Font Size: a A A

The Parallelization Of One-dimensional Electromagnetic Method Forward Modeling And The Preliminary Study Of GPU Algorithm

Posted on:2013-09-23Degree:MasterType:Thesis
Country:ChinaCandidate:F B LiuFull Text:PDF
GTID:2230330371482568Subject:Earth Exploration and Information Technology
Abstract/Summary:PDF Full Text Request
Forward problem of electromagnetic method is essentially to solve geophysicalmodel of the response function. The forward is prerequisite and condition of theinversion. Only by accurately calculated the earth’s response function, will it bepossible to obtain reliable geophysical model.Electromagnetic forward and inversion calculation experiences one-dimensional,two-dimensional to three-dimensional development process, but for any kind ofinversion method, large quantities of forward calculation was inevitable, forwardcalculation speed determines the inversion calculation speed. At present the mainresearch in the forward calculation are2D and3D finite difference, finite element andintegral equation. The one-dimensional forward calculation is less mentioned.Compared with2D and3D forward calculation, one-dimensional forward calculationof development is relatively mature, and in the actual production has been widelyused.With the development of parallel computing technology and its applications inelectromagnetic method, it is possible for researching of one-dimensional forwardcalculation process and improvement of calculation speed. CPU parallel computing isaccompanied by multiple nuclear CPU computer, CPU parallel computing technologyhas been widely applied in many fields. Although CPU parallel technology is mature,but the hardware speed is limited, so CPU parallel computing acceleration propertiesare limited. In recent years, GPU is a kind of parallel computing technology widelyapplied, the use of GPU acceleration can greatly improve the computational efficiencyof general science and engineering. Since GPU technology has appeared with rapiddevelopment, usually a GPU with dozens or even thousands of stream processors.GPU calculation ability is more than hundreds times of CPU and in a large the amount of data operation is very practical. This paper tries to apply CPU and GPU parallelcomputing into the MT one-dimensional forward calculation and the electric dipole ofCSAMT one-dimensional forward calculation. Result shows that, both CPU and GPUforward speed was improved, especially the GPU computation time greatly shortened.At the same time, based on the frequency domain dipole source, time domain dipolesource and time domain of the large fixed loop and other commonly usedelectromagnetic methods discussed, hoping that the GPU parallel algorithm is furtherpopularized in electromagnetic method.In this paper all procedures are all written by the Fortran language, consideringthat most workers using C, C++language as the main programming language, andthese two kinds of computer language has certain defects in the numerical calculation.This paper attempts to use the C, the C++language commonly used complexcomputation module, in order to check for.This article discusses restricted to some commonly used electromagnetic methodfor forward. It is hoped that the next step work can be GPU parallel computingintroduced into one-dimensional inverse calculation, and in practical work has beenapplied.
Keywords/Search Tags:Electromagnetic, One-dimensional Forward, Parallel Computing, GPU
PDF Full Text Request
Related items