Neural-Network-Based Predictive Control for PMSM Drives

Model predictive control (MPC) has gained increasing popularity in the control of motor drives and power converters. However, its control performance deteriorates when there is a mismatch in system parameters. This article introduces a novel neural-network-based predictive control (NNPC) architecture to address this issue. By means of a supervised learning mechanism based on the input-output data of the governed system, the established neural network can replace the original predictive model. In contrast to the standard MPC controller that uses constant parameters to predict future behaviors, the advantage of the proposed method lies in the fact that it can learn the non-linearity of the system and adapt to the parameter perturbations. Consequently, NNPC enhances robustness against parameter variations while maintaining the superior performance of the conventional MPC method. Finally, the proposed method is validated through comparative simulations on a surface-mounted permanent magnet synchronous motor.