Research On The Numerical Simulation Of Elastic Waves In Finite-difference Method Based On GPU

Seismic wave field numerical simulation technology use theoretical calculationmethod to research the propagation of seismic wave while given the undergroundmedium structure and parameters, and then synthetize to seismic record.The finite difference method is one of the most commonly used modelingmethods, it replaces the derivative space and time with the corresponding wave fieldfunction, which is approximate because of the discrete, inevitably reduces thenumerical simulation result resolution. Therefore, in order to reduce the numericaldispersion, commonly used method is to improve the model of the subdivisionaccuracy. But when we improve the model accuracy or increase the grid size, thefinite difference method will spend a large amount of computing time, then theconsideration of using the parallel computing method to improve the efficiency of thealgorithm is called the most effective way.The unified calculation based on GPU hardware platform CUDA developed byNVIDIA company, fusion calculation characteristics and C language programminginterface GPU in general, computational efficiency and develop friendly become ahotspot of current high performance computing.This paper takes CUDA as the application development platform, the followingmain research focus:(1) In the three-dimensional isotropic medium, we use the staggered grid finitedifference method for elastic wave numerical simulation, and PML perfectly matchedlayer boundary condition for the effective treatment.(2) Study on the effect of the finite difference method of numerical accuracyfactors simulation and suppression method of dispersion, including spatial gridprecision and difference equation of order.(3) Study the effect of accuracy and the order to enhance the efficiency of thesimulation, and to this end, the multi-thread parallel computing in numericalsimulation of elastic waves.(4) Research on finite difference based on CUDA component implementation method of numerical simulation of elastic wave, and discusses the parallel computingfor numerical simulation of the actual performance of speed to further theimprovement potential.