Seismic Noise Attenuation Based On Rank-reduction Method And Empirical Mode Decomposition

With the development of petroleum exploration,the geological environments and structures have become more and more complicated.The signal-to-noise ratio of seismic data acquired from mid-deep layers is always low and the useful signal is weak.There exist seriously random or coherent noise that would directly affect the accuracy of geological interpretation.Seismic noise attenuation is an essential and significant step in the seismic data processing flows.Stated thus,it's necessary to intensively study the seismic denoising methods.The rank-reduction method and the mode decomposition method have become hot topics at the field of seismic noise attenuation in recent years.In this paper,we remove seismic random noise through forming Hankel matrix and performing rank-reduction step.We remove random and coherent noise through f-x empirical mode decomposition.The main workflow can be summarized as follows:1.We analyze the basic principle of forming Hankel matrix approach to suppress seismic random noise in the rank-reduction method and take focus on the condition of linear events,rank parameters selection and the steps of the algorithm.On the other hand,we study the differences of singular values distribution between the pure seismic data and noisy seismic data.2.We study the four rank-reduction methods including multichannel singular spectrum analysis(MSSA)with truncated singular value decomposition(TSVD),improved damped multichannel singular spectrum analysis(DMSSA)based on a damping factor,data-driven optimal singular value shrinkage(OptShrink)based on optimal(re)weighting of the singular value vectors of the noisy measurement matrix,damped data-driven optimal singular value shrinkage(DOptShrink)which combine the damping factor of DMSSA with OptShrink.The comparison of linear model and sigmod model experiments illustrate that the denoising results based on DOptShrink can achieve higher signal-to-noise ratio and amplitude preservation than the traditional MSSA method which based on TSVD.3.A new adaptive rank-selection algorithm is introduced to solve the tricky problem that the rank-reduction parameter is not easy to determine for the complicated seismic data.The new algorithm can select the rank-reduction parameter automatically by calculating the optimal hard threshold for singular values.Then,we use this new algorithm to improve the rank-reduction methods.The adaptive rank-reduction method is the adaptive damped data-driven optimal singular value shrinkage(ADOptShrink).The model and field data experiments illustrate that the proposed new method have much more robustness and adaptability than the fixed rank parameter methods.4.Considering that the coherent noise is similar to useful signal in the singular spectrum,which means the rank-reduction methods cannot achieve the goal of attenuating coherent noise through cutting off small singular values simply.For this reason,we propose a hybrid approach based on rank-reduction method and f-x empirical mode decomposition to simultaneously attenuate seismic random and coherent noise in stacked sections.Firstly,we use f-x EMD to remove both dip coherent noise and most random noise,because of the mode mixing of EMD,there will exist some residual random noise in the results of previous step.Then,we use ADOptshrink algorithm to remove the residual random noise automatically.The new approach can obtain higher signal-to-noise ratio compared to f-x emd and can remove coherent noise compared to rank-reduction methods.Synthetic and field examples demonstrate the superiority of the proposed approach.