Abstract: Abstract: Ground Source Heat Pump (GSHP) air conditioning system has been well applied in greenhouse temperature control, which is mainly used in warming system for greenhouse during the winter. Although GSHP system is a high-efficiency and energy saving system, it still has potential for improvement in the practical application by changing operation mode to save energy. The method changing circulating pump's working frequency of GSHP system was proposed in the study. The greenhouse environment characteristics are a combination of great inertia, pure time-delay and nonlinear which make it difficult for us to obtain an accurate mathematical model of the environment. Thus it was not useable for some traditional algorithms such as PID control algorithm. To solve the problem, in this study, we introduced a grey prediction method in temperature modeling process from which we used previous temperature data sequence to forecast future temperature and the predicted future data as feedback value into the controller. As temperature was the only factor for prediction, we chose GM(1,1) as the grey prediction model. To improve the prediction precision, equal dimension and new information method were applied in the modeling process which meant that the latest data took the place of the oldest data with no change in model dimension. Model dimension also influence the prediction precision. The longer dimensions are used, the more accurate prediction can be obtained. However excess dimension can add time and complex on the calculation. To find the most proper model dimension, we compared the different prediction value under diverse dimension and concluded that six was the most suitable one from which the absolute error between actual value and predicted value was 0.34247 and the variance was 0.035974, which was an acceptable precision of grey prediction. According to the difference between the predicted temperature and the setting temperature of the greenhouse, the controller adjusted and decided the frequency of the circulating pump and based on this way, the energy consumption can be reduced. The control strategy to satisfy energy saving and greenhouse temperature requirement was decided by actual repeated experiment and revision and verified at Cuihu greenhouse of Shangzhuang, Beijing, which displayed a saving of 24% energy consumption after frequency reformation. Experimental results showed that this control method not only improved the control quality of the circulating pump, but also achieved the purpose of energy saving.