Clustering methods for occupancy prediction in smart home control

Clustering methods are deployed to extract patterns from large amounts of data. For home and building automation, usage patterns and their resulting profiles allow improving control systems with prediction capabilities. This paper shows how different clustering methods identify patterns representing the occupancy of inhabitants. Regarding the occupancy, the clustering methods are tested with real data from three kinds of rooms taken from a database of buildings monitored for five years. Later on, they are analyzed and compared using a simulated environment for the automated control of a use case dedicated to heating setpoint temperature control. As will be shown, methods based on Fuzzy C-means and eXclusive Self-Organizing Maps obtain the best performance in simulations, presenting excellent features for the application of interest.