Title

Centralized Wiener filtering for high-dimensional seismic data

单位

(1.波现象与智能反演成像研究组(WPI),同济大学海洋与地球科学学院,上海200092;2.中国石油化工股份有限公司胜利油田分公司物探研究院,山东东营257000;3.中国石油化工股份有限公司中原油田分公司物探研究院,河南濮阳457001)

Organization

(1.Wave Phenomena and Intelligent Inversion Imaging Group (WPI),School of Ocean and Earth Science,Tongji University,Shanghai 200092,China;2.Research Institute of Geophysics of Sinopec Shengli Oilfield,Dongying 257000,China;3.Research Institute of Geophysics of Sinopec Zhongyuan Oilfield,Puyang 457001,China)

摘要

两宽一高”地震勘探面向更为复杂的油藏描述,反演成像成了必要的技术手段,但数据中的噪声会严重影响地震波反演成像的质量,因此针对“两宽一高”地震数据研发有效且实用的去噪方法十分必要。在Bayes估计理论框架下,当信号满足线性(可预测性)假设且噪声满足高斯分布时,在均方误差或最小平方误差最小准则下估计最佳预测滤波器可进行有效滤波。同时随着地震数据维度的升高,高维信号的空间结构信息更丰富,因此在高维数据空间设计滤波器能够更有效地提高滤波器的信号预测能力,并且更好地压制噪声。首先将基于AR(自回归)模型的f-x域预测滤波器(前向预测滤波器和后向预测滤波器)修改为Wiener中心预测滤波器;然后在最小二乘意义下对高维地震数据进行多级Hankel矩阵排布构建法方程;再求解法方程得到Wiener中心预测滤波器,最后实现高维地震数据的Wiener中心预测滤波。为满足高维数据的局部线性假设,对复杂波场的地震数据,采用局部取窗的方式进行Wiener中心预测滤波去噪。合成理论数据和实际数据的测试结果说明该方法能够在地震数据中存在弱线性同相轴、振幅值缓慢变化以及信噪比相对较低的情况下较好地压制随机噪声,提高数据的信噪比,故该方法具有较强的实用性。

Abstract

The seismic exploration technology based on broadband,wide-azimuth and high-density (BWH) seismic data aims to describe more complex reservoirs,for which inversion imaging becomes a necessary technique.However,the noise in the filed data can seriously affect the quality of the seismic wave inversion imaging,therefore developing an effective and practical denoising method becomes necessary for the correct use of BWH seismic data.Within the framework of the Bayesian estimation theory,if the signal satisfies the linear hypothesis (signal predictability),and the noise satisfies a Gaussian distribution,the optimal prediction filter can be estimated by minimizing the mean square error or the second norm sense.With the dimension of seismic data increasing,the information carried by the spatial structure of the high-dimensional signal is richer,thus the design of filters for high-dimensional data spaces can improve the ability of predicting the signal and remove the noise more effectively.In this paper,the f-x predictive filter (forward predictive filter and backward predictive filter) based on an AR model is replaced with centralized Wiener predictive filter.The centralized Wiener predictive filtering is realized by solving the normal equation,which is constructed by a multi-level block Hankel matrix arrangement of high dimensional seismic data in the least squares sense.For seismic data with complex wavefields,in order to satisfy the local linear hypothesis of the high dimensional signal,the centralized Wiener predictive filtering denoising is carried out by setting partial spatial windows.Results obtained using both synthetic data and field data showed that the method can be effectively applied to suppress the random noise of seismic data with weak linear events,slow amplitude change,and relatively low SNR.

关键词：

Bayes估计; AR模型; Wiener中心滤波; Hankel矩阵; 随机噪声; 高维地震数据;

Keywords:

Bayes estimation,; AR model,; centralized Wiener filtering,; Hankel Matrix,; random noise,; high-dimensional seismic data;

基金项目

国家自然科学基金(41774126)和国家科技重大专项(2016ZX05024-001,2016ZX05006-002)共同资助。

DOI

10.3969/j.issn.1000-1441.2019.03.002

高维地震数据Wiener中心滤波方法