| Abstract |
This paper deals with the application of a neural observer for the estimation of induction motor state.
It focuses on a simple neural observer having neural-network (NN) architecture with six inputs, two
flux state outputs, and one hidden layer with a few nodes using sigmoidal-functions. This NN is
recurrent, i.e. the two past state outputs are fed back into the network. The other four inputs are the
á,â components of stator voltages and currents. This research is aimed at both increasing the learning
speed, using the Kalman filter to teach the neural network, and reducing the oscillations of estimated
fluxes. Theoretical results are given to show the effectiveness of the neural observer for control
purposes in induction motor drives. |
