Fast training of an MLP based neurocontroller using EBUDS

In this paper, a neural network controlled induction motor drive architecture, which replaces the indirect field oriented controller (FOC), is presented. The neurocontroller is trained to function as an FOC through a novel training method called error backpropagation using direct solution (EBUDS). The new method ensures accurate training, which is much faster than the conventional error backpropagation (EBP) method. In order for the neurocontroller to capture the full dynamic nature of the induction motor, an intensive training of the neurocontroller is made using the past values of inputs as well as the present values