Torque Ripple Suppression of PMSM Speed Regulation System Using Neural Network

In this paper, a neural network (NN) based torque compensator is combined with a third order extended state observer (ESO) in active disturbance rejection control (ADRC) system to suppress the torque ripple for PMSM low speed drives. The ESO can observe and reject low frequency disturbance, while the torque harmonics caused by cogging torque and inverter nonlinearity, etc. contain certain high frequency component. Therefore, a two layers NN compensator is proposed to further suppress the torque harmonics. After training, the NN can output a feedforward torque compensation to further reduce the torque ripple. Experiment results validate the effectiveness of the proposed method.