A Novel Asymmetric Fifteen-Level Inverter for Medium Voltage Applications

A reduced-switch asymmetric fifteen-level multilevel inverter (MLI) topology for medium-voltage applications is shown in this article, coupled with a selective harmonic elimination (SHE) control technique aided by artificial intelligence (AI). Ten unidirectional switches and three unequal DC sources are used in the suggested arrangement to limit the total standing voltage (TSV) while producing a stepped output voltage. To find the most effective switching angles for various modulation indices, a hybrid control strategy that combines an artificial neural network (ANN) and the Newton-Raphson (NR) method is employed. A low-voltage experimental prototype is used to test the suggested topology's performance, which is assessed using MATLAB/Simulink simulations. The findings show that, under the evaluated operating conditions, the inverter can produce multi-level output waveforms with low lower-order harmonic content and good conversion efficiency. Over the assessed load range, the measured efficiency ranges from 98.6% to 97.1%, and under nominal conditions, total harmonic distortion (THD) is found to be 5.47%. The suggested topology achieves a lower switch count and competitive TSV and harmonic performance under comparable operating assumptions, according to a comparison analysis with a few multilevel inverter topologies from the literature. Nevertheless, because operational circumstances and control procedures are not completely standardized across research, these comparisons are just informative. While the suggested architecture is meant for scalable extension toward medium-voltage applications, experimental validation is carried out utilizing a low-voltage prototype to confirm switching behavior and control capabilities. The results show that the topology and control technique are feasible, even though the experimental validation is carried out at lower voltage levels. To achieve wider applicability, more validation under higher voltage circumstances and standardized benchmarking frameworks are needed.