Non-parametric induction motor rotor flux estimator based on feed-forward neural network

Abstract

The conventional induction motor rotor flux observer based on current model and voltage model are sensitive to parameter uncertainties. In this paper, a non-parametric induction motor rotor flux estimator based on feed-forward neural network is proposed. This estimator is operating without motor parameters and therefore it is independent from parameter uncertainties. The model is trained using Levenberg-Marquardt algorithm offline. All the data collection, training and testing process are fully performed in MATLAB/Simulink environment. A forced iteration of 1,000-epochs is imposed in the training process. There are overall 603,968 datasets are used in this modeling process. This four-input two-output neural network model is capable of providing rotor flux estimation for field-oriented control systems with 3.41e-9 mse and elapsed 28 minutes 49 seconds training time consumption. This proposed model is tested with reference speed step response and parameters uncertainties. The result indicates that the proposed estimator improves voltage model and current model rotor flux observers for parameters uncertainties.