Abstract: Greenhouse control system needs to control the actuators to make corresponding regulations according to the change of the greenhouse climate. When the temperature is too low, the heating system will be used to heat the greenhouse; when it is too high, the ventilation facility, the sun-shading system, or the spray equipment will be employed to cool the greenhouse and avoid overheat. In most conventional greenhouse control systems, the actuators were individually controlled based on the measured value and the setting value. This kind of control systems were working in passive mode and only made regulations when the greenhouse’s climate changed. It could not predict the future status of the greenhouse and then make regulations in advance. Besides, the actuators were established and set individually and could not work together harmoniously, which resulted in over-regulations and vibrations. Therefore, the control system needs to be developed with more intelligence for the whole system management. In this study, interior and exterior environmental information of the greenhouse, crop growing period and local climate data were integrated by using the global prediction model for the development of an innovative greenhouse control system. Compared to conventional greenhouse control systems, the interior and exterior temperature, the humidity, the ray radiation, the status of each actuator and near-future local climate were considered as global variable. The BP neutral networking was employed for model prediction. The global variable obtained from the corresponding sensors were input to the model to obtain the predicted values and the control system made the regulations with the use of PID before the climate changed. In order to validate the model, the experiment was conducted in a greenhouse for area of 96 m2, Because of the coupling effects of the various parameters, the greenhouse was divided into 5 areas: heating system, crop growing region, greenhouse side windows, ceiling and outside the greenhouse. Sensors were installed in each region, the data is collected, a total of 21 temperature sensors, 16 humidity sensors and two light sensors to be used .Prediction model of the BP neutral network consisted of three layers: the input layer, the hidden layer, and the output layer. Input parameters is the data collected by sensors, the state of six actuators and a weather forecast value, 4 prediction values are output: temperature, humidity, ray radiation, concentration of CO2.Tomato at growth stage of florescence was planted in the experiment greenhouse. The optimum temperature range for tomato at florescence period is 20-25℃, the night temperature is 15-20℃, the optimum humidity range is 65%-85%. The experimental results showed that this model can be controlled greenhouse environment in the state of optimal crop growth environment. In order to further validate of the model, the PID control simulation results were used to compare the actual situation. Results showed that temperature and humidity changes in greenhouse with the prediction model were gentler than that with only the PID controller. That meant this method increased the stability of greenhouse environment control system. This study demonstrated that the model could avoid the lagging response, passive control and inharmonious regulation in conventional control systems and it was effective and rational.