A neural network for the prediction of performance parameters of transformer cores

The paper shows that Artificial Neural Networks (ANNs) may offer new possibilities for the prediction of transformer core performance parameters, i.e. no-lend power losses and excitation. Basically this technique enables simulations with respect to different construction parameters most notably the characteristics of corner designs, i.e. the overlap length, the air gap length, and the number of steps. However, without additional physical knowledge incorporated into the ANN extrapolation beyond the training data limits restricts the predictive performance.