On-line predictive control based on LS-SVM

Aim at the robustness of predictive control and on-line modeling, an on-line predictive control based on least squares support vector machine (LS-SVM) is proposed. In order to carry out on-line learning, the training data threshold is set through discussing the theory of LS-SVM and the character of control system. Then the model of the on-line predictive control system is established, and the analytical solution of control variable is deduced integrating the method of model predictive control (MPC). The results of simulation indicate that the method has enough rapid speed to establish on-line model and strong robustness to external disturbance and parameters variation.