAS allows for a more efficient search of the parameter space. Several obstacles, however, need to be addressed for a successful application of DDDAS. The first one is the computational time required to complete simulations of complex, large scale reservoir models. Another challenge is to implement the capability to manage and navigate multi-terabyte datasets from simulations and field measurements. Optimization strategies normally evaluate hundreds or even thousands of scenarios (each requiring a simulation run) in the course of searching for the optimal solution to a given management question. This process is extremely time-consuming and data-intensive [1], [2].
Grid computing is rapidly emerging as the dominant paradigm for large-scale parallel and distributed computing. A key contribution of Grid computing is the potential for seamless aggregations of and interactions among computing, data, and information resources, which is enabling a new generation of scientific and engineering applications that are self-optimizing and dynamic data driven. However, achieving this goal requires a service-oriented Grid infrastructure that leverages standardized protocols and services to access hardware, software, and information resources [3], [2].
In our previous work, we described a suite of tools and middleware that enable execution and analysis of large, distributed collections of simulations and datasets [4], [5]. In this paper, we present the infrastructure for solving optimization problems in dynamic, data-driven reservoir simulations in the Grid. The infrastructure builds on three key components; a computational engine consisting of a simulation framework (IPARS) and optimization services, middleware for distributed data querying and subsetting (STORM), and an autonomic Grid middleware (Discover/Pawn) for service composition, execution, and collaboration. We describe these components and their application in autonomic data-driven management of the oil production process [3], [6].
Section snippets
The integrated parallel accurate reservoir simulator (IPARS)
IPARS represents a new approach to parallel reservoir simulator development, emphasizing modularity, code portability to many platforms, ease of integration and interoperability with other software. It provides a set of computational features such as memory management for general geometric grids, portable parallel communication, state-of-the-art non-linear and linear solvers, keyword input, and output for visualization. A key feature of IPARS is that it allows the definition of different
Querying and subsetting of distributed data: STORM
An increasingly important issue in Grid computing is to enable access to and integration of data in remote repositories. An emerging approach is the virtualization of data sources through relational and XML models [11], [12], [13]. STORM (formerly called GridDB-Lite) [14] is a service-oriented middleware that supports data select and data transfer operations on scientific datasets, stored in distributed, flat files, through an object-relational database model. In STORM, data subsetting is done
An autonomic Grid middleware for oil reservoir optimization
Discover [18] enables seamless access to, and peer-to-peer integration of applications, services, and resources on the Grid. The middleware substrate integrates Discover collaboratory services with the Grid services provided by the Globus Toolkit using the CORBA Commodity Grid (CORBACoG) Kit [19]. It also integrates the Pawn peer-to-peer messaging substrate [20]. Pawn enables decentralized (peer) services and applications to interact and coordinate over wide area networks. Finally, the DIOS [21]
Putting it together: data-driven oil production optimization
The oil production optimization process involves (1) the use of an integrated multi-block reservoir model and several numerical optimization algorithms (global and local approaches) executing on distributed computing systems on the Grid; (2) distributed data archives for historical, experimental (e.g., data from field sensors), and simulated data; (3) Grid services that provide secure and coordinated access to the resources and information required by the simulations; (4) external services that
Policy-driven optimization
A key objective of this research is to formulate policies that can be used by the autonomic self-optimizing reservoir framework to discover, select, configure, and invoke appropriate optimization services to determine optimal well locations. The choice of optimization service depends on the size and nature of the reservoir. The SPSA algorithm is suited for larger reservoirs with relatively smooth characteristics. In case of reservoirs with many randomly distributed maxima and minima, the VFSA
Conclusion
In this paper, we presented an infrastructure and its components to support the autonomic oil production management process. Use of this infrastructure to implement Grid-enabled data-driven application support can aid in gaining better understanding of subsurface properties and decision variables. With a better understanding of these properties and variables, engineers and geoscientists can implement optimized oil production scenarios. We believe autonomic oil production management strategies
Acknowledgment
We would like to thank R. Martino and M. Peszyńska for their help in executing the VFSA-based experiments. This work is partly supported by the National Science Foundation under Grants ACI-9619020 (UC Subcontract 10152408), ANI-0330612, EIA-0121177, SBR-9873326, EIA-0121523, ACI-0203846, ACI-0130437, ACI-9982087, CNS-0305495, NPACI 10181410, ACI-9984357, EIA-0103674 and EIA-0120934, DOE ASCI/ASAP via grant numbers PC295251 and 82-1052856, Lawrence Livermore National Laboratory under Grant
Manish Parashar is associate professor of electrical and computer engineering at Rutgers University, where he also is director of the Applied Software Systems Laboratory. He received a BE degree in electronics and telecommunications from Bombay University, India in 1988, and MS and PhD degrees in computer engineering from Syracuse University in 1994. He has received the NSF CAREER Award (1999) and the Enrico Fermi Scholarship from Argonne National Laboratory (1996). His current research
References (22)
- et al.
