西南石油大学学报(自然科学版) ›› 2025, Vol. 47 ›› Issue (2): 95-104.DOI: 10.11885/j.issn.1674-5086.2023.02.27.01

• 石油与天然气工程 • 上一篇    下一篇

基于FCMFS特征选择算法的煤层气压裂效果预测

闵超1,2,3, 郭星1,2, 华青4, 张娜4, 张馨慧1,2   

  1. 1. 西南石油大学理学院, 四川 成都 610500;
    2. 西南石油大学人工智能研究院, 四川 成都 610500;
    3. 油气藏地质及开发工程全国重点实验室·西南石油大学, 四川 成都 610500;
    4. 中国石油西南油气田公司重庆气矿, 重庆 江北 400700
  • 收稿日期:2023-02-27 发布日期:2025-05-15
  • 通讯作者: 闵超,E-mail:minchao@swpu.edu.cn
  • 作者简介:闵超,1982年生,男,汉族,四川成都人,教授,博士,主要从事最优化方法与不确定理论在油气田开发中的应用研究。E-mail:minchao@swpu.edu.cn
    郭星,1999年生,男,汉族,四川达州人,硕士研究生,主要从事深度学习的研究工作。E-mail:177009824@qq.com
    华青,1989年生,女,汉族,湖南常德人,工程师,主要从事油气田开发方面的研究工作。E-mail:huaqing2013@petrochina.com.cn
    张娜,1981年生,女,汉族,湖北荆州人,高级工程师,硕士,主要从事油气田开发方面的研究工作。E-mail:zhangcoco@petrochina.com.cn
    张馨慧,1997年生,女,汉族,内蒙古赤峰人,硕士研究生,主要从事深度学习的研究工作。E-mail:1844182840@qq.com
  • 基金资助:
    四川省科技创新苗子工程资助项目(2022034);西南油气田公司科研项目(20230303-10)

Coalbed Methane Fracturing Effect Prediction Based on FCMFS Feature Selection Algorithm

MIN Chao1,2,3, GUO Xing1,2, HUA Qing4, ZHANG Na4, ZHANG Xinhui1,2   

  1. 1. School of Science, Southwest Petroleum University, Chengdu, Sichuan 610500, China;
    2. Institute for Artificial Intelligence, Southwest Petroleum University, Chengdu, Sichuan 610500, China;
    3. State Key Laboratory of Oil and Gas Reservoir Geology and Exploitation, Southwest Petroleum University, Chengdu, Sichuan 610500, China;
    4. Chongqing Gas Mine, Southwest Oil and Gasfield Company, PetroChina, Jiangbei, Chongqing 400700, China
  • Received:2023-02-27 Published:2025-05-15

摘要: 煤层气压裂效果与特征之间存在的非线性关系难以从机理层面进行分析,针对该问题,开展煤层气压裂效果特征内在联系研究,提出了一种基于FCMFS特征选择算法的煤层气压裂效果预测方法。该方法利用模糊综合评价进行标签标定,并采用遗传编程和XGBoost算法进行影响因素特征构造和筛选,包括2个新构造特征(应力比和地质施工遗传因素)以及射孔段厚度、渗透率、破裂压力、煤体结构、含气饱和度和加砂强度等6个特征。实验结果表明,基于FCMFS特征选择算法所构造和筛选的8个特征,结合多种机器学习算法进行煤层气压裂效果预测时,在准确率、召回率、F1分类评价指标上提高了约5%~10%,其中,深度森林模型在训练集和测试集上具有最优的预测分类效果,在3项分类评价指标上均达到95%和80%以上。

关键词: 煤层气, 压裂效果, 主控因素, 遗传编程, 深度森林模型

Abstract: It is difficult to analyze the nonlinear relationship between the fracturing effect and characteristics of coalbed methane from the mechanism level. Aiming at the problem, the internal relationship between the characteristics of coalbed methane fracturing effect is studied, and a prediction method of coalbed methane fracturing effect based on FCMFS feature selection algorithm is proposed. The method uses fuzzy comprehensive evaluation to calibrate the label, and uses genetic programming and XGBoost algorithm to construct and screen the characteristics of influencing factors, including two new structural features (stress ratio and genetic factors of geological construction) and six characteristics of perforation section thickness, permeability, fracture pressure, coal structure, gas saturation and sand strength. The experimental results show that based on the eight features constructed and screened by the FCMFS feature selection algorithm, combined with a variety of machine learning algorithms to predict the effect of coalbed methane fracturing, the accuracy, recall rate, and F1 classification evaluation indicators are improved by about 5%~10%. Among them, the Deep Forest algorithm has the best prediction classification effect on the training set and the test set, and the three classification evaluation indicators are all above 95% and 80%.

Key words: coalbed methane, fracturing performance, main controlling factor, gene programming, Deep Forest model

中图分类号: