Abstract: To solve the problems of control target deviation, increased relative error, and degradation of tobacco leaf quality caused by strong coupling of spatiotemporal variables in the temperature and humidity control of a tobacco leaf aging warehouse, a spatiotemporal control method for temperature and humidity based on a recurrent neural network (RNN) and extreme gradient boosting (XGBoost) was designed.First, temperature and humidity sensors were used to collect real-time data in the warehouse, and the spatiotemporal dimensional features of temperature and humidity were extracted from wireless sensor network (WSN) data packets. Then, RNN was used to capture the dynamic spatiotemporal variation patterns of temperature and humidity, and reasonable control targets were set in combination with the XGBoost algorithm. On this basis, a compensation mechanism for the decoupling control variables of temperature and humidity targets was introduced. Compensation values were calculated using the inertial time constant to replace the original objective function, making the compensated control targets approach the expected values, thereby effectively offsetting the lag effect caused by system coupling.The experimental results showed that compared with the PID method and the FNN method, the RNN-XGBoost method strictly controlled the relative error of temperature and humidity within 0-1.0%, significantly improving the control accuracy and stability.

Key words: recurrent neural network (RNN), extreme gradient boosting (XGBoost), tobacco leaf aging warehouse, temperature and humidity, spatiotemporal control