基于大数据和无监督聚类算法的岩石可钻性表征和预测方法
A Rock Drillability Characterization Method Based on Big Data and Unsupervised Clustering Algorithm
田龙 1朱智华 1王立伟 1于佳伟 2王一帆2
作者信息
- 1. 中国石油新疆油田分公司工程技术研究院,新疆克拉玛依 834000
- 2. 中国石油大学(北京)人工智能学院,北京昌平 102249
- 折叠
摘要
岩石可钻性的评估在地质勘探和钻井工程中具有重要意义.传统的评价方法主要基于岩心可钻性测试结果,但受制于岩心获取困难和费用昂贵的限制,开发新的无监督学习方法变得愈发重要.针对这一问题,提出了基于测井大数据和无监督聚类算法的连续地层可钻性评估方法.首先,利用自组织映射神经网络对大量的测井数据进行聚类,将地层特征进行有效提取和分类;然后,通过分析每个聚类对应地层的机械钻速分布,将地层分成了6个可钻性等级,从而实现了对地层可钻性的有效评估.这项研究的核心价值在于利用了大数据和先进的无监督学习算法,克服了传统方法中对大量岩心可钻性测试结果的依赖,并取得了显著的成果.通过该方法成功对测试井地层进行了可钻性分级,并验证了其有效性.研究结果显示,随着可钻性等级的增加,地层所对应的平均机械钻速逐渐降低;并且与岩心实测法相比较,模型得到的岩石可钻性等级划分结果偏差不大.这一结果进一步印证了该方法在连续地层可钻性评估中的重要性和准确性.
Abstract
The evaluation of rock drillability is of great significance in geological prospecting and drilling engineering.The traditional evaluation methods are mainly based on the core drillability testing,but due to the technical difficulties and high costs of coring,new unsupervised learning methods have become increasingly important.This study proposes a continuous formation drillability evaluation method based on well logging big data and unsupervised clustering algorithm to address this issue.Firstly,a self-organizing mapping neural network is used to cluster a large amount of well logging data and effectively extracting and classifying stratigraphic features.Then,by analyzing the penetration rate distribution of the formation corresponding to each cluster,the formation is graded by six drill-ability levels,thus achieving effective evaluation of the formation drillability.The core value of this study lies in utilizing big data and advanced unsupervised learning algorithms to overcome the reliance on a large number of core drillability test results in traditional methods,and deliver significantly improved evaluation performance.Through this method,the drillability classification of formations of the test well is successfully carried out,which validates the effectiveness of the method.The research results show that as the drillability level increases,the average penetration rate of the formation gradually decreases,and compared with the core test method,no notable deviation of the rock drillability level classification results is observed.This finding further confirms the importance and accuracy of this method in continuous formation drillability evaluation.
关键词
钻井/可钻性/机器学习/机械钻速/神经网络/大数据Key words
drilling/drillability/machine learning/rate of penetration/neural network/big data引用本文复制引用
基金项目
国家重点研发计划(2019YFA0708300)
中国石油与中国石油大学(北京)战略合作技术项目(ZLZX2020-03)
出版年
2024
NETL
NSTL
万方数据