林年添^1,3,,
张栋^2,,,
张凯¹,
王守进⁴,
付超⁵,
张建彬⁶,
张冲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. 海底科学与探测技术教育部重点实验室, 中国海洋大学海洋地球科学学院, 山东青岛 266100
3. 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
4. 中国石化石油物探技术研究院, 南京 211103
5. 山东大学 岩土与结构工程研究中心, 济南 250061
6. 山东正元建设工程有限责任公司潍坊分公司, 山东潍坊 261000

基金项目: 国家科技发展计划863项目（2013AA064201，2012AA061202）与国家自然科学基金项目（41174098）联合资助


详细信息

作者简介: 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

通讯作者: 张栋, 男, 1991年生, 中国海洋大学博士研究生, 主要从事地震储层预测及海洋地震勘探方面的研究工作.E-mail:zhangdong_mail@163.com

中图分类号: P631

收稿日期:2018-07-06
修回日期:2018-09-06
上线日期:2018-10-05

Predicting distribution of hydrocarbon reservoirs with seismic data based on learning of the small-sample convolution neural network

LIN NianTian^1,3,,
ZHANG Dong^2,,,
ZHANG Kai¹,
WANG ShouJin⁴,
FU Chao⁵,
ZHANG JianBin⁶,
ZHANG Chong¹
1. College of Geological Sciences & Engineering, Shandong University of Science and Technology, Shandong Qingdao 266590, China
2. Key Lab of Submarine Geosciences and Prospecting Techniques, MOE and College of Marine Geosciences, Ocean University of China, Shandong Qingdao 266100, China
3. Laboratory for Marine Mineral Resources, Qingdao National Laboratory for Marine Science and Technology, Shandong Qingdao 266071, China
4. Institute of Geophysical Prospecting, SINOPEC Research Institute of Petroleum Exploration and Development, Nanjing 211103, China
5. Geotechnical and Structural Engineering Research Center, Shandong University, Jinan 250061, China
6. The Shandong Zhengyuan Construction Engineering, Shandong Weifang 261000, China


More Information

Corresponding author: ZHANG Dong,E-mail:zhangdong_mail@163.com

MSC: P631

--> Received Date: 06 July 2018
Revised Date: 06 September 2018
Available Online: 05 October 2018


摘要
摘要:地震储层预测是油气勘探的重要组成部分，但完成该项工作往往需要经历多个环节，而多工序或长周期的研究分析降低了勘探效率.基于油气藏分布规律及其在地震响应上所具有的特点，本文引入卷积神经网络深度学习方法，用于智能提取、分类并识别地震油气特征.卷积神经网络所具有的强适用性、强泛化能力，使之可以在小样本条件下，对未解释地震数据体进行全局优化提取特征并加以分类，即利用有限的已知含油气井段信息构建卷积核，以地震数据为驱动，借助卷积神经网络提取、识别蕴藏其中的地震油气特征.将本方案应用于模型数据及实际数据的验算，取得了预期效果.通过与实际钻井信息及基于多波地震数据机器学习所预测结果对比，本方案利用实际数据所演算结果与实际情况有较高的吻合度.表明本方案具有一定的可行性，为缩短相关环节的周期提供了一种新的途径.
关键词: 人工智能/
深度学习/
卷积神经网络/
卷积核/
地震数据驱动/
油气藏识别

Abstract:Prediction of reservoirs with seismic data plays an important role in oil and gas explorations. However, multiple stages or long periods of work are often required to complete prediction processes, which can lower the efficiency of exploration. To solve this problem, based on features of distribution of hydrocarbon reservoirs and their seismic response, this work introduces a convolution neural network depth learning to intelligently extract, classify and identify characteristics of oil and gas from seismic data. It makes it possible to globally extract and classify features from unexplained seismic data under small sample conditions with strong applicability and generalization of the convolutional neural network. That is, the convolution kernels are obtained from the known drilling data, and then which are used to recognize hydrocarbon characteristics from seismic data driven based on the convolution neural network. This scheme is applied to the checking of model data and actual data, and the expected results are achieved. It can be seen that the calculation results of this scheme have a high degree of agreement with the actual situation by comparing the actual drilling data and the predicted results with multi-component seismic data by machine learning. It shows that the scheme has certain feasibility and provides a new way to shorten the periods of relevant steps.
Key words:Artificial intelligence/
Deep learning/
Convolution neural network/
Convolution kernel/
Seismic data driven/
Oil and gas reservoir identification



PDF全文下载地址:

http://www.geophy.cn/data/article/export-pdf?id=dqwlxb_14718