Model-Free Predictive Control for NPC Inverter Based on Nonlinear Extended State Observer

Grid-connected inverters under weak grid conditions suffer from robustness issues, while conventional model predictive control (MPC) heavily relies on accurate grid impedance parameters, limiting its practical application. To address MPC’s performance degradation under parameter mismatch, this paper proposes a model-free predictive control strategy based on a nonlinear extended state observer (NESO) for NPC-type three-level inverters, achieving output current control and DC-link neutral-point voltage balancing. Firstly, a NESO is designed to estimate the lumped disturbance in the ultra-local model of the inverter, which eliminates the dependence on the system parameters and enhances the robustness. Next, a predictive control model for neutral-point voltage balancing is established, incorporating capacitor voltage deviations into the cost function to optimize balancing performance. The optimal switching vector is selected through rolling optimization and applied in the next sampling period, ensuring coordinated current and voltage control. Simulation results demonstrate that the proposed method effectively tracks output current, maintains neutral-point balance, suppresses grid current harmonics, and exhibits strong robustness and dynamic performance under parameter variations.