Direct Model Predictive Control of a Five-Level ANPC Inverter with an Adaptive Linear Neuron-based Impedance Estimator

Abstract:
The paper proposes a direct model predictive control (MPC) method for a five-level active neutral point clamped (5L-ANPC) inverter which includes an online impedance identification through an adaptive linear neuron (ADALINE) estimator. This is adopted to improve the robustness of the underlying MPC against load parameters uncertainty and wear-out, i.e., minimizing model mismatches, while keeping the computational complexity at bay. To this aim, the formulation of a single-layer neural network and the related online training for the application at hand is derived. The MPC optimization problem is designed such that the current reference tracking and the balancing of either the flying capacitors voltages and the neutral point voltage are addressed altogether. The presented results verify the effectiveness of the proposed strategy and demonstrate the performance benefits against impedance variations.