基于卷积-长短记忆神经网络的页岩气井短期产量预测与概率性评价  

Short-term Production Prediction and Probability Assessment of Shale Gas Wells Based on Convolution Long Short-term Memory Neural Network

作　　者：郭建春[1,2] 任文希 曾凡辉[1,2] 刘彧轩 段又菁[1,2] 罗扬 GUO Jianchun;REN Wenxi;ZENG Fanhui;LIU Yuxuan;DUAN Youjing;LUO Yang(Southwest Petroleum University,Chengdu,Sichuan 610500,China;State Key Laboratory of Oil and Gas Reservoir Geology and Exploitation,Chengdu,Sichuan 610500,China;Zhenhua Oil Co.,Ltd.,Beijing 100031,China)

机构地区：[1]西南石油大学 [2]油气藏地质及开发工程全国重点实验室 [3]振华石油控股有限公司

出　　处：《钻采工艺》2025年第1期130-137,共8页Drilling & Production Technology

基　　金：国家自然科学基金面上项目“大数据驱动的深层页岩压裂参数协同优化与实时调控研究”(编号:52374045);四川省自然科学基金项目“深层页岩储层多簇射孔压裂竞争扩展多目标协同智能优化与调控”(编号2023NSFSC0424)。

摘　　要：页岩气赋存方式多样、渗流机理复杂,气井生产制度多变,准确预测页岩气井产量难度大。针对这一问题,文章基于数据驱动的思想,对历史生产数据进行了预处理,建立了由产量、油嘴尺寸、生产时间和关井时间组成的多维时间序列,结合卷积神经网络(CNN)和长短记忆神经网络(LSTM),基于混合式深度学习架构,建立了基于卷积-长短记忆神经网络的页岩气井短期产量预测模型(CNN-LSTM)。CNN-LSTM采用CNN提取高维特征之间的交互作用信息,并利用LSTM提取这些特征的时序信息,实现了交互作用信息和时序信息的融合。生产数据测试表明:CNN-LSTM考虑了生产制度的影响,因此其产量预测精度高于单变量LSTM和多变量LSTM。进一步发展了基于核密度估计理论的产量概率性预测方法,实现了产量预测结果的不确定分析,获得了未来气井产量的变化范围。研究成果有望为页岩气井生产动态分析、产量预测和生产管理提供支撑。It is difficult to predict the shale gas well production accurately because of variety of shale gas occurrence modes,and complexity of permeation mechanism and variability of gas well production system.To address this issue,based on data-driven idea,the historical production data were preprocessed,and a multidimensional time series composed of production rates,choke sizes,production times,and shut-in times established,combined with Convolutional Neural Network(CNN)and Long Short-term Memory Neural Network(LSTM),and based on the hybrid deep learning architecture,a short-term production prediction model for shale gas wells based on Convolutional Long Short-Term Memory Neural Networks(CNN-LSTM)was developed.Production data test shows that CNN-LSTM takes into account the influence of production system,so its yield prediction accuracy is higher than univariate LSTM and multivariable LSTM.Additionally,a probability prediction method based on kernel density estimation theory has been further developed,enabling uncertainty analysis of the prediction results and providing a range for future well production rates.The research findings are expected to support production forecasting,analysis and production management of shale gas wells.

关 键 词：页岩气井 产量预测 神经网络 不确定分析 数据驱动 

分 类 号：TE328[石油与天然气工程—油气田开发工程] TP183[自动化与计算机技术—控制理论与控制工程]