Robust neural network observer for induction motor control

A neural network observer for induction motor state estimation, which is robust with respect to parameter variations is presented. Robustness is obtained using a suitable training set based on a stochastic model of the motor obtained by the Price algorithm. This algorithm is used to obtain the confidence ellipsoid for the model parameters, which are then modeled as Gaussian random variables strictly contained in the ellipsoid. Simulation results show that the resulting neural observer provides a good trade off between estimates accuracy and robustness.