吉林大学学报(信息科学版) ›› 2022, Vol. 40 ›› Issue (1): 77-81.

• • 上一篇    下一篇

基于长短期记忆神经网络的油田注水预测

于志刚1, 张德政1, 宋文江1, 葛 嵩1, 辛小军2   

  1. 1. 中海石油(中国)有限公司 湛江分公司, 广东 湛江 524057; 2. 中法渤海地质服务有限公司 湛江分公司, 广东 湛江 524057
  • 收稿日期:2020-06-13 出版日期:2022-01-25 发布日期:2022-01-29
  • 通讯作者: 辛小军(1983— ), 男, 甘肃平凉人, 中法渤海地质服务有限公司湛江分公司工程师, 硕士, 主要从事试井技术研究, (Tel)86-890-68700605(E-mail)xinxj@cfbgc.com
  • 作者简介:于志刚(1982— ), 男, 河北唐山人, 中海石油(中国)有限公司湛江分公司高级工程师, 硕士, 主要从事采油工程技术研究, (Tel)86-759-3901802(E-mail)yuzhg1@cnooc.com.cn
  • 基金资助:
    中海油重点基金资助项目(CNOOC-KJ135ZDXM38ZJ)

Oil Field Water Injection Prediction Based on LSTM Neural Networks

YU Zhigang1, ZHANG Dezheng1, SONG Wenjiang1, GE Song1, XIN Xiaojun2   

  1. 1. Zhanjiang Branch, China National Offshore Oil Limited, Zhanjiang 524057, China; 2. Zhanjiang Branch,China France Bohai Geoservices Limited, Zhanjiang 524057, China
  • Received:2020-06-13 Online:2022-01-25 Published:2022-01-29

摘要: 为解决目前常用的人工智能注水预测无法考虑数据在时间上的相关性问题, 通过选取一种基于循环神经网络(RNN: Recurrent Neural Network)改进的长短期记忆(LSTM: Long Short-Term Memory Neural Network)神经网络构建油田注水预测模型。 该模型不仅能考虑到注水量和影响因素之间的联系, 还能兼顾注水量随时间变化的趋势和前后关联。 以国内某复杂断块油藏的注水预测为例, 建立 LSTM 注水预测模型, 对某单井一时段的注水量进行了预测, 并与传统 RNN 建立的预测模型进行了对比。 实验结果显示, 该模型有着更为理想的预测效果, 预测精度较高, 能有效地提高油田注水预测的准确性。

关键词: 注水预测, 长短期记忆神经网络, 循环神经网络, 人工智能

Abstract: In order to solve the problem that the commonly used artificial intelligence water-flood prediction fails to consider the correlation of data in time, by selecting an improved LSTM ( Long Short-Term Memory Neural Network) based on RNN (Recurrent Neural Network) the oilfield water injection prediction model is built. The model can take into account the relationship between the historical water injection volume and the influencing factors, and take into account the trend and correlation of water injection volume with time. Taking the water injection prediction of a complex fault-block reservoir in China for example, the LSTM water injection prediction model is established to predict the water injection volume of a single well in a period of time, and compared with the prediction model established by traditional RNN. The experimental results show that the model has more ideal prediction effect and higher prediction accuracy, which can effectively improve the accuracy of oil field water injection prediction.


Key words: water-flood prediction, long short-term memory neural network, recurrent neural network;artificial intelligence

中图分类号: 

  • TP183