Research on Torque Ripple Suppression Based on RNN Neural Network Amplitude Optimization of Pulsating Injection

The high-frequency square-wave injection approach is frequently utilized for rotor position estimation in IPMSMs operating without position sensors, particularly under low or zero speed conditions. To mitigate torque ripple induced by high-frequency excitation, this work develops an adaptive scheme that continuously adjusts injection amplitude in closed-loop systems via a Recurrent Neural Network. In this method, the RNN adjusts the injection amplitude based on d-axis current data and position error details, ensuring accurate position extraction while effectively reducing current harmonics and torque ripple. Simulation results demonstrate that compared to traditional fixed-amplitude injection methods, while ensuring the position tracking accuracy, the peak-to-peak value of torque ripple and the THD drop of current are reduced by 24.93% and 46.24% respectively.