Development of seasonally clustered building models for applying model-based predictive control in buildings

Grey-box modeling is commonly applied in model-based predictive control (MPC) for managing building heating, ventilation, and air conditioning operations. However, typical grey-box models are time-invariant, requiring multiple model sets to account for seasonal variations. To address this limitation, we introduce a novel grey-box modeling method that incorporates K-means clustering techniques. Weekly grey-box models were established using measurement data from a department store in South Korea. These 22 model sets were subjected to clustering to group similar models. The performance of the proposed clustered model in predicting indoor air temperature was compared with that of two baseline models: (1) a single-week-based model (Baseline 1), and (2) a monthly-updated model (Baseline 2). The clustered model achieved up to a 10 ∼ 35% reduction in the root-mean-square error, demonstrating improved predictive accuracy. Lastly, MPC simulations using the clustered model were conducted for the summer (July and August). The electricity consumption and operating costs were reduced by 6.78% and 36.26% in July and 2.50% and 30.47% in August, respectively compared to the baseline cases. These results validate the applicability of the proposed clustering-based modeling approach for MPC implementation. • Performed grey-box building modeling for a large building with measured data from existing sensors • Developed clustered building models for different seasons proving the limitations of using a single model • Performed MPC simulations using a proposed model, yielding high energy and cost savings