Development of Three-Level F-Type Inverter Fed PMSM Drive With ANN-Assisted $\alpha ^{\prime }\beta ^{\prime }$-Framed Virtual SVPWM for Electric Propulsion

The efficiency and performance of multilevel inverter (MLI)-based PMSM drives make them a viable option for electric vehicle (EV) traction. Among various MLI topologies, the three-level F-type neutral-point-clamped (3L FNPC) inverter stands out as a more efficient alternative for EV applications, offering lower power losses and enhanced reliability. However, in off-road scenarios with low load angles and high modulation index, this inverter suffers from neutral-point (NP) voltage imbalance. Conventional virtual space vector pulse-width modulation (VSVPWM) based on α′β′-based elliptical reference frame provides a solution by preserving the average zero-point NP current; however, its implementation is computationally expensive due to complex dwell time and sub-sector calculations. In this paper, a novel VSVPWM based on an artificial neural network (ANN) in an α′β′-framed coordinate system is proposed to address this problem. The simple architecture and low training error of the ANN ensure that it runs quickly and efficiently on commercial embedded controllers. The ANN model is trained using data from a coordinate system technique, and the effectiveness of the model is verified through Simulink simulation and experimental tests on a 3L-FNPC inverter-fed PMSM drive. The results with different speed-torque conditions show that the proposed ANN-assisted VSVPWM is a feasible and effective alternative for future electric vehicle applications.