基于径向基—多层感知器神经网络联合的复杂岩相智能识别与表征  

Intelligent identification and characterization of complex lithofacies based on radial basis-multilayer perception neural network joint model

作　　者：姜世一 孙盼科[1,2] 张林 贾浪波 何太洪 徐怀民[1,2] 艾贝贝 张何锋 饶华文[5] 丁遥 JIANG Shiyi;SUN Panke;ZHANG Lin;JIA Langbo;HE Taihong;XU Huaimin;AI Beibei;ZHANG Hefeng;RAO Huawen;DING Yao(College of Geosciences,China University of Petroleum(Beijing),Beijing 102249,China;State Key Laboratory of Petroleum Resources and Prospecting,China University of Petroleum(Beijing),Beijing 102249,China;CNPC Xibu Drilling Engineering Company Limited,Urumqi,Xinjiang 830011,China;Exploration and Development Research Institute,PetroChina Changqing Oilfield Company,Xi'an,Shaanxi 710018,China;PetroChina Tarim Oilfield Company,Korla,Xinjiang 841000,China;Gas Company,PetroChina Southwest Oil&Gasfield Company,Chengdu,Sichuan 610051,China)

机构地区：[1]中国石油大学(北京)地球科学学院,北京102200 [2]“油气资源与探测”国家重点实验室·中国石油大学(北京) [3]中国石油西部钻探工程有限公司 [4]中国石油长庆油田公司勘探开发研究院 [5]中国石油塔里木油田公司 [6]中国石油西南油气田公司燃气分公司

出　　处：《天然气工业》2022年第9期47-62,共16页Natural Gas Industry

基　　金：国家自然科学基金青年科学基金项目“湖相云质岩致密油储层非均质性特征及其定量表征方法”(编号:41902125)。

摘　　要：苏里格气田东二区二叠系石盒子组盒8段(以下简称盒8段)为典型的河流相致密砂岩储层,其强非均质性及复杂的储层结构导致该区面临“甜点”储层优选困难等关键技术瓶颈。为此,在分析盒8段储层岩相类型及组合特征、岩相约束下测井数据特征的基础上,建立了一种契合岩相及其组合特征、测井数据特征、人工智能算法原理的径向基—多层感知器神经网络联合模型,并开展了储层岩相的精确识别与表征研究。研究结果表明:(1)盒8段发育块状层理砾岩相、槽状交错层理粗砂岩相、板状交错层理粗砂岩相、板状交错层理中砂岩相、平行层理中砂岩相、交错层理细砂岩相、波状层理粉砂岩相、块状层理泥岩相8种岩相类型;(2)盒8上亚段曲流河相储层岩相密度偏小、岩相频率偏高、对应测井数据分布较分散,盒8下亚段辫状河相储层岩相密度偏大、岩相频率偏低、对应测井数据分布较集中;(3)建立的径向基—多层感知器神经网络联合模型识别准确率可达89.06%,相较于单一神经网络模型、交会图、主成分分析和决策树等方法识别准确率明显提高。结论认为,建立的径向基—多层感知器神经网络联合模型不仅克服了现有岩相识别方法准确率低且难以推广的缺陷,而且对实现河流相强非均质性致密砂岩储层高效开发具有重要意义。The eighth member of Permian Shihezi Formation(hereinafter referred to as He8 Member) in the East Ⅱ Block of Sulige Gas Field is a typical fluvial sedimentary tight sandstone reservoir, and its strong heterogeneity and complex reservoir architecture lead to the key technical bottlenecks in the study area, such as difficult selection of sweet spot reservoir. To solve these key problems, this paper analyzes the characteristics of lithofacies and lithofacies combination and the logging response characteristics under the constraint of lithofacies in He8 Member, and then establishes a radial basis-multilayer perception neural network joint model that fits the characteristics of lithofacies and lithofacies combination,logging response characteristics and artificial intelligence algorithm principle. In addition, the accurate identification and characterization of reservoir lithofacies are carried out. And the following research results are obtained. First, there are eight lithofacies types in He8 Member, including massive bedding conglomerate facies, trough cross-bedding coarse sandstone facies, plate cross-bedding coarse sandstone facies, plate crossbedding medium sandstone facies, parallel bedding medium sandstone facies, cross-bedding fine sandstone facies, wavy bedding siltstone facies,and massive bedding mudstone facies. Second, the meandering river facies reservoir in the upper sub-member of He8 Memberis characterized by low lithofacies density, high lithofacies frequency and scattered logging data distribution, while the braided river facies reservoir in the lower submember of He8 Member is characterized by high lithofacies density, low lithofacies frequency and concentrated logging data distribution. Third,the identification accuracy of the radial basis-multilayer perception neural network joint model is 89.06%, which is much higher than that of single neural network model, cross plot, principal component analysis and decision tree. In conclusion, the radial basis-multilayer perception neural network joint mod

关 键 词：苏里格气田东二区 盒8段 河流相 致密砂岩储层 岩相类型 径向基—多层感知器神经网络 智能化 岩相识别 

分 类 号：TE37[石油与天然气工程—油气田开发工程]