Scholarlink

<p>Lithology identification is essential for reservoir characterization and downhole accidents prevention in the field of drilling engineering. However, the complexity and variability of logging data represent significant challenges for accurate lithology identification. In this paper, an intelligent lithology identification model integrating a bidirectional gated recurrent unit (Bi-GRU) and multi-head attention mechanism is proposed. This model includes multi-source data standardization and imbalance handling through a cost-sensitive approach, feature extraction using correlation analysis and random forest feature selection, and lithology identification incorporating Bi-GRU, multi-head attention, and the RMSprop optimizer. Input features consist of depth, natural gamma ray, neutron logging, porosity, acoustic time difference, shallow array induction, gray matter content, and density logging. The results demonstrate that the accuracy of the improved Bi-GRU model reached 92.39%, which was 3.71% higher than that of a model without data imbalance processing. Compared with 8 traditional machine learning methods such as random forest, backpropagation neural network, and long short-term memory neural network model, the generalization ability of this model increased by 12.84–39.59%, indicating strong generalization capability.</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

Attention-Based Bidirectional Gated Recurrent Unit Neural Networks for Lithology Identification from Well-Logging Data

  • 期刊论文

  • 发表日期：

    2026年1月26日

查看全文
  • Xiaohui Sun,
  • Lina Zhang,
  • Jintang Wang,
  • Zhiyuan Wang,
  • Baojiang Sun

摘要

Lithology identification is essential for reservoir characterization and downhole accidents prevention in the field of drilling engineering. However, the complexity and variability of logging data represent significant challenges for accurate lithology identification. In this paper, an intelligent lithology identification model integrating a bidirectional gated recurrent unit (Bi-GRU) and multi-head attention mechanism is proposed. This model includes multi-source data standardization and imbalance handling through a cost-sensitive approach, feature extraction using correlation analysis and random forest feature selection, and lithology identification incorporating Bi-GRU, multi-head attention, and the RMSprop optimizer. Input features consist of depth, natural gamma ray, neutron logging, porosity, acoustic time difference, shallow array induction, gray matter content, and density logging. The results demonstrate that the accuracy of the improved Bi-GRU model reached 92.39%, which was 3.71% higher than that of a model without data imbalance processing. Compared with 8 traditional machine learning methods such as random forest, backpropagation neural network, and long short-term memory neural network model, the generalization ability of this model increased by 12.84–39.59%, indicating strong generalization capability.