Inverter Nonlinearity Compensation in the Presence of Current Measurement Errors and Switching Device Parameter Uncertainties

This paper proposes a compensation strategy of the unwanted disturbance voltage due to the inverter nonlinearity employing an emerging learning technique called support vector regression (SVR). SVR constructs the motor dynamic voltage model by a linear combination of current samples in real time, which exhibits fast observer dynamics and robustness on observation noise. Then the disturbance voltage is estimated by subtracting the constructed voltage model from the current controller output. The proposed method compensates all of the inverter nonlinearity factors at the same time and all the processes in estimating distortions are independent of the dead time and power device parameters. From the analysis of the effect on current measurement errors, it is confirmed that the sampling error has little negative impact on the proposed estimation method. Experiments demonstrate the superiority of the proposed method in suppressing voltage distortions caused by inverter nonlinearity