Multiphase Flow Modeling Using Process Knowledge Integrating Temporal Graph Convolution Network

The accurate description of the flow characteristics for a multiphase pump is highly desirable but still intractable. A process knowledge integrated with temporal graph convolution network (PKTGCN) is proposed to depict multiphase flow characteristics of a whole discharge process. The process knowledge of multiphase pumps is first adopted to recognize three stages of the multiphase flow. Subsequently, each stage is elaborately redescribed using its individual model, respectively. Consequently, different flow characteristics can be captured suitably with a divide-and-conquer strategy. Experimental results show that this approach can enhance the prediction performance of the model built with limited samples. Particularly, the interpretability of the proposed model is demonstrated to help engineers in exploring the inner transportation mechanism of the multiphase pump.