天然气勘探与开发2024，Vol.47Issue(4)：63-71,9.DOI:10.12055/gaskk.issn.1673-3177.2024.04.007
基于MAE神经网络的测井曲线地层自动识别方法
An automatic identifying method for strata via logging curves based on MAE neural network
摘要
Abstract
This paper presents a new method to automatically identify strata based on masked autoencoder(MAE)neural network in an effort to address two problems of low accuracy in computerized automatic identification and poor efficiency in artificial identifica-tion.As characteristic variables,the spontaneous potential,natural gamma,acoustic slowness,resistivity,well coordinates,and kelly bushing are input into the trained simulation which is predicted by optimizing logging curves.Furthermore,all simulation performance is evaluated using loss function,accuracy and precision,as well as recall and F1 values.It is found that,this trained simulation makes the prediction accuracy up to 95.54%.Some experimental contrasts demonstrate that,with the increasing accuracy by 8.3%and 6.32%,respectively,the simulation is better in its performance and prediction accuracy than those of both convolutional neural network(CNN)and boundary-guided CNN.Without any stratigraphic disturbance,this new method enjoys an obvious cutting edge in stratigraphic identification.In addition,after its application to unknown wells,the accuracy achieved 98.07%,showing a better effect.In general,this method may provide theoretical support and helpful discussion for stratigraphic identification based on self-supervised neural net-work algorithms.关键词
掩码自编码器/自监督/神经网络/测井曲线/地层识别Key words
MAE/Self-supervised/Neural network/Logging curve/Stratigraphic identification引用本文复制引用
白薷,王世玉,张璐,张亮,杜炜,耿代,姚振杰..基于MAE神经网络的测井曲线地层自动识别方法[J].天然气勘探与开发,2024,47(4):63-71,9.基金项目
国家重点研发计划项目(编号:2022YFE0206700)、陕西延长石油(集团)有限责任公司科技项目(编号:ycsy-2022jcts-B-45,ycsy2022jcts-B-47). （编号:2022YFE0206700）
