生成对抗网络算法在电成像测井裂缝空白条带填充中的应用  

Application of a generative adversarial network algorithm to filling blank strips of fractures in formation microresistivity imaging images

作　　者：康正明 武辰升 杨国栋 武迪生 王瑞飞[5] 杨湘雨 甘伟 KANG Zhengming;WU Chensheng;YANG Guodong;WU Disheng;WANG Ruifei;YANG Xiangyu;GAN Wei(School of Electronic Engineering,Xi’an Shiyou University,Xi’an 710065,China;Sichuan Hengyi Petroleum Technology Service Company,Chengdu 610051,China;No.11 Oil Production Plant,Changqing Oilfield Company,Qingyang 745000,China;College of Petroleum Engineering,Xi’an Shiyou University,Xi’an 710065,China)

机构地区：[1]西安石油大学电子工程学院,陕西西安710065 [2]中国石油川庆钻探地质勘探开发研究院,四川成都610051 [3]四川恒溢石油技术服务有限公司,四川成都610051 [4]长庆油田分公司第十一采油厂,甘肃庆阳745000 [5]西安石油大学石油工程学院,陕西西安710065

出　　处：《煤田地质与勘探》2025年第3期197-207,共11页Coal Geology & Exploration

基　　金：国家自然科学基金项目(42304138);陕西省教育厅重点科学研究计划项目(22JY053);西安石油大学研究生创新与实践能力培养计划(YCS23215348)。

摘　　要：【目的】电成像测井仪器极板之间存在间隔,测得的井壁电阻率图像存在空白条带,对井壁附近裂缝参数评价影响较大。针对裂缝空白条带填充没有全井眼图像,填充质量评价难的问题,提出一种模拟数据与真实数据相结合构造数据集,基于生成对抗网络的电成像测井裂缝空白条带填充方法。【方法】首先,基于三维有限元法对裂缝地层的电阻率测井响应进行模拟。采用过渡边界条件代替实体裂缝的方法提升了多尺度模型的计算效率,满足了深度学习对样本的需求。其次,将模拟得到的裂缝与实际测井图像相结合构建了全井眼图像,用电成像测井图像制作的掩码作为空白条带。通过图像填充指标评价裂缝空白条带填充能力的方式优选超参数,获得最佳空白条带填充效果。最后,采用不同占比的掩码进行裂缝空白条带填充,并评价裂缝填充质量。【结果和结论】结果表明:模型能够在不同占比的空白条带情况下进行填充,对粗糙裂缝有较好的填充效果且裂缝形态平滑,能更准确地恢复边缘轮廓与细节。在真实测井数据上的填充进一步验证了其适用性,填充后的图像自然且能有效恢复裂缝特征,有助于裂缝提取及定量计算,为裂缝储层的精准评价提供基础,支撑油气产能的准确预测。[Objective]Gaps between the electrodes of formation microresistivity imaging(FMI)imagers lead to blank strips in the resistivity images of borehole walls,significantly influencing the parameter assessment for fractures near borehole walls.The absence of full-borehole images renders it challenging to assess the blank strip filling quality for fractures in FMI images.Using a dataset constructed utilizing both simulated and actual data,this study proposed a generative adversarial network(GAN)-based method for filling the blank strips of fractures in FMI images.[Methods]First,the resistivity logging responses of a fractured formation were simulated using the 3D finite element method.Actual fractures were simplified using the transition boundary condition,improving the computational efficiency by avoiding multi-scale models and meeting the requirements of deep learning for substantial samples.Second,full-borehole images were attained by integrating the simulated fractures with actual FMI images,and masks generated from FMI images were used as blank strips.To achieve the optimal filling effects,hyperparameters were optimized by assessing the capacity of blank strip filling for fractures using image filling indicators.Finally,the blank strips of fractures were filled using masks with varying proportions,followed by a filling quality assessment.[Results and Conclusions]The results indicate that the proposed method allows for the filling of blank strips with varying proportions.Notably,it yields encouraging filling effects and smooth morphology for coarse fractures and can restore their margins and details more accurately.Its applicability was further verified using actual log data,revealing that it produced natural-looking filled images that can effectively restore fracture features.The proposed method facilitates the extraction and quantitative analysis of fractures,laying a foundation for the precise assessment of fractured reservoirs and supporting the accurate prediction of hydrocarbon productivity.

关 键 词：电成像测井 空白条带填充 深度学习 生成对抗网络 有限元法 井壁裂缝 

分 类 号：P631[天文地球—地质矿产勘探]