结合LSTM自编码器与集成学习的井漏智能识别方法  

An Intelligent Lost Circulation Recognition Method Using LSTM-Autoencoder and Ensemble Learning

作　　者：孙伟峰[1] 冯剑寒 张德志[2] 李威桦 刘凯[2] 戴永寿[1] SUN Weifeng;FENG Jianhan;ZHANG Dezhi;LI Weihua;LIU Kai;DAI Yongshou(College of Oceanography and Space Informatics,China University of Petroleum(East China),Qingdao,Shandong,266580,China;College of Control Science and Engineering,China University of Petroleum(East China),Qingdao,Shandong,266580,China)

机构地区：[1]中国石油大学(华东)海洋与空间信息学院,山东青岛266580 [2]中国石油大学(华东)控制科学与工程学院,山东青岛266580

出　　处：《石油钻探技术》2024年第3期61-67,共7页Petroleum Drilling Techniques

基　　金：国家自然科学基金项目“基于深度学习的深地叠前时空域地震子波提取方法研究”(编号:42274159)资助。

摘　　要：为了解决传统的井漏智能识别模型因井漏样本数量受限导致其识别准确率低的问题,提出了一种长短期记忆(long short-term memory,LSTM)网络与自编码器(auto-encoder,AE)相结合、集成LSTM-AE的井漏智能识别方法。首先,采用正常样本训练多个包含不同隐藏层神经元数目的LSTM-AE模型,利用重构得分筛选出识别效果较好的几个模型作为基识别器;然后,采用集成学习对多个基识别器的识别结果进行加权融合,解决单一模型因对样本局部特征过度学习导致的误报与漏报问题,提高模型的识别准确率。从某油田18口井的钻井数据中选取了6000组正常钻进状态下的立压、出口流量、池体积数据,对集成LSTM-AE模型进行训练和测试,结果表明,提出方法的识别准确率达到了94.7%,优于其他常用的智能模型的识别结果,为井漏识别提供了一种新的技术途径。To enhance the low recognition accuracy of traditional intelligent lost circulation models,which suffer from limited samples,this study combined the long short-term memory(LSTM)network and auto-encoder(AE)to create an integrated LSTM-AE-based intelligent lost circulation recognition model.Initially,multiple LSTM-AE models with varying numbers of hidden neurons were trained using normal samples.Several models with better recognition performance were selected as base recognizers based on their reconstruction scores.Subsequently,the recognition results from these base recognizers were fused using ensemble learning.This approach addresses the tendency of a single model to produce false alarms and missed alarms due to overlearning of local sample characteristics,thereby improving the recognition accuracy of the model.The integrated LSTM-AE model was trained and tested using 6000 sets of stand pipe pressure,outlet flow,and mud pit volume data from 18 wells under normal drilling conditions in an oilfield.The results show that the proposed method achieves a recognition accuracy of 94.7%,surpassing the recognition results of other commonly used intelligent models.This approach offers a novel method for lost circulation recognition.

关 键 词：井漏识别 长短期记忆网络 自编码器 集成学习 

分 类 号：TE28[石油与天然气工程—油气井工程]