Abstract
In this paper, in order to solve the problem of mismatch between the noise covariance matrix of the traditional Kalman filter and the actual system, a Sage-Husa noise estimator is incorporated to adaptively tune the system and process noise matrices, thus improving the robustness and convergence speed. An extended inverse potential observer based on Kalman filtering is used to construct the PMSM full-order state equation, which combines NQPLL and LPF filtering techniques. A modification of the Sage-Husa inertial weight equation introduces stabilization coefficients to ensure the semi-positive character of the noise covariance matrix, thus improving the system stability. In addition, sliding mode control is integrated into the velocity loop to improve immunity and response speed, while the high torque per ampere module optimizes the torque-to-current ratio control to minimize the motor losses. Simulation and experimental results show that the proposed observer achieves high estimation accuracy and adaptability in all velocity domains.