基于深度学习的页岩气藏压裂缝网反演方法研究  

Research on deep learning-based fracture network inversion method for shale gas reservoirs

作　　者：陈维铭 蒋琳 罗彤彤 李悦 汪健华 CHEN Weiming;JIANG Lin;LUO Tongtong;LI Yue;WANG Jianhua(Chongqing Shale Gas Exploration and Development Co.,Ltd.,Chongqing 401121,China;Optical Science and Technology(Chengdu)Ltd.,Chengdu,Sichuan 611700,China;Geological Exploration and Development Research Institute of CNPC Chuanqing Drilling Engineering Co.,Ltd.,Chengdu,Sichuan 610056,China)

机构地区：[1]重庆页岩气勘探开发有限责任公司,重庆401121 [2]中油奥博(成都)科技有限公司,四川成都611700 [3]中国石油川庆钻探工程有限公司地质勘探开发研究院,四川成都610056

出　　处：《油气藏评价与开发》2025年第1期142-151,160,共11页Petroleum Reservoir Evaluation and Development

基　　金：国家自然科学基金项目“一种复杂缝网的能量断裂准则及其在致密砂岩压裂模拟中的应用”(11672333)。

摘　　要：页岩气储集层致密性强,非均质性显著,自然产量极低,必须采用水力压裂技术进行增产改造才能获得工业气流,而评估压裂作业成效及优化工艺参数的关键在于获取准确的压裂缝网参数。传统裂缝监测技术(如微地震监测)费用高昂,无法实现井区全覆盖监测,而数值模拟预测模型需要大量的工程地质参数,导致地质资料不完整或缺失井段预测效果不佳,亟须一种经济高效地获取缝网参数的新方法。为此,提出一种基于深度学习的页岩气藏压裂缝网反演方法,其核心是以现场施工压裂曲线数据为基础,对压裂曲线特征参数进行量化分析,以缝网参数的强相关性指标作为输入,以微地震监测缝网参数(包括缝网长度、宽度、高度、体积)作为目标输出,建立BP(误差反向传播)神经网络反演模型,实现压裂缝网参数精确反演。根据渝西地区页岩气井现场450段压裂曲线,对模型进行了训练和参数优化,测试集缝网参数反演结果平均相对误差低于15%,验证了这种新方法对页岩气藏压裂缝网反演的可行性。Shale gas reservoirs are characterized by high compactness and significant heterogeneity,with naturally low production that necessitates hydraulic fracturing technology for enhanced productivity to achieve industrial gas flow.The key to evaluating the effectiveness of fracturing operations and optimizing process parameters lies in obtaining accurate fracture network parameters.Traditional fracture monitoring techniques,such as microseismic monitoring,are costly and cannot achieve full coverage monitoring of well areas.Numerical simulation prediction models require a large number of engineering geological parameters,leading to poor prediction effects for geological data that are incomplete or missing well sections.There is an urgent need for a new method that is economically efficient in obtaining fracture network parameters.To address this,a shale gas reservoir fracture network inversion method based on deep learning was proposed.The core of this method is to quantitatively analyze the fracturing curve characteristic parameters based on the site fracturing curve data,using strongly correlated indicators of fracture network parameters as inputs and microseismic monitoring fracture network parameters(including length,width,height,and volume)as target outputs.A back-propagation(BP)neural network inversion model was established to achieve accurate inversion of fracture network parameters.The model was trained and optimized using 450 fracturing curve segments from shale gas wells in western Chongqing,with the average relative error of fracture network parameter inversion results in the test set being below 15%,which verified the feasibility of this new method for inversion of shale gas reservoir fracture networks.

关 键 词：页岩气 压裂曲线 缝网参数预测 BP神经网络 反演 

分 类 号：TE377[石油与天然气工程—油气田开发工程]