基于LOFUnet深度卷积神经网络低序级断层多属性识别方法  

Multi-attribute recognition method for low-order faults based on LOFUnet deep convolutional neural network

作　　者：马玉歌 苏朝光[1] 丁仁伟 颜世磊[1] 张玉洁 韩天娇 闫绘栋 MA Yuge;SU Chaoguang;DING Renwei;YAN Shilei;ZHANG Yujie;HAN Tianjiao;YAN Huidong(Geophysical Research Institute,Shengli Oilfield Company,SINOPEC,Dongying 257022,China;College of Earth Sciences and Engineering,Shandong University of Science and Technology,Qingdao 266590,China)

机构地区：[1]中国石化胜利油田分公司物探研究院,东营257022 [2]山东科技大学地球科学与工程学院,青岛266590

出　　处：《物探化探计算技术》2024年第3期272-283,共12页Computing Techniques For Geophysical and Geochemical Exploration

基　　金：中国石化胜利油田分公司项目(YKY2405)。

摘　　要：低序级断层控制圈闭及油气富集,对油气勘探开发具有重要的意义。但其识别描述难度大、效率低,严重制约了该类油藏的勘探开发进程。随着人工智能的发展,深度学习为低序级断层识别提供了新的途径。这里在样本集构建及方法上都有创新之处:建立了同相轴错动、扭动、微扭动地震响应特征的低序级断层样本集,为智能识别奠定了良好的样本库;LOFUnet网络是在UNet基础上进行的改进,可以获取样本中更多低序级断层信息的特征。笔者通过方差属性、倾角属性和振幅属性融合获得新的断层体,用构建的LOFUnet网络进行低序级断层识别。网络通过残差块构建编码器端可以获取更多的低序级断层特征,解决梯度消失问题,提高模型的收敛速度,增强模型的稳定性以及低序级断层检测的精度和效率。选用正演模拟数据和实际地震数据分别对UNet模型、LOFUnet模型进行测试,结果表明,基于LOFUnet深度卷积神经网络低序级断层多属性识别方法提取的信息更加丰富,提高了低序级断层识别的准确度。Low-order faults control traps and hydrocarbon enrichment,which are significant for oil and gas exploration and development.However,its identification and description are complicated and inefficient,which seriously restricts such reservoirs'exploration and development process.With the development of artificial intelligence,deep learning provides a new way to identify low-order faults.LOFUnet network is an improvement based on UNet,which can obtain more features of low-order fault information in the sample.In this paper,a new fault body is obtained through the fusion of variance attribute,dip attribute,and amplitude attribute,and the LOFUnet network is constructed to identify low-order faults.The network in this paper can obtain more low-order fault features at the encoder end,solve the problem of gradient disappearance,improve the model's convergence speed,enhance the model's stability,and improve the accuracy and efficiency of low-order fault detection.The forward simulation and actual seismic data are used to test the UNet and LOFUnet models,respectively.The results show that the multi-attribute recognition method of low-order faults based on the LOFUnet depth convolution neural network can extract more information and improve the accuracy of low-order fault recognition.

关 键 词：低序级断层 Unet网络 LOFUnet网络 多属性识别 模型试算 

分 类 号：P618.13[天文地球—矿床学]