西湖凹陷平北斜坡带ResU-Net复杂断裂识别技术及应用  

ResU-Net Complex Fault Identification Technology and Application in Pingbei Slope Zone of the Xihu Depression

作　　者：王红岩 赵洪 蔡坤 邵宇蓝 赵天亮 杨敏 Wang Hongyan;Zhao Hong;Cai Kun;Shao Yulan;Zhao Tianliang;Yang Min(Shanghai Branch Institute,CNOOC Co.,Ltd.,Shanghai 200030,China)

机构地区：[1]中海石油(中国)有限公司上海分公司研究院,上海200030

出　　处：《工程地球物理学报》2025年第2期161-171,共11页Chinese Journal of Engineering Geophysics

基　　金：“十四五”国家重大科技项目(编号:KJGG-2022-0302)。

摘　　要：平北斜坡带似花状断裂发育区,断层数量多、类型广、产状变化快,断层之间相互切割、搭接成圈,进而导致构造圈闭落实难度极大。针对以上问题,本文利用基于ResU-Net混合神经网络的断层智能预测技术开展复杂断层识别。首先,利用人工合成的带噪音断层样本训练ResU-Net神经网络模型,形成断层预测初始模型。然后,利用ResU-Net模型对实际地震数据进行去噪处理,提高资料信噪比,增强断裂反射特征。最后,使用少量人工解释的断层标签对神经网络进行迁移学习,以增强网络的泛化能力,优化网络的识别结果,在此基础上完成断层预测工作。实践证明,ResU-Net混合神经网络、预训练模型、迁移学习是智能断层预测技术的核心,该技术有效地解决了似花状断层预测的难题,预测吻合率由73%提高到92%,工作效率提高了95%。The Pingbei Slope is a flower-shaped fault development area characterized by a large number of faults,diverse types,and rapid changes in trend.The mutual cutting and overlapping of faults make it extremely difficult to implement structural traps.In response to the above issues,this article utilizes intelligent fault prediction technology based on the ResU-Net hybrid neural network to carry out complex fault identification.Firstly,artificially synthesized noisy fault samples are used to train the ResU-Net neural network model,forming an initial model for fault prediction.Then,the ResU-Net model is used to denoise the actual seismic data to improve the signal-to-noise ratio of the data and enhance the fault reflection characteristics.Finally,transfer learning is performed on the neural network using a small amount of manually interpreted fault labels.The purpose is to enhance the generalization ability of the network,optimize the recognition results of the network,and finally complete the fault prediction work.Practice has proven that the ResU-Net neural network,pre-trained model,and transfer learning are the core of intelligent fault prediction technology.This technology effectively solves the problem of flower-shaped fault prediction,with a prediction accuracy rate increased from 73%to 92%and work efficiency improved by 95%.

关 键 词：平北斜坡带 复杂断裂 ResU-Net 深度学习 

分 类 号：P631.4[天文地球—地质矿产勘探]