Robust Two-Layer Model Predictive Control for Full-Bridge NPC Inverter-Based Class-D Voltage Mode Amplifier

Abstract

Finite control set model predictive control (FCS-MPC) is able to handle multiple control objectives and constraints simultaneously with good dynamic performance. However, its industrial application is limited by its high dependence on system model and the huge computational effort. In this paper, a novel robust two-layer MPC (RM-MPC) with strong robustness is proposed for the full-bridge neutral-point clamped (NPC) voltage mode Class-D amplifier (CDA) aiming at this problem. The errors caused by the parameter mismatches or uncertainties of the LC filter and the load current are regarded as lumped disturbance and estimated by the designed Luenberger observer. The robust control can be achieved by compensating the estimated disturbance to the used predictive model. In order to reduce computation of the controller, a two-layer MPC is proposed for the full-bridge NPC inverter with an LC filter. The first layer is used to calculate the optimal output level which minimizes the tracking error of the output voltage. The second layer is used to determine the switching state for the purpose of capacitor voltage balancing. The experimental results show that the lumped model error is observed centrally through only one observer with low complexity. The two-layer MPC further reduced the computation without affecting the dynamic performance.