Performance Investigation of Deadbeat Predictive Controllers for Three-Level Neutral Point Clamped Inverter

This paper aims to improve the performance of deadbeat (DB) predictive controller for a three-level neutral point clamped (NPC) inverter. The effect of the number of effective vectors considered for the cost function evaluation on steady-state and dynamic performance is investigated. To do that, three DB predictive controllers with different numbers of effective vectors, namely, 19 vectors-based, 6 vectors-based, and 3 vectors-based DB, are compared beside the conventional current-based model predictive control (MPC). The neutral-point voltage is balanced using the redundant vectors. Simulations and experimental tests are performed to evaluate the performance of the competing MPC algorithms in terms of four main criteria, namely: neutral-point (NP) voltage balancing error, total harmonic distortion (THD), the computational effort required, average switching frequency, power losses, and sensitivity to parameters mismatch. Compared to conventional MPC, the experimental results show that the three vectors-based DB predictive controller has the best steady-state and dynamic performance with a reduction of computational burden up to 60.