基于数据驱动模型加迁移学习的油田注水管网泄漏诊断方法  

The Method for Diagnosing Leakage in Water Injection Pipeline Networks Based on Data-driven Model Plus Transfer Learning

作　　者：刘书张 张艳[1,3] 申建非 陈冠男 任永良 张勇[3] 张新成 LIU Shu-zhang;ZHANG Yan;SHEN Jian-fei;CHEN Guan-nan;REN Yong-liang;ZHANG Yong;ZHANG Xin-cheng(School of Mechanical Science and Engineering,Northeast Petroleum University;Guangyuan China Nuclear Vocational and Technical College;Intelligent Manufacturing Institute,Taizhou University;Daqing Oilfield Water and Environmental Protection Company)

机构地区：[1]东北石油大学机械科学与工程学院 [2]广元中核职业技术学院 [3]台州学院智能制造学院 [4]大庆油田水务环保公司

出　　处：《化工机械》2024年第6期835-840,878,共7页Chemical Engineering & Machinery

基　　金：国家自然科学基金区域创新发展联合基金重点支持项目(批准号:U21A20104)资助的课题;大庆石油管理局项目(批准号:2022SWX191)资助的课题;台州市科技计划项目(批准号:23gyb10)资助的课题。

摘　　要：为解决油田注水管网泄漏诊断机器学习方法准确率不高的问题,基于数据驱动模型结合迁移学习的思路,提出一种油田注水管网泄漏诊断方法。研究结果表明:通过Epanet软件可在已知故障数据的基础上对泄漏故障进行模拟以实现数据增强;经过迁移学习的预训练和二次训练后,对数据驱动模型的准确率进行对比,5种模型中CNN卷积神经网络模型为最佳解决方案,其注水管网泄漏诊断准确率可达94.12%。Considering low accuracy of the machine learning methods in diagnosing the leaks in oilfield water injection pipelines,having the data-driven model plus transfer learning method based to propose a method for diagnosis of the leakage in water injection pipeline networks was implemented.The results show that,the Epanet software can simulate the leakage fault based on the known fault data to realize data en-hancement.After pre-training and secondary training of the transfer learning,comparing the accuracy of data-driven models indicates that,among the five models,the convolutional neural network(CNN)model is the best solution,and the accuracy of leakage diagnosis of water filling pipe network can reach 94.12%.

关 键 词：注水管网 泄漏诊断 数据驱动模型 迁移学习 卷积神经网络 数据增强 

分 类 号：TQ055.81[化学工程]