SelfMPC: Automated Data-Driven MPC Design for a Class of Nonlinear Systems

Manual tuning required for Model Predictive Control (MPC) schemes can be labor-intensive and prone to errors due to the requisite domain expertise. In this paper, we propose a new procedure called SelfMPC: an automated, data-driven method for tuning MPC for an unknown system within a specific nonlinear class. We pursue a maximum likelihood approach using Gaussian processes to uncover system dynamics and to optimize a tracking cost function. We show the effectiveness of our approach through extensive simulations on a benchmark case study, illustrating its superior performance over traditional manual tuning techniques. Furthermore, we offer formal assurances regarding the stability and robustness of the resulting controller, ensuring its versatility across diverse operating conditions and uncertainties within the system.