In thermal power plant, the superheated steam temperature affect both the security and thermal efficiency. It possesses the non-linear features under load changes. A nonlinear model predictive control strategy is proposed. A linear parameter varying (LPV) model is used to represent this nonlinear model, which is identified using linear models of selected working-points and other data. In each sample time, the model is linearized through the scheduling variable, and the linear predictive controller is designed. This method keeps better control performance than traditional PID and linear MPC. The effectiveness is verified in simulation platform of real 200MW power plant.