ScholarWorks@성균관대학교

초록

This study introduces a lightweight, rule-based optimal control strategy that preserves the predictive strengths of MPC while enabling real-time operation in resource-constrained residential environments. The proposed approach is tailored to residential radiant floor heating (RFH) systems and directly determines daily ON/OFF heating schedules. This distinguishes it from existing MPC-to-ANN imitation studies, which mainly emulate continuous control actions for air-based HVAC systems. A 24R–8C grey-box thermal model was developed to perform MPC simulations under diverse occupancy patterns and winter weather conditions. The ideal MPC outputs were converted into binary ON/OFF signals, which were then used to train artificial neural networks (ANNs). The resulting rule-based, near-optimal control algorithm autonomously determines heating activation and deactivation timings without requiring real-time optimization (e.g., 80% reduced computational time). Simulation results indicate that the proposed control approach delivers practical accuracy compared to baseline ON/OFF control, with a root mean square error (RMSE) ranging from 7.8 to 28.7 min. Furthermore, it achieved up to 17.9% reduction in heating energy consumption and over 90% reduction in thermal comfort violations compared with baseline ON/OFF control, approaching the performance of ideal MPC. These results suggest the potential of the proposed method for practical implementation in large-scale residential RFH systems using low-cost embedded control hardware.

키워드