新疆石油地质

新疆石油地质 ›› 2026, Vol. 47 ›› Issue (2): 222-232.doi: 10.7657/XJPG20260211

• 应用技术 • 上一篇    下一篇

基于极度随机树算法的火山岩岩性识别——以长岭断陷查干花地区火石岭组为例

王晔磊1,2(), 慈兴华1,2(), 杜焕福1,2, 侯文辉1,2, 汪志峰3, 王顺晔4, 王春伟1,2   

  1. 1 中石化经纬有限公司 地质测控技术研究院， 山东 青岛 266000
    2 中石化测录井重点实验室， 山东 青岛 266000
    3 中国石化 东北油气分公司 勘探开发研究院， 长春 130062
    4 廊坊师范学院 电子信息工程学院， 河北 廊坊 065000
  • 收稿日期:2025-02-20 修回日期:2025-04-23 出版日期:2026-04-01 发布日期:2026-04-08
  • 通讯作者: 慈兴华（1969-），男，山东东营人，教授级高级工程师，博士，综合地质研究及录井技术推广应用，（Tel）19707786685（Email）cixinghua.osjw@sinopec.com
  • 作者简介:王晔磊（1989-），男，河北辛集人，助理研究员，博士，非常规油气储层综合评价，（Tel）15610046181（Email）wangyl26925.osjw@sinopec.com
  • 基金资助:
    国家科技重大专项(2025ZD1404600);中石化经纬有限公司博士后项目(MWBG2400270001)

Lithology Identification of Volcanic Rocks Based on Extra-Trees Classifier: A Case Study of the Huoshiling Formation in the Chaganhua Subsag, Changling Fault Depression, Songliao Basin

WANG Yelei1,2(), CI Xinghua1,2(), DU Huanfu1,2, HOU Wenhui1,2, WANG Zhifeng3, WANG Shunye4, WANG Chunwei1,2   

  1. 1 Geological Measurement and Steering Technology Research Institute, Sinopec Matrix Co., Ltd., Qingdao, Shandong 266000, China
    2 Sinopec Key Laboratory of Well Logging and Mud Logging, Qingdao, Shandong 266000, China
    3 Research Institute of Exploration and Development, Sinopec Northeast Oil and Gas Company, Changchun, Jilin 130062, China
    4 School of Electronic Information Engineering, Langfang Normal University, Langfang, Hebei 065000, China
  • Received:2025-02-20 Revised:2025-04-23 Online:2026-04-01 Published:2026-04-08

RichHTML

7

PDF

23

摘要：

火山岩储层是松辽盆地长岭断陷近年来勘探重点之一，研究火山岩复杂岩性测录井响应是梳理储层四性的关键。以火山岩复杂岩性显微分析为基础，通过薄片鉴定、X射线衍射全岩分析、RoqScan矿物自动识别等定量分析，对长岭断陷查干花地区火石岭组火山岩岩性进行标定；将标定段常规测井数据与元素录井数据分为训练集和测试集，利用训练集拟合目标岩性，将测试集带入模型计算得到预测结果，并利用模型进行盲井预测。研究表明，长岭断陷查干花地区火石岭组火山岩可划分为火山熔岩、火山碎屑熔岩、火山碎屑岩、沉火山碎屑岩和火山碎屑沉积岩5大类13小类，岩性复杂多变；对比决策树、LightGBM、随机森林、神经网络、K近邻和极度随机树6类不同算法区分岩性准确度，6类算法准确度均在77%以上，其中极度随机树综合性能最优，准确率达90%；该模型泛化能力较强，对盲井的预测准确率达到 89%的同时对原始岩屑录井进行了修正，能够对研究区火山岩岩性进行准确识别与预测，为后续火山岩油气勘探开发提供智能化支持。

关键词: 松辽盆地, 长岭断陷, 火石岭组, 火山岩, 岩性, 机器学习, 极度随机树

Abstract:

Volcanic rock reservoir is one of the key exploration targets in the Changling fault depression of the Songliao Basin in recent years. The logging responses of complex volcanic lithologies are crucial to clarifying the reservoir properties (lithology, physical property, electrical property, and oil-bearing property). Based on the microscopic analysis of complex volcanic lithology, the lithology of volcanic rocks in the Huoshiling formation of the Chaganhua subsag in the Changling fault depression was calibrated through thin-section examination, whole-rock X-ray diffraction (XRD) analysis, and quantitative analysis using the RoqScan mineral auto-identification system. The conventional logging data and elemental logging data of the calibrated interval were divided into training set and test set. The training set was used to fit the target lithology, and the test set was loaded into the model calculation for prediction. Moreover, the model was employed in blind well testing. The results show that the volcanic rocks in the Huoshiling formation of the Chaganhua subsag in the Changling fault depression can be categorized into 5 classes such as volcanic lava, pyroclastic lava, pyroclastic rock, sedimentary pyroclastic rock, and pyroclastic sedimentary rock, indicating complex and varying lithologies. Six algorithms, i.e. decision tree, LightGBM, random forest, neural network, K-nearest neighbor (KNN), and extra-trees classifier (ETC), were compared for distinguishing lithology, revealing the accuracy of above 77% for all algorithms. ETC exhibits the best performance, with an accuracy up to 90%. This model has a strong generalization ability and yields an accuracy of 89% in blind well testing while correcting the results of original cutting logging. It can accurately identify and predict the lithology of volcanic rocks in the study area and provide intelligent support for subsequent volcanic oil and gas exploration and development.

Key words: Songliao Basin, Changling fault depression, Huoshiling formation, volcanic rock, lithology, machine learning, extra-trees classifier

中图分类号: