基于FL-XGBoost算法的砂泥岩识别方法——以胜利油田牛庄地区为例  

FL-XGBoost algorithm-based method for identifying sandstone and mudstone:A case study of Niuzhuang area in Shengli Oilfield

作　　者：彭英[1] 李克文[2] 朱应科[1] 徐志峰 杨澎涛[1] 孙秀玲 PENG Ying;LI Kewen;ZHU Yingke;XU Zhifeng;YANG Pengtao;SUN Xiuling(Geophysical Exploration Research Institute of SINOPEC Shengli Oilfield Company,Dongying City,Shandong Province,257000,China;College of Computer Science and Technology,China University of Petroleum(East China),Qingdao City,Shandong Province,266580,China;Shandong Shengruan Technology Co.,Ltd.,Dongying City,Shandong Province,257000,China)

机构地区：[1]中国石化胜利油田分公司物探研究院,山东东营257000 [2]中国石油大学(华东)计算机科学与技术学院,山东青岛266580 [3]山东胜软科技股份有限公司,山东东营257000

出　　处：《油气地质与采收率》2023年第1期76-85,共10页Petroleum Geology and Recovery Efficiency

基　　金：国家自然科学基金项目“储层天然气水合物相变和渗流多场时空演化规律”(51991365);山东省自然科学基金项目“基于多源数据融合的浊积岩有效储层预测方法”(ZR2021MF082)。

摘　　要：砂泥岩识别任务通常基于测井曲线,依据经验公式、实地岩心取样、交会图和聚类分析等传统方法实现,但这些方法难以充分利用测井曲线所包含的砂泥岩特征,且精度低、效率低,人为影响因素大。为此,以测井和录井资料为基础,综合砂泥岩识别的关键技术难点,对测井参数进行敏感性分析,以选取适当的影响因素,通过多项预处理操作构建完整的训练数据集,并根据测井标签稀疏性的特点,引入Focal Loss函数,提出FL-XGBoost模型,进而开展胜利油田牛庄地区砂泥岩识别。研究结果表明,采用FL-XGBoost算法的砂泥岩识别模型对研究区砂泥岩识别的准确率达到了0.827。通过5种公开分类数据集设计对比实验,证明FL-XGBoost算法在识别分类领域上具有强泛化能力。sandstone and mudstone identification tasks are usually based on logging curves and rely on traditional methods such as empirical formulas,field core sampling,cross plots,and cluster analysis,but these methods fail to make full use of the sandstone and mudstone features contained in the logging curves. At the same time,these traditional methods have low accuracy and slow efficiency and are greatly affected by human factors. To address the above problems,this paper uses logging data as the basis,combines the key technical difficulties of sandstone and mudstone identification, and conducts sensitivity analysis on logging parameters,so as to select appropriate influencing factors and construct a complete training data set through several pre-processing operations. In addition,the paper introduces the Focal Loss function and proposes the FL-XGBoost model according to the sparsity of logging labels and carries out sandstone and mudstone identification in Niuzhuang area of Shengli Oilfield. The experimental results show that the sandstone and mudstone identification model using the FL-XGBoost algorithm achieves an accuracy of 0.827 in identifying the sandstone and mudstone in the study area. Finally,the strong generalization ability of the FL-XGBoost algorithm in the identification classification field is verified through five publicly classified dataset design comparison experiments.

关 键 词：FL-XGBoost算法 迭代决策树 机器学习 砂泥岩识别 测井资料 

分 类 号：TE319[石油与天然气工程—油气田开发工程]