A Gaussian Process Based Model Predictive Controller For Nonlinear System With Uncertain Input-output Delays

In industrial processes, many systems can be described as a nonlinear system withtime delay, under external disturbances. However, since the stability analysis for linearcontrol systems is well developed, there are still many industrial processes usinglinearization theory to make nonlinear system controlled by approximation, which leads tothe poor robustness. So it’s necessary to research such kind of nonlinear dynamics withuncertain time delay, under external disturbances.For the above problems, A Gaussian Process based Model Predictive Control isproposed. The method contains two parts: Gaussian Processes and Model PredictiveControl. Firstly a Gaussian Process is a kind of probability model forecasting systemoutput by prior and likelihood. On the basis of that, we propose a single-step predictivemodel and implement the multi-step prediction by the method of iteration. On the otherhand, Model Predictive Control is based on prediction model, moving optimization andfeedback correction and obtain control signal by computing maximum of cost function.In the experiment parts, we chose a first order system and Continuous stirringReactor model as experimental plants and apply Gaussian Process model, RBF neuralnetwork respectively to have the simulation results compared. Through the resultcomparisons, we come to the conclusion that a Gaussian Process base Model PredictiveController can achieve good effect and guarantee the good robustness of the system.