Building occupation modelling using motion sensor data

In smart building environments, both office and residential buildings, it is important to have some information about the use and occupation. Today this is normally solved by a fixed time schedule meaning the occupants must adapt to the system, not the other way around. This paper discuss the usage of a top hat probability models, based on a four weeks history from inexpensive sensor devices, for prediction of the occupation in the next week. The model was divided into seven groups, one group for each of day of the week. A software system, based on several modules, was developed. One module was used to record the information from the motion sensors and stored the data as historical data. One module was used to create the model, and another module was used to prediction of occupation for the next days, up to a week. The models are working satisfactory as long as the behavior patterns are similar for the training and prediction period. However, the models are sensitive to changes in the daily behavior pattern of the occupants, like holidays or taking a day off.