ALOS: Automatic learning of an occupancy schedule based on a new prediction model for a smart heating management system

Abstract

In our day and age, lowering energy consumption in buildings is a must. Smart-buildings will provide the answer if and when they can adjust the required indoor temperature to the occupancy. Developing an occupancy model that forecasts the time of arrival and departure is therefore mandatory. Our article deals with the occupancy prediction model of a building meant for an inteligent heating management system. The prediction also integrates short and long duration of occupation/unoccupation. ALOS is based on an unsupervised clustering method (to classify the events ‘departure’ and ‘arrival’) and on the EM (Expectation Maximisation) algorithm with a new mixture model to determine short and long duration of the events. While most previous studies focused on either the residential or the tertiary building, our approach predicts occupancy in both types of buildings. In order to demonstrate the efficiency of our approach, it was tested on real occupancy datasets (familly consisting of 4 people and elderly person living alone). The results indicate that ALOS achieves excellent average prediction accuracies, notebaly from 80% up to 90%, which makes it efficient and provides easy implementation. Finally, a major strength of the ALOS method is that it only needs just under a week to integrate a change of the occupants' habits.