Abstract: In the field of smart home, accurately predicting the potential use behavior of users is a key technology to improve the user experience. Improper air conditioning operation will reduce the effect of air conditioning, increase energy consumption and even reduce the service life of air conditioning. At present, air conditioning risk identification technology is limited, cannot cope with complex scenarios, and lacks personalized recommendations. A technical framework based on Bayesian-Transformer hybrid model (BTHNM)is presented. By integrating user historical behavior data, motion sensing data and environmental data, the BTHNM model can accurately predict potential risky operations and give reasonable operation suggestions. Compared with traditional rule methods, BTHNM model can capture the deep correlation between user behavior, time and environment, and the accuracy and timeliness of risk event identification are greatly improved. Through comparative analysis of experiments, the accuracy of risk event identification verified by BTHNM model is 40% higher than that of traditional rule method, and the accuracy of suggestion is 41% higher.

Key words: Transformer, Bayesian network, Experimental research, Feature extraction, Event recognition, Operation recommendation