| 注册
首页|期刊导航|天然气勘探与开发|基于Attention+Bi-LSTM神经网络算法的页岩气井关键测井曲线补全方法

基于Attention+Bi-LSTM神经网络算法的页岩气井关键测井曲线补全方法

吴磊 黄小惠 姜巍 周一博 吴林 徐云贵 曹国威

天然气勘探与开发2025,Vol.48Issue(3):45-53,9.
天然气勘探与开发2025,Vol.48Issue(3):45-53,9.DOI:10.12055/gaskk.issn.1673-3177.2025.03.005

基于Attention+Bi-LSTM神经网络算法的页岩气井关键测井曲线补全方法

A completing method for key logging curves in shale gas wells based on Attention+Bi-LSTM neural network algorithm

吴磊 1黄小惠 1姜巍 1周一博 1吴林 2徐云贵 3曹国威3

作者信息

  • 1. 中国石油集团川庆钻探工程有限公司页岩气项目经理部 四川 成都 610051
  • 2. 斯伦贝谢中国有限公司 北京 100015
  • 3. 西南石油大学地球科学与技术学院 四川 成都 610500
  • 折叠

摘要

Abstract

Both hole collapse and complexities easily happen in shale gas drilling due to exclusive bedding structure in shale,further to bring about incomplete logging data,like density logs.Therefore,to accurately predict logs just serves as one way to improve the prediction accuracy for shale gas reservoirs especially in certain areas with complex well conditions or with poor logging quality.Adopted some quality training samples from logging data in appraisal wells in a shale gas block of Sichuan Basin,a bi-directional long short-term memory(Bi-LSTM)neutral network model based on attention mechanism(Attention+Bi-LSTM model),which the existing density logs were conducted as label samples for deep learning,was built and trained to complete logging data.Moreover,conventional logs of non-radioactive sources in well correlation with density logs were selected,and attention mechanisms were incorporated into the Attention+Bi-LSTM model to reinforce characteristic learning.Results show that the density data predicted from acoustic,natural gamma ray,gamma ray without uranium and resistivity logging,and uranium element content,have an average correlation coefficient with actual density up to 0.94,also an increase of 13.6%than before.It is concluded that,with better flexibility in logs prediction in shale gas blocks,the Attention+Bi-LSTM model can effectively complete logging data,reduce radioactive-source risks as needed,and save time and costs.This built method is worth popularizing.

关键词

复杂页岩气井区/测井数据缺失/测井曲线补全/双向长短期记忆/神经网络模型/注意力机制/密度曲线

Key words

Complex shale gas well area/Incomplete logging data/Logging-curve completion/Bi-LSTM/Neural network model/Attention mechanism/Density log

引用本文复制引用

吴磊,黄小惠,姜巍,周一博,吴林,徐云贵,曹国威..基于Attention+Bi-LSTM神经网络算法的页岩气井关键测井曲线补全方法[J].天然气勘探与开发,2025,48(3):45-53,9.

基金项目

中国石油川庆钻探工程有限公司重点攻关项目(编号:CQ2024B-39-1-1). (编号:CQ2024B-39-1-1)

天然气勘探与开发

1673-3177

访问量0
|
下载量0
段落导航相关论文