Efficient Current Sensorless Model Predictive Control for Matrix Converter-Fed PMSM Drives

Abstract

The model predictive control (MPC) for matrix converters (MCs) typically requires multiple sensors and extensive computational resources to measure various quantities and evaluate control objectives. To address this problem, an efficient MPC approach has been developed for MC-fed motor drives, which entails fewer sensors and less calculation than existing MPC techniques. The proposed method uses one lookup table and a new source current reference calculation to significantly reduce the number of switching state candidates, model predictions, and cost function evaluations. Moreover, Luenberger observers are designed to eliminate sensors for all currents and load torque, under the challenge that MCs directly couple currents and voltages at the source and load sides. With the gain selection structurally optimized, their robustness against electrical and mechanical parameter variations has been rigorously analyzed and validated. Finally, the proposed MPC and benchmark methods are executed and compared in simulation and experiments. The results demonstrate that the proposed method reduces almost half the sensor requirements and 12% computational overhead of the existing simplified method without significantly compromising the system performance, leading to a cost-effective design. The proposed method enables the MPC to work at higher switching frequencies than the traditional ones, thereby mitigating the total harmonic distortion (THD) increases.