无监督地层倾角智能计算方法及应用效果

打开文本图片集

中图分类号：P631 文献标识码：A DOI：10. 13810/j. cnki. issn. 1000⁃7210. 20240337

Abstract：In seismic geometric attributes，stratigraphic dip is the basis for calculating attributes such as curva⁃ ture and coherence，and has been widely applied in seismic data interpretation. However，traditional multi⁃win⁃ dow scanning algorithms are inefficient，and the existing intelligent algorithms based on end ⁃ to ⁃ end supervised training are constrained in their generalization and transferability due to the diversity of seismic data. There⁃ fore，this paper proposes an unsupervised training method for intelligent dip calculation using deep neural net⁃ works. This method is based on a three⁃dimensional convolutional neural network（3D CNN）and achieves un⁃ supervised optimization of the deep neural network by establishing and solving an optimization objective for the structure tensor. It does not require the prior creation of a large number of labels，and combined with transfer learning and fine⁃tuning for actual work area，achieves efficient and stable 3D dip angle calculation based on the efficient computation of seismic feature vectors. Extensive applications on models and actual data have shown that this intelligent method significantly improves computational efficiency while maintaining stable computation results. Specifically，the geometric curvature obtained based on intelligent dip calculation results exhibits more advantages in expressing fracture information.

Keywords：deep neural network（DNN），unsupervised learning，seismic feature vector，stratigraphic dip郭锐，文若冲，梁琰，等 . ［J］. 石油地球物理勘探，2025，60（3）：598‑605.GUO Rui，WEN Ruochong，LIANG Yan，et al. Unsupervised intelligent stratigraphic dip calculation and itsapplication effects［J］. Oil Geophysical Prospecting，2025，60（3）：598‑605.

0 引言

地层倾角被地震资料处理解释流程中广泛使用[1]，它可用于解释地震反射的同相性特征，进而成功应用于多项地球物理任务中。（剩余10450字）