Online Support Vector Regression for Non-Linear Control

Model Predictive Control (MPC) can provide robust control for non-linear processes. In this paper, we propose an MPC using online Support Vector Machine (SVM) which incorporates learning guided by error minimization as new data samples arrive. The tracking errors of future plant behavior are minimized by Differential Evolution, a global optimizer. With these features, an online SVR based MPC controller can be implemented to improve the model and controller performance in the presence of disturbances. The simulation results for the cases of two chemical processes are shown to demonstrate the performance of the proposed scheme.