特种油气藏 ›› 2022, Vol. 29 ›› Issue (1): 38-45.DOI: 10.3969/j.issn.1006-6535.2022.01.006

• 地质勘探 • 上一篇    下一篇

基于机器学习的火山岩岩性智能识别及预测

刘凯1, 邹正银1, 王志章2, 蒋庆平1, 常天全1, 王伟方2, 杨笑2,3   

  1. 1.中国石油新疆油田分公司,新疆 克拉玛依 834000;
    2.中国石油大学(北京)油气资源与探测国家重点实验室,北京 102249;
    3.中国石油长庆油田分公司,陕西 西安 710018
  • 收稿日期:2020-08-25 修回日期:2021-10-25 出版日期:2022-02-25 发布日期:2023-01-10
  • 作者简介:刘凯(1989—),男,工程师,2012年毕业于长安大学地球物理学专业,2015年毕业于中国石油大学(北京)地球物理学专业,获硕士学位,现主要从事油气地质综合研究工作。
  • 基金资助:
    中国石油重大科技专项“火山岩油藏效益开发关键技术研究与应用”(2017E-0405);中国石油科技项目“准噶尔盆地石炭系火山岩分类评价与规模效益建产关键技术研究及工业化应用”(kt2017-18-05)

Intelligent Identification and Prediction of Lithology of Volcanic Reservoirs Based on Machine Learning

Liu Kai1, Zou Zhengyin1, Wang Zhizhang2, Jiang Qingping1, Chang Tianquan1, Wang Weifang2, Yang Xiao2,3   

  1. 1. PetroChina Xinjiang Oilfield Company, Karamay, Xinjiang 834000, China;
    2. State Key Laboratory of Petroleum Resources and Prospecting, China University of Petroleum (Beijing), Beijing, 102249, China;
    3. PetroChina Changqing Oilfield Company, Xi′an, Shaanxi 710018, China
  • Received:2020-08-25 Revised:2021-10-25 Online:2022-02-25 Published:2023-01-10

摘要: 针对准噶尔盆地金龙2井区佳木河组火山岩油气藏岩性多变,常规方法难以准确识别的问题,利用机器学习中的决策树、随机森林、梯度提升树、贝叶斯4种算法对研究区岩性进行智能识别,在分析研究区火山岩储层地质特点的基础上,结合不同岩性测井响应特征,确定M、N等8个对火山岩岩性极为敏感的特征参数。研究结果表明:随机森林法模型最优,准确率达到90%以上,模型泛化能力最强,可作为利用常规测井曲线识别火山岩岩性的有效方法。该模型可以高精度地进行火山岩岩性识别及预测,为后续火山岩油藏的勘探与开发奠定基础。

关键词: 火山岩, 岩性特征, 机器学习, 智能识别, 决策树, 随机森林

Abstract: To address the problem that the lithology of the volcanic reservoirs of Jiamuhe Formation in Well Block Jinlong 2, Junggar Basin is variable and difficult to be accurately identified by conventional methods, four algorithms in machine learning, including decision tree, random forest, gradient boosting tree and Bayes, were adopted to intelligently identify the lithology of the study area, and then determine eight characteristics parameters such as M and N, which are extremely sensitive to the lithology of volcanic rocks based on the analysis of the geological characteristics of volcanic reservoirs in the study area and the logging response characteristics of different lithologies. The results of the study proved that the random forest method was preferred with the best model, an accuracy rate of more than 90% and the highest model generalization. It was an effective method to identify volcanic rock lithology based on conventional logging curves. According to this study, the volcanic rock lithology can be identified and predicted with a high precision, laying a foundation for subsequent exploration and development of volcanic rock reservoirs.

Key words: volcanic rock, lithological characteristics, machine learning, intelligent recognition, decision tree, random forest

中图分类号: