Surface-related Multiple Suppression Method Investigation Based On Wave Equation

Multiple widely exists in seismic data, especially the marine data. At present, majority processing methods only use primary imaging, the existence of multiple can seriously cover up primary reflection information, improper multiple processing will directly affect the imaging and seismic data interpretation. Therefore, it is greatly significant to do multiple attenuation for seismic data.Surface-related multiple suppression method based on wave equation, also known as SRME, is a data-driven method without any prior information, Using the feedback iterative method can effectively suppress multiples in complex media. The SRME method can be divided into two steps: surface-related multiple prediction and adaptive subtraction. At present, under the influence of computation and data storage, the SRME method based on wave equation is mostly limited to two-dimensional, or use two-dimensional algorithm to do approximate calculation on three-dimensional seismic data through a series of assumptions,this goes against the situation of underground medium, the suppression result is not good. The characteristics of matched filtering method based on L2 norm is fast and stable. Its successful application requires two assumptions: primary and multiple are orthogonal and record after multiple suppression has the minimum energy.This paper have completed the conventional 2D SRME method and matched filtering method based on L2 norm, innovative completed five aspects: One is 3D SRME prediction multiple method, the method breaks through the limitation of the two-dimensional algorithm,considering the spatial propagation of seismic wave in underground medium, it can predict multiple more accurate. Two is 3D SRME based on parabolic sparse inversion method without interpolation, the method uses parabolic Radon transform instead of crossline direction summation of multiple contribution gathers. The method avoids interpolation and extrapolation of seismic data in crossline direction, and reduces spatial aliasing. Three is matched filtering method based on L1 norm, this method can avoid two assumptions based on L2 norm filtering method, the robustness of the L1 norm remain stable for the abnormal value. Four is matched filtering method based on L1/L2 combined norm to attenuate multiple,this paper makes use of the characteristics of L1 norm and L2 norm matched filtering method,proposed L1/L2 norm combined matching filtering method to attenuate multiple. This method can overcome disadvantages of the above two methods, get the convergence wienerfilter in relatively short time, making predicted multiple and real multiple fit better. Five is the GPU(Graphics Processing Unit) and CPU(Central Processing Unit) parallel acceleration technology, it is successfully applied in the above algorithms, which greatly improves the computational efficiency of multiple prediction and adaptive subtraction.