Scholarlink

<p>Prestack Amplitude variation with offset (AVO) inversion enables the quantitative identification of reservoirs and geological structures by analyzing variations in different elastic parameters. However, AVO inversion faces numerous challenges, including nonlinearity and non-unique solutions. To identify oil and gas reservoir with higher accuracy, this study first derives an AVO approximation that involves only bulk modulus, shear modulus, and density to calculate the reflection coefficients. Then, the study introduces two deep-learning-based AVO inversion methods: one based on physics-constrained neural networks (PCNN) and the other based on automatic differentiation. By incorporating the AVO equation into the inversion as a physical constraint, the inversion adheres to geophysical principles. These methods significantly improve inversion accuracy and provide more reliable support for reservoir prediction. Compared to traditional AVO inversion methods, the proposed deep-learning-based AVO inversion methods both demonstrate better inversion performance. Moreover, the PCNN-based inversion method demonstrates superior inversion capabilities, offering distinct advantages in seismic inversion.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Prestack AVO Inversion Based On Physics-constrained Deep Learning Method

  • 期刊论文

  • 发表日期：

    2025年3月17日

查看全文
  • Silin Wang,
  • Cai Liu,
  • Chao Song,
  • Pengfei Zhao,
  • Zhiqi Guo

摘要

Prestack Amplitude variation with offset (AVO) inversion enables the quantitative identification of reservoirs and geological structures by analyzing variations in different elastic parameters. However, AVO inversion faces numerous challenges, including nonlinearity and non-unique solutions. To identify oil and gas reservoir with higher accuracy, this study first derives an AVO approximation that involves only bulk modulus, shear modulus, and density to calculate the reflection coefficients. Then, the study introduces two deep-learning-based AVO inversion methods: one based on physics-constrained neural networks (PCNN) and the other based on automatic differentiation. By incorporating the AVO equation into the inversion as a physical constraint, the inversion adheres to geophysical principles. These methods significantly improve inversion accuracy and provide more reliable support for reservoir prediction. Compared to traditional AVO inversion methods, the proposed deep-learning-based AVO inversion methods both demonstrate better inversion performance. Moreover, the PCNN-based inversion method demonstrates superior inversion capabilities, offering distinct advantages in seismic inversion.