An inverter nonlinearity identification method based on physics-informed network for different working conditions

The compensation of inverter nonlinearity effects (INEs) can significantly mitigate current distortion, motor vibration, and losses within the motor. In industrial applications, compensation voltages derived from the INE model, as well as offline identification methods and online INE observers, are commonly employed. However, the INE model often lacks precision due to the difficulty in obtaining various inverter parameters. Offline identification is time-consuming, as it necessitates determining the compensation values under a range of working conditions, including varying DC-bus voltages, dead-times, and switching frequencies. Online identification methods tend to be complex because they require the integration of additional identification components into the control system. In this paper, a physics-informed network is proposed to identify the INE compensation value in different working conditions, and an INE model is proposed to provide physics information. This approach aims to reduce identification time and circumvent the complexities associated with traditional control system calculations. The proposed network requires only experimental data of two working conditions from offline identification experiments and can accurately predict compensation voltages for 45 working conditions. Experimental results demonstrate the effectiveness of the proposed method.