Accurate gas–water two-phase flow measurement is critical for monitoring and optimizing industrial production processes. In industrial sites, since the severe working conditions and the complex flow mechanism, accurately determining the volume flowrate of gas–water two-phase flow has perennially presented a challenge. Traditional methods for this issue can only work under specific conditions. In order to address this, we propose a method based on multi-sensor fusion and Convolutional Flowrate Attention Network (CFAN). The multi-sensor fusion strategy is to comprehensively capture the flow features, involving collecting data using a Venturi tube, a gradiometer, and Electrical Capacitance Tomography (ECT). In our case, using the multi-sensor fusion strategy, we collected the dataset from the Chuanqing Gas Field, which is located in Sichuan, China, during which the Gas Volume Fraction (GVF) ranges from 95.6% to 99.9%. Each sample in the dataset is a one hour sequence, and each frame in the sequence is a vector that contains measured data from sensors. The proposed CFAN is integrated with the Convolutional Block Attention Module (CBAM) on ResNet to improve the network’s flowrate feature extraction capability. ResNet combines the advantages of 1D and 2D convolutional layers, allowing the network to extract features that represent both the temporal evolution and the spatial distribution. Through comparative experiments, we found that both CBAM and multi-sensor fusion improved the accuracy of flowrate prediction, especially under low flowrate conditions. Through multi-sensor fusion based on CFAN, the gas phase flow accuracy achieved 95.1% at ± 10% confidence level and the water phase flow accuracy achieved 84.5%, which meets the industrial error standard. This confirms the superior performance and robustness of the proposed CFAN in industrial data of gas–water two-phase flow tasks.
Song et al. (Thu,) studied this question.