Abstract

Surface-related multiple prediction by SRME call for much space convolution computation, which show matrix multiplication at each frequency component. In the abstract, we exploit the computation advantage of GPU to calculate matrix multiplication. We arrange concentrated matrix multiplication on GPU to accelerate surface-related prediction, and other computation on CPU sequentially. Surface-related multiple will be suppressed by cooperated and parallel computation on GPU/CPU, which improve computation efficiency for surface-related multiple prediction greatly. Shot data and complex Pluto synthetic data model are tested by the proposed method. Comparison result show the application of GPU computation technology accelerates multiple prediction greatly.