Continuous Control Set Model Predictive Control Method for Three-Level Active Power Filter Based on NN-RLS Parameter Identification

Abstract

To mitigate the harmonic pollution caused by the widespread integration of distributed generation and nonlinear loads, this paper proposes a continuous-control-set model predictive control (CCS-MPC) method for a three-level active power filter (APF) based on neural-network-assisted recursive least-squares (NN-RLS) online parameter identification. A three-phase, three-wire shunt neutral-point-clamped (NPC) three-level APF is considered, and its equivalent mathematical and discrete predictive models are established in the $\alpha \beta $ reference frame. CCS-MPC is combined with space-vector pulse-width modulation (SVPWM) and delay compensation to achieve accurate compensation-current tracking. To address predictive-model mismatch caused by branch-parameter drift, an NN-assisted RLS identification scheme adaptively corrects the model parameters online. Simulation and hardware-in-the-loop results show that the proposed identifier tracks parameter changes faster and with higher steady-state accuracy than conventional RLS. Compared with finite-control-set MPC (FCS-MPC), the complete NN-RLS+CCS-MPC scheme provides a fixed switching frequency, lower steady-state current ripple, improved harmonic-current tracking, and better output waveform quality.