摘要
提出了一种自注意力机制-变分自编码器-长短期记忆网络(SA-VAE-LSTM)模型,用于实现气液两相流中气含率与气相流速的测量。首先,采用16电极阵列电导传感器实时采集流动信号,通过变分自编码器(VAE)对采集信号进行特征提取;随后引入并行自注意力机制对关键流动特征进行自适应增强;最后,利用长短期记忆网络(LSTM)对提取的时序特征建模,实现气含率和气相流速的测量。测试结果表明:SA-VAE-LSTM模型在两项预测任务中均取得优异表现,预测值与实测值的决定系数均为0.9999,平均绝对误差分别为0.0005和0.0004。与VAELSTM等基准模型相比,所提方法在特征表征与时序建模精度方面更具优势,显著提升了预测性能。
A self-attention variational autoencoder long short-term memory network(SA-VAE-LSTM)model is proposed for the measurement of gas volume fraction and gas velocity in gas-liquid two-phase flow.Firstly,the model utilizes a 16-electrode array conductivity sensor to acquire real-time flow signals.Secondly,a variational autoencoder(VAE)is employed to extract representative features from the multi-channel input signals,followed by a parallel selfattention mechanism to adaptively enhance key flow-related features.Finally,a long short-term memory(LSTM)network is used to capture the temporal dependencies of the extracted features,enabling accurate prediction of gas volume fraction and gas velocity.Experimental results demonstrate that the proposed SA-VAE-LSTM model achieves excellent performance in both prediction tasks,with coefficients of determination reaching 0.9999 and mean absolute errors of 0.0005 and 0.0004,respectively.Compared with baseline models such as VAE-LSTM,the proposed approach exhibits superior feature representation and temporal modeling capabilities,leading to significantly improved predictive accuracy.
作者
顾恬文
张立峰
GU Tianwen;ZHANG Lifeng(Department of Automation,North China Electric Power University,Baoding,Hebei 071003,China)
出处
《计量学报》
北大核心
2025年第11期1591-1597,共7页
Acta Metrologica Sinica
基金
国家自然科学基金(61973115)。
关键词
流量计量
气液两相流
SA-VAE-LSTM模型
阵列电导传感器
气含率
流速
测量
flow metrology
gas-liquid two-phase flow
SA-VAE-LSTM model
electrode array conductivity sensor
gas volume fraction
liquid velocity
measurement
