Artificial neural network approximations of linear fractional neutron models

In this paper the artificial neural network (ANN) approximation for fractional neutron point kinetics (FNPK) model and fractional reduced-order model (F-ROM) is presented. The input-output data of the step response generated using the closed-loop stable fractional-order models was used for the training the ANN models. The ANN topology with various layers and different number of neutrons was tried to obtain the best approximation for the fractional-order model. The results confirm that the designed ANNs provide a good approximation to linear FNPK and F-ROM models. It was also observed that the convergence and learning of ANN is greatly affected by the type of model (FNPK or F-ROM), value of anomalous diffusion coefficient (fractional-order derivative) and the relaxation time.