Parameter identifiability for multi-zone building models

Abstract

Model-based control strategies are widely advocated to improve the energy efficiency of building systems but they require accurate dynamic building models, which are often difficult to obtain in practice. While modeling a multi-zone lab for model-based control, we observed that the experimental data often does not have sufficient quality for system identification. To address this problem, we propose a Building Data-Dependent Identifcation (BDDI) algorithm to calculate the numerical identifiability for high-order RC models. The algorithm is a closed-loop “active identification” architecture that dynamically improves experimental design for better data quality. We conclude that the experimental conditions needed for accurate identification of building systems fall outside normal building operation. Therefore, systematic design of experiments with minimal disturbances (such as occupancy) are necessary to reliably model building dynamics.