Abstract: In order to provide reference for vegetable growth management and environmental optimization regulation, the dynamic changes of greenhouse temperature and humidity were predicted by using ARIMA model and RBF neural network. In this study, the data from July 28 to 29 were taken as the validation set, and the data from the previous 5 to 25 days were taken as the sample set. Three input variables were set to explore the temperature and humidity model prediction effects. Based on the measured data of air temperature, relative humidity and light intensity in the vegetable greenhouse, and according to the difference in the predictive ability of ARIMA model and RBF neural network for linear and nonlinear problems, temperature and humidity prediction models based on weight combination of ARIMA and RBF neural network were constructed. The results showed that the temperature and humidity in the greenhouse had more obvious linear and nonlinear variation characteristics, respectively, and the single models with better prediction effects were the ARIMA model and the RBF model. The optimal single temperature model was the model ARIMA(0,1,1)(1,1,1) 24 with 25 days sample set. The optimal single humidity model was the RBF model (3-6-2) with 25 days sample set and input variable A (temperature, humidity and light intensity). Compared with the optimal single models, the ARIMA-RBF neural network weight combination model had higher prediction accuracy and better stability. The MAE, MAPE and RMSE were 1.04℃, 2.95%, 1.21℃ for the optimal temperature combination model and 0.35 percentage point, 0.36%, 0.55 percentage point for the optimal humidity combination model. The weight combination model gives full play to the ability of single models to process data from different characteristics through an appropriate weighting strategy, and can better evaluate the temperature and humidity status in the greenhouse, which provides a reference for establishing a more universal greenhouse environmental factor model.

Key words: Greenhouse, Temperature, Humidity, Model, ARIMA, RBF