ANN based Auto-Tuned Optimized FCS-MPC for Grid-Connected CSC Inverter

Abstract

This paper proposes an auto-tuned finite control set-model predictive control (FCS-MPC) for a grid-tied singlephase crossover switches cell (CSC) inverter. The multilevel inverter (MLI) under study generates 9 voltage levels. The FCSMPCobjective is to minimize the total harmonic distortion (THD) of the current fed to the grid with unity power factor while regulating the capacitor voltage at its reference value to maintain the 9 voltage levels. The switching losses are reduced by managing the redundant switching states selection. Artificial Neural Network (ANN) based on the Bayesian regularized feedforward learning technique is applied to predict the optimal weighting factor of the FCS-MPC with respect to the measured reference current value. The effect of using a dynamic weighting factor on the current THD for different reference current peak values (ranging from 2A to 8A) is studied through MATLAB/Simulink simulation. The presented simulation is intended to show that the application of a dynamic weighting factor can significantly enhance the current THD compared to the use of a fixed weighting factor.