Model-Free Predictive Control: A Method to Always Improve the Performance Robustness of Power Electronic Systems?

Abstract

Model predictive control (MPC) typically suffers from performance degradation due to model uncertainties. To address this issue, model-free predictive control (MFPC) based on the ultra-local model has been proposed, which promises perfect performance robustness against model uncertainties compared to conventional MPC. However, simulation and experimental results show that this advantage does not always hold in practice. In this paper, we discuss the question of whether MFPC is always more performance robust than MPC, based on classical control theories. Two representative MFPC methods are analyzed and compared with MPC. And a criterion to improve their performance robustness is proposed. Such a criterion will provide a powerful tool for the development of more robust MFPC methods in the future. The criterion is validated using a buck converter with a Pareto Fronts analysis in simulations and experiments simultaneously.