Computationally Efficient Multistep Model Predictive Control for Multilevel Converters

Recently, multistep direct model predictive control (MPC) has emerged as an effective control alternative for low-switching high-power converters. These multistep MPC approaches often rely on the Sphere Decoding Algorithm (SDA) to effectively solve the associated optimization problem. However, as analyzed in this work, the computational effort required by the SDA-based optimization grows significantly with the number of available output voltage levels in the converter. To address this issue, the work at hand proposes a computationally efficient multistep MPC strategy for multilevel converters. The key idea is to reformulate the converter dynamic model by considering voltage level changes as control inputs, instead of actual voltage levels. This reformulation enables the SDA to search over a finite and reduced set of possible transitions, yielding execution times that are independent of the converter's voltage levels. Simulation results for a three-phase, 7-level converter are provided to validate the proposed multistep MPC for multilevel converters.