A Unified Computationally Efficient Model Predictive Control for Multilevel Inverter–Fed Multiphase PMSM Drives for Electric Vehicles

Multiphase PMSM drives fed by multilevel inverters are attractive for transportation electrification because they deliver high reliability, smooth torque, and good voltage quality. However, finite-control-set MPC in these multiphase, high-level converters suffers from an exponential growth of switching candidates, which jeopardizes real-time implementation on automotive-grade controllers. This paper aims to develop a generalized MPC scheme that preserves dynamic performance while drastically reducing computational burden. The proposed method restricts the evaluation to at most 2m (with m denoting the number of phases) neighbouring switching states around the present vector, independent of inverter level, thereby shrinking the search space from n^m (where n is the number of inverter levels) to a small fixed set. An outer ultra-local model-free speed controller generates the i_q^* reference directly from speed and current measurements, removing PI retuning across machines and converter configurations. Simulation and dSPACE-based experimental results on three- and five-phase PMSM drives with three-, five-, and seven-level inverters show balanced, nearly sinusoidal currents with reduced current THD and torque ripple comparable to or better than conventional MPC, while keeping execution time within the < 100 µs budget for 10 kHz operation. These results confirm the real-time feasibility, harmonic-quality improvement, and scalability of the proposed strategy for multiphase, multilevel traction drives.