储层粒度神经网络预测模型研究  

Research on Neural Network Prediction Model of Reservoir Particle Size

作　　者：王利华[1] 楼一珊[1] 马晓勇[2] 程福山[1] 陈宇 

机构地区：[1]油气资源与勘探技术教育部重点实验室.长江大学,湖北荆州434023 [2]中国石化胜利石油工程有限公司,山东东营257064 [3]北京锦辉博泰科技有限公司,北京昌平102249

出　　处：《西南石油大学学报（自然科学版）》2016年第1期53-59,共7页Journal of Southwest Petroleum University(Science & Technology Edition)

基　　金：国家科技重大专项(2008ZX05056-002-03;2008ZX05024-003-01)

摘　　要：国内外多年的研究表明,储层粒度特征值(d_(50))、非均质系数(d_(40)/d_(90))是防砂设计的基础。常规获取粒度分布范围的方法主要有激光粒度测试法(LDA)与筛析法(SA),两种方法均需要通过岩芯粒度测试来获取数据,而在制定开发井的完井防砂措施时往往没有实际开采层位的岩芯,只能参照探井粒度数据进行设计,从而导致较大的误差。针对该问题,从测井的角度出发,开展了储层粒度与多种测井曲线的响应关系的研究,采用神经网络技术,建立了探井伽马、密度测井项与实测粒度特征值三者样本库,训练出满足工程需要的学习网络,进而结合开发井测井资料,获得了整个粒度纵向分布剖面,为防砂分层设计提供准确的基础数据支撑。目前,该方法在中国海上多个油田的分层防砂优化设计中获得了成功应用。预测误差可控制在10%以内。According to researches at home and abroad, sand control design is based on reservoir particle size characteristic value. LDA and SA are the conventional methods used to analyze particle size distributions. Both methods requires data through core particle size testing. But sand control design can only use test well data, because no core at actual producing position can be used when sand control measure is established, which can result in major errors. This article elaborates the relevance about median grain size and gamma ray logging or density logging through researches on reservoir particle size and variety of log curve response relation. And then through establishing sample pool of gamma ray logging or density logging and characteristic value, and by neural network technology we trained learning network satisfing engineering requirements. Then the particle size longitudinal distribution profile can be established according to development well logging data. This profile supplied data basis for sand control layering design. At present, this method has been successfully used in several Chinese offshore oil field in sand control optimization with errors below 10%.

关 键 词：分层防砂 粒度特征值 神经网络 伽马密度测井 样本库 

分 类 号：TE257[石油与天然气工程—油气井工程]