--> Abstract: Reducing the Need for Up-Scaling: High Resolution 3-D Modeling of Reservoir Processes, by Marcin Krotkiewski, Marcin Dabrowski, and Yuri Y. Podladchikov; #90066 (2007)

Datapages, Inc.Print this page

Reducing the Need for Up-Scaling: High Resolution 3-D Modeling of Reservoir Processes

Marcin Krotkiewski, Marcin Dabrowski, Yuri Y. Podladchikov
Physics of Geological Processes, University of Oslo, PO Box 1048 Blindern, 0316 Oslo, Norway e-mail: [email protected]

The geology and structure of reservoirs are commonly very well described through 3-D seismics and well data. Gaining a quantitative understanding of the processes occurring inside reservoirs proves to be more challenging, as reservoir simulators often fail to resolve the small-scale heterogenic structures. Possible ways out of this dilemma include up-scaling techniques or direct handling of high resolution input data. We follow the later approach to simulate porous flow and seismic wave propagation inside reservoirs by adapting fast algorithms to new computer architectures. In particular, we make use of modern 64-bit, shared memory multiprocessor systems (SMP), which make substantially more memory (RAM) available to a single CPU. Fast and direct access to huge amounts of memory makes it possible to efficiently solve problems of sizes previously requiring expensive clusters. Parallel applications for SMPs are also easier to develop, as they do not require time consuming explicit communication programming. Suitability of an 8 dualcore AMD Opteron machine with 128GB of RAM for application in computational challenges regularly encountered in modeling of geological processes is explored. We solve two standard problems in reservoir simulations: 3D diffusion (parabolic, oil pumping out of a reservoir) and 3D wave propagation (hyperbolic, synthetic seismics) using operator splitting techniques. The emphasis is put on simulations with strong material heterogeneities. We present efficiency results for both sequential and parallel runs of our code. Performance on a single CPU reaches over 1 GFLOPS and the parallel implementation is scalable. We run simulations with up to 8 billion grid points, which is sufficient to accurately resolve the fine-scale structures of reservoirs.


AAPG Search and Discover Article #90066©2007 AAPG Hedberg Conference, The Hague, The Netherlands