特种油气藏 ›› 2021, Vol. 28 ›› Issue (6): 62-69.DOI: 10.3969/j.issn.1006-6535.2021.06.008

• 油藏工程 • 上一篇    下一篇

基于随机森林的K-近邻算法划分火成岩岩性

赖强1, 魏伯阳2,3, 吴煜宇1, 潘保芝2, 谢冰1, 郭宇航2   

  1. 1.中国石油西南油气田分公司,四川 成都 610041;
    2.吉林大学,吉林 长春 130026;
    3.河南省煤炭地质勘察研究总院,河南 郑州 450046
  • 收稿日期:2020-10-06 修回日期:2021-10-13 出版日期:2021-12-25 发布日期:2022-02-16
  • 作者简介:赖强(1979—),男,高级工程师,2002年毕业于西南石油学院应用地球物理专业,2006年毕业于中国石油大学(北京)地球探测与信息技术专业,获硕士学位,现从事测井资料处理及解释方法研究工作。
  • 基金资助:
    中国石油西南油气田科技重大专项“四川盆地二叠系火成岩成藏地质理论与勘探开发关键技术研究”(2019ZD01-04)

Classification of Igneous Rock Lithology with K-nearest Neighbor Algorithm Based on Random Forest (RF-KNN)

Lai Qiang1, Wei Boyang2,3, Wu Yuyu1, Pan Baozhi2, Xie Bing1, Guo Yuhang2   

  1. 1. PetroChina Southwest Oil and Gasfield Company, Chengdu, Sichuan 610041, China;
    2. Jilin University, Changchun, Jilin 130026, China;
    3. Henan General Research Institute of Coal Geology and Exploration, Zhengzhou, Henan 450046, China
  • Received:2020-10-06 Revised:2021-10-13 Online:2021-12-25 Published:2022-02-16

摘要: 针对火成岩油气藏火成岩岩性划分难,岩性划分准确率受薄片鉴定样本数量影响大的问题,利用随机森林(RF)算法分析不同的测井曲线与火成岩岩性相关性,再利用K-近邻(KNN)算法划分小样本薄片鉴定情况下的火成岩岩性。将研究成果应用于川西地区二叠系火成岩地层,结果表明:测井曲线与岩性相关程度从高到低依次为GR、RtDEN、CNL、AC;KNN算法划分火成岩岩性,k取值受分类数量和训练样本数量2个因素控制,样本数量较小时后者影响程度大于前者;k为3时,24个火成岩训练样本(5种岩性)KNN法回判准确率为87.5%,14个火成岩(5种岩性)测试样本测试准确率为92.5%。对比图版划分火成岩岩性,KNN算法受人为影响小,参数调节简便。该研究对小样本情况下火成岩岩性划分有重要指导意义。

关键词: 火成岩油气藏, 岩性划分, 薄片鉴定, KNN, 随机森林

Abstract: To address the problems that it is difficult to classify igneous rock lithology in igneous rock reservoirs and the lithology identification accuracy is greatly affected by the number of slice identification samples,the correlation between different logging curves and igneous rock lithology was analyzed by random forest (RF) algorithm,and then igneous rock lithology was classified by the the K-nearest neighbor (KNN) algorithm according to the slice sample identification.The study results were applied to the Permian igneous rock formation in Western Sichuan,and the results showed that the correlation between logging curves and lithology was decreased in order of GR, Rt, DEN,CNL and AC.The igneous rock lithology was classified with the KNN algorithm, and the value of k was controlled by two factors: the number of classifications and the number of training samples.When there were less samples,the effect of the latter was greater than that of the former.When k was 3, the backcasting accuracy of KNN algorithm was 87.5% for 24 igneous rock training samples (5 types of lithology),and the testing accuracy was 92.5% for 14 igneous rock samples (5 types of lithology).In the classification of igneous rock lithology with comparison of charts,there was less man-made influence on the KNN algorithm and the parameter adjustment was simple.This study provides an important guide to the classification of igneous rock lithology with small samples.

Key words: igneous rock reservoir, lithology classification, slice identification, KNN, random forest

中图分类号: