2025 lEEE International Conference on Predictive Control of Electrical Drives and Power Electronics (PRECEDE 2025)

The Modular Multilevel Matrix Converter (M3C) is an important solution in the field of renewable energy grid integration due to its ability to directly perform AC-AC conversion, as well as its good scalability and fault tolerance. However, the finite-control-set model predictive control (FCS-MPC) scheme for M3C needs to account for the coupling between clusters and the potential mismatches in the system model, resulting in high computation burden and low accuracy. To eliminate the coupling between modules and improve the robustness of the model predictive control scheme, this paper proposes a prediction control scheme based on an online high-gain observer (HGO) combined with neural network. Specifically, the cluster is treated as a separate system, using a neural network to estimate external disturbances and eliminate the coupling term between clusters, while the HGO is used to eliminate the internal state errors of the cluster, thereby achieving robust and fast predictions, maintaining high accuracy even at low switching frequencies. Finally, the simulation and experimental results for the M3C grid integration confirm the effectiveness of the proposed method.

modular multilevel matrix converter (M3C),robustness,Finite Control Set Model Predictive Control,Neural network