Abstract: Abstract: The control effect of the conventional control method to the greenhouse temperature depends on the accuracy of the plant model and interference measurement. However, an accurate model of the greenhouse is difficult to obtain because of characteristics of the greenhouse such as uncertainty, imprecision, time-varying, multi-disturbance, etc., with the interference being particularly difficult to accurately measure. For example, the conventional PID control algorithm, widely used in many respects with a good performance record, has poor adaptability and weak robustness when used in greenhouses, and the smith predictive control, used in time delay systems to compensate temperature hysteresis, requires precise mathematical object model. Thus, the usual PID+Smith predictor controller often has overshoot and oscillation, generating a large amount of energy consumption in the process of temperature adjustment when used in the greenhouse temperature control system. Therefore, the grey prediction compensation control algorithm is adopted for compensating the aforementioned characteristics of the greenhouse. The advantage of the proposed control strategy is its getting rid of the dependence on conventional control algorithms in the plant model accuracy and interference measurement accuracy, and bypassing the theoretical and technical obstacles in obtaining the object model and interference. Both the simulation and actual operation indicated that the proposed control strategy achieves satisfactory control effect and the system accuracy is significantly improved. Statistical analysis indicated that the correlation coefficients between the estimated value and the true value of the uncertainty and interference grey parameters is 0.9968, 0.9804, and 0.9938, respectively, and the coefficient of determination between them is 0.9935, 0.9585, and 0.9871, respectively. The grey parameters absolute error mean is -0.11510, -0.26733, and -0.31035, and the variance is 0.05150, 0.16324, and 0.09474, the grey parameters relative error mean is -1.68%, -8.06%, and -8.73%, and the variance is 0.01368, 0.00533, and 0.00581. The correlation coefficient between the measured temperature curve and the simulation temperature curve is 0.973972, and the coefficient of determination between them is 0.948621. Also, the overshoot and oscillation in the process of temperature regulation is weakened or eliminated, so the energy consumption is greatly reduced, which not only meets the temperature requirements, but also achieves energy-savings.