Model free method of nonlinear system predictive control is investigated in this paper. The universal model is taken as predictive model of a class of nonlinear system. Through the multi-layer hierarchical forecasting, feature vector in the universal model is predicted, so that predictive model based on universal model is worked out. Through the optimization of performance index, a kind of gas stove modeling method based on model free predictive control algorithm for nonlinear system is obtained. The simulation examples show the effectiveness of the algorithm, and through comparison of model free predictive control and model free control, it is concluded that model free predictive control with functional combination module has the same advantages with model free control.