Weighting Factors Autotuning of FCS-MPC for Hybrid ANPC Inverter in PMSM Drives Based on Deep Residual Networks

Hybrid active neutral-point-clamped (HANPC) inverters have been recently considered as an attractive solution for high-power applications, while their control with multiple objectives remains quite complicated. The model predictive control (MPC) is an optimal control technique for multilevel inverters due to its powerful ability to handle the multiobjective optimization and nonlinear constrains. However, the tuning of weighting factors (WFs) in MPC is quite challenging and requires additional computational resources. For this motivation, this article proposes a deep residual network optimization approach for the dynamical WFs tuning of the finite control set (FCS) MPC in the HANPC inverters. The proposed method utilizes the residual connection to avoid the problem of gradient vanishing, so as to improve the predictive accuracy. To realize the dynamic adjustment of WFs, a look-up table is formulated with the predicted WFs, then this look-up table is integrated into the FCS-MPC. Consequently, the proposed method occupies less logic resources in the MPC algorithm, providing the desired behavior with fast dynamic response. Comparative studies are presented to evaluate the training and prediction performance of the networks. Finally, the experimental validations are conducted on a platform of HANPC-inverter-fed permanent magnet synchronous motor drive.