基于KNN分类算法的微生物白云岩岩相测井综合识别——以四川盆地GM地区灯四段为例  

Comprehensive logging identification of microbial carbonate lithofacies based on KNN classification algorithm:a case study of Dengying Formation in GM area,Sichuan Basin

作　　者：李昌 王鑫 冯周[3] 宋连腾[3] LI Chang;WANG Xin;FENG Zhou;SONG Lianteng

机构地区：[1]中国石油杭州地质研究院 [2]中国石油集团碳酸盐岩储层重点实验室 [3]中国石油勘探开发研究院

出　　处：《海相油气地质》2023年第4期433-440,共8页Marine Origin Petroleum Geology

基　　金：中国石油天然气集团有限公司“十四五”前瞻性基础性战略性技术攻关课题“人工智能测井储层评价新方法研究”(编号:2021DJ3806)资助。

摘　　要：微生物碳酸盐岩的岩-电关系复杂,目前常规测井与电成像测井结合是最有效且精度最高的识别手段。针对人工智能测井岩相识别方法存在的不同维度的测井数据融合难、取心资料有限而训练样本数量不充足的问题,提出基于适应小样本的机器学习法——K邻近分类算法(KNN),对常规测井与电成像测井分别训练和识别,再将识别结果融合的技术方法。首先,基于岩心资料分别建立岩相分类方案和岩石构造特征分类方案,建立岩心训练样本参数库;然后,基于KNN方法,应用常规测井识别的岩相类型,应用电成像测井识别岩石构造特征类型;最后,根据专家经验对2种识别结果进行融合,获得细分类的岩相类型。以四川盆地GM地区灯影组四段为例,应用上述方法分别识别6种岩相类型和7种岩石构造特征类型,在此基础上根据专家经验融合,最终识别9种细分类的岩相类型。该方法总体识别符合率在85%以上,有效支撑了GM地区灯四段沉积微相的精细研究,推动了该区的勘探和开发工作。该方法发挥了常规测井和电成像测井的优势,能够实现高效率、高精度的岩相测井识别,可推广应用。Microbial structures are developed in microbial carbonate rocks,with strong diagenesis superimposed,and their lithology-electrical property relationship is more complex.Conventional logging has been unable to distinguish microbial structure characteristics.Although electric imaging logging has high resolution and can identify microbial structures,there is also a problem of multiple solutions.At present,the combination of conventional logging and electrical imaging logging is the most effective and accurate identification method.The main methods of combination include chart method and artificial intelligence learning method.However,the efficiency of chart method is low,and artificial intelligence methods also have two problems:(1)there is difficulty in integrating logging data from different dimensions;(2)the core sampling data is limited,and the number of training samples is insufficient.Therefore,this article selects the K-Neighbor Classification Algorithm(KNN),a machine learning method that adapts to few samples,and proposes a method of separate training and recognition,and re-fusion of recognition results.Firstly,based on core data,we establish lithofacies classification schemes and rock structure feature classification schemes respectively,and establish a core training sample parameter library,and then use KNN method to identify lithofacies types with conventional logging and rock structure types with electrical imaging logging.Finally,based on expert experience,we fuse the two recognition results to obtain finely classified lithofacies types.Taking the Dengying Member 4 in the GM area of Sichuan Basin as an example,6 types of lithofacies and 7 types of rock structural feature types were identified.Based on expert experience fusion,9 types of finely classified lithofacies were finally identified,with a recognition accuracy rate over 85%.This study has effectively supported the fine research work on sedimentary microfacies of the Dengying Member 4 in the GM area and promoted the exploration and development wor

关 键 词：微生物碳酸盐岩 KNN算法 常规测井 电成像测井 特征参数 岩相识别 

分 类 号：TE122.2[石油与天然气工程—油气勘探]