Neural Network-Enhanced Control Strategy for Multiport Multilevel Inverters with a Parallel Structure in Distributed Photovoltaic Generation

The high computational burden of Finite Control Set Model Predictive Control (FCS-MPC) restricts its application to complex topologies like parallel Cascaded H-Bridge (CHB) inverters in photovoltaic (PV) systems, despite its performance advantages. This paper proposes a Neural Network-enhanced FCS-MPC (NN-MPC) to overcome this bottleneck. By training a Multi-Layer Perceptron (MLP) offline with expert FCS-MPC data, the computationally intensive online exhaustive search is replaced by a fast forward-pass prediction. Simulations on a seven-level parallel CHB inverter validate the proposed strategy. It significantly reduces computational time while maintaining comparable performance to conventional FCS-MPC in terms of low grid current THD, submodule voltage balancing, and fast dynamic response under balanced conditions. The approach also demonstrates stable operation under PV power imbalance. The NN-MPC thus offers a practical and computationally efficient solution for advanced control of multiport multilevel inverters in distributed PV applications.