Nonlinear generalized predictive control based on online least squares support vector machines

For a class of nonlinear discrete systems with unknown parameters, an adaptive direct generalized predictive control method based on online least squares support vector machines (OLS-SVM) is proposed. In the method, the OLS-SVM is used to design the controller directly, and an improved projection algorithm based on the tracking error is introduced to adjust the weights of the OLS-SVM adaptively, so the inverse matrix is avoided in the process of online real-time control. It is proved that the proposed method can make the tracking error converge to a small neighborhood of the origin. Simulation results have shown the correctness and effectiveness of the proposed method.