摘要： 南海西部Y盆地L地区中新统地层呈高温、超高压特征,钻井难度大、取心资料少,岩屑录井反映岩性的精度较低,难以满足岩性精细识别的要求。以Y盆地L地区黄流组二段深层复杂碎屑岩为例,首先,利用有限的壁心粒度分析、录井、测井等资料,优选出表征岩性的粒度参数: 粒度中值Md和对粒度变化敏感的自然伽马、密度、中子、声波时差、电阻率5条测井曲线,构建粒度中值Md-测井5变量数据集; 其次,采用K-MEANS聚类方法,将数据集根据“误差平方和与聚类数”最优关系划分成了4类(简称“粒度分类”),分类后优化了粒度中值Md与测井响应的相关性,且获得不同类别的测井响应特征和相应岩性类型; 然后,在实际井资料处理过程中,应用Fisher判别方程来判别未知深度点所属的粒度分类类型; 最后,建立粒度分类下基于XGBoost算法的粒度中值测井智能计算模型,依据不同岩性对应粒度中值的数值范围,实现了井筒剖面上根据测井反演粒度中值Md曲线进而达到岩性精细识别的目的。研究结果表明: L地区黄流组二段考虑粒径的差异将砂岩岩性划分为: 粉砂岩、细砂岩、中砂岩、粗砂岩,其中细砂岩和中砂岩是最主体发育的岩性,粒度中值Md与不同粒径岩性的关系最密切,是最能反映不同粒径岩性的粒度参数; 粒度分类后基于XGBoost算法的粒度中值测井智能计算模型预测效果优于多元回归预测模型,计算粒度中值与实测值的相关系数达0.9397,平均绝对误差MAE为0.0195 mm,平均相对误差MRE为17.52%。该模型是一种有效实现深层复杂碎屑岩岩性精细识别的方法,也为纵向剖面上沉积粒序分析和储层构型精细解释、有效性评价奠定了基础。

关键词: 南海, 中新统, 复杂岩性, 粒度中值, 测井, 反演, 机器学习, XGBoost算法

Abstract: The Miocene strata in the L area of the Y Basin in the western part of the South China Sea are characterised by high temperature and ultra-high pressure,which makes drilling difficult and core data rare. In addition, the accuracy of rock chip logging in reflecting lithology is relatively low,making it difficult to meet the requirements of fine identification of lithology. The deep clastic rocks in the second section of Huangliu Formation in the L area of Y basin are used in this study,firstly,using the limited data of core size analysis,rock chip logging and logging,we selected the particle size parameter characterizing lithology: median Md and five logging curves of natural gamma,density,neutron,acoustic time difference and resistivity which are sensitive to changes in the particle size,and constructed the data set of five variables of the median Md and logging,and then we used K-MEANS secondly,using K-MEANS clustering method,the dataset was divided into four classes according to the optimal relationship between “sum of error squares and the number of clusters”(referred to as “granularity classification”),which optimised the correlation between the median Md-granularity and the logging response,and obtained the logging response characteristics of the different classes and the corresponding lithological types. Then,in the actual well data processing process,Fisher's discriminant equation is applied to determine the type of particle size classification to which the unknown depth point belongs,and finally,the intelligent calculation model of median particle size logging based on XGBoost algorithm is established under the particle size classification,and based on the numerical range of median particle size corresponding to different lithologies,it realises the purpose of fine identification of lithology by inverting the median Md curve according to the logging on the wellbore profile. The purpose of fine identification of lithology is achieved by inverting the Md curve on the wellbore profile according to the logging.The results show that the sandstone lithology in the second section of Huangliu Formation in L area is divided into: siltstone,fine sandstone,medium sandstone and coarse sandstone considering the difference of grain size,among which the fine sandstone and medium sandstone are the most dominantly developed lithologies,and the median Md of grain size has the closest relationship with the lithology of different grain sizes,and it is the most reflective of the different grain sizes of lithologies;the intelligent calculation of the median Md of grain size logging model based on XGBoost algorithm is better than that of multiple regression algorithm after the classification of the grain sizes. The prediction effect of the model is better than that of the multiple regression prediction model,and the correlation coefficient between the calculated median particle size and the measured value reaching 0.9397,the average absolute error(MAE) is 0.0195,and the average relative error MRE is 0.1752. The model is an effective method for the fine identification of the lithology of the deep complex clastic rocks,and it also lays a foundation for sedimentary grain sequence analysis and fine interpretation of the reservoir configuration,and the evaluation of the validity on the vertical profile. It also lays the foundation for sedimentary grain sequence analysis,fine interpretation of reservoir configuration and validity evaluation in longitudinal section.

Key words: South China Sea, Miocene, complex lithology, median grain size, logging, inversion, machine learning, XGBoost algorithm

中图分类号: 

• P631.8