Multirate Predictive Control for Diode Clamped Inverters with Data-Based Learning Implementation

Abstract

Diode clamping is a widely used topology for power inverters due to the low number of components, low cost, and general efficiency of the circuit, but its main drawback is the difficulty in controlling a multilevel configuration. For that, finite-control-set model predictive control was chosen, as it is a multivariable optimal controller that takes into account constraints and deals with multiple objectives. However, it has high computational burden and its control solution is held constant during the sampling period if not using modulation. This work first proposes a multirate enhancement for this control technique to address the latter issue, and then uses a novel learning method to allow the online implementation and overcome the high computation time. This method is tested in simulation in a three-phase, five-level diode-clamped inverter. Comparisons and performance gains of the multirate technique are provided.