基于XGBoost算法的煤体结构测井识别技术研究  

Research on logging recognition technology of coal structure based on XGBoost algorithm

作　　者：丁阳阳 赵军龙 李兆明[3] 姚晓莉 王朝阳 李根敏 DING YangYang;ZHAO JunLong;LI ZhaoMing;YAO XiaoLi;WANG ZhaoYang;LI GenMin(School of Earth Sciences and Engineering,Xi'an Shiyou University,Xi'an 710065,China;Shaanxi Key Laboratory of Petroleum Accumulation Geology,Xi'an Shiyou University,Xi'an 710065,China;Linfen Branch of PetroChina Coalbed Methane Co.,Ltd.,Linfen 030000,China;School of Computing,Xi'an Shiyou University,Xi'an 710065,China)

机构地区：[1]西安石油大学地球科学与工程学院,西安710065 [2]西安石油大学陕西省油气成藏地质学重点实验室,西安710065 [3]中国石油煤层气有限责任公司临汾分公司,临汾030000 [4]西安石油大学计算机学院,西安710065

出　　处：《地球物理学进展》2022年第3期998-1006,共9页Progress in Geophysics

基　　金：西安石油大学研究生创新与实践能力培养计划资助(YCS20212090)资助。

摘　　要：煤体结构是影响煤层气开发的因素之一,为厘定DN区块煤体结构类型,服务于该区煤层气勘探开发,本文在文献调研基础上,根据收集到的煤岩宏观描述资料开展煤体结构类型划分,结合多种测井响应特征,利用图版法开展煤体结构识别;为了改善识别效果,引入XGBoost机器学习方法进行煤体结构识别.研究表明,本区的煤体结构类型可以划分为块煤、块粉煤、粉煤三类;双井径幅度差和双侧向电阻率幅度差交会图版识别煤体结构类型效果较好;XGBoost机器学习特征重要性分析显示,双井径幅度差、扩径率、双侧向电阻率幅度差、声波时差、自然伽马等测井响应参数对煤体结构识别模型的贡献较大,交叉验证结果显示该模型精度达到97.2%,XGBoost机器学习方法可以实现多种测井信息融合,有利于提高煤体结构的综合判识效果,实现煤体结构的高精度和高效率识别,效果好于常用图版法.Coal structure is one of the factors that affects coalbed methane development,to set the DN block coal structure types,in the service of the coalbed methane exploration and development,this article on the basis of literature research,macro description coal and rock according to the collected data to carry out the coal structure type,in combination with a variety of logging data characteristics,using engraving method to carry out the coal structure identification;In order to improve the recognition effect,XGBoost machine learning method is introduced for coal structure recognition.The study reveals that the coal structure in this area can be divided into three types:lump coal,lump coal and powder coal.The cross-plot of the amplitude difference of double borehole diameter and double lateral resistivity is effective in identifying coal structure types.XGBoost importance analysis shows that machine learning features dual caliper separation,expanding rate,dual laterolog resistivity separation,acoustic time,natural gamma ray logging parameters,the contribution to the coal structure identification model is bigger,cross validation results show that the model accuracy reached 97.2%,XGBoost machine learning method can achieve a variety of logging information fusion,It can improve the comprehensive identification effect of coal structure and realize the identification of coal structure with high precision and high efficiency,which is better than the common drawing method.

关 键 词：煤体结构 测井响应 图版法 XGBoost机器学习 识别 

分 类 号：P631[天文地球—地质矿产勘探]