Abstract:

In order to study the simulation ability of BP neural network on the average temperature and average relative humidity in the facility greenhouse under three weather types of sunny, cloudy and overcast days, the meteorological data of the four-span plastic greenhouses in Songjiang district, Shanghai from March 1, 2021 to February 28, 2022 and the meteorological station of the same period were selected, and the average temperature, average relative humidity, average wind speed and solar altitude angle outside the greenhouse were used as input factors to construct temperature and humidity BP neural network prediction models for three weather types at the annual and seasonal scales, and the models were validated. The results showed that: (1) at the annual scale, the RMSE of the average temperature model was 0.8−1.1℃, and the R²≥0.9890 (P<0.001). The RMSE of the average relative humidity model was 2.9−4.0pp (percent point), and the R²≥0.9434 (P<0.001). Overcast days were the best simulations results, followed by cloudy days and sunny days at the lowest levels. (2) At the seasonal scale, the average RMSE value of the average temperature model was 0.7−0.9℃, and the average R²≥0.9765 (P<0.001). The average RMSE value of the average relative humidity model was 2.2−3.2pp (percent point), and the average R²≥0.9451 (P<0.001). Of these, the simulation accuracy was best in summer, followed by autumn, and the lowest in spring and winter. The simulation performance of the seasonal scale model was better than that of the annual scale model. (3) The average temperature and average relative humidity models at the annual and seasonal scales passed the independent sample test of the significant level of 0.001, indicating that the model has a good simulation effect and can be used to simulate the temperature and humidity in the greenhouse and practical application.

Key words: Four-span plastic greenhouse, Weather type, Temperature and humidity simulation, BP neural network, Time scale