Leveraging distributed AI for multi-occupancy prediction in Cognitive Buildings

Cognitive Buildings are autonomous smart environments capable of setting themselves according to some self-learned rules. Such rules are inferred according to, e.g., the inhabitants’ behaviors, users’ needs, and specific policies for optimizing security, energy, and comfort management. To do this, it is of foremost importance to gather information about users’ habits like room occupancy. Indeed, Cognitive Buildings can effectively exploit information about sensors in the different rooms, thus being able to detect, learn, and forecast the presence of users in the buildings and act in accordance with these predictions. In this direction, this paper proposes an innovative approach for multi-occupancy prediction in Cognitive Buildings, incorporating a multi-layer hierarchy for Federated Learning, the utilization of IoT devices at the Edge, the implementation of long short-term memory neural network models, and the exploitation of Edge Computing. The approach also introduces a versatile design template for developing real distributed systems for occupancy prediction. The proposed approach uses a distributed paradigm to safeguard data privacy so that the collected data is used to train separate local deep learning models, which are then merged in the Cloud. The paper validates the approach by providing a preliminary prototype realized at ICAR-CNR, Rende Italy, and presents a performance analysis, which shows that the occupancy is predicted with an 84.5% accuracy.