Frontiers in Built Environment (Aug 2022)
Parameter identification approach to represent building thermal dynamics reducing tuning time of control system gains: A case study in a tropical climate
Abstract
As one of the main consumers of primary energy globally, buildings have been among the main targets for implementing energy efficiency solutions, such as building control strategies that maintain occupant comfort and reduce operating costs. The design of such control schemes relies on a thermal model of the building to predict indoor temperature. The model should be sufficiently accurate to describe building dynamics but simple enough to remain optimal for control purposes. This paper proposes a methodology to identify thermal RC networks to model building thermal dynamics of a residential buildings located in humid and rainy climates, a topic not widely covered in current literature. The candidate models for the methodology are determined through a parameter dispersion study, which consists of training the models multiple times and checking if the parameters converge to a single value regardless of their initial value. Then the effect of the training dataset characteristics on model performance is studied. The methodology is established and then tested in a residential case study in Panama from these conclusions. Results show that a linear model with few parameters and trained with only 10 days of data can successfully represent a system with prominent nonlinear phenomena. The model with the best performance during active operation has a validation root mean square error of 0.36°C, which is satisfactory for controller design purposes. The model is then used to tune a proportional integral derivative controller, successfully employed to maintain the desired indoor temperature. Using the identified model to calibrate the controller avoids tedious trial and error procedures.
Keywords
