A Smoothed Kalman filter Trained MRAS Based Recurrent Neural Observer for Indirect Vector Controlled Induction Motor Drive

Abstract

A smoothed Kalman filter trained recurrent neural network is proposed as an observer for sensorless vector control of three phase induction motor. Recurrent neural networks which are capable of online training is used here. The speed and flux are estimated using this observer for closed loop vector control. The proposed observer shows good performance in transient and steady state and also in variations in load and speed. Simulations are performed on a 0.75HP induction motor drive and results are compared with those of an extended Kalman filter trained recurrent neural observer.