基于深度AUC最大化算法的井漏风险预测  

Lost circulation prediction based on deep AUC maximization

作　　者：罗俊如 丁言瑞 徐明华[1] 胡超[1] 刘炳官[2] 孔维军 马强维 石林[1] LUO Junru;DING Yanrui;XU Minghua;HU Chao;LIU Bingguan;KONG Weijun;MA Qiangwei;SHI Lin(Aliyun School of Big Data,Changzhou University,Changzhou 213164,China;Jiangsu Oil Field Branch,China Petroleum and Chemical Corporation,Yangzhou 225009,China;Jiangsu Rutong Petro-Machinery Co.,Ltd.,Nantong 226400,China)

机构地区：[1]常州大学阿里云大数据学院,江苏常州213164 [2]中国石油化工股份有限公司江苏油田分公司,江苏扬州225009 [3]江苏如通石油机械股份有限公司,江苏南通226400

出　　处：《常州大学学报（自然科学版）》2024年第3期34-44,共11页Journal of Changzhou University：Natural Science Edition

基　　金：中国石油-常州大学创新联合体资助项目(2021DQ06);江苏省双创博士研究资助项目(JSSCBS20210885);常州大学阿里云大数据学院研究资助项目(ZMF21020012)。

摘　　要：井漏是石油天然气钻井过程中经常面临的一项重要挑战,它的发生会极大的降低钻井效率,增加钻井成本。运用人工智能技术实现井漏风险的精准预测具有重要意义。文章将钻井液泄露分类问题转化为不平衡分类问题。类别间的非均衡性和钻探特征间缺乏高度相关性,对传统的深度学习模型提出了挑战。在这种情况下,传统的准确率度量很难正确评估模型性能。此外,研究引入了一种称为FAUC-S的深度AUC最大化(DAM)算法实现井漏风险预测,该算法是通过关注困难样本的AUC损失来训练组合深度学习模型。实验中还应用了一些经典的深度学习模型实现井漏风险的分类。实验结果表明,与其他3个模型相比,FAUC-S获得了最高的精确度、召回率和F 1分数,实验验证了FAUC-S模型优越的分类性能。因此,该深度模型的成功应用可以有效地帮助钻井队解决钻井问题。Lost circulation is a significant challenge in oil and gas drilling,which can lead to various costly and time-consuming problems.It is of great significance to use artificial intelligence technology to accurately predict the risk of lost circulation.The lost circulation prediction problem was converted into an imbalanced classification problem,which pose challenges to traditional deep learning models due to the imbalance between categories and the lack of high correlation between drilling features.Accuracy is not an appropriate measurement for imbalanced classification algorithms.A deep AUC maximization(DAM)algorithm,which is called FAUC-S,is introduced in this paper.It trains a combination deep learning model by focusing on the AUC loss of hard samples(FAUC-S).Several traditional deep learning methods are also applied to classify lost circulation risk during oil exploration in the experiments.The result shows that the FAUC-S method achieved the highest accuracy,recall,and F 1 score among the other three models.This confirms that the FAUC-S model has superior classification performance.Therefore,the successful implementation of this deep model can help drilling teams effectively solve drilling problems.

关 键 词：井漏 非均衡分类 深度学习 AUC最大化 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]