ARX models inspired by physics as a service for building's users

Recently, in the field of building energy efficiency, many works focused on model based energy management services but developing models consistent with a measured reality is still an issue. Fine physical models with many parameters cannot be adjusted while non-physical models cannot extrapolate to situations never met in the training data. In this paper, pure data models will be implemented with ARX models. In a second time, physical knowledge is added in order to improve the accuracy especially in the cross-validation process. These models are applied to a mono-zone study case in order to forecast the indoor CO
₂
concentration.