Abstract: In 2014-2016, plastic greenhouses and glass greenhouses in different districts of Jiangsu Province were selected for monitoring. Cosine segmentation function, sinusoidal piecewise function, sine-exponential piecewise function, first-order function and neural network model were used to simulate inside hourly temperature in different seasons and different weather conditions (clear day and rainy day). The results showed that all five models can simulate hourly air temperature inside greenhouse through the highest and lowest temperature of the day. The neural network simulation accuracy was the higher (RMSE=0.69℃) and was less affected by the type of greenhouse, weather conditions, and seasonal changes, the universality was higher. The sinusoidal-exponential piecewise function had the best accuracy (RMSE=0.43℃) and was less affected by weather and seasons, but it was affected by the characteristics of the greenhouse itself and the region. The cosine piecewise function (RMSE=0.8℃) and the sinusoidal function (RMSE=0.78℃) had similar simulation results and was affected by the weather and region. The accuracy of a piecewise function is low (RMSE=0.90℃) and the error varies greatly. Models had higher simulation accuracy in plastic greenhouses then it in glass greenhouses. Seasonal variation of the model's simulation accuracy was different between the models and the types of greenhouse, but usually, the simulation error in spring and winter was greater than in autumn, and the error in summer was the smallest.

Key words: Plastic greenhouse, Glass greenhouse, Simulation of air temperature, Greenhouse, Daily variation of temperature