Transient thermal conditions in the breathing zone created by pulsating air supply, can improve thermal comfort compared with steady air supply under stratum ventilation. This study proposes a dynamic physical-empirical hybrid model for the transient air temperature in a stratum-ventilated classroom, concerning specific locations and dynamics in the breathing zone, accuracy to the second. The model is validated through experiments of four cases and twelve measurement points at the height of 1.1 m each case. The raw air temperature signal is divided into an approximation component maintaining the major fluctuations and a detail component representing the stochastic volatility, by a validated denoising structure using wavelet decomposition. Firstly, the dynamic model can accurately predict the raw air temperature, with the root mean square errors (RMSE) of 0.04–0.07 °C. Secondly, the rate of air temperature variation positively correlates to the air velocity variation between the idle/duty period and the steady state, as well as the supply air velocity deviation, with the Pearson correlation coefficient value of about 0.7 (p < 0.01). Thirdly, the dynamic models for the approximation component are consistent with that for the raw air temperature signal, with an improved coefficient of determination of 0.91–0.97, and the RMSE of 0.02–0.05 °C. The decomposition level is the key hyperparameter to denoise the air temperature signal for approaching the modelled value. Lastly, the first 20% of the approximation component is used to model the entire period, reducing measurement costs. The model performs well on accuracy, simplicity and robustness.