Abstract: To enhance the practical consistency and interpretability of deep learning approaches in power flow (PF) calculation, this letter proposes a virtual graph constrained message passing neural ...