Occupancy Prediction in Buildings: An approach leveraging LSTM and Federated Learning

Nowadays, the energy used in commercial, residential, and office buildings represents a significant amount of the total energy spent worldwide. In these contexts, energy can be dramatically reduced by understanding when there is a waste of such an important resource. This can allow both a meaningful saving on energy costs and a significant reduction in CO2 emissions. In this field, occupancy prediction can help limit energy waste by allowing clever use of appliances and systems according to the real presence of the final beneficiaries. The aim of the paper is twofold. On a side, it wants to propose an approach based on Federated Learning (FL) and Long Short-Term Memory neural networks for the occupancy prediction in several rooms of a building. On the other side, it wants to show how FL helps in the occupancy predictions for the spaces in which the training of a specific model was not already performed. Some simulation experiments will show the effectiveness of the proposed approach.