Solving scalability issues in PV hosting capacity assessment through physics-informed neural networks

Accurate assessment of a distribution network’s hosting capacity (HC) is crucial for effectively integrating and utilizing distributed energy resources (DERs). While existing HC methods based on Monte Carlo sampling enable comprehensive analyses by simulating diverse input scenarios, they often incur significant computational costs due to iterative load flow calculations. This paper presents a novel approach for fast and accurate PV HC assessment using a physics-informed graph neural network (PIGNN)-based power flow solver. The proposed PIGNN framework integrates power flow equations and network topology to produce precise voltage estimates for unseen scenarios with substantially reduced time complexity. To validate its efficacy, we evaluate the HC of three realistic distribution feeders under varying PV generation levels. Results demonstrate that the proposed approach achieves a 30-fold speed up while maintaining 95% accuracy compared to traditional numerical methods. This scalable and efficient methodology enables rapid, large-scale planning studies for active distribution grids.