Model Predictive Control Algorithm for Converter Based on a Convolutional Neural Network

In the finite control set model predictive control (FCSMPC) algorithm for a converter based on a neural network, the optimal control variables computed by neural network controllers achieve decoupling between the optimal FCSMPC algorithm design and online computational burden. However, the limited generalization capability of neural network controllers leads to degraded control performance when converter load types vary, so it is essential to design switching rules for neural network controller model parameters tailored to different load types and rapidly identify the converter load type. To address this issue, this article designs a switching strategy for neural network controller model parameters of converters and employs a convolutional neural network (CNN) to identify the converter load type. The CNN-based identification achieves adaptive switching of controller model parameters based on detected load types, ensuring consistent control performance across different converter load types. Simulation results demonstrate that load-type identification based on the CNN achieves alignment between neural network controller model parameters and load types, and the adaptability of converter neural network controllers is enhanced significantly. The effectiveness and feasibility of the proposed method are validated.