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

作　　者：白薷 王世玉 张璐[2] 张亮 杜炜 耿代 姚振杰 BAI Ru;WANG Shiyu;ZHANG Lu;ZHANG Liang;DU Wei;GENG Dai;YAO Zhenjie(Research Institute,Shaanxi Yanchang Petroleum(Group)Co.Ltd.,Xi’an,Shaanxi 710065,China;Xingzichuan Oil Production Plant,Yanchang Oilfield Co.Ltd.,Yan’an,Shaanxi 717400,China)

机构地区：[1]陕西延长石油(集团)有限责任公司研究院,陕西西安710065 [2]延长油田股份有限公司杏子川采油厂,陕西延安717400

出　　处：《天然气勘探与开发》2024年第4期63-71,共9页Natural Gas Exploration and Development

基　　金：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)。

摘　　要：针对目前计算机自动分层识别准确率低,人工分层效率不高的问题,提出了一种基于掩码自编码器(MAE)神经网络算法的地层分层识别新方法。通过测井曲线优选,以自然电位、自然伽马、声波时差、电阻率、井位坐标和补心海拔作为输入特征变量进行模型训练及预测,再利用损失函数、准确率、精确率、召回率、F1值进行模型性能评估。研究结果表明:(1)基于MAE神经网络训练得到地层分层识别模型,预测准确率能够达到95.54%;(2) MAE神经网络模型与卷积神经网络和关注分层边界的卷积神经网络模型进行地层分层实验对比,MAE神经网络模型的性能和预测精度均较高,准确率分别提高了8.30%和6.32%,且无地层紊乱情况出现,具有明显分层优势;(3) MAE神经网络模型应用于未知井的地层分层中,自动分层预测准确率为98.07%。表明该方法具有较高的地层识别效果,为油田地层识别提供了一种基于自监督神经网络算法的理论支持和有益探索。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 

分 类 号：TE151[石油与天然气工程—油气勘探] TP183[自动化与计算机技术—控制理论与控制工程]