合肥工业大学校徽

合肥工业大学学报自科版

合肥工业大学学报(自然科学版)

JOURNAL OF HEFEI UNIVERSITY OF TECHNOLOGY (NATURAL SCIENCE)

主管：中华人民共和国教育部

主办：合肥工业大学

ISSN 1003-5060, CN 34-1083/N

基于双分支卷积神经网络的气水动态分析

Gas-water dynamic analysis based on two-branch convolutional neural network

期刊信息

合肥工业大学（自然科学版），2025年6月，第48卷第6期：828-832，838

DOI: 10.3969/j.issn.1003-5060.2025.06.018

作者信息

李道伦，吕茂春，查文舒，沈路航

（合肥工业大学数学学院，安徽合肥230601）

摘要和关键词

摘要: 传统气水分析方法主要有数值分析方法、实验模拟法等,但数值分析方法需要大量难以测量的数据,实验模拟法难以表征复杂油气开发现场,为此,文章基于深度学习神经网络提出一种新的气水分析方法。该方法根据气水动态物理模型建立的双分支卷积神经网络分别对排水井和产气井进行建模,个性化表征生产井和排水井的动静态数据;将产气井的气水产量数据作为输出,实现井组动态耦合关联,建立井组气水动态分析的深度学习网络模型。主动排水井组动态生产数据分析表明,该双分支卷积神经网络可实现3口生产井的日产气量和日产水量的高质量预测,揭示了主动排水井组中的复杂关联,可进行气水关系动态分析,从而为油气藏工程师提供了一种方便快捷的分析方法。

关键词: 气水动态分析；主动排水井组；双分支卷积神经网络；日产气；日产水

Authors

LI Daolun, LYU Maochun, ZHA Wenshu, SHEN Luhang

(School of Mathematics, Hefei University of Technology, Hefei 230601, China)

Abstract and Keywords

Abstract: Traditional gas-water analysis methods mainly include numerical analysis methods and experimental simulation methods, but numerical analysis methods require a large amount of difficult-to-measure data, and experimental simulation methods are difficult to characterize complex oil and gas development sites. For this reason, this paper proposes a new gas-water analysis method based on deep learning neural networks. Based on the dynamic physical model of gas and water, a two-branch convolutional neural network is established to model drainage well and production well separately, and personalize the dynamic and static data of production and drainage wells; the gas and water production data from the production wells are then used as the output to achieve dynamic coupling and correlation of the well group, thereby building a deep learning network model for the dynamic analysis of gas and water in the well group. The analysis of the dynamic production data of the active drainage well group shows that the two-branch convolutional neural network can achieve high-quality prediction of daily gas and water production of the three production wells. This suggests that the two-branch convolutional neural network can reveal complex correlations of wells in active drainage well group and perform dynamic analysis of gas-water relationships, thus providing a convenient and fast analysis method for oil and gas reservoir engineers.

Keywords: gas-water dynamic analysis; active drainage well group; two-branch convolutional neural network; daily gas production; daily water production

基金信息

国家自然科学基金资助项目（12172115）

前一篇后一篇

返回期刊目录