Font Size: a A A

Parallel Algorithm Research Of2-D Waveform Forward And Inverse In Frequency Domain Based On MPI+CUDA

Posted on:2013-06-24Degree:MasterType:Thesis
Country:ChinaCandidate:X Q WangFull Text:PDF
GTID:2230330371482384Subject:Earth Exploration and Information Technology
Abstract/Summary:PDF Full Text Request
With the seismic exploration targets increasingly complex and the requirementsof exploration increasingly fine, seismic exploration data and calculated amount areincreasing, which makes it a higher demand on high-performance computing.Full-waveform inversion method use the full wave field information, so it has thepotential to accurately portray the details of the model, it will be the developmenttrend of the future seismic exploration. However, due to the full-waveform inversionand waveform forward numerical simulation has a large amount of storage andcomputational limitations, this method is still not widely used in the real seismic dataprocessing. Therefore, to develop high-performance computing in the full-waveforminversion of seismic exploration is very necessary. once overcome the difficulties ofcalculation, the full-waveform inversion method will get satisfactory results.Waveform numerical simulation of different frequencies in the frequency domainis calculated independently, and facilitate the implementation of the multi-scalewaveform inversion strategy from the low frequency to high frequency. So weadopted the2-D frequency domain full-waveform inversion theory, derived isotropicmedium2-D acoustic wave equation finite difference scheme in the frequency domain,derived the Jacobian matrix, the gradient direction and approximate Hessian matrixformula of the gradient inversion method on the basis of the previous. We can see thatdomain waveform numerical simulation and waveform inversion methods cann’t bewidely used because of the huge coefficient matrix storage restriction and the solvingproblem of the large-scale ribbon coefficient matrix.In view of the above problems, the coordinate format was used to store the largesparse ribbon coefficient matrix, and the iterative method library accelerated by GPUtechnology was called to solve the linear equations, while the MPI technology was applied to independently calculate the frequency domain waveform forward of eachfrequency and the frequency-domain waveform inversion of Multi-RHS, to achievethe frequency-domain waveform forward and inversion parallel algorithm. Finally,verified the effectiveness and computational efficiency of the parallel algorithm bysome waveform forward and inversion cases of several theoretical models in thefrequency domain.
Keywords/Search Tags:Full-waveform inversion, Frequency-domain waveform forward, Sparse linear equations, parallel algorithm, MPI/GPU
PDF Full Text Request
Related items