基于数据增广和伪标签生成的波阻抗反演

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. 20240387

Abstract：Wave impedance inversion based on deep learning often requires a large number of label data to drive the model for network training，but in practice，the acquisition of label data（logging data）is difficult and costly，and usually only a small amount of label data is used for training. Therefore，a semi⁃supervised wave im ⁃ pedance inversion method based on data augmentation and pseudo ⁃label generation is proposed. Firstly，the la⁃ bel wave impedance data is interpolated by cubic spline interpolation method and then randomly resampled，and then the seismic records corresponding to the augmented wave impedance are calculated by forward modeling method. The amplified seismic record and wave impedance are used as the network training set to train the net⁃ work and predict the wave impedance. High quality prediction data is selected as pseudo⁃labels，and the pseudo⁃ labels are augmented to greatly expand the training data set. However，Temporal Convolutional Network （TCN）has advantages in time series data modeling，which can capture the long ⁃ term dependency of data and complete the wave impedance inversion task well. In this paper，the Marmousi model test results show that the proposed method is suitable for the wave impedance inversion of a small amount of label data，has good anti ⁃ noise performance，and still has good inversion accuracy for different label distributions. The inversion results of the actual exploration data show that this method can effectively solve the problem of seismic impedance in version.

Keywords：wave impedance inversion，temporal convolutional networks，data augmentation，pseudo⁃label，semi⁃ supervised learning

程程，赵岩 . 基于数据增广和伪标签生成的波阻抗反演［J］. 石油地球物理勘探，2025，60（3）：642‑654.

CHENG Cheng，ZHAO Yan. Wave impedance inversionbased on data augmentation and Pseudo⁃label genera⁃ tion［J］. Oil Geophysical Prospecting，2025，60（3）：642‑654.

0 引言

地震勘探是解决油气勘探问题广泛应用的技术方法之一[1]，地震勘探技术经历了从认识地下的构造形态到当前直接通过地震信息判断岩性、分析岩相、定量计算岩层的物性参数及寻找油气显示等[2-3]。（剩余17883字）