Artificial Neural Network-based Simplified Model Predictive Control for Enhancing Voltage Source Inverters: CMV Reduction and Weighting Factor Optimization

Controlling voltage source inverters (VSIs) is crucial for ensuring the efficient operation of inverter-based systems. Model predictive control (MPC), notably finite control set MPC (FCS-MPC), is increasingly recognized as an effective control strategy within power electronics, particularly for addressing multi-objective control problems. This article proposes a novel approach for controlling voltage source inverters using a simplified FCS-MPC scheme with common-mode voltage (CMV) reduction, limited switching frequency, and optimized weighting factors (WFs). The proposed method simplifies the MPC controller by replacing the computational burden of current predictions with a single voltage reference estimation, thereby reducing computational complexity. Moreover, offline optimization based on the non-dominated sorting genetic algorithm II (NSGA-II) is employed to determine optimized weighting factors for the cost function, as well as enhance the controller's performance. For further alleviation of computational burden, the controller is imitated using an artificial neural network (ANN) trained offline with data labeled by the original FCS-MPC algorithm. A simulation study is presented to control both conventional two-level VSI (2L-VSI) and three-level neutral point clamped (3L-NPC) inverters to demonstrate the effectiveness and validity of the proposed method. The results underline that the ANN imitator of simplified MPC reduces computational complexity while maintaining performance comparable to optimal control.