Linear-nonlinear Learning Feedforward Control for PMSM Based on Deterministic Learning

Yiming Fei, Zixian Zhao, Weijian Li, Jiangang Li

Abstract: In this paper, a linear-nonlinear learning feedforward controller is proposed for the recurrent trajectory tracking control of permanent magnet synchronous motor (PMSM) system. Based on the simplified model of PMSM, a linear adaptive feedforward controller is proposed to address the main tracking error which is proportional to the speed of the motor. To address the tracking error caused by the nonlinear friction, a deterministic learning-based radial basis function neural network (RBFNN) feedforward controller is proposed and combined with the linear feedforward controller. As a result of the coordination between the linear feedforward controller and RBFNN controller, the neuron number and computational load are reduced effectively. Under the persistent excitation (PE) condition, the parameters of the proposed composite feedforward controller will converge to fixed values which represent the learned knowledge, i.e., the real dynamics of the PMSM system along the reference trajectory. Therefore, by recalling the learned knowledge, the adaptive process is no longer required when the motor repeats the same trajectory tracking task. Experimental results demonstrate the effectiveness of the proposed feedforward controller.