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0.999 9，平均绝对误差分别为0.000 5和0.000 4。与VAE-LSTM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 self-attention 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 0.999 9 and mean absolute errors of 0.000 5 and 0.000 4， 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.