电路到图：基于多维泛化的功率变换器建模

Accurate modeling is crucial in the optimal design and efficient operation of power converters. However, existing data-driven modeling methods lack generalization across diverse operating scenarios, including operating conditions (OCs), modulation strategies (MSs), and circuit topologies (CTs). Furthermore, these methods typically require extensive data and lack interpretability. To address these challenges, this article proposes the Circuit-to-Graph (C2G) method for performance modeling of power converters. The method leverages graph neural networks (GNNs) to comprehensively capture the global graph features of power converters and combines with domain adaptation (DA) technique to achieve efficient generalization modeling across multiple operating scenarios, which maintains high prediction accuracy even with limited data. The comprehensive algorithm experiments show that the mean squared error (MSE) of the proposed C2G method is reduced by 90.21% compared to the second-best algorithm in the source domain, and its MSE is reduced by an average of 52.11% compared to the best-performing baseline algorithms in each target domain. Additionally, compared to the second-best algorithm in the target domain, C2G reduces the total training time by 59.75% in both the source domain and target domains. The proposed C2G method requires an average of 1.425 ms to predict in specific operating scenario. Finally, 1-kW hardware experiments further verify the practical feasibility and superiority of this method.