A Model Predictive Control Guided Partial Neural Network Compensation Method for Permanent Magnet Synchronous Motor Control

Permanent magnet synchronous motor (PMSM) usually contains a current inner loop (CIL) and a speed outer loop (SOL). By identifying the PMSM parameters, the dynamic process of the CIL can be satisfactory, however, the SOL often deviates from its design point, resulting in the slow speed tracking or the current overshoot during transient regulation. This paper proposes a Model Predictive Control guided Partial Neural Network (MPC-PNN) compensation method to improve the dynamic process of SOL. The innovation and advantages of the proposed method are: (1) only a simple NN is added to the existing SOL for compensation purpose, so the original SOL design is not impacted; (2) the training of the PNN is guided by MPC, makes it achieve good control performance with less data; (3) the PNN is trained offline in PC or cloud, which reduces the computational burden; This method does not require substantial changes to the existing PMSM controller, and is low computational burdens for real-time implementation. The experimental results also verify the effectiveness of this method.