Nonlinear neural-network modeling of an induction machine

Abstract

Presents an approach to identify the nonlinear model of an induction machine. The free acceleration test is performed on a 5-HP induction machine, and the resulting stator voltages, stator currents and rotor angular velocity are measured. Using the maximum likelihood (ML) algorithm, the parameter sets of the nonlinear model at various operating conditions are estimated. Then the nonlinear model parameters are represented by feedforward neural networks (FNNs). For validation, the simulated responses of the identified model using the measured and the simulated input patterns for the FNN models are performed. The identified model can be utilized for power system transient stability analysis and for online computer controlled electric drives.