Abstract:The public weather forecast and observed meteorological data in the plastic greenhouse in Cixi city, Zhejiang province have been used to set up a BP neural network model, in order to predict daily extreme temperatures in the plastic greenhouse, whose input variables were daily maximum and minimum temperatures, relative humidity, maximum wind force scale, day and night weather condition, and whose output variables were the maximum and mininum temperatures in the plastic greenhouse. The results show that the root mean squared error(RMSE) and absolute error(AE) between trained and measured values of daily maximum temperatures in plastic greenhouse were 4.0℃ and 3.2℃, while the daily minimum temperature were 1.3℃ and 1.0℃. Furthermore, RMSE between predicted and measured values of the daily maximum and minimum temperatures were 3.6℃ and 1.2℃, while the AE were 3.0℃ and 1.0℃, respectively. With easy access to data and wide practicability, this model could accurately predict the coming extreme temperatures in the plastic greenhouse and provide scientific basis for greenhouse management and environment regulation.