基于改进U-Net的多属性地震相识别方法  

Multi-attribute seismic facies identification method based on improved U-Net

作　　者：韩旭东[1,2] 张广智 周游[1,2] 闫凯[3] 张金淼 朱振宇 李超 HAN XuDong;ZHANG GuangZhi;ZHOU You;YAN Kai;ZHANG JinMiao;ZHU ZhenYu;LI Chao(Key Laboratory of Deep Oil and Gas,China University of Petroleum(East China),Qingdao 266580,China;School of Geosciences,China University of Petroleum(East China),Qingdao 266580,China;First Institute of Geographic Information Cartography,Ministry of Natural Resources,Xi'an 710054,China;CNOOC Research Institute Co.,Ltd.,Beijing 100028,China)

机构地区：[1]中国石油大学(华东)深层油气重点实验室,青岛266580 [2]中国石油大学(华东)地球科学与技术学院,青岛266580 [3]自然资源部第一地理信息制图院,西安710054 [4]中海油研究总院有限责任公司,北京100028

出　　处：《地球物理学进展》2024年第1期344-354,共11页Progress in Geophysics

基　　金：国家自然科学基金项目(U19B2008,U19B6003,42074136,41674130);国家科技重大专项(2016ZX05027004-001);中国石油前瞻性基础性项目(2021DJ0606)联合资助.

摘　　要：地震相识别技术是进行沉积环境分析与储层预测的有力工具.传统的人工地震相识别方法不仅工作量大,而且效率非常低.目前利用深度学习方法可以大幅度提高地震相识别的效率,但是受限于有限的数据集和网络提取特征能力,对样本数量少的地震相识别效果较差.针对上述问题,本文提出了基于改进U-Net的多属性地震相识别方法.首先通过弹性形变算法来扩增数据集,将经过属性选择后的多属性数据体作为输入数据,提高输入数据的数量和质量;其次通过引入注意力机制对网络提取的特征添加权重,提高U-Net网络提取特征的能力;并在损失函数中引入Dice指数,解决了样本不均衡问题.经过数值实验表明,基于改进U-Net模型可有效提高地震相预测准确率.Seismic facies identification technology is a powerful tool for sedimentary environment analysis and reservoir prediction.The traditional artificial Seismic facies identification method not only has a large workload but also has very low efficiency.At present,using the deep learning method can greatly improve the efficiency of Seismic facies identification,but limited by the limited data sets and network feature extraction ability,the recognition effect of seismic facies with a small number of samples are poor.To solve the above problems,a multi-attribute Seismic facies identification method based on an improved U-Net is proposed in this paper.Firstly,the elastic distortion algorithm is used to augment the data set,and the multi-attribute data volume after attribute selection is used as the input data to improve the quantity and quality of the input data.Secondly,the attention mechanism is introduced to add weight parameters to the features extracted by the network to improve the feature extraction ability of the U-Net network.And the Dice index is introduced into the loss function to solve the problem of sample imbalance.Numerical experiments show that the accuracy of seismic phase prediction can be effectively improved based on the improved U-Net model.

关 键 词：地震相识别 U-Net 地震属性 注意力机制 数据扩增 

分 类 号：P631[天文地球—地质矿产勘探]