A Personalized Federated Learning Scheme for Operational Parameter Determination of PV Smart Inverters

With the increasing integration of distributed photovoltaic (PV) systems into distribution networks, voltage regulation has become a significant challenge. Smart inverters (SIs) that allow remote switching of reactive/active power control parameters via communication commands provide a flexible solution for voltage regulation in PV-dominated systems. However, this requires the distribution system operator (DSO) to determine appropriate operational parameters for individual inverters to ensure fair and effective operation, avoiding potential loss of PV generation opportunities. This study proposes a personalized federated learning framework to optimize the SI control parameters. The DSO utilizes statistical data on voltage and potential PV generation at each point and shares control sensitivities among the connection points with similar power flow conditions. This approach efficiently derives the tailor-made control parameter for each inverter, improving voltage control, minimizing PV generation opportunity losses, and ensuring fairness among PV owners. Numerical simulations on a high-PV penetration distribution model demonstrate the framework's potential to enhance voltage regulation and PV utilization.