A Comparative Analysis of Room Air Temperature Modelling for Control Purposes
Abstract
The use of accurate models is a relevant point for simulation, optimization and control purposes. More specifically, they represent a cornerstone for the development of different based on model control strategies, as Model-based Predictive Control (MPC) or Internal Model Control (IMC). These control strategies help to obtain high thermal comfort levels inside buildings in an efficient way by an optimal combination with passive strategies. For control purposes, the selection of an appropriate kind of model will depend on its complexity, aim and available resources. In this work, a comparison between the complexity and accuracy of several models is performed. To do that, three different room-level indoor air temperature models have been developed: i) a Linear Time-Invariant (LTI) model estimated by means of a Pseudo-Random Binary Sequence (PRBS) signal; ii) a nonlinear model based on Artificial Neural Networks (ANNs); and iii) a nonlinear first principles model. These models have been calibrated and validated using real data from a characteristic office room of a bioclimatic building. The obtained results show as the three approaches provide good results with an NMAE error less than 14% in the worst case (LTI model) and approximately equal to 5% in the best one (first principles model), and thus, they could be used to develop appropriate control strategies.
Keywords
Room dynamic modelling, System identification, Thermal comfort control, Indoor temperature