CIESC Journal ›› 2019, Vol. 70 ›› Issue (S2): 311-321.DOI: 10.11949/0438-1157.20190352

Previous Articles     Next Articles

Prediction for dynamic liquid level of sucker rod pumping using generation of multi-scale state characteristics in oil field production

Yanbin HOU1(),Xianwen GAO1(),Xiangyu LI2   

  1. 1. College of Information Science and Engineering, Northeastern University, Shenyang 110819, Liaoning, China
    2. School of Automation and Electrical Engineering, Shenyang Ligong University, Shenyang 110168, Liaoning, China
  • Received:2019-04-03 Revised:2019-06-14 Online:2019-09-06 Published:2019-09-06
  • Contact: Xianwen GAO

采油过程多尺度状态特征生成的有杆泵动态液面预测

侯延彬1(),高宪文1(),李翔宇2   

  1. 1. 东北大学信息科学与工程学院,辽宁 沈阳 110819
    2. 沈阳理工大学自动化与电气工程学院,辽宁 沈阳 110168
  • 通讯作者: 高宪文
  • 作者简介:侯延彬(1984—),男,博士研究生,houyanbin1984@gmail.com
  • 基金资助:
    国家自然科学基金项目(61573088)

Abstract:

Dynamic liquid level is an important parameter to reflect the liquid supply capacity of oil deposit. In the actual production site, dynamic liquid level is usually detected by the echo meter. This method is limited by the low detection efficiency, discrete measurement, and safety hazards. In addition to this, the existing prediction methods based on data-driven have difficulties in modeling due to the lack of historical data caused by various reasons. In response to the problem about lack of historical data, the multi-scale state characteristics which are closely connected with dynamic liquid level by analyzing the mechanism has been found out in this paper. Further, the state characteristics have been generated by using generative adversarial network (GAN). Simulation experiments have shown that the generated data can be used to establish dynamic liquid level prediction model and solve the problem about lack of historical data. On this basis, support vector regression (SVR) has been adopted as the modeling method and used in oil field production. Finally, the actual prediction results show that the proposed method is effective and meets the application requirements of oilfield engineering.

Key words: dynamic liquid level, soft-sensing, multi-scale state characteristics, generative adversarial networks, support vector regression

摘要:

动态液面是反映油藏供液能力的重要参数,生产现场多采用回声仪进行检测,其检测效率低,无法连续检测,存在安全隐患。针对现有基于数据驱动的预测方法由于各种原因造成历史数据不足,进而导致建模困难的问题,通过对机理进行分析,找出与动态液面关系密切的多尺度状态特征;采用生成对抗网络生成状态特征,解决历史数据不足的问题;仿真实验表明生成的数据可以用于建立动态液面预测模型,并确定采用支持向量回归作为建模方法;最后在油田生产现场使用本文提出的建模方法,实际预测结果表明本方法的有效性,满足油田工程应用要求。

关键词: 动态液面, 软测量, 多尺度状态特征, 生成对抗网络, 支持向量回归

CLC Number: