Speed Estimation of Induction Machines Using Square Root Unscented Kalman Filter

The application of a new evolutional EKF algorithm - SRUKF to the speed estimation of induction machines is investigated in this paper. Compared with the EKF, the impacts of the sampling time, the process noise covariance matrix Q and the measurement noise covariance matrix R on the performances of the two Kalman filters are analyzed numerically. The results show that for the EKF the increasing of the sampling time mainly influences the dynamic performance when the rotor speed is changed quickly, and for the SRUKF, it mainly has evident effect on the stationary error, and that the ratio of r₁₁ to q₅₅ affects the stationary error and dynamics of the filters evidently, and that the tuning of Q and R for both the EKF and the SRUKF is difficult under a larger sampling time, but under a smaller sampling time the tuning for the EKF is easier than that for the SRUKF. Then, the performances of the filters, such as stationary error, dynamic performance, induction machine parameter sensitivity, noise sensitivity, method complexity and computational cost with the optimized parameters under the different sampling times are evaluated. The comparison researches show that the EKF is still the more efficient and feasible estimation algorithm for the speed estimation of induction machines. This conclusion is also supported by the experimental results