摘要
地震资料解释是油气勘探的关键环节之一,其成果直接服务于油气田的勘探开发.随着油田精细化勘探的需求不断加深,地震解释工作量逐年增加.常规的地震层位自动解释方法在面对复杂构造时存在解释精度较差,工作量大等问题,因此,为解决上述问题,本文创新性地将一种基于图像分割技术的U-Net网络应用于地震层位解释工作中.通过输入地震数据及少量人工解释的标签数据,利用该网络进行监督学习,多套层位同时训练建模,实现地震层位自动识别,并应用于海外Parihaka地震三角洲沉积地区和国内海域工区.实际工区应用表明该技术在多层识别模型中的性能稳定,多层同时识别准确率达到90%以上,与常规地震层位自动解释方法相比,基于U-Net卷积神经网络的智能算法在小层、弱层识别方面优势明显,同时具有较高的效率与准确性.
Seismic data interpretation is one of the key processes in oil and gas exploration,and its results directly serve the exploration and development of oil and gas fields.With the ever-increasing demand for refined oilfield exploration,the workload of seismic interpretation is increasing year by year.When faced with complex structures,there are some problems with the conventional automatic tacking techniques of horizon interpretation,such as poor interpretation accuracy and large workload.Therefore,in order to solve the above problems,this paper innovatively applies a U-Net network based on image segmentation technology to horizon interpretation.By inputting seismic data and a small amount of manually interpreted label data,the network is used for supervised learning,and multiple sets of horizons are trained and modeled at the same time to realize the automation of horizon interpretation.The technology is applied to the overseas Parihaka seismic delta depositional area and the domestic sea area.The application in the field data shows that the performance of this technology is stable,and the accuracy of multi-layer simultaneous recognition is more than 90%.Compared with the conventional automatic tacking techniques of horizon interpretation,the proposed method of intelligent algorithm based on U-Net convolution neural network has obvious advantages in the recognition of small and weak layers,and has higher efficiency and accuracy.
作者
朱振宇
黄小刚
丁继才
王清振
李超
ZHU ZhenYu;HUANG XiaoGang;DING JiCai;WANG QingZhen;LI Chao(CNOOC Research Institute Co.,Ltd.,Beijing 100028,China)
出处
《地球物理学进展》
CSCD
北大核心
2023年第4期1722-1738,共17页
Progress in Geophysics
关键词
层位解释
深度学习
多层识别
卷积神经网络
Horizon interpretation
Deep learning
Multi-layer recognition
Convolutional neural network
