天然气勘探与开发2025，Vol.48Issue(6)：104-117,14.DOI:10.12055/gaskk.issn.1673-3177.2025.06.009
基于Transformer模型的页岩气井积液预测方法
A Transformer-model-based method for predicting liquid loading in shale gas wells
摘要
Abstract
The large number of wells in shale gas fields and frequent liquid loading in individual wells bring challenges to refined production management.To address the challenges,and improve the timeliness and accuracy of liquid loading prediction,this study takes shale gas wells in the southern Sichuan Basin as an example.Based on the Transformer model,a triple-attention mechanism prediction method,comprising data dimensionality reduction,feature fusion,and threshold reconstruction,is established by integrating production time-series data and geological engineering parameters.The model was trained using data from 44 wells in shale gas reservoirs of Changning block in the basin,and subsequently validated on 14 wells.The following results are obtained.(i)Compared with the classical critical liquid-carrying model,the Transformer model can more effectively capture long-range dependencies in multi-year production time-series data of shale gas wells,overcoming the limitations of traditional prediction methods in timeliness,accuracy,and convenience.(ii)The anomaly detection technique based on the Encoder-Decoder framework and dynamic threshold reconstruction error can predict liquid loading 10 days in advance,achieving an accuracy of 83.3%on the test set.(iii)The Transformer model demonstrates excellent engineering applicability,with rapid training convergence and a single inference time of only 0.1 s,facilitating refined management on large-scale well cluster.(iv)A"cloud-edge collaborative architecture"enables real-time processing of production data and model computation,making it promising for industrialization.It is concluded that the Transformer-model-based method for predicting liquid loading in shale gas wells offers a novel solution for tracking and early warning of liquid loading,and refined production management in large-scale development of shale gas fields in the southern Sichuan Basin.关键词
页岩气井/积液/时序数据/机器学习/Transformer/精细化采气管理Key words
Shale gas well/Liquid loading/Time-series data/Machine learning/Transformer/Refined production management分类
能源科技引用本文复制引用
刘军,任静思,胡南,李琴,段洋,孔德蔚然,罗昱暄..基于Transformer模型的页岩气井积液预测方法[J].天然气勘探与开发,2025,48(6):104-117,14.基金项目
中国石油天然气集团有限公司科研项目"页岩气规模增储上产与勘探开发技术研究"(编号:2023ZZ21). （编号:2023ZZ21）
