Heating Optimisation of a Multi-Zone Building’s Thermal Comfort Under Stochastic Condition using Data-Driven Model Predictive Control

Abstract

Model Predictive Control (MPC) has gained popularity in recent years and is widely adopted in building control. This study proposes a novel data-driven robust MPC to make the optimal heating plan, specifically for the multi-zone single-floor building. In this study, the room temperature and relative humidity (RH) will be highly valued in the optimisation decision. To better incorporate RH into the state-space model (SSM), the linear relations between RH and other room temperature parameters in the thermal zones are formulated, ensuring the better linear fitting of SSM to the original nonlinear model. Afterward, k-means clustered, principal component analysis (PCA), and kernel density estimation (KDE) based data-driven uncertainty set is constructed and applied to MPC. The other three kinds of MPC’s are compared to our proposed data-driven robust MPC (RMPC), including conventional RMPC, k-means clustered, data-driven RMPC (KM-DDRMPC), PCA and KDE based data-driven RMPC (PKDDRMPC). The results demonstrate that the optimality of our proposed k-means clustered, PCA and KDE based data-driven RMPC (KM-PKDDRMPC), which consumes 9.8 % to 17.9 % less energy in controlling both temperature and RH, compared to other data-driven robust MPC’s, and essentially follow the constraints which certainty equivalent MPC and conventional RMPC cannot conform.