Federated Learning for Residential Load and PV Production Forecast

This paper explores the application of federated learning in predicting energy consumption and solar production for households equipped with Enphase smart home solar systems. Traditional centralized model training for energy prediction entails significant communication overhead, particularly when dealing with distributed data sources. To mitigate this challenge, federated learning is proposed as a privacy-preserving alternative, wherein each client trains a local model on its data and transmits model updates to a central server for aggregation. We propose a neural network-based energy demand predictor trained with federated learning, leveraging distributed data while maintaining prediction accuracies. Federated learning ensures data security and confidentiality by transmitting only model updates, preserving individual privacy. The datasets used are anonymized, and the actual data cannot be extracted from the shared gradients. Through experiments, we demonstrate the feasibility and effectiveness of federated learning in energy forecasting, highlighting its potential to enhance prediction capabilities without compromising privacy.