论文详情
相比传统的声波逆时偏移方法,弹性波逆时偏移(ERTM)可以提供更多地下结构的物理信息,然而其偏移中的串扰噪声及各种非物理噪声严重降低了成像质量。为了获得更高分辨率的地震成像,介绍了一种在ERTM生成的倾角域成像道集上使用卷积神经网络(CNN)估计叠加孔径,实现稳相叠加从而压制偏移噪声的方法。该方法通过在倾角域识别主要的反射波能量,剔除对成像贡献不大的部分,从而压制了偏移中的各种噪声,提高了ERTM成像质量。CNN是一个端到端的深度学习过程,一旦网络经过训练得到适合权系数和偏置,可以替代人工实现自动拾取。BGP盐丘模型数据和SEG起伏地表模型数据测试结果表明:利用CNN实现自动拾取的算法在只对少量道集拾取并作为标签数据,对神经网络训练后,可较好实现海量道集的自动拾取。基于CNN的倾角域弹性波逆时偏移噪声压制方法效果好、效率高。
The elastic reverse time migration (ERTM) technique has become increasingly common thanks to the development of high-performance computing.ERTM can provide more physical information on subsurface structures compared with conventional methods;however,its imaging resolution is sensitive to migration artifacts.To obtain a high-resolution seismic image through ERTM,additional constraints are introduced by using a convolutional neural network (CNN) in the dip-angle domain.The beneficial effect of the proposed CNN-ERTM method relies on the identification of dominant reflection events and the concomitant rejection of migration artifacts.In particular,a CNN-based end-to-end deep learning process is implemented into the automatic aperture function which is used in the picking process.The proposed method consists of three steps.First,the dip-angle gathers for the ERTM are generated using Poynting vectors shot by shot.Then,all the dip angle gathers are stacked over all the shots.Finally,a CNN is employed to predict the constraint function in the dip-angle domain.The CNN-based picking is an automatic process which does not require human intervention once the network is well-trained.Tests on the BGP salt model and SEG foothill model dataset demonstrated that the developed method could achieve successful training and automatic picking by using only a small part of the dip-angle gathers.
基金项目：自然科学基金(41804129)与中央级公益性科研院所基本科研业务费项目(DZLXJK202006)共同资助。