基于支持向量机的潜山储层岩性识别  

Lithology Identification of Buried Hill Reservoirs Based on Support Vector Machine

作　　者：高永德 吴进波 孙殿强 GAO Yongde;WU Jinbo;SUN Dianqiang(Zhanjiang Branch,CNOOC China Limited,Zhanjiang,Guangdong 524000,China)

机构地区：[1]中海石油(中国)有限公司湛江分公司,广东湛江524000

出　　处：《测井技术》2025年第1期77-87,共11页Well Logging Technology

基　　金：国家自然科学基金项目“基于‘矿物-岩性-测井’响应机制的火成岩潜山岩性建模井孔硬数据质量增强方法研究”(42472213)。

摘　　要：潜山基岩储层作为一种非常规储层,由于其基岩岩石结构、构造特征、化学成分、矿物成分的多样性,与传统的碎屑岩储层相比,其岩性识别难度极大。常规测井资料和传统的测井评价方法在此类储层的岩性识别中效果有限,难以有效区分不同岩性类型。为解决这一问题,选取HZ凹陷为研究区,提出了一种基于支持向量机算法(Support Vector Machine,SVM)的岩性识别模型。SVM算法具有优异的高维数据处理能力和较强的泛化能力,同时实现相对简单,适合于复杂储层岩性的识别。在模型构建过程中,考虑到研究区测井曲线较强的非线性特征,结合实验数据和录井信息,将计算得到的地层矿物含量作为模型训练的约束条件。为了进一步增强SVM模型对岩性特征的学习能力,选取对岩性敏感的自然伽马、声波时差和补偿密度等测井曲线数据进行特征提取,并将其用于SVM模型的训练。通过该方法,建立了一个高效的岩性识别模型,并将其应用于HZ地区7口井的岩性识别任务。实验结果表明,SVM模型的平均识别准确率达到了94.46%,与常规的交会图法和随机森林算法相比,支持向量机算法在该地区的应用表现出显著优势,能够有效地解决潜山储层复杂的岩性识别问题。As an unconventional reservoir,the potential mountain bedrock reservoir presents significant challenges for lithology identification compared to traditional clastic reservoirs due to its complex bedrock structure,tectonic features,chemical composition and mineralogy.Conventional logging data and traditional evaluation methods are limited in their effectiveness for identifying lithology in such reservoirs,making it difficult to distinguish between different lithology types.To address this issue,we selecte the HZ depression as the study area and propose a lithology identification model based on the support vector machine(SVM)algorithm.SVM is well-suited for handling high-dimensional data and exhibits strong generalization capability,while being relatively simple to implement,making it ideal for identifying complex reservoir lithology.During the model development process,considering the strong non-linear characteristics of the logging curves in the study area,we combine experimental data and driling information,and use the calculate formation mineral content as a constraint condition for model training.To further enhance the lithological feature learning capability of the SVM model,we selecte logging curve data that are sensitive to lithology,such as natural gamma,acoustic transit time,and compensate density,for feature extraction and use them for training the SVM model.Using this approach,an efficient lithology identification model is established and applied to the lithology identification task of seven wells in the HZ area.The experimental results indicate that the SVM model achieves an average identification accuracy of 94.46%.Compared with the conventional crossplot method and Random Forest algorithm,the SVM algorithm shows significant advantages in this area and effectively addresses the complex lithology identification challenges of the potential mountain reservoir.

关 键 词：潜山储层 岩性识别 支持向量机 矿物含量 HZ凹陷 

分 类 号：P631.84[天文地球—地质矿产勘探]