The Method Of Multiple Attenuation Based On Independent Component Analysis

In seismic data processing,the interference of multiple reflections is still a large problem,especially for marine dataset.Strong multiples can lead to fault seismic interpretation and finally lead to dry-well.So it is important and necessary for multiple suppression.Tradional technology for multiple based on the principle of the least energy of output signals and deduced by second order statistical data,need to hypothesize that singal reflection and multiple must be orthogonal for optimality solution,which is not in accordance with the real situation of seism exploration.Independent component analysis(ICA) drived from blind source separation(BBS) is a new method for finding underlying factors or components from multivariate statistical data.The goal of ICA is to recover the unobserved source signals without any prior information given only the sensor observations that are unknown linear mixtures of the independent sources.And ICA do not need the hypothesization,which is more in accordance with the real situation.This paper raised the ICA model for blind separation of multiples,based on studying and nanlysing the theory of ICA and the common technology of multiple attenuation,and the structure of seismic data,and designed the seoaration algorithm with the principle of non-Gauss maximization,which solved the uncertainly problem about the amplitude and order of singal reflection handily,and achieved the goal of separation of singal reflection and multiple,taking useful of the prior knowledge about mixed matrix.We compound the seismic data in section according to the t-x equation of singal reflection and multiple,and simulated,compared with the common method,and got better result,which proved the new mthod effective.ICA overcome the defect that common method hypothesize that singal reflection and multiple are orthogonal,remove multiple effectively from seismic data,which reveal the application value and future of ICA in seismic data processing.