For off-line model hardly adapt to the real-time change of nonlinear object in nonlinear model predictive control (NMPC), an NMPC based on online support vector regression(OSVR) is developed. This algorithm utilizes OSVR off-line training and on-line learning to built nonlinear dynamical model with on-line adjusted capability. Furthermore, the multi-step control rules are obtained by the rolling optimization of gradient decent rule. Simulation results for a nonlinear system show
the good adaptability and effectiveness of the proposed algorithm.