An adaptive feedforward controller for permanent magnet synchronous motor based on physics-guided neural network model

In order to improve the servo accuracy and dynamic performance of permanent magnet synchronous motors (PMSM), a physics-guided neural network (PGNN) model is proposed, which can be used to establish a dynamic inverse model of PMSM-based servo systems. First, based on the characteristics of PMSM, a PGNN model is constructed, which is divided into a preprocessing model part, a physical model part, and a neural network part. Then, the above model is applied to the design of feedforward controller to improve servo performance. Considering that the dynamic characteristics of servo systems are variable, this paper further proposes an adaptive method that can periodically update model parameters based on input and output data of the system, thereby improving control effectiveness. The simulation experiment results show that the PGNN model proposed in this paper has high accuracy and good generalization ability, and can accurately describe the nonlinear characteristics of PMSM. The control effect of the feedforward controller based on this model is better than traditional modeling method and the model established in our previous work. Therefore, the control method proposed in this paper holds significant value and is crucial for advancing the intelligent development of PMSM systems.