基于FMI图像深度学习的砂砾岩体沉积微相识别方法——以东营凹陷北带Y920区块沙四上亚段为例  

A Method for Identifying Sedimentary Microfacies in a Sandy Conglomerate Body on Deep Learning of FMI Images:Case study of upper submember of the Fourth member,Shahejie Formation in Y920 block,northern zone,Dongying Sag

作　　者：罗歆 闫建平[1,2,3] 王军 耿斌[4] 王敏[4] 钟广海[5] 张帆 李志鹏[4] 高松洋[7] LUO Xin;YAN JianPing;WANG Jun;GENG Bin;WANG Min;ZHONG GuangHai;ZHANG Fan;LI ZhiPeng;GAO SongYang(State Key Laboratory of Oil&Gas Reservoir Geology and Exploitation,Southwest Petroleum University,Chengdu 610500,China;School of Geoscience and Technology,Southwest Petroleum University,Chengdu 610500,China;Key Laboratory of Tectonics and Petroleum Resources(China University of Geosciences),Ministry of Education,Wuhan 430074,China;Institute of Exploration and Development,Shengli Oil Field,SINOPEC,Dongying,Shandong 257015,China;Shale Gas Research Institute,PetroChina Southwest Oil and Gas Field Company,Chengdu 610500,China;Research Institute of Exploration and Development,PetroChina Xinjiang Oilfield Company,Karamay,Xinjiang 834000,China;Research Institute of Exploration and Development,PetroChina Daqing Oilfield Company,Daqing,Heilongjiang 163712,China)

机构地区：[1]油气藏地质及开发工程国家重点实验室(西南石油大学),成都610500 [2]西南石油大学地球科学与技术学院,成都610500 [3]中国地质大学构造与油气资源教育部重点实验室,武汉430074 [4]中国石化胜利油田分公司勘探开发研究院,山东东营257015 [5]中国石油西南油气田公司页岩气研究院,成都610500 [6]中国石油新疆油田公司勘探开发研究院,新疆克拉玛依834000 [7]中国石油大庆油田公司勘探开发研究院,黑龙江大庆163712

出　　处：《沉积学报》2023年第4期1138-1152,共15页Acta Sedimentologica Sinica

基　　金：国家科技重大专项(2017ZX05072-002,2017ZX05049-004);国家自然科学基金项目(41830431);中国石油—西南石油大学创新联合体科技合作项目(2020CX020000);高等学校学科创新引智计划(111计划)(D18016)。

摘　　要：砂砾岩体属于近源快速堆积,扇体多期叠置,岩性多样、非均质性强,常规测井曲线受粗碎屑岩石组构、流体复杂性的影响,往往对沉积微相识别有难度。以东营凹陷北带Y920区沙四上亚段砂砾岩体为例,将岩心刻度FMI(全井眼微电阻率扫描成像测井)图像,总结各沉积微相的FMI图像特征;分析不同沉积微相与岩性、物性、含油性的关系;利用灰度共生矩阵图像处理手段提取不同沉积微相FMI图像的对比度、相关度、角二阶矩、同质性4种纹理参数,将4种纹理参数与取心段不同沉积微相FMI图像分别作为K最近邻分类算法(KNN)和卷积神经网络(CNN)的学习样本,训练机器学习和深度学习网络来开展沉积微相的分类和识别。研究结果表明:Y920区沙四上亚段砂砾岩体可细分为扇根主水道、扇中辫状水道、扇中辫状水道间和扇端泥4种微相类型。其中扇中辫状水道为最优势微相,FMI图像大多具有亮色、块状且砾径分布较均匀的特征,其岩石分选性、物性、含油性较好,是有利的储集层发育带。KNN分类算法和CNN网络模型都可对沉积微相进行判识,相较于传统KNN机器学习,CNN深度学习模型识别微相误差更小、鲁棒性更高。FMI图像特征提取及CNN方法应用深化了砂砾岩体沉积微相的内部结构认识,为沉积微相精细刻画与有效储层预测提供了依据。Sandy conglomerate bodies are the result of rapid near-source accumulation of sediment,with multi-phase superposition of fan bodies,and diverse,non-homogeneous lithological properties.Conventional logging is adversely affected by coarse clastic rock formations and fluid complexity,which often hinder the identification of sedimentary microphases.In this study,core-scale full borehole micro-resistivity scanning imaging(FMI)of the sand and gravel body was used to summarize the properties of each sedimentary microphase.The relationships between microphases and lithology,physical properties and oil content were analyzed.The contrast,correlation,angular second-order moments,and the four texture parameters and FMI images of the core sections were used as learning samples for the knearest neighbors classification algorithm(KNN)and convolutional neural network(CNN).The machine-learning and deep-learning networks were trained to classifiy and identify sedimentary microphases.The results show that this sand and gravel body is divided into four microphase types:fan-root main channel,mid-fan braided channel,interfan braided channel and fan-end mud.The KNN classification algorithm and CNN network model were both able to identify the sedimentary microphases,but the CNN deep-learning model showed smaller error and greater robustness than traditional KNN machine learning.The application of FMI image-feature extraction together with CNN deepens the understanding of the internal structure of sedimentary microphases in gravels,and provides a basis for fine characterization of sedimentary microphases and effective reservoir prediction.

关 键 词：FMI图像 砂砾岩体 沉积微相 东营凹陷 深度学习 

分 类 号：P618.13[天文地球—矿床学]