Multi-Zone Grey-Box Thermal Building Identification with Real Occupants

Model-based building control can optimize energy use and thermal comfort. To control each room individually and account for occupants' preferences, a multi-zone model is required to represent each room. For multi-zone models with grey box structures, the rooms can either be modeled uncoupledly (decentral) or coupledly by considering the heat exchange between rooms (central). Therefore, this paper analyzes (i) central and decentral model structures, (ii) different identification algorithms to estimate their parameters, and (iii) occupants and disturbances during the identification process. For each grey-box model structure, we evaluate the performance of a Genetic Algorithm (GA) and different gradient-based identification algorithms. The thermal dynamics of this study remain largely undisturbed on weekends when the building is unoccupied, but are greatly disturbed on weekdays by the occupants, e.g., by the opening of windows. Our results show that the occurrence of the disturbances during the identification has the most negative impact on grey-box model performance. Furthermore, we show that the lowest identification error can be attained with gradient-based optimization methods and that the model structure has a minor influence on the identification error.