本文已被:浏览次 下载次 |
码上扫一扫! |
|
基于小波变换和卷积神经网络的地震储层预测方法及应用 |
张国印1,2,王志章2,林承焰1,王伟方2,李令2,李诚1
|
(1. 中国石油大学(华东)地球科学与技术学院,山东青岛 266580;2.中国石油大学(北京)地球科学学院,北京102249)
|
|
摘要: |
提高储层预测的分辨率和准确性一直是油气藏表征的一个关键问题。将频谱分解与深度学习相结合,提出基于小波变换和卷积神经网络的地震岩性、储层类型预测方法。小波变换能够提供包含高频和低频信息的二维时频谱图,卷积神经网络具有超强的二维图像特征提取和分类能力,时频谱图作为卷积神经网络的输入,有助于充分挖掘地震数据高频和低频信息进行岩性和储层预测。将提出的方法应用于川西沙溪庙组储层预测中,首先利用叠后地震数据预测得到河道砂体分布,然后利用叠前地震数据在河道内部预测储层类型分布。结果表明,深度学习反演预测岩性和储层类型的分辨率和精度更高,能够识别小河道砂体,与生产测试情况更加吻合,优于常规地震反演方法。 |
关键词: 地震储层预测 岩性预测 深度学习 卷积神经网络 时频谱图 |
DOI:10.3969/j.issn.1673-5005.2020.04.010 |
分类号::P 631 |
文献标识码:A |
基金项目:国家自然科学基金项目(41772139);国家科技重大专项(2017ZX05009001,2017ZX05072);中国石油大学(华东)自主创新科研计划项目(20CX06053A) |
|
Seismic reservoir prediction method based on wavelet transform and convolutional neural network and its application |
ZHANG Guoyin1,2, WANG Zhizhang2, LIN Chengyan1, WANG Weifang2, LI Ling2, LI Cheng1
|
(1.School of Geosciences in China University of Petroleum(East China), Qingdao 266580, China;2. School of Geosciences in China University of Petroleum(Beijing), Beijing 102249, China)
|
Abstract: |
Improving the resolution and accuracy of seismic reservoir prediction has long been a key problem in reservoir characterization. A novel seismic reservoir prediction method based on wavelet transform and convolution neural network is proposed, which is driven by prestack seismic, poststack seismic and well logging data. Wavelet transform can obtain two-dimensional time-frequency spectrum including both high-frequency and low-frequency information. Convolutional neural network is good at feature extraction and classification of two-dimensional image-like data. Using the spectrum as the input of convolutional neural networks is helpful for making full use of high-frequency and low-frequency information of seismic data for lithology and reservoir prediction. The proposed method is applied to the reservoir prediction of the Shaximiao Formation in the Western Sichuan Basin. Firstly, the distribution of the channel sand body is predicted by poststack seismic data, and then the distribution of reservoir type is predicted by prestack seismic data within the channel. The results show that the proposed method has higher resolution and accuracy and can predict more small-scale channel sands than conventional inversion method. |
Key words: seismic reservoir prediction lithology prediction deep learning convolutional neural network time-frequency spectrum |