基于伪地震数据模式学习的多次波自适应相减方法

Adaptive multiple subtraction based on pattern-learning method using pseudo-seismic data

相比于传统匹配相减方法,模式学习方法将多次波衰减分成多次波模式学习和自适应相减的两个独立过程,因而能够更好地保护一次波。然而,由于多次褶积的影响,预测多次波模型存在相位、频带、振幅误差。为了在不损伤一次波的基础上,尽可能减少多次波残留,提出了基于伪地震数据的模式学习方法。利用预测地震道,得到其Hilbert变换道以及预测道和Hilbert变换道的一阶导数,作为伪地震数据的4个分量,补充预测多次波模型中相位和高频信息。模式学习阶段利用主成分分析法分别学习伪地震数据4个分量的多次波字典矩阵,然后按列拼接得到联合字典矩阵;自适应相减阶段基于学习到的联合字典矩阵,利用裂步Bregman算法从原始地震数据中重构多次波,实现一次波和多次波的分离。模拟数据和实际资料处理结果表明,该方法在保护一次波的同时,能有效压制多次波,模型数据的一次波信噪比提升了1dB。

Diffracted multiples are a major problem in which zones are characterized by steep slopes and rugged seafloors.It is difficult to predict out-of-plane diffracted multiples via SRME owing to insufficient near-offset and negative-offset information,which results in phase,frequency,and amplitude errors in the predicted multiples.To suppress multiples without damaging the primaries,we propose an alternative method that applies pattern learning to pseudo-seismic data.Compared with traditional matching filter methods,pattern-learning methods suppress multiples via pattern coding and adaptive suppression,which better protects the primaries.Pseudo-seismic data,also known as mathematical adjoints of a predicted trace,include its first derivative,the Hilbert transform,and the derivative of the Hilbert transform,which provide additional phase and high-frequency information.Specifically,the proposed method was implemented in a two-stage process.First,in the pattern-coding stage,the pattern dictionary was obtained using principal component analysis for each component of the pseudo-seismic data,and the joint dictionary matrix consisted of all sets of key patterns from the pseudo-seismic data.Second,in the adaptive attenuation stage,the joint dictionary matrix from the pseudo-seismic data was utilized to estimate the multiples in the recorded data to separate the primaries and multiples,using the split Bregman algorithm.The proposed method,on the one hand,inherits the advantages of the pattern-learning method and protects the primaries better.On the other hand,additional phase and high-frequency information from the pseudo-seismic data can be utilized to suppress the residual diffracted multiples.We validated this approach using synthetic and field datasets.Our method yielded an SNR improvement of 1dB for synthetic data.In addition,for the field data,the residual diffracted multiples were substantially reduced using the proposed method.