基于集成机器学习的测井曲线大尺度差异超分辨  

Large-Scale Difference Super-Resolution of Logging Curves Based on Integrated Machine Learning

作　　者：曹志民[1,2] 丁璐 韩建[1,2] 郝乐川 Cao Zhimin;Ding Lu;Han Jian;Hao Lechuan(NEPU Sanya Offshore Oil&Gas Research Institute,Sanya 572024,Hainan,China;School of Physics and Electronic Engineering,Northeast Petroleum University,Daqing 163318,Heilongjiang,China;Postdoctoral Work Station,Daqing Oil Field,Daqing 163318,Heilongjiang,China;Postdoctoral Mobile Station,Northeast Petroleum University,Daqing 163318,Heilongjiang,China)

机构地区：[1]东北石油大学三亚海洋油气研究院,海南三亚572024 [2]东北石油大学物理与电子工程学院,黑龙江大庆163318 [3]大庆油田博士后工作站,黑龙江大庆163318 [4]东北石油大学博士后流动站,黑龙江大庆163318

出　　处：《吉林大学学报(地球科学版)》2025年第2期670-685,共16页Journal of Jilin University:Earth Science Edition

基　　金：海南省科技厅重点研发项目(ZDYF2022GXJS222,ZDYF2022GXJS220)。

摘　　要：精细储层描述一直是非常规油气资源开发和生产的重点,但常规测井曲线的纵向分辨率难以满足对厘米级甚至毫米级储层的有效识别。针对这一问题,本文以集成机器学习技术为核心,从多视多尺度的角度出发,提出了一种两级知识迁移的测井曲线大尺度差异超分辨方法提高测井曲线的纵向分辨率,实现低成本情况下的储层精细描述;选取地层反映较好的微球电阻率、自然伽马、声波时差曲线作为目标曲线,实现高分辨成像电阻率曲线信息到目标测井曲线映射模型的构建,进而实现目标测井曲线的大尺度差异超分辨,并将超分辨结果与不同超分辨方法进行对比。结果表明,本文方法得到的超分辨曲线与真实高分辨曲线相关系数大于0.9,与对比方法相比提高了3.6%~16.0%,均方误差、均方根误差、平均绝对误差、平均绝对百分比误差、对称平均绝对百分比误差分别降低了28.9%~90.8%、15.7%~69.8%、24.4%~74.7%、25.0%~74.2%、25.2%~77.4%。本文方法能够在一定程度上实现现有常规测井曲线的毫米级超分辨处理,得到的超分辨曲线能够大致地捕捉到地层的变化,降低了精细储层有效识别问题的难度。Fine reservoir description has always been the focus of development and production of unconventional oil and gas resources,but the vertical resolution of conventional logging curves is difficult to satisfy the effective identification of thin layers at centimeter or even millimeter scale.Aiming at this problem,this paper proposes a two-level knowledge migration super-resolution method for large-scale difference in logging curves to improve the vertical resolution of logging curves,thus realizing high-resolution target reservoir fine description in low-cost cases,using integrated machine learning as the basic tool and the perspective of multi-view and multi-scale as the core.The microsphere resistivity,natural gamma ray,and acoustic time difference curves with better formation response are selected as the target curves,and the construction of a mapping model from the information of high-resolution imaging resistivity curves to the target logging curves is realized,which in turn realizes the large-scale difference super-resolution of target logging curves,and the super-resolution results are compared with different super-resolution methods.The results show that the correlation coefficients between the super-resolution curves obtained by the method of this paper and the real high-resolution curves are greater than 0.9,which are improved by 3.6%to 16.0%compared with the comparison methods,and the mean square error,the root mean square error,the mean absolute error,the mean absolute percentage error,the symmetric mean absolute percentage error are reduced by 28.9%to 90.8%,15.7%to 69.8%,24.4%to 74.7%,25.0%to 74.2%,and 25.2%to 77.4%,respectively.Therefore,the method of this paper is able to largely realize the millimeter-level super-resolution processing of the existing conventional logging curves,and the obtained super-resolution curves are able to roughly capture the formation changes,which alleviates the difficulty of the problem of effective identification of fine reservoirs.

关 键 词：精细储层描述 测井曲线 集成机器学习 大尺度 超分辨 

分 类 号：P631.8[天文地球—地质矿产勘探]