沉积相智能地震识别技术研究及应用

Research and application of intelligent seismic identification technology of sedimentary facies

通常利用地震反射同相轴的几何形状、横向连续性、振幅、频率和层速度等参数描述沉积相类型。传统的沉积相地震识别方法利用人工解释目标沉积体,受限于解释人员对目标工区的地质认识,工作强度大、效率低,存在多解性。在很多地震勘探工区只能够标定少量标签,这些标签不足以支撑完成强监督学习,此时小样本学习可以作为好的解决方案。为此,以小样本学习为切入点,主要研究碳酸盐岩丘滩体沉积相预测方法,探讨了基于小样本学习方法在沉积相识别方面的应用效果。在小样本弱监督学习方面,首先根据地震反射构型以及钻井信息建立了典型地震剖面地震相标签库,在川中地区灯影组地震数据中共解释了14条地震剖面作为训练标签,占总数的2.8%。其次,基于层序地层格架控制的沉积相智能分类方法,利用地震层序格架构建隐式标量场,引入地震层位的空间变化信息,避免了深度学习在地震相预测过程中未引入地质信息的缺陷,提高了沉积相智能预测精度。利用所提方法刻画了四川盆地川中地区灯影组碳酸盐岩微生物丘滩体,结果表明:礁体受控于台地边缘斜坡,对礁体边界等的预测结果与地震信息高度吻合,符合地质规律;在平面上,礁体呈条带状分布,与台地边缘的分布基本吻合,与沉积相一致。

Generally,sedimentary facies types are described by seismic parameters such as geometry,lateral continuity,amplitude,frequency,and interval velocity.Limited by the interpreters'geological understanding of the target area,the traditional seismic identification method of sedimentary facies using manual interpretation has high work intensity,low efficiency,and multiple solutions.Only few labels can be made in many seismic survey regions,which cannot support the completion of strongly supervised learning.At this time,few-shot learning can be a good solution.In this work,the application effect of few-shot learning in sedimentary facies identification is discussed mainly through the research on the prediction me-thod of carbonate mound-shoal complexes.In terms of weakly supervised learning,a label library for seismic facies of typical seismic profiles is constructed according to seismic reflection configurations and drilling information.A total of 14seismic profiles are interpreted as training labels in seismic data of Dengying Formation in the central Sichuan Basin,accounting for 2.8%of the total.Then,the intelligent identification method of sedimentary facies controlled by the sequence stratigraphic framework is employed.The implicit scalar field is constructed with the seismic sequence framework,and the spatial change information of seismic horizons is introduced,which avoids the lack of geological information in the prediction of seismic facies and improves the accuracy of intelligent prediction of sedimentary facies.The pro posed method is used to depict the carbonate microbial mound-shoal complexes of Dengying Formation in the central Sichuan Basin.The results show that the organic reef is controlled by the platform margin slope,and the prediction results of reef boundaries are highly consistent with seismic information and conform to geologic laws.On the plane,the reefs are distributed in strips,which is consistent with platform margin distribution and sedimentary facies.