基于深度学习的页岩孔隙类型自动识别方法  

Automated Identification Method of Shale Pore Types Based on Deep Learning

作　　者：毕飞宇 肖占山 张学忠 张元中[1] 赵建斌 方朝强 BI Feiyu;XIAO Zhanshan;ZHANG Xuezhong;ZHANG Yuanzhong;ZHAO Jianbin;FANG Chaoqiang(School of Geophysics,China University of Petroleum(Beijing),Beijing 102249,China;Geological Research Institute,China National Logging Corporation,Beijing 102200,China;Well Logging Key Laboratory,China National Petroleum Corporation,Xi’an,Shaanxi 710077,China;Suligenan Operation Branch,PetroChina Changqing Oilfield Company,Ordos,Inner Mongolia 017000,China)

机构地区：[1]中国石油大学(北京)地球物理学院,北京102249 [2]中国石油集团测井有限公司地质研究院,北京102200 [3]中国石油天然气集团有限公司测井重点实验室,陕西西安710077 [4]中国石油长庆油田公司苏里格南作业分公司,内蒙古鄂尔多斯017000

出　　处：《测井技术》2022年第4期439-445,共7页Well Logging Technology

基　　金：中国石油天然气集团有限公司科学研究与技术开发项目“测井数字岩石技术研究”(2021DJ4003)。

摘　　要：页岩孔隙研究对页岩油甜点预测和储层评价具有重要意义,与常规储层相比,页岩储层的孔隙类型更为多样,孔隙结构更为复杂,纳米尺度的孔隙广泛发育。目前,常规的岩石物理实验在页岩储层参数表征方面遇到困难,难以满足页岩等复杂岩石类型评价的需求。基于多分辨率的数字岩心技术,在数据规则化的基础上,利用高分辨率的数字岩心图像,采用深度学习算法,对页岩储层的孔隙类型进行自动智能识别。该算法识别精度达到0.65(mAP@0.5),极大提升了页岩孔隙类型识别的时效性,为非常规储层孔隙类型的表征提供了新的方法和手段。Shale pore research is important for shale oil dessert prediction and reservoir evaluation.Compared with conventional reservoirs,shale reservoirs have more diverse pore types and more complex pore structures,and nano-scale pores are widely developed.At present,conventional petrophysical experiments encounter difficulties in characterizing the parameters of shale reservoirs,and it is difficult to meet the needs of evaluation of complex rock types such as shales.Based on the multi-resolution digital core technology,the pore type of shale reservoir is automatically and intelligently identified by deep learning algorithm,which is based on the data regularization and using high-resolution digital core images.The recognition accuracy of the algorithm reaches 0.65(mAP@0.5),which greatly improves the timeliness of pore type identification of shale and provides a new method and means for the characterization of pore type of unconventional reservoirs.

关 键 词：测井解释 非常规储层 数字岩石物理 深度学习 卷积神经网络 孔隙识别 

分 类 号：P631.84[天文地球—地质矿产勘探]