History-Matching in Parallel Computational Environments
The overall objective of the proposed research is to develop a new numerical methodology for improved performance prediction of petroleum reservoirs. The strategy combines a stochastic approach for updating the underlying geological model of reservoir heterogeneity with a domain-decomposition technique for distributing the flow simulations across multiple processors.
University of Texas
Researchers established a general procedure for gradually updating geological models within an assisted history-matching framework. A generic, simulator-independent method of estimating sensitivities via multiple realizations was developed and shown to perform as well as the principal-components analysis. Both methods are more robust ways to adjust permeabilities within the spatial domain. The researchers demonstrated this approach on a realistic 3-D test case. A functional prototype of middleware has been tested. The middleware enables a user to apply the history matching algorithm in conjunction with any reservoir simulator.
The goal of history-matching is to obtain a model of a reservoir from which reliable forecasts of future production can be obtained. History-matching-which entails choosing a large set of parameters (e.g., local permeability values) so that a small data set (e.g, well flow rates as a function of time)-is under-constrained. A solution to this problem that makes geological sense is more likely to provide reliable forecasts.
The second guiding idea is that it must be possible to obtain insight from a computer implementation of the history-matching in a practical length of time (e.g., overnight). The time scale for decision-making in many industrial applications does not allow for lengthy calculations. Currently, history-matching is the most time-consuming aspect of any flow simulation project; organizations routinely dedicate several man-months of personnel time to the task. The proposed procedure is automatic and can be scheduled as a job running in the background on whichever computing platforms are available, whether parallel or distributed.
History-matching is an ill-posed problem, and the parameter set that results in minimizing the deviation from data is not unique. Mathematically, the history-matching problem can be posed in an optimization context, i.e., the minimization of a complex, least-squares objective function in a parameter space populated by multiple local minima. Two broad approaches for solving the problem are trial-and-error and gradient-based methods. In contrast, the methodology of this project quantifies the information in production data pertaining to the reservoir heterogeneity in a probabilistic manner. The proposed procedure has only a few deformation parameters to be determined. This results in a computationally efficient history-match procedure.
Project work has been completed. The project has completed these activities:1. tested and validated a stochastic approach to integrating production datata.They have also developed , 2. developed and implemented a domain –decomposition scheme for flow simulation, 3. demonstrated the applicability of the at the approach on realistic data sets and 4. and commence on the development of middleware that facilitates the application of the metho with typical reservoir simulators
Current Status (January 2009)
All of the proposed project work has been completed and the final report is available below under "Additional Information".
Project Start: September 1, 2003
Project End: August 31, 2006
Anticipated DOE Contribution: $393,305
Performer Contribution: $115,296 (20% of total)
NETL - Chandra Nautiyal (firstname.lastname@example.org 918-699-2021)
U.T. Austin - Steve Bryant (email@example.com or 512-471-3250)
Final Project Report [Word-3.13MB]
Srinivasan, S., and Bryant, S: "Integrating Dynamic Data in Reservoir Models Using a Parallel Computational Approach," SPE 89444, SPE/DOE Improved. Oil Recovery, Conference, Tulsa, OK, April 2004.
Yadav, S., Srinivasan, S., Bryant, S., and Barrera, A., History Matching Using Probabilistic Approach in a Distributed Computing Environment," SPE93399, Annual Technology Conference and and Exhibition, ATCE, Dallas, Oct. 2005.
Yadav, S. History Matching Using Probabilistic Approach in a Distributed Computing Environment, M.S. Thesis, The University of Texas at Austin, 2005.
Yadav, S., Bryant, S. and Srinivasan, S. “”Ranking of Geostatistical Reservoir Models and Uncertainty Assessment Using Real-Time Pressure Data”, SPE 100403, Proceedings of 2006 SPE Western Regional/AAPG Pacific Section/GSA Cordilleran Section Joint Meeting held in Anchorage, Alaska, U.S.A., 8–10 May 2006
Demonstration of the new history matching method. Shown is one layer of a five-layer 3D domain: (left) the actual permeability field (middle) the initial guess (right) the history matched result.
The principal component analysis of sensitivities of a permeability field (above) yields geologically reasonable domains (below) that are used for the probabilistic updating.