Building energy consumption represents a significant portion of global energy use, demanding efficient control strategies. Traditional rule-based methods lack adaptability, while data-driven approaches struggle with generalisation. This study presents the machine learning-based indoor environment control system (MIECS), which integrates reinforcement learning, deep learning, and edge computing within a three-layer architecture. By modelling device-environment interactions with heterogeneous graphs, MIECS improves sample efficiency and convergence speed. Experimental results show a 23.7% reduction in energy consumption and an 18.5% increase in user comfort compared to conventional methods, providing a scalable, adaptive solution for intelligent building management.
Shen Mengqi (Thu,) studied this question.