Acta Scientiarum Naturalium Universitatis Pekinensis ›› 2024, Vol. 60 ›› Issue (3): 453-463.DOI: 10.13209/j.0479-8023.2024.031

Previous Articles     Next Articles

Prestack Surface Multiple Suppression Method Based on Matching Algorithm with Unsupervised Neural Network

LIU Lichao1, HU Tianyue1,†, LI Xixi1, LIU Yimou2, LIANG Shanglin3, HUANG Jiandong4   

  1. 1. School of Earth and Space Sciences, Peking University, Beijing 100871 2. Petroleum and New Resources Company, PetroChina, Beijing 100011 3. Research Institute of Petroleum Exploration and Development, Beijing 100083 4. Department of Mathematical Sciences, Tsinghua University, Beijing 100084
  • Received:2023-06-12 Revised:2023-09-15 Online:2024-05-20 Published:2024-05-20
  • Contact: HU Tianyue, E-mail: tianyue(at)pku.edu.cn

基于无监督神经网络匹配算法的叠前表面多次波压制方法

刘立超1, 胡天跃1,†, 李徯徯1, 刘依谋2, 梁上林3, 黄建东4   

  1. 1. 北京大学地球与空间科学学院, 北京 100871 2. 中国石油油气和新能源分公司, 北京 100011 3. 中国石油勘探开发研究院, 北京 100083 4. 清华大学数学科学系, 北京 100084
  • 通讯作者: 胡天跃, E-mail: tianyue(at)pku.edu.cn
  • 基金资助:
    国家自然科学基金(42274163)、国家重点研发计划(2018YFA0702503)和中国石油天然气集团有限公司–北京大学基础研究项目资助

PDF

Abstract:

To effectively suppress surface multiples in marine seismic data and then correctly image the exploration target, a prestack surface multiple attenuation algorithm based on unsupervised neural networks is proposed, which combines neural network methods with surface-related multiple elimination (SRME) methods. By continuously decreasing the learning rate, the unsupervised neural network replaces the matching filter operator for suppressing surface multiples in the prestack seismic data. This method requires neither traditional matching algorithms nor training on labeled datasets. The application of simple synthetic data, Sigsbee model data and field data verifies the effectiveness of the proposed method for surface multiple wave suppression.

Key words: unsupervised neural networks, surface multiple suppression, prestack seismic data, matching algorithm

摘要：

为了有效地压制海上地震勘探数据中的表面多次波, 实现勘探目标的正确成像, 提出一种基于无监督神经网络的叠前表面多次波匹配算法, 将神经网络方法与表面相关多次波压制方法相结合, 通过设定学习率不断下降, 用无监督神经网络取代匹配滤波算子, 对叠前地震数据进行表面多次波的压制, 既不需要传统的匹配算法, 也不需要在标签数据集上进行训练。在简单合成数据、Sigsbee模型数据和实际数据上的应用结果验证了该方法对表面多次波压制的有效性。

关键词: 无监督神经网络, 表面多次波压制, 叠前地震数据, 匹配算法

Copyright © Acta Scientiarum Naturalium Universitatis Pekinensis, All Rights Reserved.
Powered by Beijing Magtech Co., Ltd.