Low Order Grey-box Models for Short-term Thermal Behavior Prediction in Buildings

Abstract Low order grey-box models are suitable to be used in predictive controls. In real buildings in which the measured quantities are few the reliability of these models is crucial for the control performance. In this paper an identification procedure is analyzed to investigate the accuracy of different order grey-box models for short-term thermal behavior prediction in a real building, part of a living smart district. The building has a low number of zones and a single indoor temperature measuring point. The models are identified on the data acquired in 31 days during the winter 2015. The second order model shows the best performance with a root-mean-square error (RMSE) less than 0.5°C for a prediction horizon of 1-hour and a RMSE less than 1 °C for a prediction horizon of 3-hours.