Abstract: In order to provide a method for real-time understanding of the temperature and humidity distribution characteristics in different regions within a box-type flue-cured tobacco production process, a Long Short-Term Memory (LSTM) network prediction model was established based on deep learning methods and in combination with historical temperature and humidity data monitored by actual sensors inside and outside the box. The learning rate and the number of neuron nodes were screened, and the model's prediction performance was evaluated using the coefficient of determination R², root mean square error (e_RMSE), and mean absolute percentage error (e_MAPE). The results indicated that when the learning rate of the model is 0.008 and the number of neuron nodes is 500, the model can achieve the optimal performance. Through the verification of the prediction results, it was shown that the change trend of the predicted values in the upper, middle and lower parts inside the curing box during each baking period is consistent with the measured values, and the determination coefficients of the predicted values are above 0.900. The maximum error of the predicted value and the true value of the temperature and relative humidity in the upper middle and lower parts of the curing box’s interior appeared between 45 h and 75 h, of which the maximum error of the middle part of the box was the largest, the maximum error of the temperature was 1.87 ℃, and the maximum error of the relative humidity was 12.54 percentage points. The LSTM-based temperature and humidity prediction model proposed in this study can accurately depict the changes in temperature and relative humidity at various levels within the curing box, providing a method for monitoring and controlling the temperature and humidity in the box-type flue-curing barn environment.