基于CNN-LSTM-Attention的气井井筒积液诊断  

Diagnosis of Wellbore Fluid Accumulation in Gas Well based on CNN-LSTM-Attention

作　　者：徐子鸿 王仪 XU Zihong;WANG Yi(School of Computer Science and Engineering,Sichuan University of Science and Engineering,Yibin 644000,China)

机构地区：[1]四川轻化工大学计算机科学与工程学院,四川宜宾644000

出　　处：《成都工业学院学报》2025年第1期14-20,共7页Journal of Chengdu Technological University

基　　金：四川省科学技术厅项目(2019YFG0200);泸州市科学技术局重点研发项目(2021-GYF-4)。

摘　　要：为了解决传统积液诊断模型存在的诸多问题,如选择缺乏定性标准、计算结果差异大以及无法满足实际工程需求等,提出一种基于神经网络的气井井筒积液诊断方法,该模型将卷积神经网络(CNN)和长短期记忆网络(LSTM)的结合,使其能够有效捕捉气井在不同工况下的动态特征,增强模型对复杂数据的处理能力,并在此基础上引入注意力机制自动聚焦于输入数据中最相关的信息,从而提升特征的权重。在实验中,使用真实气井生产相关数据集,对比分析多个模型与所提出的CNN-LSTM-Attention模型的相关性能指标。实验结果显示,所提模型的准确率高达97.6%,多次试验结果方差值明显优于其他深度学习模型和传统方法。这一显著的性能提升,验证了模型的有效性,并对气田生产具有一定的指导作用。To address the various issues present in traditional fluid accumulation diagnosis model,such as the lack of qualitative criteria for selection,significant differences in calculation results and the inability to meet the actual engineering requirements,a neural network-based method for diagnosing wellbore fluid accumulation in gas well was proposed,which combines convolutional neural network(CNN)and long short-term memory network(LSTM)to effectively capture the dynamic characteristics of gas wells under different working conditions,and enhances the model's ability to process complex data.On this basis,the attention mechanism was introduced to automatically focus on the most relevant information within the input data,so as to improve the weight of features.In the experiment,the real gas well production related data-set was used,and the relevant performance metrics of several models and the proposed CNN-LSTM-Attention model were compared and analyzed.Experimental results show that the accuracy of the proposed model is as high as 97.6%,and the variance value of multiple experiments is significantly better than other deep learning models and traditional methods.This significant performance improvement not only verifies the effectiveness of the model,but also has a certain guiding role in gas field production.

关 键 词：卷积神经网络 长短期记忆网络 注意力机制 气井积液诊断 

分 类 号：TE377[石油与天然气工程—油气田开发工程]