SVD与EMD联合去噪方法在地震勘探数据处理中的研究与应用

将基于倾角扫描的奇异值分解与经验模式分解法相结合应用到地震资料随机噪声压制中。首先利用经验模式分解法消除部分噪声,增强地震道有效信号的相关性,再利用奇异值分解对地震信号进行同相轴自动追踪,截取小时窗数据体,并进行同相轴拉平处理,经SVD计算小时窗数据中心点的值来代替计算样点的值,最终实现随机噪声的压制。理论模型试算和实际资料处理表明,本文提出的EMD-SVD方法简单易行,比单一的SVD方法去噪效果更显著有效地消除了地震资料中的随机噪声,提高了地震资料的信噪比,并改善了叠加剖面的质量。

Application of Joint Denoising Using Empirical Mode Decomposition and Singular Value Decomposition in Seismic Data Processing

Random noise widely exists in seismic data, either in prestack or poststack data. The dip-scanning singular value decomposition (SVD) algorithm has been proven to be very effective for eliminating seismic data noise, especially for data with complex deep structures. However, limited volumes of data, especially data with strong noise, in a small window cannot completely reflect the strong correlation among traces. Therefore, the dip-scanning SVD application results are severely constrained. By examining factors that restrict the full utilization of SVD, we developed a new joint denoising approach that uses empirical mode decomposition (EMD) and dip-scanning SVD to eliminate random noise in seismic data. First, this method uses EMD to reconstruct a signal to both reduce noise variance and enhance the correlation of effective signals among traces. Second, it automatically tracks seismic events with dip-scanning SVD to solve the singular value selection problem. Finally, it intercepts small data volumes, flattens an event, and identifies noise points so that dip-scanning SVD can be used on horizontal events to effectively and efficiently eliminate noise. Through the development of a theoretical model and real data application, we prove that the EMD-SVD joint denoising method is a more efficient algorithm when compared with conventional dip-scanning SVD. Simulated and field data results show that the EMD-SVD method can effectively eliminate random noise and significantly increase the signal-to-noise ratio of seismic data, thereby significantly improving the quality of a stack section. For this purpose, proper noise-identifying threshold values should be set according to the features of real seismic data. Moreover, the direction parameter applied by dip-scanning SVD may be modified depending on the dip angle of events. Seismic data random noise can be efficiently and automatically eliminated with relatively short window lengths and fewer constrained conditions of this approach.