Applying soft computing methods to improve the cómputational tractability of a subsurface simulation–optimization problem
J. Petroleum Sci. Eng.
(2001) - et al.
Distributed processing of very large datasets with DataCutter
Parallel Comput.
(2001) - et al.
Autonomic optimization of an oil reservoir using decentralized services
- V. Bhat, V. Matossian, M. Parashar, M. Peszynska, M. Sen, P. Stoffa, M.F. Wheeler, Autonomic oil reservoir optimization...
- J. Saltz, et al., Driving scientific applications by data in distributed environments, in: Dynamic Data Driven...
- M. Parashar, R. Muralidhar, W. Lee, M.F. Wheeler, D. Arnold, J. Dongarra, Enabling interactive oil reservoir...
- W. Bangerth, H. Klie, V. Matossian, M. Parashar, M.F. Wheeler, An autonomic reservoir framework for the stochastic...
- IPARS: Integrated Parallel Reservoir Simulator, The University of Texas at Austin,...
- et al.
Global optimization methods in geophysical inversion
Introduction to stochastic search and optimization:estimation, simulation and control
(2003)
Cited by (23)
Well placement optimization under time-dependent uncertainty using an ensemble Kalman filter and a genetic algorithm
2013, Journal of Petroleum Science and EngineeringCitation Excerpt :Including uncertainty can further increase the number of required simulations. The use of parallel computing to distribute the simulations over a network of CPUs has been suggested previously (e.g., Bangerth et al., 2005; Cullick et al., 2005; Parashar et al., 2005) and we implement this here. During the EnKF update process, data is acquired from the commercial simulator in P parallel streams, where P is the number of required slave processes.
Well placement optimization: A survey with special focus on application for gas/gas-condensate reservoirs
2012, Journal of Natural Gas Science and EngineeringCitation Excerpt :Although this is a reliable technique, an excessive number of simulations needed in some of the optimization techniques (e.g. genetic algorithm) can make them too computationally expensive for application in large-scale fields. Some researchers suggested the use of parallel computing to distribute the simulations on a network of computers (e.g. Cullick et al. (2005), Bangerth et al. (2005), Parashar et al. (2005)). Other researchers combined finite difference with various proxy models or response surface models (RSM) to reduce necessary reservoir simulations (e.g. least squares by Pan and Horne (1998), kriging by Pan and Horne (1998), Guyaguler and Horne (2000) and Ozdogan et al. (2005), artificial neural networks by Centilmen et al. (1999) and Yeten et al. (2003), statistical proxy model by Onwunalu et al. (2008), and neuro-fuzzy proxy model by Zarei et al. (2008)).
Simulation-based optimization of multiple-task GRID applications
2008, Future Generation Computer SystemsCitation Excerpt :For example, an application has to declare how many nodes or processors it needs, and it has no way to “tune” this choice taking into account the system load: the resource manager simply blocks it until all the requested nodes or processors are available. On the other hand, developers try to design flexible and scalable applications, trying to optimize performance even on systems whose computational power can change during their life cycle (for example after hardware expansion) [9–12]. In other words, resource schedulers may be able to optimize administrator-oriented performance metrics, such as resource utilization, but usually cannot optimize user-oriented performance metrics, such as application response time.
Dynamic data driven application system: Recent development and future perspective
2007, Ecological ModellingOptimal task partition and distribution in grid service system with common cause failures
2007, Future Generation Computer SystemsCitation Excerpt :C. Li and L. Li [22] and Buyya et al. [3] also introduced the optimal task/resource scheduling problems and showed the significant improvement by a good schedule strategy. Some other optimization schemes, proposed for grid or cluster, include [28,14,26]. However, none of them consider the reliability factor when solving the optimization problems.
