Proposta de utilização de redes neurais feedforward multicamadas para a otimização de padrões de recarga do combustível em um reator PWR

Abstract

The management of refueling in a power reactor has as its main focus the refueling pattern in the core in order to achieve better performance in the cycle observing all the safety parameters adopted. In this paper we present the strategy of an algorithm based on multi-layered feedforward neural network based on multi-valued neurons that will be used to develop a methodology capable of choosing the best combinations that satisfy the radial power peak factor and maximize the effective multiplication factor atthe beginning of the cycle and also satisfy the ratio of minimum critical power and maximum rate of heat generation in the end of the cycle.