| 本文已被：浏览次 下载次 |
码上扫一扫！ |
|
|
| 专注智能油藏储量预测的深度时空注意力模型 |
|
李宗民1,2,李亚传1,赫俊民3,张益政3,姚纯纯1,刘玉杰1
|
|
(1.中国石油大学(华东)计算机科学与技术学院,山东青岛 266580;2.中国石油大学胜利学院, 山东东营257061;3.中国石化胜利油田物探院, 山东东营 257022)
|
|
| 摘要: |
| 现有油藏储量预测方法的精度远不能满足实际应用的需求。受循环神经网络和注意力机制的启发,提出一种专注智能油藏储量预测的深度时空注意力模型。该模型通过时间注意力模型来捕获输入数据之间的关键信息,空间注意力模型捕获隐藏状态之间的关系紧密程度,能够缓解数据波动对预测结果的不利影响,从而大幅减小预测误差。结果表明,相比传统方法和已有的深度学习方法,该模型预测精度有显著提高,为今后油藏储量预测提供一种更优的选择。 |
| 关键词: 油藏储量预测 循环神经网络 注意力机制 深度时空注意力模型 |
| DOI：10.3969/j.issn.1673-5005.2020.04.009 |
| 分类号::TE 155 |
| 文献标识码:A |
| 基金项目:国家自然科学基金项目(61379106)；山东省自然科学基金项目(ZR2013FM036,ZR2015FM011)；国家重点研发计划(2019YFF0301800) |
|
| A deep spatio-temporal attention model focusing on intelligent reserve prediction |
|
LI Zongmin1,2, LI Yachuan1, HE Junmin3, ZHANG Yizheng3, YAO Chunchun1, LIU Yujie1
|
|
(1.School of Computer Science and Technology in China University of Petroleum(East China), Qingdao 266580, China;2.Shengli College of China University of Petroleum, Dongying 257061, China;3.Geophysical Research Institute of Shengli Oilfield, SINOPEC, Dongying 257022, China)
|
| Abstract: |
| The accuracy of existing reservoir reserve prediction methods is far from satisfying the needs of practical application. Inspired by recurrent neural network(RNN) and attention mechanism, this paper proposes a deep spatio-temporal attention model focusing on intelligent forecasting of reservoir reserves, in which the time attention model can capture the key information within the input data, and the spatial attention model can capture the link between the hidden states. Therefore, the adverse effect of data fluctuation can be mitigated on the prediction results to greatly reduce the prediction error. Experimental results on the real data of a large oil field show that compared with traditional methods and existing deep learning methods, the prediction accuracy of the model by this method is significantly improved. Our approach therefore provides a better choice for reserve prediction in the future. |
| Key words: reservoir reserve prediction recurrent neural network attentional mechanism deep spatio-temporal attention model |