Symbiotic Simulation System (S3) for Industry 4.0
2019, Springer Series in Advanced Manufacturing
Manish Parashar is associate professor of electrical and computer engineering at Rutgers University, where he also is director of the Applied Software Systems Laboratory. He received a BE degree in electronics and telecommunications from Bombay University, India in 1988, and MS and PhD degrees in computer engineering from Syracuse University in 1994. He has received the NSF CAREER Award (1999) and the Enrico Fermi Scholarship from Argonne National Laboratory (1996). His current research interests include autonomic computing, parallel, distributed and Grid computing, networking, scientific computing, and software engineering. Manish is a member of the executive committee of the IEEE Computer Society Technical Committee on Parallel Processing (TCPP), part of the IEEE Computer Society Distinguished Visitor Program (2004–2006), and a member of ACM. He is also the co-founder of the IEEE International Conference on Autonomic Computing (ICAC). Manish has co-authored over 130 technical papers in international journals and conferences, has co-authored/edited 5 books/proceedings, and has contributed to several others in the area of parallel and distributed computing.
Hector Klie got a PhD degree in computational science and engineering at Rice University, 1996, he completed his master and undergraduated degrees in computer science at the Simon Bolivar University, Venezuela in 1991 and 1989, respectively. Hector Klie’s main research interests are in the mathematics, numerical analysis, and high performance computational issues related to the solution of transport and flow of porous media equations. He devotes particular interest to the development of parallel iterative methods for the solution of large scale linear, nonlinear and coupled problems. He is currently pursuing novel optimization approaches for oil reservoir simulation coupled with seismic, geomechanics and other different physical models (multiphysics) for instrumented and automatic oil field management. He also has extensive industrial experience in various geophysical problems such as tomography, wave modeling in anisotropic media and converted wave data processing. He has written several papers in these geoscientific research areas. He currently holds the position of associate director and senior research associate in the Center for Subsurface Modeling at the Institute of Computational Science and Engineering at The University of Texas at Austin.
Umit Catalyurek is an assistant professor in the Department of Biomedical Informatics at The Ohio State University. His research interests include graph and hypergraph partitioning algorithms, grid computing, and runtime systems and algorithms for high-performance and data-intensive computing. He received his PhD, MS and BS in computer engineering and information science from Bilkent University, Turkey, in 2000, 1994 and 1992, respectively.
Tahsin Kurc is an assistant professor in the Department of Biomedical Informatics at the Ohio State University. His research interests include runtime systems for data-intensive computing in parallel and distributed environments, and scientific visualization on parallel computers. He received his PhD in computer science from Bilkent University, Turkey, in 1997 and his BS in electrical and electronics engineering from Middle East Technical University, Turkey, in 1989.
Wolfgang Bangerth is a postdoctoral research fellow at both the Institute for Computational Engineering and Sciences, and the Institute for Geophyics, at The University of Texas at Austin. He obtained his PhD in applied mathematics from the University of Heidelberg, Germany in 2002. He is the project leader for the deal.II finite element library (http://www.dealii.org). Wolgang is a member of the ACM.
Vincent Matossian obtained a masters in applied physics from the French Université Pierre et Marie Curie. Vincent is currently pursuing a PhD degree in distributed systems at the Department of Electrical and Computer Engineering at Rutgers University under the guidance of Manish Parashar. His research interests include information discovery and ad-hoc communication paradigms in decentralized systems.
Joel Saltz is professor and chair of the Department of Biomedical Informatics, professor in the Department of Computer and Information Systems and a senior fellow of the Ohio Supercomputer Center. Prior to coming to Ohio State, Dr. Saltz was professor of Pathology and Informatics in the Department of Pathology at Johns Hopkins Medical School and professor in the Department of Computer Science at the University of Maryland. He received his MD and PhD in computer science from Duke University in 1985 and 1986, respectively. He earned his BS in mathematics and physics from University of Michigan in 1978. His research interests are in the development of systems software, databases and compilers for the management, processing and exploration of very large datasets.
Dr. Mary Fanett Wheeler has directed the Center since its inception as the Subsurface Modeling Group at Rice University from 1990 to 1995, when the group moved to The University of Texas at Austin. She has directed over 25 PhD graduates who are presently employed in the petroleum and environmental industries, universities, and government laboratories. Her primary research interest is in the numerical solutions of partial differential systems with applications to flow in porous media, geomechanics, surface flow, and parallel computation. Her numerical work includes formulation, analysis and implementation of finite-difference/finite-element discretization schemes for nonlinear, coupled PDE’s as well as domain decomposition iterative solution methods. Dr. Wheeler has published more than 100 technical papers and edited seven books. She is currently an editor of six technical journals and managing editor of Computational Geosciences. In 1998 she was elected to the National Academy of Engineering and currently holds the Ernest and Virginia Cockrell Chair in Engineering. She is currently professor in the Aerospace Engineering and Engineering Mechanics and in Petroleum and Geosystems Engineering.