PRESSURIZED WATER-REACTOR CORE PARAMETER PREDICTION USING AN ARTIFICIAL NEURAL NETWORK

Abstract

In pressurized water reactors, the fuel reloading problem has significant meaning in terms of both safety and economics. The local power peaking factor must be kept lower than a predetermined value during a cycle, and the effective multiplication factor must be maximized to extract the maximum energy. If these core parameters could be obtained in a very short time, the optimal fuel reloading patterns would be found more effectively and quickly. A very fast core parameter prediction system is developed using the back propagation neural network. This system predicts the core parameters several hundred times as fast as the reference numerical code, within an error of a few percent. The effects of the variation of the training rate coefficients, the momentum, and the hidden layer units are also discussed.