石油与天然气地质 ›› 2024, Vol. 45 ›› Issue (6): 1524-1536.doi: 10.11743/ogg20240602

• 油气地质 • 上一篇    下一篇

基于ResNet残差神经网络识别的深部煤层显微组分和微裂缝分类

刘大锰1,2(), 王子豪1,2, 陈佳明1,2, 邱峰1,2, 朱凯1,2, 高羚杰1,2, 周柯宇1,2, 许少博1,2, 孙逢瑞1,2   

  1. 1.中国地质大学(北京) 能源学院,北京 100083
    2.非常规天然气地质评价与开发北京市重点实验室,北京 100083
  • 收稿日期:2024-06-10 修回日期:2024-10-11 出版日期:2024-12-30 发布日期:2024-12-31
  • 第一作者简介:刘大锰(1965—),男,教授、博士研究生导师,煤层气储层地质学。E-mail: dmliu@cugb.edu.cn
  • 基金项目:
    国家自然科学基金项目(42130806)

Classification of macerals and microfractures in deep coal seams based on ResNet: A case study of the No.8 coal seam of the Carboniferous Benxi Formation in the Ordos Basin

Dameng LIU1,2(), Zihao WANG1,2, Jiaming CHEN1,2, Feng QIU1,2, Kai ZHU1,2, Lingjie GAO1,2, Keyu ZHOU1,2, Shaobo XU1,2, Fengrui SUN1,2   

  1. 1.School of Energy Resources,China University of Geosciences (Beijing),Beijing 100083,China
    2.Beijing Key Laboratory of Unconventional Natural Gas Geological Evaluation and Development Engineering,Beijing 100083,China
  • Received:2024-06-10 Revised:2024-10-11 Online:2024-12-30 Published:2024-12-31

摘要:

显微组分和微裂缝是煤储层重要的微观特征,影响煤储层产气能力和力学性质。采集鄂尔多斯盆地深部煤层气井石炭系本溪组8#煤层样品,运用ResNet残差神经网络识别方法,研究了显微组分和微裂缝发育特征。在煤样305个显微组分和65个微裂缝图样本研究的基础上,建立了基于残差神经网络识别的煤岩显微组分和微裂缝识别方法,并利用残差神经网络技术对镜下数据进行反演,构建了深部煤储层显微组分和微裂缝的识别和分类模型。结合地质特征和聚类算法结果联合验证,模型具有可靠性。显微组分预测准确率为0.90,微裂缝预测准确率为0.80,可以有效预测煤岩显微组分和微裂缝类型。模型识别与预测表明裂缝形态与显微组分具有相关关系。裂缝的发育与显微组分中的镜质组关系最大,裂缝类别和数量的预测结果与显微组分发育的相吻合。

关键词: 分类模型, 残差神经网络, 显微组分, 微裂缝, 深部煤储层, 煤层气, 石炭系, 鄂尔多斯盆地

Abstract:

Macerals and microfractures are identified as important microscopic characteristics of coal reservoirs, as they are factors affecting the reservoirs’ gas production capacity and mechanical properties. Based on coal samples from the No. 8 coal seam of the Carboniferous Benxi Formation in deep coalbed methane wells in the Ordos Basin, we investigate the developmental characteristics of macerals and microfractures using a residual neural network (ResNet). With 305 maceral and 65 microfracture sample points from the sampled coal, we develop a ResNet-based methodology for identifying macerals and microfractures in coals, and construct an identification and classification model for macerals and microfractures in deep coal reservoirs through the inversion of microscopically observed data using the ResNet technique. The results indicate that the model is reliable, as jointly corroborated by geological characteristics and clustering algorithm-derived results. The model demonstrates a prediction accuracy of 0.90 for macerals and 0.80 for microfractures, enabling the effective prediction of macerals and microfractures in coals. The identification and prediction results of the model reveal correlations between fracture morphologies and macerals. Notably, the fracture formation is the most closely correlated with vitrinites in macerals, with the predicted fracture types and numbers agreeing well with macerals.

Key words: classification model, residual neural network (ResNet), maceral, microfracture, deep coal reservoir, coalbed methane, Carboniferous, Ordos Basin

中图分类号: