Research on Capacitor Voltage-Balancing Control of an NPC Five-Level Inverter Based on Model-Free Predictive Control

To address the problem whereby traditional model predictive control suffers from mismatches between the model and actual parameters due to system parameter variations in the capacitor voltage-balancing control of a neutral-point-clamped (NPC) five-level inverter, an improved model-free predictive control strategy based on particle swarm optimization and the deadbeat principle is proposed. Firstly, an ultra-local model of the inverter is established, and a particle swarm optimization algorithm with an adaptive inertia coefficient is employed to self-tune the control gain of the ultra-local model, thereby reducing current control error. Secondly, the electrical angle of the reference voltage is calculated using the deadbeat principle, and a simplified vector set is constructed for voltage vector traversal. Control is applied only to the capacitor with the largest voltage deviation from the balance value, which reduces computational burden while achieving current tracking and capacitor voltage balancing. Finally, the simulation results show that under steady-state conditions, the output current total harmonic distortion (THD) is 0.28%, and the DC-side capacitor voltage fluctuation is 0.01%, demonstrating a significant improvement in control performance compared with the extremum-seeking control and Kalman filtering methods. Under transient conditions, the proposed control strategy achieves a response time of 0.7 ms while maintaining good control performance and strong robustness. These results verify the effectiveness of the proposed control strategy.