付超^1,2,,
林年添^1,2,,,
张栋³,
文博⁴,
魏乾乾⁵,
张凯¹
1. 山东省沉积成矿作用与沉积矿产重点实验室, 山东科技大学地球科学与工程学院, 山东青岛 266590
2. 海洋国家实验室海洋矿产资源评价与探测技术功能实验室, 山东青岛 266071
3. 海底科学与探测技术教育部重点实验室, 中国海洋大学地球科学学院, 山东青岛 266100
4. 山东科瑞机械制造有限公司, 山东东营 257000
5. 山东省煤田地质局, 济南 250104

基金项目: 国家高技术研究发展计划（863）项目（2013AA064201，2012AA061202）和国家自然科学基金项目（41174098，41374126）联合资助


详细信息

作者简介: 付超, 男, 1992年生, 山东科技大学硕士研究生, 主要从事地震属性反演及储层预测方面的研究工作.E-mail:further_01@163.com

通讯作者: 林年添, 男, 1962年生, 博士, 教授, 主要从事地震信号处理、成像、反演方法及地质与地球物理综合研究.E-mail:linnt@sina.com

中图分类号: P631

收稿日期:2017-04-05
修回日期:2017-10-30
上线日期:2018-01-05

Prediction of reservoirs using multi-component seismic data and the deep learning method

FU Chao^1,2,,
LIN NianTian^1,2,,,
ZHANG Dong³,
WEN Bo⁴,
WEI QianQian⁵,
ZHANG Kai¹
1. Shandong Provincial Key Laboratory of Depositional Mineralization and Sedimentary Minerals, College of Geological Sciences and Engineering, Shandong University of Science and Technology, Shandong Qingdao 266590, China
2. Laboratory for Marine Mineral Resources, Qingdao National Laboratory for Marine Science and Technology, Shandong Qingdao 266071, China
3. Key Lab of Submarine Geosciences and Prospecting Techniques, MOE and College of Marine Geosciences, Ocean University of China, Qingdao 266100, China
4. Kerui Machinery Manufacturing Company Limited of Shandong, Shandong Dongying 257000, China
5. Shandong Bureau of Coal Geology, Ji'nan 250104, China


More Information

Corresponding author: LIN NianTian,E-mail:linnt@sina.com

MSC: P631

--> Received Date: 05 April 2017
Revised Date: 30 October 2017
Available Online: 05 January 2018


摘要
摘要:有机并有效利用纵波与转换横波在油气储层敏感度上存在的差异，有助于突出地震油气储层特征，有助于提高地震油气储层分布边界刻画的精度.基于此，本文设计了一种卷积神经网络与支持向量机方法相结合的多波地震油气储层分布预测的深度学习法（Deep Learning Method）.首先，利用莱特准则剔除所生成的多波地震属性中可能存在的异常值降低网络变体数量.然后，通过能突出多波地震油气储层特征的聚类算法和无监督学习算法构建隐藏层，用于增加网络共享，提取油气特征.最后，将增加网络罚值后的井点样本作为支持向量机预测的输入样本，以降采样后的C3卷积层属性作为学习集，进行从已知到未知的地震油气储层的预测.本方案应用于HG地区晚三叠统HGR组的碳酸盐岩油气储层预测，所预测的地震油气储层边界更加清晰，预测结果与实际情况基本吻合.应用结果表明：本论文方案不仅具有可行性，且具有有效性.
关键词: 多波地震/
卷积神经网络/
支持向量机/
深度学习/
油气储层预测

Abstract:The accuracy of forecasted characteristics and distribution of hydrocarbon reservoirs is a key in quantitative interpretation of seismic data, which can be improved by analyzing the sensitivity difference of compressive P and converted PS waves in seismic properties of reservoirs. Based on this consideration, this work proposed a deep learning method to predict the distribution and features of hydrocarbon reservoirs with a combination of a convolution neural network and a support vector machine. All possible outliers of multi-component seismic attributes, which might reduce the number of variants in the network, are eliminated firstly in accordance with the Wright criterion. Second, using the verified clustering and unsupervised algorithms, the hidden layer is constructed to enhance network sharing and extract hydrocarbon features. Finally, with the network penalized well-log data as the input of the support vector machine, the C3 convolution layer attributes after downsampling are adopted as the learning set to predict the desired hydrocarbon characteristics. This novel method has been applied to predict carbonate rock reservoirs of late Triassic in XGR Formation of the HG structure in an oil filed, resulting in enhanced resolution and improved lateral distribution of hydrocarbon reservoirs which coincides roughly with the real drilling data. The feasibility and effectiveness of this approach is corroborated by the application.
Key words:Multi-component seismic data/
Convolution neural network/
Support vector machine/
Deep learning/
Prediction of hydrocarbon distribution



PDF全文下载地址:

http://www.geophy.cn/data/article/export-pdf?id=dqwlxb_14317