基于机器学习的复杂火山碎屑岩储层流体性质测井识别:以海拉尔盆地乌尔逊凹陷铜钵庙组为例  

Logging identification of complex pyroclastic reservoir fluid properties based on machine learning:taking Tongbomiao Formation of Wuerxun depression in Hailar Basin as an example

作　　者：赵小青 张哲清[1] 于继崇 李明慧 姜艳娇 ZHAO Xiaoqing;ZHANG Zheqing;YU Jichong;LI Minghui;JIANG Yanjiao(School of Geosciences,Northeast Petroleum University,Daqing 163318,Heilongjiang,China;Key Laboratory of Continental Shale Oil and Gas Accumulation and Efficient Development of Ministry of Education,Northeast Petroleum University,Daqing 163318,Heilongjiang,China;Daqing Branch,China Petroleum Logging Company Limited,Daqing 163318,Heilongjiang,China)

机构地区：[1]东北石油大学地球科学学院,黑龙江大庆163318 [2]东北石油大学陆相页岩油气成藏及高效开发教育部重点实验室,黑龙江大庆163318 [3]中国石油集团测井有限公司大庆分公司,黑龙江大庆163318

出　　处：《世界地质》2023年第3期488-500,共13页World Geology

基　　金：中国石油集团测井有限公司大庆分公司项目(ZYCJ-DQ-2022-JS-3442)。

摘　　要：为了更准确地识别海拉尔盆地乌尔逊凹陷铜钵庙组储层的流体性质,经母岩成分、孔隙结构和岩心实验分析,笔者认为饱和纯水电阻率(R 0)受岩性、物性和孔隙结构的影响。通过神经网络预测R 0值消除岩性和孔隙结构对电阻率值的影响,对比接受者操作特性曲线(ROC)及自验证准确率,优选储层流体性质识别相关性较高的测井曲线组合:声波时差(DT24)、自然伽马(GR)、密度(ZDEN)、中子(CNC)、深双侧向电阻率(RLLD)、R 0。结合试油井段,根据决策树、支持向量机、神经网络3种机器学习方法识别结果,得出海拉尔盆地乌尔逊凹陷铜钵庙组储层流体性质识别的机器学习架构方法,其中包含3层隐含层且激活函数为Relu的BP神经网络,储层流体性质识别准确率达92.5%。In order to more accurately identify the fluid properties of Tongbomiao Formation reservoir of Wuerxun depression in Hailar Basin,the authors consider that the saturated pure water resistivity(R 0)is influenced by lithology,physical properties and pore structure through experimental analysis of parent rock composition,pore structure and core experiments.In this paper,the value of R 0 is predicted by neural network to eliminate the influence of lithology and pore structure on resistivity value,and comparison of receiver operating characteristic curves(ROC)and self-validation accuracy to optimize the combination of logging curves with high relevance for reservoir fluid properties identification:interval transit time(DT24),natural gamma(GR),density(ZDEN),neutron(CNC),deep investigate double lateral resistivity(RLLD),and R 0.The results of three machine learning methods,namely,decision tree,support vector machine and neural network,were combined with the test well sections to derive a machine learning architecture method for fluid properties identification in Tongbomiao Formation reservoir of Wuerxun depression in Hailar Basin,the BP neural network with three hidden layers and activation function of Relu has an accuracy of 92.5%for identifying fluid properties.

关 键 词：火山碎屑岩 流体性质识别 机器学习 神经网络 海拉尔盆地 

分 类 号：P618.13[天文地球—矿床学]