Recurrent Neural Network Based Nonlinear State Estimation for Induction Motors

Abstract

This study presents a recurrent neural network (RNN) based nonlinear state estimator which uses an Elman neural network structure (ENN) for state estimation of a squirrelcage induction motor. Proposed algorithm only uses the measurements of the stator currents and the rotor angular speed, and learns of the dynamic behavior of the state observer from these measurements, through prediction error minimization. A squirrel-cage induction motor was fed from sinusoidal, six-steps, and Pulse Width Modulation (PWM) supply sources at different times in order to observe the performance of the proposed estimator for different operation conditions. Estimation results showed that the proposed algorithm is capable of estimating the states of an induction motor and it performs better than Extended Kalman Filtering (EKE) in terms of accuracy and convergence speed. Copyright (C) 2011 Praise Worthy Prize S.r.l. - All rights reserved.