Data-driven occupant actions prediction to achieve an intelligent building

An intelligent building has to know the specificities of the occupants and determine their drivers to perform actions so that it can optimize the building operation. Five windows of different rooms of the same dwelling were analysed in-depth to understand the specificities and variations of occupants' behaviour. Logistic regressions were used as a machine learning method to predict occupants' actions. The windows opening prediction models were formulated by taking into account continuous and categorical variables. An evaluation of the required data length that allows obtaining the prediction models with results identical to those obtained with the complete year was performed. It was concluded that the best option was to use at least 15 days in summer and 15 days in winter to have a reliable prediction for the full year. The model constructed for each window did not show good prediction success when applied in another room of the same dwelling. This study shows that the specificity of humans needs do not allow a generalization of their behaviours in the built environment. Thus, it is necessary to adapt the algorithms of the building automation systems through data-driven machine learning techniques.