PhaseDGAT for dynamic modeling and analysis of gas–liquid two-phase flow

Gas–liquid two-phase flow is a common phenomenon in industries such as petroleum, chemical processing, and nuclear energy, where accurate modeling of flow dynamics is crucial for safety and optimization. This study employs complex network theory and graph neural networks to analyze vertical gas–liquid flow behavior. Signals captured by a multi-sector conductivity sensor are transformed into high-dimensional representations through a phase-space projection network, enabling the identification of dynamic features across different flow states. On this basis, we propose a phase-space graph attention adaptive dynamic network, which extracts key features from network topologies and integrates a dynamic graph structure to effectively capture nonlinear and time-varying interactions within the flow system. This innovative approach significantly enhances the accuracy of gas void fraction prediction, providing a robust framework for monitoring and controlling two-phase flows in industrial applications.