Accelerating Multiple-Point Geostatistical Simulation with GPU

Óli D. Jóhannsson; Thomas Mejer Hansen

doi:10.3997/2214-4609.201801725

Accelerating Multiple-Point Geostatistical Simulation with GPU

Óli D. Jóhannsson, Thomas Mejer Hansen

Climate and Geophysics

Abstract

Geostatistical simulation methods are widely used to characterize reservoir models in the Earth Sciences. Multiple-Point Geostatistics (MPS) is a class of methods of sequential simulation, that unlike two-point statistics, can reproduce complex spatial features, such as connectivity, curviliniarity, and features of multiple scale orders. In practice the utilization of MPS based simulation is hampered by their relatively high computational costs. Here a Graphics Processing Unit (GPU) based parallel implementation of IMPALA is presented in which the pattern statistics is scanned only once, and stored on the GPU.

Original language	Danish
Publication date	1 Jun 2018
Number of pages	4
DOIs	https://doi.org/10.3997/2214-4609.201801725
Publication status	Published - 1 Jun 2018
Event	80th EAGE Annual Conference and Exhibition 2018 - Bella Center, København, Denmark Duration: 10 Jun 2018 → 15 Jun 2018 https://events.eage.org/en/2018/eage-annual-2018

Conference

Conference	80th EAGE Annual Conference and Exhibition 2018
Location	Bella Center
Country/Territory	Denmark
City	København
Period	10/06/2018 → 15/06/2018
Internet address	https://events.eage.org/en/2018/eage-annual-2018

Access to Document

10.3997/2214-4609.201801725

Johannsson and Hansen, 2018 - Accelerating Multiple-Point Geostatistical Simulation with GPUFinal published version, 305 KB

Cite this

@conference{1499f8f45ff04d5ca2767b40d107b140,

title = "Accelerating Multiple-Point Geostatistical Simulation with GPU",

abstract = "Geostatistical simulation methods are widely used to characterize reservoir models in the Earth Sciences. Multiple-Point Geostatistics (MPS) is a class of methods of sequential simulation, that unlike two-point statistics, can reproduce complex spatial features, such as connectivity, curviliniarity, and features of multiple scale orders. In practice the utilization of MPS based simulation is hampered by their relatively high computational costs. Here a Graphics Processing Unit (GPU) based parallel implementation of IMPALA is presented in which the pattern statistics is scanned only once, and stored on the GPU.",

author = "J{\'o}hannsson, {{\'O}li D.} and Hansen, {Thomas Mejer}",

year = "2018",

month = jun,

day = "1",

doi = "10.3997/2214-4609.201801725",

language = "Dansk",

pages = "1",

note = "80th EAGE Annual Conference and Exhibition 2018, EAGE 2018 ; Conference date: 10-06-2018 Through 15-06-2018",

url = "https://events.eage.org/en/2018/eage-annual-2018",

}

TY - ABST

T1 - Accelerating Multiple-Point Geostatistical Simulation with GPU

AU - Jóhannsson, Óli D.

AU - Hansen, Thomas Mejer

PY - 2018/6/1

Y1 - 2018/6/1

N2 - Geostatistical simulation methods are widely used to characterize reservoir models in the Earth Sciences. Multiple-Point Geostatistics (MPS) is a class of methods of sequential simulation, that unlike two-point statistics, can reproduce complex spatial features, such as connectivity, curviliniarity, and features of multiple scale orders. In practice the utilization of MPS based simulation is hampered by their relatively high computational costs. Here a Graphics Processing Unit (GPU) based parallel implementation of IMPALA is presented in which the pattern statistics is scanned only once, and stored on the GPU.

AB - Geostatistical simulation methods are widely used to characterize reservoir models in the Earth Sciences. Multiple-Point Geostatistics (MPS) is a class of methods of sequential simulation, that unlike two-point statistics, can reproduce complex spatial features, such as connectivity, curviliniarity, and features of multiple scale orders. In practice the utilization of MPS based simulation is hampered by their relatively high computational costs. Here a Graphics Processing Unit (GPU) based parallel implementation of IMPALA is presented in which the pattern statistics is scanned only once, and stored on the GPU.

U2 - 10.3997/2214-4609.201801725

DO - 10.3997/2214-4609.201801725

M3 - Konferenceabstrakt til konference

SP - 1

T2 - 80th EAGE Annual Conference and Exhibition 2018

Y2 - 10 June 2018 through 15 June 2018

ER -