A new nonlinear predictive controller is proposed, which combines the idea of model-free control, particle swarm optimization(PSO) and predictive control, and the convergence is also analyzed. The controller takes the universal model with error correction as predictive model and the PSO algorithm with high speed convergence as online optimization. It not only avoids calculating the complicated inverse matrix in the nonlinear predictive control, but also improves the speed of convergence and enhances the real-time implement. The simulation results show the effectiveness of the proposed controller.