Hybrid model-based predictive HVAC control through fast prediction of transient indoor temperature fields

Efforts to reduce energy demand in the building sector have prompted a focus on the operational control of HVAC systems. Despite extensive research on HVAC control based on temperature prediction models, existing approaches often rely on node-based or average temperature predictions, which lack the detailed temperature distribution data necessary for accurate control, especially in transient situations with both spatial and temporal variations. This study introduces a precise HVAC control method based on a fast temperature field prediction model. By combining the single-step prediction response coefficient (SPRC) method with Convolutional Neural Network (CNN) architecture, sub-temperature field prediction models for multiple independent heat sources were constructed and integrated to achieve fast temperature field predictions. Subsequently, utilizing the predicted temperature field, air conditioning operation parameters were optimized and controlled to minimize energy consumption. Application of the proposed method in real building scenarios demonstrated the temperature field predictions closely aligned with computational fluid dynamics (CFD) simulations, achieving a mean absolute error (MAE) of 0.27 °C and a root mean square error (RMSE) of 0.24 °C. Furthermore, this model achieved a notable 57.8 % improvement in prediction accuracy compared to models relying solely on single-step prediction responses. Additionally, the model predictive control based on the hybrid model's temperature field predictions significantly reduced the runtime of the HVAC system by 18.18 % while maintaining temperatures within the comfort range throughout the operation period. The method presents a promising avenue for optimizing HVAC operations and minimizing energy consumption in building environments, thereby contributing to sustainable building management practices.