The main steam temperature in the thermal power plant boiler has the characteristics such as large delay, strong inertial, time varying parameters and so on. When the condition changes greatly, it is difficult to maintain satisfactory control results using traditional PID control based on a single model. To solve this problem, this paper presents a predictive control method based on multimodel switching. Firstly, the object models are established under a number of typical conditions. Secondly, the best dynamic matrix controller under different models are designed. Thirdly, the switching strategy is designed to select the most appropriate controller to mitigate the effects of varied operating conditions (especially load changes) on the main steam temperature system. Finally, the validity of the proposed method is verified through Matlab simulation. The simulation results show that the proposed method is superior to the traditional PID control, and the multimodel switching improves the robustness of the fixed parameter model predictive control.