Power converters play a critical role in modern power systems and electronic devices. Accurate performance prediction is essential for enhancing system reliability and efficiency. Traditional modeling approaches rely on complex physical models or empirical formulas, which are difficult to adapt to variable operating conditions and complex circuit topologies. In recent years, with the rapid development of artificial intelligence (AI) technology, data-driven methods have been applied to the performance modeling of power converters. However, existing data-driven methods still have many deficiencies in terms of generalization capability, computational cost and model interpretability. To address these issues, this paper proposes a Graph Neural Networks (GNN) -based modeling method for power converters, which achieves accurate prediction of the dynamic performance of power converters by efficiently capturing the complex nonlinear relationships within the circuits by utilizing GNN and Multi-Layer Perceptron. This study applies this method to model the current stress of the LC-type resonant dual active bridge and compares it with other commonly used AI algorithms. The proposed modeling method can be generalized to a wide range of operating conditions, with mean absolute percentage error of less than 1% when applied to out-of-domain scenarios, compared to models that are specifically trained for those conditions. 1-kW hardware experiments comprehensively validate the feasibility of the proposed method for performance modeling.