Nonlinear model predictive control based on support vector regression

This paper proposes a novel method to train the nonlinear predictive model, which is used in nonlinear statistical model predictive control. The accuracy of the predictive model for the nonlinear process is improved by using support vector regression (SVR). Simulation results show that the identification ability of SVR is comparable to that of the neural network (NN), and the generation ability of SVR outperforms that of NN. Moreover, the control performance of nonlinear model-based predictive control (NMPC) is improved by using SVR instead of the traditional used NN.