Performance Evaluation of Federated Learning for Residential Energy Forecasting

Short-term energy-consumption forecasting plays an important role in the planning of energy production, transportation and distribution. With widespread adoption decentralised self-generating systems residential communities, short-term load is expected to be performed a distributed manner preserve privacy ensure timely feedback perform reconfiguration distribution network. In this context, edge computing enabling technology decentralized data collection, management, processing delivery. At same time, federated learning emerging paradigm that fits naturally such edge-computing environment, providing AI-powered privacy-preserving solution for time-series forecasting. paper, we present performance evaluation different federated-learning configurations resulting levels forecast consumption with collected by real smart meters. To aim, experiments are run using Flower (a popular framework) data. Our results allow us demonstrate feasibility approach study trade-off between accuracy prediction, which characterizes quality service system final users.