Artificial neural networks for spatial distribution of fuel assemblies in reload of PWR reactors

Abstract

An artificial neural network methodology is being developed in order to find an optimum spatial distribution of the fuel assemblies in a nuclear reactor core during reloading. The main bounding parameter of the modeling was the neutron multiplication factor, keff. The characteristics of the network are defined by the nuclear parameters: cycle, burnup, enrichment, fuel type, and average power peak of each element. As for the artificial neural network, the ANN Feedforward Multi_Layer_Perceptron with various layers and neurons were constructed. Three algorithms were used and tested: LM (Levenberg-Marquardt), SCG (Scaled Conjugate Gradient) and BayR (Bayesian Regularization). The artificial neural network has implemented using MATLAB 2015a version. As preliminary results, the spatial distribution of the fuel assemblies in the core using a neural network was slightly better than the standard core.