Deep Canonical Variate Analysis with Interpretable Attribute Guidance for Three-Phase Flow Process Monitoring

Oil-gas-water three-phase flow has multiple flow states that exhibits dynamic, nonlinearity and instantaneity. Real-time monitoring and analysis of flow state are crucial for ensuring safe operation of industries. However, there is no universal definition for three-phase flow states due to complex flow characteristics and structures. Hence, labeling three-phase flow states by categories is impractical. For a comprehensive monitoring and analysis, the flow states should be described from multiple aspects, such as the dominant phase and interphase structure. To address the above issues, a method called deep canonical variate analysis with attribute guidance (Ag-DCVA) is proposed, which provides an innovative process monitoring paradigm that designs attributes for process description. To effectively guide the model for feature embedding, an attribute guidance module is introduced to increase supervision information at both the class and attribute levels. Then, the convolutional neural networks (CNN) are developed to learn nonlinear transformations with an objective of maximizing serial correlation for effectively encoding attributes and characterizing flow states. Finally, the attribute evolution heat map offers a convincing and comprehensive analysis of the flow process. Besides, the extracted features can be input to classifiers to identify typical flow states. Extensive experiments demonstrate that Ag-DCVA outperforms existing methods, and it can monitor the flow process comprehensively and minutely while providing reasonable explanations.