Abstract: The mechanism models of the greenhouse temperature system are analyzed,and then the temperature system is described respectively with auto regressive moving average models with external input(ARMAX) and auto regressive integrated moving average models with external input(ARIMAX).The outside air temperature,relative humidity,global solar radiation and wind speed are used as the input variables of the system,and the inside air temperature is used as the output variable of the system.Statistical hypothesis test and model fitness analysis are used together to select the model structure,gradually oblivious recursive extended least squares method is adopted to identify the model parameters on line,and an intelligent supervisory segment is devised to monitor the on-line modeling process.Finally it is investigated to what extent ARMAX and ARIMAX with 4 input variables or 3 input variables(wind speed is omitted) can be used to describe the greenhouse inside air temperature system.The results of the modeling and simulation experiments indicate that the gradually oblivious recursive extended least squares with intelligent supervisory segment can satisfactorily characterize the dynamic features of the greenhouse temperature system.