特种油气藏 ›› 2022, Vol. 29 ›› Issue (6): 141-149.DOI: 10.3969/j.issn.1006-6535.2022.06.018

• 油藏工程 • 上一篇    下一篇

四川盆地威远区块页岩气单井产量预测方法及应用

韩珊1, 车明光1,2, 苏旺1, 肖毓祥1, 吴忠宝1, 陈建阳1, 汪莉彬1   

  1. 1.中国石油勘探开发研究院,北京 100083;
    2.中国科学技术大学,安徽 合肥 230027
  • 收稿日期:2022-02-18 修回日期:2022-10-13 出版日期:2022-12-25 发布日期:2023-01-10
  • 作者简介:韩珊(1994—),女,助理工程师,2016年毕业于中国石油大学(北京)石油工程专业,2020年毕业于该校油气田开发工程专业,获硕士学位,现主要从事致密油气(含页岩油气)开发研究工作。
  • 基金资助:
    中国石油重大专项“深层页岩气有效开采关键技术攻关与试验——深层页岩气体积压裂技术现场试验”(2019F-31-04)

Prediction Method and Application of Single Shale Gas Well Production in Weiyuan Block, Sichuan Basin

Han Shan1, Che Mingguang1,2, Su Wang1, Xiao Yuxiang1, Wu Zhongbao1, Chen Jianyang1, Wang Libin1   

  1. 1. China Petroleum Exploration and Production Research Institute, Beijing 100083, China;
    2. University of Science and Technology of China, Hefei, Anhui 230027, China
  • Received:2022-02-18 Revised:2022-10-13 Online:2022-12-25 Published:2023-01-10

摘要: 针对四川盆地威远区块页岩气单井产量主控因素不明的问题,基于该区132口投产1 a以上气井的地质与工程数据及生产数据,使用灰色关联法对影响页岩气单井产量的主控因素进行分析研究。研究表明:影响页岩气单井首年累计产量的主控因素为支撑剂量、压裂改造段数、水平井垂深的中值、压裂段长度、压裂液量、孔隙度、压力系数和加砂强度。通过机器学习方法与传统经验公式法对比预测首年累计产量及初期产量,明确了机器学习法精度更高。同时,基于主控因素分析的基础上,优选出适用于研究区的机器学习法为支持向量机法,其预测精度高于90%。研究结果对同类页岩气区块产能评价具有重要意义。

关键词: 页岩气, 主控因素, 单井产量, 机器学习, 支持向量机, BP神经网络, 四川盆地

Abstract: To address the problem that the main controlling factors of single shale gas well production in Weiyuan Block, Sichuan Basin are unknown, based on the geological and engineering data and production data of 132 gas wells which have been put in production for more than a year in the area, an analytical study was conducted by the gray correlation method. The study shows that, the main controlling factors affecting the first-year cumulative production of a single shale gas well are proppant dose, number of fracturing stages, median vertical depth of horizontal wells, fracturing section length, fracturing fluid volume, porosity, pressure coefficient and sanding intensity. It was clear that the machine learning method was higher in accuracy after comparison of the machine learning method and the traditional empirical formula method to predict the first year's cumulative output and initial output. Meanwhile, based on the analysis of the main controlling factors, the machine learning method applicable to the study area was preferably selected as the support vector machine method, and its prediction accuracy was higher than 90%. The study has an important implication to the productivity evaluation of similar shale gas blocks.

Key words: Shale gas, main controlling factor, single-well production, machine learning, support vector machine, BP neural network, Sichuan Basin

中图分类号: