基于分频智能反演的曲流河点坝与废弃河道识别

曲流河点坝、废弃河道级次构型表征对丰富曲流河储层构型模式、指导油气田高效开发具有重要意义。为此，以渤海湾盆地孤岛油田中12—斜检3011井区馆陶组为例，基于分频智能反演技术识别曲流河点坝与废弃河道。首先，通过地震资料分频处理，依据振幅与砂体厚度之间的关系优选最佳频段地震资料，采用支持向量回归(SVR)的机器学习算法进行分频反演；其次，在利用反演数据体平面属性刻画复合河道砂体分布规律的基础上，根据河道边界的地震、测井等响应特征预测单一曲流带；最后，以废弃河道泥质半充填的样式为指导，选取目的层上、中、下部的反演属性切片进行RGB融合，建立废弃河道识别模板，并在定量模式约束下识别点坝和废弃河道。研究结果表明：(1)基于机器学习的分频反演技术能够充分利用不同频段地震信息与测井信息，提高了反演结果的分辨率，可指导河道边界识别；(2)采用RGB融合技术融合河道不同位置的反演属性切片，能够辅助判别砂体之间的空间组合关系，有助于井间废弃河道识别；(3)在地震资料主频为38 Hz的情况下，利用基于分频智能反演的曲流河点坝与废弃河道识别技术在研究区目的层复合曲流带中共识别了4个单一曲流带、13个废弃河道和...

Identification of point bar and abandoned channel of meandering river by spectral decomposition inversion based on machine learning

Characterizing the hierarchical architecture of point bars and abandoned channels of a meandering river is of great significance for enriching the architecture pattern of the meandering river reservoir and guiding the efficient development of oil and gas fields. Therefore, the Guantao Formation of Zhong12-Xiejian 3011 well block in Gudao Oilfield in Bohai Bay Basin is taken as an example, and the point bars and abandoned channels of the meandering river are identified based on spectral decomposition inversion technology. Firstly, with the seismic data processed by spectral decomposition, the optimal seismic data of frequency bands are selected according to the relationship between the amplitude and the thickness of the sand body, and the machine learning algorithm of support vector regression (SVR) is used for spectral decomposition inversion. Then, on the basis of describing the distribution law of sand bodies in composite channels by the plane attribute of inversed data volume, the single meander belt is predicted according to the seismic, well logging, and other response characteristics of channel boundaries. Finally, guided by the muddy semi-filling pattern of abandoned channels, this paper selects the inversion attribute slices of the upper, middle, and lower parts of the target layer for RGB fusion, so as to establish the identification pattern of abandoned channels and identify point bars and abandoned channels under the constraints of the quantitative model. The results show that:① The spectral decomposition inversion technology based on machine learning can make full use of seismic information and logging information of different frequency bands, which improves the resolution of inversion results and can guide channel boundary identification; ② RGB fusion technology is used to fuse the inversion attribute slices at different positions of the channel, which can help recognize the spatial combination relationship between sand bodies and identify inter-well abandoned channels; ③ Based on the seismic data with a dominant frequency of 38 Hz, four single meander belts, 13 abandoned channels, and 15 point bars are identified in the composite meander belts of the target layer by using the identification technology of point bars and abandoned channels of meandering river based on spectral decomposition inversion.The injection well data verifies the accuracy of this method, which has a good application prospect.