Comparative Study on the Numerical Methods for View Factor Computation for Packed Pebble Beds: Back Propagation Neural Network Methods Versus Monte Carlo Methods

Abstract

It's an unsolved problem to calculate the thermal radiation view factors among fuel pebbles as accurately and quickly as possible in the simulation of the temperature fields within the pebble-bed. In this study, a series of fully connected neural networks (FCNs) has been developed to realize the fast calculation of view factors. In order to verify the accuracy and effects of the networks, the neural networks are compared with the Monte Carlo (MC) algorithm. The results show that, in most cases, the relative errors of the FCN method can be controlled within 1.0%, and the prediction accurate probability is up to 99%. In comparisons of specific examples, the temperature errors of the FCN method and the MC method are less than 1 K within the range neural networks have covered. In addition, the time of neural networks for a single calculation is about 2–20 μs, which is even less than 10−4 of the time taken by the MC algorithm. In conclusion, neural networks can greatly improve computational efficiency while keeping the same accuracy as the MC algorithm, which makes real-time simulation of the temperature fields possible.