基于数据驱动的水平井压裂裂缝扩展动态智能表征方法  

A Data-driven Dynamic Intelligent Characterization Method for Fracture Propagation in Horizontal Well Fracturi

作　　者：袁彬[1] 赵明泽 戴彩丽[1] 张伟[1] 吴淑红[2] 范天一 YUAN Bin;ZHAO Mingze;DAI Caili;ZHANG Wei;WU Shuhong;Fan Tianyi(School of Petroleum Engineering,China University of Petroleum(East China),Qingdao,Shandong 266580,China;PetroChina Research Institute of Petroleum Exploration&Development,Beijing 100083,China)

机构地区：[1]中国石油大学(华东)石油工程学院,山东省青岛市266580 [2]中国石油勘探开发研究院

出　　处：《钻采工艺》2025年第1期138-146,共9页Drilling & Production Technology

基　　金：国家重点研发计划项目“油/水/固界面浸润调控智能流体提高采收率关键材料与机理研究”(编号:2019YFA0708700);山东省泰山学者建设工程人才支持项目“复杂油藏动态监测与提高采收率”(编号:ZX20210178);中国石油天然气股份有限公司科技项目“油气勘探开发人工智能关键技术研究”(编号:2023DJ84)。

摘　　要：水力压裂是提高地质能源开发的一项关键技术,实现压裂裂缝扩展准确高效预测对地质能源的开发至关重要。深度学习方法为压裂裂缝扩展快速预测提供了新技术,但现有神经网络结构不适用于受多因素耦合影响的水力压裂场景。为实现压裂裂缝扩展的高效智能预测,文章耦合快速傅里叶算法、并行卷积层和U-Net框架,建立了AttFC-U-Net网络结构。基于储层参数非均质性和压裂设计等参数,Att-FC-U-Net能够高效预测水平井各压裂段中裂缝扩展的三维形态。模型评价指标交叉熵损失(CE)低于0.0001,F1分数超过0.93。研究结果表明,与数值模拟方法相比,Att-FC-U-Net在预测裂缝扩展方面表现出极强的学习性和高效性,为压裂智能化提供了新思路,有望成为辅助甚至替代数值模拟技术的新技术。Hydraulic fracturing is a key technology for enhancing geological energy development.Accurate and efficient prediction of fracture propagation is crucial for the development of geological energy resources.The deep learning provide a new technology for rapid prediction of fracture propagation,yet existing neural network structures are not suitable for the multi-factor-coupled hydraulic fracturing scenarios.To achieve efficient and intelligent prediction of fracture propagation,this study integrates the Fast Fourier Transform algorithm,parallel convolution layers,and the U-Net framework to establish the Att-FC-U-Net network structure.Based on reservoir parameter heterogeneity and fracturing design parameters,Att-FC-U-Net can efficiently predict the three-dimensional morphology of fracture propagation in various fracturing stages of horizontal wells.The model evaluation metrics show a cross-entropy loss(CE)below 0.000,1 and an F1 score exceeding 0.93.The results demonstrate that Att-FC-U-Net exhibits superior learning capability and efficiency in predicting fracture propagation compared to numerical simulation methods,providing a novel approach for the intelligentization of fracturing and holding potential as a supplementary or even alternative technology to numerical simulation techniques.

关 键 词：三维裂缝扩展 水平井压裂 天然裂缝 交叉熵损失 数据驱动 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TE357.1[自动化与计算机技术—控制科学与工程]