A Computationally Efficient FCS-MPC Imitator for Grid-Tied Three-Level NPC Power Converters Based on Sequential Artificial Neural Network

Abstract

Finite control set model predictive control (FCS-MPC) has attracted lots of attention recently thanks to its intuitive design and ease to include multiple control objectives and constraints. However, its computational demand surges exponentially with the number of independent switching pairs and the prediction horizons. To reduce the computational burden, we propose a fast FCS-MPC imitator based on two sequential shallow artificial neural networks (SANN). Compared with the existing ANN based strategies, the proposed solution has much reduced number of neurons in hidden layers, and is much suitable for redundant voltage vector selection. The proposed solution is tested with a grid-tied three-level neutral point clamped power converter (3L-NPC). Simulation results confirm that, i) the computational complexity is much reduced than both FCS-MPC and ANN-MPC while achieving similarly qualified control performance, ii) SANN-MPC shows better imitation accuracy than the classical ANN based MPC.