Evaluation of SVM speed and position observers for sensorless PMSM in start-up region

In this paper, a support vector machine regression (SVR) observer is presented to estimate the rotor speed and position in start-up region for sensorless vector control of permanent magnet synchronous motor (PMSM). To avoid the saliency reliance and the estimation error caused by parameter inaccuracy in the existing sensorless methods, the SVR observer, describing the nonlinear relationship between the input stator currents and voltages and the output rotor speed and position, is established based on machine learning theory, which is independent to the structure of object system. The structure of the SVR observer is simplified with a modified kernel function, and the dimension of training input is reduced by merely collecting the one-phase current and the sample size is minimized under an acceptable accuracy to reduce computational complexity. In order to eliminate the divergence caused by the harmonics of the input current, a current filter is inserted before the simplified SVR speed and position observers. The simulation results show the effectiveness of the proposed observer and the pre-placed current filter in suppressing the chattering problems of the estimation results.