Abstract

The effectiveness of Model Predictive Control (MPC) heavily depends on the precision of the model in accurately representing the dynamics of the plant. However, identifying such models can be challenging due to complex dynamics and unknown or changing disturbances. One strategy is to utilize a simple and fixed nominal model while learning the unknown deviation between this nominal model and the actual plant dynamics. Gaussian Process (GP) regressions have demonstrated their value as a reliable tool for predicting disturbances and model mismatches, thereby facilitating their incorporation into MPC predictions. This paper introduces a framework for learning the dynamics of load disturbance in the DC motor servo system. We evaluate and compare the performance of the GP-MPC controller with that of a conventional MPC controller. Results indicate that the GP-MPC controller outperforms the conventional MPC controller in position servo performance during load disturbance while adhering to input saturation and state constraints.