Federated Learning for Short-Term Residential Load Forecasting

Load forecasting is an essential task performed within the energy industry to help balance supply with demand and maintain a stable load on electricity grid. As transitions towards less reliable renewable generation, smart meters will prove vital component facilitate these tasks. However, meter adoption low among privacy-conscious consumers that fear intrusion upon their fine-grained consumption data. In this work we propose explore federated learning (FL) based approach for training models in distributed, collaborative manner whilst retaining privacy of underlying We compare two approaches: FL, clustered variant, FL+HC against non-private, centralised fully private, localised approach. Within approaches, measure model performance using RMSE computational efficiency. addition, suggest FL strategies are followed by personalisation step show can be improved doing so. achieve ~5% improvement ~10x reduction computation compared learning. Finally provide advice private aggregation predictions building end-to-end application.