Enhancing Short-Term Energy Forecasting in Distributed Systems Through Federated Learning

Energy consumption forecasting plays an important role in smart buildings, enabling energy monitoring, and planning. Especially nowadays it is a useful tool that can alert the consumers for upcoming energy usage peaks. In the literature, there is a gap in energy usage forecasting for educational institutions. They are suitable study cases because they are research centers that develop and create novel concepts and serve as examples for energy-efficient buildings. This paper proposes an approach for short-term energy consumption forecasting for buildings that are property of Ss Cyril and Methodius University in Skopje. The proposed model is trained on 25 different university department buildings' data from January 1, 2021 to November 16, 2023, with a resolution of one hour without threatening data privacy. Energy usage and positional encoding of hour, day of week, and month are used as input variables. The proposed approach utilizes federated learning technique to develop a generalized deep learning model for short-term forecasts. Additionally, a comparison between recurrent neural networks in centralized and decentralized learning approach is presented. The model performance is evaluated using the following metrics: RMSE and MAE. The obtained results show that federated learning achieves great performance.