Seismic Exploration In2D Frequency-domain Forward Based On Parallel Algorithms

With the development of seismic exploration,Exploring in complex geologicalcondition for gas and oil has been the major objective of geophysical exploration,full-waveform inversion method use the full seismic wave field information, it haspotential for revealing geologic structure and reservoir physical properties in complexgeological conditions. But the data volume and computational cost is too big to bewidely used. Therefore, it is meaningful to develop a high performance computingmethod.Forward modeling is an important method of researching the kinematics anddynamics information of seismic wave propagation. and it is the foundation offull-waveform inversion.Efficient and accurate forward modeling directly affectswaveform inversion result.This paper is based on theory of2d frequency-domainwaveform forward modeling. We used9-point finite difference scheme in mixed-gridstencil to discretize visco-acoustic wave equation. Construct a large sparse matrixequation, we compress and store this equation and used direct method to solve theforward modeling equation. We used PML(Perfectly Matched Layers) and/or45°paraxial condition for absorbing boundary conditions.Frequencies and multi-source simulations computing independently infrequency-domain forward modeling, it is suitable to use MPI parallel method tospeedup. So,this paper use a combination of coarse-grained parallel algorithm basedon MPI, we put different frequency group of forward modeling in different MPIcommunicators, make the frequencies computing independently. in eachcommunicator, we use MUMPS(Multi-frontal Massively Parallel Sparse Direct Solver)based on MPI to speedup the method of solving the equations. we achieve thefrequency-domain waveform forward methods algorithm based on multi-MPIcommunicators. This paper develop different size of FD grid to test one shot and multi-shotsmodels in one computer and in PC clusters. We compare the computational efficiencyof one communicator to multi-communicators, from the test, we verified theeffectiveness and computational efficiency of our algorithm in multi-shots models.Themore MPI communicators we make the shorter time we spend if the hardwareconditions is possible.