融合注意力机制的Bi-LSTM页岩气产能预测方法  

Bi-LSTM Shale Gas Productivity Prediction Method Integrating Attention Mechanism

作　　者：朱仲义 刘洪[2] 岳圣杰 周鸿 朱怡晖 文宏川 ZHU Zhongyi;LIU Hong;YUE Shengjie;ZHOU Hong;ZHU Yihui;WEN Hongchuan(Shale Gas Research Institute of PetroChina Southwest Oil and Gas Field Company,Chengdu 610051,China;School of Safety Engineering,Chongqing University of Science and Technology,Chongqing 401331,China;Shunan Gas of PetroChina Southwest Oil and Gas Field Company,Luzhou Sichuan 646001,China)

机构地区：[1]中石油西南油气田分公司页岩气研究院,成都610051 [2]重庆科技学院安全工程学院,重庆401331 [3]中石油西南油气田分公司蜀南气矿,四川泸州646001

出　　处：《重庆科技学院学报（自然科学版）》2023年第3期52-57,共6页Journal of Chongqing University of Science and Technology：Natural Sciences Edition

基　　金：重庆市自然科学基金重点项目“深层页岩气压裂复杂裂缝形成机理及工艺技术研究”(CSTC2020JCYJ-ZDXMX0001)。

摘　　要：传统的经验公式法和数学解析法对页岩气产能预测的误差较大,为此,提出一种融合注意力机制的双向长短时记忆神经网络(Attention-Bi-LSTM)对页岩气产能进行预测。在Bi-LSTM的网络层架构中添加注意力机制后,神经网络输入子集的能力、模型处理时间序列数据的能力均得到了提升。采用该模型对页岩气井的日产气量进行预测,相对于单一的Bi-LSTM、LSTM,此模型的预测效果更好。The traditional empirical formula method and mathematical analysis method have large errors in shale gas productivity prediction.Therefore,a bidirectional long short memory neural network(Attention-Bi-LSTM)integrating attention mechanism is proposed to predict shale gas productivity.After the attention mechanism module is added to the network layer architecture of Bi-LSTM,the ability of the neural network input subset and the ability of the model to process time series data are improved.This model is used to predict the daily gas production of shale gas wells,which is better than the single Bi-LSTM and LSTM models.

关 键 词：页岩气产能预测 注意力机制 双向长短时记忆神经网络 

分 类 号：TE377[石油与天然气工程—油气田开发工程]