摘要
在湿气计量标准装置上,使用长喉径文丘里管在管道压力2.0~3.0 MPa条件下进行了多组单气相与湿气测试。使用卷积神经网络算法解析多维时序信号,建立了识别管道中单相/多相流动状态的高准确度神经网络算法。针对湿气流动在长喉径文丘里中的信号特点,基于虚高理论分别建立了气相与液相流量的迭代预测算法。由此搭建可对工业湿气流动进行实时识别并准确分类计量的湿气流量模型。模型检测结果显示:单气相时模型可显著降低由于流态误判导致的计量偏差(气相准确度提升达9%,液相修正误判达0.6 m^(3)/h),湿气条件下气相与液相流量预测的平均绝对百分比误差分别为4.9%与12.45%.
Multiple sets of single-phase and wet gas tests were conducted using a long-throat Venturi meter on a wet gas metering standard device under pipeline pressures of 2.0~3.0 MPa.A convolutional neural network(CNN)algorithm was employed to analyze multidimensional time-series signals,establishing a high-precision neural network algorithm for identifying single-phase/multiphase flow states in the pipeline.Based on the characteristics of signals in long-throat Venturi meters for wet gas flow,iterative prediction algorithms for gas and liquid flow rates were developed using the concept of virtual height.Consequently,a wet gas flow model capable of real-time identification and precise classification and measurement of industrial wet gas flow was constructed.The model testing results showed that for single-phase gas,the model significantly reduced measurement deviations caused by flow pattern misjudgment(improving gas phase accuracy by up to 9%and correcting liquid phase misjudgment by 0.6 m^(3)/h).Under wet gas conditions,the mean absolute percentage errors(MAPE)for gas and liquid flow rate predictions were 4.9%and 12.45%,respectively.
作者
于培宁
魏来
陆兴
李轶
邹海鑫
张强
YU Peining;WEI Lai;LU Xing;LI Yi;ZOU Haixin;ZHANG Qiang(Shenzhen Institute of Information Technology,Shenzhen,Guangdong 518172,China;Engineering Technology Branch,CNOOC Energy Development Co.,Ltd.Tianjin 300450,China;Petro China Xinjiang Oilfield Company,Karamay,Xinjiang 834000,China;Tsinghua Shenzhen International Graduate School,Shenzhen,Guangdong 518055,China;Research Institute of Natural Gas Technology,PetroChina Soutwest Oil&Gasfeld Company,Chengdu,Sichuan 610213,China)
出处
《计量学报》
北大核心
2025年第11期1574-1580,共7页
Acta Metrologica Sinica
基金
国家自然科学基金(61603207)。
关键词
湿气流量计量
多相流计量
长喉径文丘里管
卷积神经网络
虚高理论
流态判别
wet gas flow measurement
multiphase flow metering
lengthened throat Venturi
convolutional neural networks
virtual height theory
flow pattern discrimination
