CF-PV: Centralized vs. Federated Edge-based Prediction Models for PV Energy Production
Narges Mehran; Peter Dorfinger; Nicola Leschke; Frank Pallas (2025): CF-PV: Centralized vs. Federated Edge-based Prediction Models for PV Energy Production In: 2025 IEEE PES Innovative Smart Grid Technologies Conference Europe (ISGT Europe).
Federated learning (FL) is a promising approach for resolving the conflict between accurate prediction models and privacy requirements in prosumer-driven photovoltaic (PV)-based energy production. However, so far, its adoption in this domain remains limited, and existing efforts only consider a narrow range of machine learning (ML) methods.To shed light on the practical applicability of federated, edge-based learning in this context, we present CF-PV, a comparative study on Centralized and Federated edge-based prediction models for photovoltaic (PV) energy production using real-world data. CF-PV incorporates four different ML model types (e.g., ensembles and neural networks) and emulates up to 12 federated prosumers. Results demonstrate how federated learning enables more sophisticated trade-offs among privacy, prediction quality, and computational overheads, thereby heightening the practical feasibility of distributed, prosumer-driven PV production.
