| 本文已被：浏览次 下载次 |
码上扫一扫！ |
|
|
| 基于深度卷积生成对抗神经网络预测气窜方向 |
|
冯其红1,2,李玉润1,2,王森1,2,任佳伟1,2,周代余3,范坤3
|
|
(1.中国石油大学(华东)石油工程学院,山东青岛 266580;2.非常规油气开发教育部重点实验室(中国石油大学(华东)),山东青岛 266580;3.中国石油塔里木油田公司勘探开发研究院,新疆库尔勒 841000)
|
|
| 摘要: |
| 注气开发是目前油田开发最有效的EOR方法之一,但注气开发面临见气时间早、气体突进严重等一系列问题。通过气窜方向预测能够及时调整工作制度,避免问题发生。利用深度卷积对抗神经网络建立渗透率场和注气后气相饱和度分布的动态映射关系,通过输入渗透率场的数据进行图像映射,得到不同时间的气相饱和度分布,预测气窜方向。结果表明:深度卷积方法在提取渗透率特征方面表现出良好性能;采用图像的结构相似性指数(SSIM)作为检验指标,将用对抗神经网络方法建立的气相饱和度分布与商业数值模拟器预测结果进行对比,二者结构相似度大于0.9;深度卷积生成对抗网络(DC-GAN)能够有效地预测注入气体在油藏中的气窜方向。 |
| 关键词: 深度卷积 对抗神经网络 结构相似性指数 气相饱和度 |
| DOI：10.3969/j.issn.1673-5005.2020.04.003 |
| 分类号::TE 357.72 |
| 文献标识码:A |
| 基金项目:中国石油天然气集团有限公司重大科技项目(ZD2019-183-007) |
|
| Predicting gas migration development using deep convolutional generative adversarial network |
|
FENG Qihong1,2, LI Yurun1,2, WANG Sen1,2, REN Jiawei1,2, ZHOU Daiyu3, FAN Kun3
|
|
(1.School of Petroleum Engineering in China University of Petroleum (East China), Qingdao 266580, China;2.Key Laboratory of Unconventional Oil ＆ Gas Development(China University of Petroleum(East China)), Ministry of Education, Qingdao 266580, China;3.Research Institute of Exploration ＆ Development, PetroChina, Tarim Oilfield Company, Korla 841000, China)
|
| Abstract: |
| Gas flooding is one of the most effective field development EOR methods, but it is faced with a series of problems, such as seeing gas early and premature gas breakthrough. Through the gas channeling prediction, we aim to adopt a proper working strategy to avoid that problem. A deep convolutional generative adversarial network model was formulated to build up the dynamic mapping between permeability and gas saturation distribution after injection gas. The gas saturation distribution at different time was obtained by image mapping with the input data of the permeability field, then the gas channeling can be predicted. The results indicate that the deep convolutional generative adversarial network(DC-GAN) method has superior performance in permeability feature representations. Using the structural similarity index (SSIM) as a test index, comparing the mapping relationship gained from DC-GAN and the results calculated from the commercial reservoir simulator, the SSIM is bigger than 0.9. The DC-GAN model can effectively predict the channeling direction of gases injected in reservoir. |
| Key words: deep convolution adversarial neural network structural similarity index gas saturation |
