Design and Analysis of the Model Predictive Control Implemented by the ANN Technique for MMC-Based Rectifier With Improved Grid Adaptability

Model predictive control (MPC) has been widely adopted as a modular multilevel converter (MMC)-based active rectifier due to its simplicity, flexibility, and ability to manage multiple control objectives. Nevertheless, the practical implementation of MPC is hindered by the absence of accurate mathematical models. When there are changes in the grid or MMC parameters, the control performance of MPC will deteriorate, indicating weak grid adaptability. For this matter, the employment to implement the MPC by the artificial neural network (ANN) technique is explored to improve the grid adaptability in this article. As ANN is based on data-driven learning, it is not affected by model and grid parameters. More importantly, the strong fault tolerance and generalization of ANN allow it to handle a variety of nonideal grid conditions and different grid impedances, thus improving grid adaptability. Furthermore, the substance regarding why the fault tolerance and generalization of ANN can improve grid adaptability is revealed. In addition, a delay compensation approach for the ANN is proposed to ensure the control performance. Finally, simulation and experiment are performed to verify that the proposed control manner can effectively improve grid adaptability in various grid conditions, including ideal, distorted, unbalanced, and varied impedance grids.