Optimisation of the extended Kalman filter for speed estimation of induction motor drives
A speed sensorless drive requires the elimination of the sensor; therefore a speed estimator is required. Speed estimation using the Extended Kalman Filter (EKF) is investigated. The use of an EKF as an observer for a sensorless induction motor has been a longstanding subject of research. However, little attempt has been made to optimise the filter performance. First some speed estimation results are presented where the commonly used Trial and Error method is used for tuning the EKF. The performance of the EKF is strictly dependant on the choice of the covariance matrices. Therefore to improve the performance of the EKF, a guided random search technique, Simulated Annealing is proposed. The work concentrates on finding the EKF parameters by the Simulated Annealing algorithm in both low and high performance drives, for constant V/f and vector control. A Genetic Algorithm is also a guided random search technique and in this work the algorithm has been used for comparison purposes on optimising the EKF. The robustness of the EKF parameters tuned by Genetic Algorithm, Simulated Annealing and Trial and Error is compared. The results presented show that Simulated Annealing is more robust against machine parameter variations. Despite the large computation time Simulated Annealing does have the potential of being an alternative method for optimising the EKF. These novel results presented here show that Simulated Annealing is capable of tuning the EKF in the induction motor drives application.