摘要
随着勘探目标的构造和地表地质条件的日趋复杂,地震数据经常存在不规则和不完整的问题,给后续的处理带来严重困难。针对这一难题,文中提出了一种基于XGBoost算法的地震数据重建方法。该方法从局部学习的角度出发,针对随机缺失的地震道,在其周围选择一定数量的相邻地震道作为参考。通过构建这些参考地震道的道号、采样点号与数值之间的回归模型,能够精确学习并重建出缺失地震道的数据。为全面评估该方法的性能,对模拟数据不同地震道缺失情况下进行了实验,并与基于U-net卷积神经网络和基于凸集投影的Curvelet算法等重建方法进行比较。实验结果表明,基于XGBoost算法的重建方法对随机缺失地震数据重建具有较高的精度。实际数据处理结果表明,该方法能够为后续地震资料处理提供高精度的规则炮集数据。
With the increasing complexity of the structure and the surface geological conditions of the exploration target,the problems of irregular and incomplete data often occur in the process of seismic data acquisition,which brings serious difficulties to the follow-up data processing.To solve this problem,this paper proposes a seismic data reconstruction method based on the XGBoost algorithm.From the perspective of local learning,this method selects a certain number of adjacent seismic traces around the randomly missing seismic traces as a reference.By constructing the regression model between the trace numbers,sampling point numbers and their values of the reference seismic traces,the missing seismic trace data can be accurately learned and reconstructed.In order to fully evaluate the performance of the proposed method,the experiments are performed on simulated data with different missing seismic traces,and the reconstruction methods such as U-net convolutional neural network and Curvelet algorithm based on projection onto convex sets are compared.The experimental results show that the reconstruction method based on the XGBoost algorithm presents high accuracy in the reconstruction of randomly missing seismic data.The actual data processing results show that this method can provide high-precision regular shot gather for the follow-up seismic data processing.
作者
李山
田仁飞
刘涛
LI Shan;TIAN Renfei;LIU Tao(College of Geophysics,Chengdu University of Technology,Chengdu,Sichuan 610059,China;Hulunbuir Subsidiary of PetroChina Daqing Oilfield Co.Ltd.,Daqing,Heilongjiang 163712,China)
出处
《石油地球物理勘探》
EI
CSCD
北大核心
2024年第5期965-975,共11页
Oil Geophysical Prospecting
基金
国家自然科学基金项目“准噶尔盆地春光区块岩性油藏倒频域烃类检测方法研究”(41304080)资助。
