Neural Network-based Model Predictive Control Approach for Modular Multilevel Converters

Abstract

This paper proposes a neural network (NN) approach as an alternative to the computationally burdensome model predictive control (MPC) in controlling modular multilevel converters (MMCs). Simulation results demonstrate the effectiveness of our approach, with a back propagation (BP) neural network model successfully trained and the NN controller performing comparably to the MPC controller in suppressing circulating currents and stabilizing submodule voltages, while reducing the computational burden of the controller. Additionally, our approach provides an effective control solution for MMCs that can be implemented in real-time power electronic applications, and has important implications for power system optimization and future research in neural network control beyond MMCs.