Prediction of window handle state using machine learning

The project described in this paper investigates the energy-relevant behavior of window control actions of the occupants of an office building in Regensburg, Germany. The extensive data monitoring regarding energy consumption, indoor as well as outdoor climate, and window control actions (state of the window handle) started in 2017. Different machine learning classification algorithms are used together with the measured data to train models for the prediction of window openings and closings. The procedure is designed to identify the potentials and limitations of the realistic forecasting of occupant behavior based on the available data.