An indoor thermal environment control model based on multimodal perception and reinforcement learning

Abstract

Achieving intelligent control and operation of building air conditioning systems to enhance indoor thermal comfort depends on accurately assessing occupant thermal status. However, traditional identification techniques, limited to single-dimensional parameters, often fail to promptly respond to various environmental and physiological factors influencing occupant thermal sensation. To bridge the gaps, this study integrates physiological heat exchange, cardiovascular, and brain nervous system responses to thermal environments to create a dynamic thermal sensation prediction model. An intelligent temperature control strategy employing reinforcement learning integrates this prediction model and occupant behavioral intention probabilities to effectively regulate indoor temperature settings. Experiment results demonstrate that compared to single parameter thermal models, the new method significantly improves prediction accuracy under conditions of drifting and step temperature changes. Furthermore, under these two different operating conditions, employing this strategy for temperature control reduces thermal discomfort accumulation by 26.46 % and 37.15 %.