Abstract:1.1%。Aiming at the existed problem that greenhouse environment control systems can not accurately predict the next stage air temperature inside the greenhouse and provide the basis for the control system to control the temperature optimally, a temperature forecast model based on time series method was developed. The temperature series collected from Jun. 6, 2001 to Sept. 16, 2002 in a hemispherical-roof greenhouse was studied. Firstly, the greenhouse temperature series was annual and first-order differenced in order to get a stationary greenhouse temperature series. Secondly, according to the characteristics of the autocorrelation coefficient and partial correlation coefficient of the first order annual difference series of greenhouse temperature, the ARMA(p,q) model was put forward to fit the greenhouse temperature. Finally, according to the minimum principles of variance estimate and the sum of squares of errors, an ARMA (4,4) model was determined as the 1-step forecast model of greenhouse temperature in summer. The test results show that the most absolute error and relative error of the forecast temperature of the 1-step forecast model are 0.8℃ and 3.2% respectively, and that the average absolute error and relative error are 0.2℃ and 1.1% respectively.