Abstract

A finite control-set model predictive control (FCS-MPC) strategy is widely recognized as an interesting research topic in both theoretical and practical architectures. One barrier to the widespread application of the FCS-MPC is its sensitivity to the accuracy of the system model. Notice that it is an underexplored issue on how to attenuate such a restriction. To this end, we continue this topic and focus on a novel FCS-MPC methodology subject to parametric uncertainty, which can be realized by incorporating a fuzzy approximation-based autoregressive with exogenous variable (ARX) model into an intelligent two-horizon robust FCS-MPC architecture. However, it introduces a prohibitively high-computational burden, which makes it unsuitable for online implementation. To remedy this, a supervised imitation learning technique, which is inspired by artificial intelligence (AI), is leveraged herein to approximate the developed controller as a black box, thus facilitating a feasible computational load. Our modification is able to simultaneously mitigate the problems of model parametric uncertainties and increased online computational demand as well as weighting factor selection inherent in the existing approach, which ensures the optimized system performance with efficient online implementation and low switching frequency (SF) operation. Finally, remarkable performance and superiority for our proposal are experimentally confirmed for power converters.