Greenhouse temperature system is a typical hybrid system, and its inputs include discrete equipment control quantities and a number of outdoor environmental disturbances that can be measured and not controlled. A hybrid system was proposed for greenhouse temperature, a switching system model was established, and a multi-input predictive control was designed based on this model. A determined device state can be considered as a subsystem, and the modeling of the greenhouse system can also be transformed into the modeling of all subsystems. There were numerous greenhouse environments, so it was needed to simplify the input variables. By correlation analysis, outside temperature, outside humidity and solar radiation had obviously strong correlation with inside temperature. The ARMAX model was used to describe the model, and the augmented recursive least square method with forgetting factor was used to identify the model parameters, and the model accuracy was verified. The ARMAX model was used to design a predictive control controller to solve each device’s action sequence, which was an NP-hard problem, and it was solved by optimized pruning simplifies the calculation process. At each sampling time, the set value predictive control was determined, however, under uncontrollable external environmental factors, the output fluctuations were large. If a fixed set-point was used, the system would be switched frequently, and thus increasing equipment loss. In order to solve this problem, the dual-period accumulative temperature method was utilized to dynamically adjust the set-point of predictive control according to the long-period average value of the indoor temperature and the current value, so as to reduce the unnecessary switching of the equipment and reduce the loss.