A Simple ANN-Aided Virtual-Space-Vector PWM Strategy for Three-Level NPC Traction Inverters With Coordinate-Data Mapping

The three-level neutral-point-clamped (3L-NPC) inverter is a mature topology that tends to be a good candidate in high-power traction applications, such as electric vehicles (EVs). However, the wide operating range under off-road scenarios inevitably renders a high modulation index and lower load angle, which affects the neutral-point (NP) voltage imbalance of the 3L-NPC inverter. To address this demerit, the prior-art virtual-space-vector pulse-width-modulation (VSVPWM) strategy has been explored due to average-zero NP currents for all ranges of load conditions. Nevertheless, this solution raises execution costs due to the complicated subsector and determination of dwell-time. To this end, in this paper, a novel artificial neural network (ANN)-aided VSVPWM is therefore proposed by leveraging the sextant-coordinate system. The designed ANN attains excellent training performance with negligible errors. More importantly, all the trained nets are designed with simple structures for running efficiently on commercial digital signal processors (DSPs). This makes the presented artificial intelligence (AI)-based modulation algorithm possible to be executed in a commercial controller of future EV powertrains. Based on the training data collected by coordinate-based derivations and the trained nets, the feasibility and effectiveness of the presented ANN-aided PWM technique were validated by simulation study through Simulink/PLECS and experimental results from a 3L-NPC traction inverter.