Speed-sensorless control of induction motors based on adaptive EKF

The noise covariance matrices Q and R are set as constant values in the traditional extended Kalman filter (TEKF). They are determined by trial and error. This process is very complicated and the optimal matrices are difficult to determine. In addition, when the characteristic of noise changes, the matrices cannot be adjusted correspondingly, and the performance of the TEKF deteriorates. Therefore, an adaptive EKF algorithm based on the maximum likelihood estimation criterion with limited memory exponential weighting (EW-MLE-AEKF) is proposed in this paper. In the proposed EW-MLE-AEKF algorithm, the windowing method is adopted to save the posterior residual sequences in the previous N calculation periods. Then, these sequences are used to estimate and update the process noise covariance matrix Q in real time. To speed up the convergence speed of the estimation, a limited memory exponential weighting algorithm is added to the windowing method, which can increase the importance of recent data. Through real-time experiments, the superiority of the proposed EW-MLE-AEKF algorithm is verified.