Physics-Informed Neural Network for Modeling and Simulation of Phase-Locked Loop Dynamics

Inhalt:
The increasing integration of inverter-based resources (IBRs) into modern power systems introduces new stability challenges, largely driven by the fast dynamics of converter control loops. In grid-following (GFL) converters, synchronization via a Phase-Locked Loop (PLL) can cause rapid transients during grid faults, potentially triggering instability. Exhaustively evaluating all operating scenarios with traditional numerical solvers is computationally prohibitive. Electromagnetic transient (EMT) simulations and reduced-order models can offer valuable insights but remain computationally demanding. Machine learning (ML) methods provide faster alternatives, yet they typically require large labeled datasets and may produce physically inconsistent predictions. To address these limitations, this paper proposes a Physics-Informed Neural Network (PINN) to model the dynamic behavior of the PLL in a GFL converter. By embedding the system’s physical equations directly into the training process, the PINN reduces the dependence on labeled data while enforcing physics-consistent outputs. The proposed approach is benchmarked against a reduced-order model and a conventional neural network, demonstrating its ability to accurately approximate system trajectories while significantly lowering data requirements and computational costs.