关键词: 家用中央空调, 灰盒模型, 物理信息神经网络, 强化学习, 可解释人工智能, 节能控制

Abstract: The intelligent upgrade of residential central air conditioning puts higher demands on control technology, and the evolution from data-driven "black-box" models to physics-data fusion "gray-box" models has become a current research hotspot. Focusing on the technical chain of "perception—modeling—control—assurance—application" and taking "the evolution from black-box to gray-box" as the main thread, First, summarize the development status of load identification technology from the perspective of multi-source environmental perception and three-dimensional modeling. Secondly, it comparatively analyzes the modeling mechanisms and applicable characteristics of black-box and gray-box models, and reviews typical methods such as physics-informed neural networks (PINN) and reinforcement learning. Furthermore, from the three dimensions of interpretability, cross-scenario generalization ability, and system stability, the key supporting technologies such as federated learning and explainable artificial intelligence are summarized. On this basis, the engineering practice prospects in directions such as dynamic regulation, multi-device coordination, and health intervention are discussed. The results show that gray-box models integrated with physical constraints have significant advantages in energy efficiency, control accuracy, and interpretability, and can provide references for research and engineering applications in intelligent control technology for residential central air conditioning.

Key words: Household central air conditioning, Grey-box model, Physics-informed neural networks, Reinforcement learning, Explainable artificial intelligence, Energy-saving control